Distinct kinematics and micromorphology for symmetrical rowing and sliding on water in ripple bugs and water striders

Locomotion on the water surface presents a unique set of physical challenges that have driven repeated and diverse evolutionary solutions among insects¹. The semiaquatic bugs (Gerromorpha) provide an example of adaptive radiation into this novel environment. These insects evolved distinct morphological, behavioral, and anatomical traits that enable movement on the air-water interface². These adaptations reflect multiple, lineage-specific strategies for solving similar functional problems–offering a model system for studying how alternative solutions involving morphology and behavior evolve under ecological constraints of physical environment–the water surface. However, the details of the co-evolutionary and functional interplay between leg microstructures, stroke kinematics, and thrust mechanics across major independent lineages of this adaptive radiation remain poorly understood. To provide more insights into this question, we compare two major independently evolved solutions for symmetrical rowing in Gerromorpha. Here, we focus on how alternative physical mechanisms–drag-based versus surface-tension-based thrust–are realized through contrasting yet functionally convergent morphologies.

Notably, two lineages—Veliidae (a polyphyletic family typically found in fast-flowing streams) and Gerridae (typically inhabiting slow or still waters)—have independently evolved symmetrical backward rowing by midlegs for forward thrust^2,3,4. While both rely on midlegs for thrust and fore- and hindlegs for support and sliding, their physical modes of thrust generation differ: Veliidae exploit hydrodynamic drag (and potentially lift), while Gerridae generate thrust primarily through surface tension forces^5,6. Despite their independent origins and these contrasting mechanisms, the leg kinematics and microstructural adaptations that support these behaviors have not been systematically compared.

Within Veliidae, species such as those in the genus Rhagovelia possess specialized midleg pretarsal structures known as swimming fans, which function as oars^2,5,7,8,9. These structures are assumed to be actively controlled via a claw retractor muscle⁵, though recent observations of isolated fans spreading in water have led to the hypothesis of passive elasto-capillary spreading^8,9. The nature of fan manipulation in intact, behaving animals remains unresolved. Similarly, the ventral microstructures on midlegs (involved in thrust) and on fore- and hindlegs (involved in support and sliding) have not been systematically examined in Rhagovelia, though their role in generating surface tension-based forces via water dimples is likely.

In contrast, Gerridae species such as Gerris latiabdominis do not possess swimming fans but instead have elongated, hairy midlegs that generate thrust through the creation of asymmetric dimples on the water surface. These midlegs, along with highly hydrophobic ventral surfaces, are critical for surface-tension-based propulsion⁶. While some aspects of locomotory performance have been compared³, detailed kinematic analyses and high-resolution comparisons of leg micromorphology between Gerridae and Veliidae are lacking. Prior studies have presented images of leg hair arrangements^2,10, but a focused comparative analysis across functional leg regions has not been conducted since Andersen (1976).

Following the framework set by Andersen (1976, 1982) and Crumière et al. (2016), we compare Rhagovelia distincta (Veliidae) and Gerris latiabdominis (Gerridae), two small-bodied representatives of their respective clades. We examine their midleg microstructures and stroke kinematics to evaluate how different physical thrust mechanisms are supported by contrasting anatomical features. We further investigate the hypothesized passive versus active mechanisms of fan control in R. distincta. Finally, we describe and compare fore- and hindleg microstructures used in support and sliding to identify potential convergences across taxa with different thrust mechanisms. Our results provide a foundation for future work on the functional evolution of water-surface locomotion in Gerromorpha.

Overview of figures and supplementary materials

The results are presented through figures cited in sequential order for clarity and traceability. Figure 1A1–D6 portrays typical thrust phase in Rhagovelia distincta, followed by corresponding views for Gerris latiabdominis in Fig. 2A1–D6. Figure 3A–S illustrates leg seta types. Figures 4A1–C4 and 5A1–C4 show SEMs of leg microstructures. Figure 6A–E compares contact angles and droplet shapes. Stroke kinematics and interspecific comparisons are shown in Figs. 7A–J and 8A1–D, with measurement summaries in Fig. 9A–H.

Supplementary figures follow the same order. Figure S1A–B depicts resting leg posture on water. Figure S2A–H shows leg-surface interaction and dimple formation. Figure S3A–E quantifies fan geometry and angles. Figures S4A–D and S5 capture fan dynamics during strokes. Figures S6A–G and S7A–C detail fan positioning within the tarsal cleft. Figures S8A–E and S9A–E present dissected fan–claw complexes. Figure S10A–S expands seta classification. Figures S11, S12A–H, S13–F, S14A–F describe ultrastructural features of fan setae and claws. Figures S15A–F, S16A–D, and S17A–B highlight ventral hook and spoon setae. Figures S18A–D and S19A–H show fore- and hindleg structures in R. distincta. Figures S20A–F and S21A–C provide analogous views for G. latiabdominis. Figures S22A–D and S23A–E show additional fore- and hindleg adaptations. Figure S24A–E and (a–e) detail droplet behavior and surface wetting (contact angles). Figures S25A–D summarize net force estimates during strokes. Figures S26A–D and S27 present multivariate and theoretical analyses of kinematics and fan leakiness.

Behavioral observations: leg use at rest

R. distincta body is supported on foreleg and hindleg tarsus with minimal contribution from the midleg tarsal tips (Fig. 1A1, B1, C1, D1; Figure S1). Occasionally, a very small portion of the midleg`s fan is protruded from the tarsal tip into the water body through the surface it`s in contact with (Fig. 1B1; Video S3 Part 4). G. latiabdominis body is supported on foreleg tarsus, hindleg tibia and tarsus, and midleg intermediate-distal tibia and tarsus (Fig. 2A1, B1, C1, D1; Video S2), which create dimples without piercing the water surface (dimples cast shadows on the bottom of the container; Fig. 2C1, D1; Video S2 Part 4).

Behavioral observations: leg use at locomotion

During a typical initial thrust phase (Fig. 1), R. distincta moves its midlegs forward–either above the water or lightly contacting the surface–then places the tarsi onto the water (Fig. 1C2), which is associated with fan extension into the water (Fig. 1B2). Alternatively, the midlegs may advance while the tarsal tips remain in contact with the surface and a small portion of the fan protrudes underwater (Fig. 1B6; Figure S2D; Video S3 Part 4). Observations from 110 slow-motion videos (each capturing 1–4 fan opening and closing events) suggest that R. distincta actively controls the timing, extent, and duration of fan protraction and retraction, regardless of midleg position on the water surface (SI Part 2; Figure S4; Video S3). For example, we frequently observed the fan opening and closing without any respective decrease or increase in the gap between the tarsus and the water surface (Fig. 1B1). This disagrees with the recently proposed passive fan actuation hypothesis⁹, which posits that lowering the tarsus onto the water surface is required to initiate fan unfolding through fan-water elastocapillary interactions (SI Part 2).

As the fan rapidly protracts (Fig. 1B2; Figure S2C, D) and the midlegs are pushed backward (Fig. 1C4, C5), the tarsus is simultaneously pressed downward (Fig. 1C3; Figures S2B and S5). This generates growing anteroposterior asymmetrical dimples (Fig. 1B4), visible as shadows that expand from the initial miniscule circles at the tarsal tips to ovals extending distally from the tibiotarsal joints (Fig. 1D4). Contrast to G. latiabdominis (Fig. 2D), R. distincta shows an expanded anterior dimple region (Fig. 1C3), likely caused by water displaced beneath the surface by the fan (Fig. 1B4; Figure S2G, H). Strong strokes can produce surface waves (Fig. 1C4, C5; 17 of 99 strokes; Video S1 Part 1), but all strokes transition into a passive sliding phase, during which the midlegs either disengage from the surface or trail behind with minimal fan protrusion (Fig. 1B6; Video S1). Observed midleg disengagement suggests adhesive forces and surface tension are overcome in this process (Video S3). Throughout, the fore- and hindleg tarsi remain in contact with the surface, providing support during sliding.

Duration of fan protraction ranged between 6 and 23 ms (12.3 ± 3.4 ms; n = 69) while the duration of fan retraction ranged between 3 and 15 ms (8.4 ± 2.6 ms; n = 73) (Figures S3 and S4). Shorter durations were often associated with strokes beginning or ending with partially protracted fans, where 2–4 distal setae tips protruded into the water. Fully protracted fans had an average projected area of 0.89 ± 0.04 mm², radius of 0.85 ± 0.01 mm, and protracted angle of 139.13 ± 3.64° (n = 6; Figure S3A). The longitudinal axis of wetted midleg and protracted fan typically lay in the same plane, slanted at 80.2 ± 6.0° (n = 26; Figure S3E).

G. latiabdominis generated thrust without piercing the surface (Fig. 2), using midlegs slightly rotated so that anteroventral gap-row microstructures pressed backward against the water (Figure S2; 27–36 ms into the stroke). Backward movement of the midlegs produced anteroposterior asymmetrical dimples and backward-moving surface waves (Fig. 2D; Figure S2E, F; Video S2). Midleg disengagement occurred when the wetted midleg aligned nearly parallel to the direction of body movement and proceeded gradually from proximal to distal segments. Some instances of disengagement appeared smooth (Video S2), while others suggested adhesive forces and surface tension were overcome during the process (Video S2; Figure S2E, 95 ms). Hindlegs contributed weakly to thrust, as indicated by faintly asymmetrical shadows and small wave bows during the initial stroke phase (Fig. 2D4). Hindleg tarsi and tibiae provided support during sliding, shown by shadows aligned with the body axis (Fig. 2D5, D6). Forelegs also supported the body, particularly near the end of the stroke, as indicated by prominent shadows (Fig. 2D6), except at mid-stroke when support was reduced (Fig. 2D4).

Overview

The legs of R. distincta had a less dense hair layer than those of G. latiabdominis. We identified 17 setae (hair) types: five shared by both species, five unique to R. distincta, and seven unique to G. latiabdominis (Fig. 3; Figure S10; Table S3). Our analysis focused on the ventral microstructures of leg segments that interact with the water surface, and respective nanometer sized details specifically in R. distincta (Figs. 4 and 5; Figures S11–S23).

Ventral microstructures for thrust generation

The swimming fan of R. distincta (Fig. 4B1, B2; Figures S6–S9 and S11–S14) consists of anterior and posterior claws and a fan made up of 17 (Figures S8, S9) to 21 (Fig. 4B2; Figure S11) setae. Each seta bears setulae along its axis at 8–12 μm intervals, forming a feather-like structure (Fig. 4B3, B4; Figures S12, S13). When protracted, the distance between adjacent setae ranges from ~ 20 to ~ 100 μm, and between setulae from several to ~ 20 μm, with typical setula spacing of 4–10 μm (Figure S13A, B). The fan is anchored at the inner proximal corner of the cleft between the two lobes of tarsomere 3 (Fig. 4B2). The surfaces of the setae, setulae, and claws lack nanogrooves (Figure S13C–F). The setae resemble flat beams or boards (Fig. 4B5; Figure S12C) with near-elliptical (Fig. 4B6; Figure S12D) or slightly triangular (Figure S12E) cross-sections, measuring 2–4 μm × 7–10 μm. The orientation of the narrow edges suggests that they face the water during thrust generation (Fig. 4B5; Figures S12, S13). The setulae are also flat, 1–2 μm wide and 300–700 nm thick, with a hollow center (Figure S12H), and their orientation further suggest that they press against the water with the narrow edges. Transverse sections of fan setae show lamellar outer layers and a central hollow (~ 700 nm in diameter) containing internal rods (~ 400 nm in diameter) (Fig. 4B6; Figure S12). Claw cross-sections are 1.5–2.5 μm thick and consist of external lamellar layers and multiple internal layers with nanofibers (100 nm) and cluster of nanofibers (200–400 nm) running in various directions (Fig. 4B8; Figure S14).

Passive elastocapillary expansion of the fan in water (Figures S8 and S9) was observed only when the fan was completely dissected–either with or without the anterior claw–and removed from its natural position in the cleft (SI Part 3B). In contrast, observations of intact fans and claws anchored naturally within the cleft indicate that fan expansion into the water at the onset of use is not passive (Videos S1 and S3; SI Part 3B). Moreover, we were able to induce fan protraction by mechanically pulling the ut tendon connected to the base of the fan-and-claw structure (Figures S8 and S9), supporting the involvement of active muscular control.

Along the ventral edge of the posterior lobe (Figures S15A–F and S16A–D), a structure composed of three rows of H2 setae—spaced 6–8 μm apart within each row and separated by two 2.5–5 μm wide gaps—forms a band that presses against the water surface without breaking it during a stroke (Fig. 4B9; Figure S16). This pattern resembles the “gaps and rows” arrangement seen on the tarsus of G. latiabdominis (Fig. 5B). Similarly, the ventral edge of the anterior lobe, which also contacts the water surface during a stroke, is lined with a band of 3–4 rows of H1 setae (Fig. 4B9; Figure S16); however, only lateral views were available, limiting precise row counts. On ventral tarsomere 2, which also interacts with the water surface without breaking it (Figure S5), we observed three rows: a posterior row of H1 setae and two anterior rows of Sp setae, separated by gaps (Fig. 4B12, B13; Figure S17).

In G. latiabdominis, a “gaps and rows” arrangement of setae was observed along ventral midleg sections that interact with the water surface (Fig. 5B; Figures S20A–F and S21A–C). This pattern resembles the structure on the ventral edge of the lobe in R. distincta (Fig. 4B9). It is especially prominent on the tarsus, where a main ventral gap separates a row of cuspidate setae (C) from a posterior row of thorn-like T2 setae, with a second T2 row positioned further posteriorly to form a T2–T2 gap (Fig. 5B4–B9). Anterior to the C row, a row of M1 setae creates a C–M1 gap (Figure S20E). The tips of these three setae types bend distally along the leg`s longitudinal axis, particularly the cuspidate setae, whose long, flat distal sections may contact each other. In contrast, setae on the lateral and dorsal leg surfaces are relatively straight (Figure S21B). On tarsomere 2, a second, less regularly arranged anterior M1 row creates an M1–M1 gap (Figure S20C, E). This arrangement is less distinct on the tibia, where T2 setae exhibit intermediate morphology between T2 and M1, and the C setae are replaced by M2 (Fig. 5B3). At the tip of the tarsus, the pattern disappears, replaced by a ventral concentration of grass-blade-like setae (g) (Fig. 5B12).

Ventral microstructures for support and sliding

In R. distincta, specialized microstructures involved in support and sliding were found on the ventral tarsi of the forelegs (Fig. 4C), hindlegs (Fig. 4A), and at the distal tips of the midleg tarsus (Fig. 4B10, B11). These structures consist of one or more rows of Sp setae with flattened, bent tips that overlap to form a “beam-like” surface ~ 20–25 μm below the leg cuticle, oriented toward the water. Two Sp rows were observed on the ventral forelegs (Fig. 4C3, C4; Figure S18) and 1 on the hindlegs (Fig. 4A1; Figure S19), with the flattened tips especially prominent near the leg ends (Figs. 4A5, A6; 4C4). Sp rows are flanked—particularly posteriorly—by H1 setae on the forelegs (Fig. 4C; Figure S18) and M2 setae on the hindlegs (Fig. 4A; Figure S19). At the tips of the midlegs, which also provide support, a dense cluster of Sp setae—likely modified H1 types—was observed, with long, flat, overlapping tips (Fig. 4B11; Figure S15F, G).

In G. latiabdominis, specialized microstructures involved in support and sliding were found on the ventral forelegs and hindlegs. On the ventral foreleg tarsus, a band of grass-blade setae (g) with overlapping flat, bent distal tips was observed ~ 15–20 μm from the leg cuticle, facing the water surface (Fig. 5C4; Figure S22). An entangled cluster of web setae (W) was also present, especially near the tibiotarsal joint (Fig. 5C2; Figure S22). On the ventral side of the hindleg tibia and tarsus, 2–3 rows of leaf-blade setae (L) were arranged in an orderly manner (Fig. 5A; Figure S23). Their overlapping distal tips formed a “beam-like” surface with nanogrooves, ~ 20 μm above the water surface (Fig. 5A2; Figure S23), and were accompanied by a posterior row of large thorn setae (T3) (Fig. 5A6, A8; Figure S23A, E). The ventral sides of joints were covered with bundles of L and leaf-like l setae (Fig. 5A1, A6), with l setae also present at the tarsal tips (Fig. 5A9).

Contact angle on midleg sections used in thrust generation

Midleg tarsal surfaces in G. latiabdominis were generally more hydrophobic than those in R. distincta (Fig. 6). In R. distincta, water droplets rapidly lost their spherical shape after contacting the hair layer, spreading 10–40% and showing decreased shape indices. Contact angles on the dorsal and ventral midleg surfaces progressively decreased from approximately 130.3° and 75.5°, respectively, to as low as 21.9° as droplets collapsed, indicating relatively high surface wettability (Fig. 6A, B; Figure S24). This effect was especially pronounced on the ventral side, where the swimming fan and associated microstructures interact with water during locomotion (Fig. 6B, C; Figure S24). Contact angle measurements revealed that the ventral side of R. distincta`s midleg—particularly the region housing the swimming fan—exhibits clear hydrophilic properties, in contrast to the more hydrophobic surfaces observed in G. latiabdominis. This elevated wettability likely facilitates water surface penetration during fan deployment, reducing resistance and promoting stable submersion of the fan. Such localized wetting, combined with the fan`s structural stiffness, inferred from setae and setulae microstructure, may enhance the effectiveness of drag-based thrust generation. In contrast, droplets on the dorsal and ventral midleg surfaces of G. latiabdominis retained their spherical shapes with minimal spreading (shape index ≈ 1.5), and contact angles remained high throughout dissipation, ranging from 132.1° to 109.3°, consistent with strong hydrophobicity (Fig. 6D, E; Figure S24; SI Part 4). These uniformly hydrophobic surfaces align with G. latiabdominis` reliance on surface tension, supporting the leg`s ability to retain air layers and resist surface penetration during sliding or thrust.

Kinematics profiles during a stroke

In both species (Fig. 7), the midleg femur angle increased gradually during the stroke, but more steeply in R. distincta than in G. latiabdominis. The initial angle was more acute in R. distincta (~ 20°) compared to G. latiabdominis (~ 60°), with both reaching a final angle of ~ 120° (Fig. 7A, B). Peak femur angular velocity was higher in R. distincta and occurred mid-stride at a femur angle of 85°, whereas in G. latiabdominis it peaked at 100° (Fig. 7A, B). In G. latiabdominis, the femur–tibia and tibia–tarsus angles remained relatively small throughout the stroke. In contrast, R. distincta showed more pronounced changes: the femur–tibia angle increased from 30° to 60° (Fig. 7C), and the tibia–tarsus angle ranged from 8° to 16° (Fig. 7D). These patterns suggest that R. distincta engages in coordinated rotations at the coxa–femur, femur–tibia, and—though to a lesser extent—tibia–tarsus joints, while in G. latiabdominis, motion is largely concentrated at the coxa–femur joint.

In both species, leg velocity during slower (longer) strokes remained below the theoretical critical velocity (~ 0.23 m/s; Fig. 7E), which marks a transition threshold above which thrust generating legs of semiaquatic organisms generate capillary-gravity waves to achieve higher propulsion force^11,12,13. In faster (shorter) strokes, leg velocity exceeded this threshold within ~ 5 ms and reached higher peak values in G. latiabdominis than in R. distincta (Fig. 7E). In R. distincta, the leg velocity vector was briefly aligned with the body movement axis only during mid-stroke, when the femur was approximately perpendicular to the trajectory (Figs. 7G and 9G). For the remaining stroke, the fan–acting as an oar blade–moved in a direction misaligned with the body axis and not perpendicular to its movement direction. In contrast, in G. latiabdominis, the leg velocity vector remained nearly parallel to the body movement axis (Figs. 7F and 9E) and nearly perpendicular to the wetted leg section (Figs. 7H and 9G) for a much larger portion of the stroke (rectangles on x-axis). In both species, the highest net force per stroke, indicated by peak body acceleration (Fig. 7J), coincided with intervals of high “effectiveness” indices (Fig. 7F–H; SI Part 5).

Kinematics comparisons of a stroke

A small subset of G. latiabdominis strokes showed notably higher average leg linear velocities, body velocities, and peak accelerations (Fig. 8). However, when all strokes were analyzed together, there were no significant differences between the two species in average midleg velocity (Fig. 8B2; Table S4), final body velocity (Fig. 8A1; Table S4), or maximal body acceleration (Fig. 8A2; Table S4). Strokes by R. distincta were shorter in duration (Fig. 8C3; Table S4) and involved faster angular femur movements (Fig. 8B1; Table S4). The legs traveled a shorter distance across the water surface during each stroke compared to G. latiabdominis (Fig. 8C1, C2). This difference is notable because R. distincta has shorter legs (Tables S1 and S2), and additional angular movements were observed at the tibia and tarsus (Fig. 7C, D). Since the kinematic variables are intercorrelated (Figure S26D), estimates of their individual effects on body speed (Figure S26A–C; Tables S5, S6 and S7) are not independent. To address this, we performed a principal component analysis and extracted two components (Table 1). Strokes by R. distincta were characterized by shorter duration and higher femur angular velocity (higher RC2 values), along with shorter distance and slower leg speed on the water surface (lower RC1 values) (Fig. 8E), distinguishing them from G. latiabdominis.

Estimated force output for thrust generation

Although we could not directly measure resistance forces during sliding and thus cannot determine total thrust, the body acceleration profiles during stokes allowed us to estimate the net horizontal thrust vector and the peak net force generated per stroke (Figure S25). Given that approximately 85–95% of total thrust typically translates into forward body momentum in water striders¹², our net force estimates likely underestimate the true thrust force by 5–15%.

In R. distincta, the maximum horizontal net force per stroke averaged 152 µN with an absolute maximum of 324 µN (Figure S25C). This force was generated through the symmetrical action of two pretarsal swimming fans, each with a projected area of 0.89 ± 0.04 mm² (1.78 mm² total; Figure S3), along with contribution from wetted tarsi measuring 1.80 ± 0.06 mm (Table S1) in length. Assuming the primary thrust originates from the fans, this translates to an average force of 85 µN/mm² of fan surface area per stroke with an absolute maximum of 183 µN/mm².

In G. latiabdominis, the maximum horizontal net force per stroke average 360 µN with an absolute maximum of 745 µN (Figure S25D). This force was generated using wetted midlegs that depress the water surface to create dimples, without piercing it. Based on average wetted leg length of 7.76 ± 0.11 mm (Table S2), this corresponds to an average of 24 µN/mm of leg length with an absolute maximum of 48 µN/mm. To compare per-unit-area force outputs, we approximated the wetted leg segment as a cylindrical surface. With a midleg diameter ranging from 80 to 110 μm (Figures S20 and S21), we used an average radius of 0.05 mm. Assuming half the lateral surface of a 1 mm-long cylinder interacts with the water, the effective thrust-generating area is ~ 0.157 mm². Based on this, G. latiabdominis generated an average of 148 µN/mm² of interacting leg surface with an absolute maximum of 305 µN/mm².

These estimates reflect functional differences in thrust-generation strategies: drag-based propulsion in R. distincta through fan employment and surface-tension-based propulsion in G. latiabdominis through longitudinal row-and-gaps ventral setal structures engagement. Importantly, these values offer a quantitative baseline for future mechanical or computational modeling efforts aimed at linking leg microstructure to propulsion performance.

Observations of live R. distincta suggest that thrust during a stroke results from a combination of two forces: hydrodynamic forces generated by the oar-like motion of the fan, as proposed in previous studies^5,7,9, and additional capillary forces arising from an anteroposteriorly asymmetrical dimple beneath the tarsus, consistent with surface-tension-based mechanisms described in Gerridae⁶. In contrast, behavioral evidence from G. latiabdominis aligns with the surface-tension-based thrust mechanism characterized in detail for Aquarius paludum⁶. The observed differences between these species appear closely tied to variations in leg microstructures, respective internal nano-structural properties, and motion kinematics that reflect their distinct ecological contexts and thrust-generation strategies.

In R. distincta, several microstructural features suggest specialization for hydrodynamic thrust. The hydrophilic properties of the claw likely facilitate surface penetration during stroke initiation, as the fan and claw extend downward from the cleft`s internal compartment. Similar surface textures between the claw and fan setae indicate that the fan may also be hydrophilic, enhancing its ability to submerge through the water surface. The fan`s setae and setulae are oriented to press against the water with their narrow edges, a configuration that minimizes deformation under hydrodynamic forces, as predicted by beam theory¹⁴ (SI Part D).

The internal architecture of the fan setae and setulae appears well-adapted for their role as underwater oars in thrust generation. Each seta comprises a hollow core reinforced with columnar nanofibers, while the associated claw features a lamellar structure—both resembling engineered designs such as sandwich and lamellar composites^15,16, which are known to enhance stiffness and fatigue resistance. Polymer nanofibers are widely recognized for their high strength due to a high surface-area-to-volume ratio, and thinner fibers are particularly associated with greater flexibility and mechanical resilience^17,18. In R. distincta, the small diameter of nanofibers within the internal walls of these flat and hollow beams likely contributes to their flexibility. Combined with the hydrophilic properties of the surface, this structure may underlie the elastocapillary behavior responsible for the fan`s characteristic “fan-like” shape when immersed in water—an effect observed both in our experiments (SI Part 3B; Figures S8 and S9) and in Ortega-Jimenez et al. (2025). Furthermore, the multidirectional growth pattern of nanofibers likely reinforces the setae across multiple planes, enhancing their ability to resist water forces. As with synthetic hollow nanofiber systems<sup>19</sup>, the central hollow core may facilitate stress distribution, controlled deformation under load, and reduced material weight—all contributing to the fan`s structural integrity and functional performance resulting in net thrust outputs of ~ 300 µN per stroke and 180 µN/mm² of fan surface area.

Additionally, the H1 and H2 setae rows positioned at the cleft entrance likely act as a barrier against water intrusion while assisting fan deployment through elastocapillary interactions. Based on the observed stroke speed, setula spacing, and thickness, we estimate the fans to operate at Reynolds numbers (dimensionless quantity expressing the ratio of inertial to viscous forces in a fluid) ranging from ~ 0.03 to 0.20 (Table S8) and fan`s leakiness (degree to which fluid flows through, rather than around, a porous or bristled structure, depending on geometry and flow regime) to range from ~ 0.3 to 0.6 (Figure S27), suggesting it behaves as a “leaky paddle” rather than a solid blade. This is roughly like bristled appendages in copepods and barnacle larvae^20,21. Comparable fan-like structures with potential “leaky paddle” functionality are found in other Veliidae genera such as Tetraripis and Trochopus. In contrast, “Veliidae” species that depend primarily on surface-tension-based thrust, like Velia sp., tend to exhibit more developed ventral “gaps and rows” arrangements and less-developed fan structures⁵, supporting the hypothesis of divergent functional adaptations.

In G. latiabdominis, a different set of microstructural adaptations supports thrust generation primarily through surface tension. The species exhibits denser and more hydrophobic setae on its midlegs than R. distincta, likely reflecting its reliance on surface-tension-based propulsion. Deep, asymmetrical dimples beneath the wetted portion of the leg, along with prominent bow waves, contribute to increased thrust. Interestingly, even the lower hydrophobicity observed in R. distincta can still support thrust via surface tension through dimple formation, though to a lesser degree.

Across both species, leg surfaces involved in surface-tension-based thrust display linear arrangements of distally bending setae forming “gaps and rows.” These are absent from other leg surfaces and may serve specialized functions, potentially related to air retention during dimple formation, which prevents surface penetration and facilitates thrust. Theoretical work^22,23 suggests such arrangements can trap air and maintain smooth water contact, providing theoretical support for our hypothesis that these structures appear to function like pressurized air pockets that resist water surface breakage under thrust loads. Additionally, the smoother longitudinal gaps may reduce adhesion during stroke recovery. This arrangement is more pronounced in G. latiabdominis, where the setae also exhibit nanogrooves known to enhance hydrophobicity^24,25. The posterior concentration of thicker T2 and T3 setae in mid- and hindlegs may be a specific adaptation to withstand the higher pressure during backward strokes, which is exemplified by higher thrust output per leg surface area in G. latiabdominis than R. distincta. These structural reinforcements are aligned with previous findings on jumping and propulsion in surface-dwelling insects^26,27.

Kinematic data further underscores how thrust generation mechanisms are integrated with species-specific leg movement strategies. In R. distincta, hydrodynamic thrust is supported by a midleg stroke that begins from a more acute femur angle and involves backward rotations across multiple leg joints. The resulting fan movement vectors deviate from the body movement axis similar to patterns characteristic to Xenopus frogs, known to incorporate lift²⁸. This suggests potential contribution of lift-like forces to thrust, akin to paddling strategies in human kayaking^29,30 and animal locomotion^28,31. The observed combination of Rhagovelia`s leg movement pattern and strokes of shorter duration would be inefficient for surface-tension-based propulsion on stagnant waters but is well-suited to fast-flowing environments where high-frequency of short strokes with minimal surface contact time is advantageous. The ability of R. distincta to actively control fan protraction and retraction through muscle action⁵ (SI Parts 2 and 3B) could allow for greater flexibility during maneuvering and stroke timing, compared to the hypothetical passive fan deployment proposed in recent studies of another Rhagovelia species^8,9, but not supported by our observations.

Conversely, G. latiabdominis initiates midleg movement at a less acute femur angle, with primary rotation occurring at the coxa–femur joint, and maintains a nearly straight femur–tibia segment. This configuration enhances wetted surface area, allowing for more efficient surface-tension-based thrust. The leg stroke is longer in duration and follows a nearly parallel trajectory to the body axis, facilitating the formation of asymmetrical dimples critical to curvature force production. The backward movement of the leg, nearly perpendicular to its axis, promotes directional asymmetry in the dimple, optimizing forward propulsion typical for this species.

Forelegs and hindlegs support the insect body during thrust and sliding, and both species share a distinct ventral setal arrangement that likely contributes to standing and sliding on the water surface. Flattened, overlapping setae form a beam-like structure along the underside of wetted legs, similar to those described in Gerris and Aquarius^5,10. This configuration minimizes surface penetration and drag, offering support without water breakage. The beam`s alignment with body motion also likely facilitates sliding and steering. In the heavier G. latiabdominis, these beams feature hydrophobic nanogrooves that enhance their supporting function. In contrast, the lighter R. distincta lacks such grooves, suggesting lower support demands. Moreover, the spoon-like setae at the tips of R. distincta`s midlegs may help resist displacement by currents, offering a stabilizing advantage in fast-flowing habitats.

Our results demonstrate that Rhagovelia distincta and Gerris latiabdominis, despite both having independently evolved symmetrical rowing, employ fundamentally different thrust-generation strategies: R. distincta relies primarily on hydrodynamic drag (and possibly lift) via actively controlled pretarsal fans functioning as “leaky paddles,” while G. latiabdominis generates thrust through surface tension. Detailed morphological and kinematic analyses revealed that fan protraction and retraction in R. distincta involve muscle action, whereas elastocapillarity contributes only to fan conformation once submerged⁹. The fan`s nanostructured architecture supports this function through its stiffness, flexibility, and surface wettability. We also show that both species exhibit morphological differences linked to their respective thrust strategies and similarities in ventral microstructures—such as longitudinal setal rows-and-gaps and beam-like setal structures—linked to the mechanisms of support and sliding on water surface. These findings challenge simplified models of leg micromorphology in surface-dwelling insects and suggest that precise microstructural arrangements are critical to locomotor performance. By providing detailed, testable hypotheses on structure-function relationships, our study lays a foundation for future mechanical modeling and for comparative evolutionary analyses connecting microstructure, movement, and ecological adaptation in semiaquatic bugs.

Field sites and study species

In January and February of 2020 and 2023, specimen collections and detailed observations of Rhagovelia distincta (body weight: 4–14 mg; Fig. 9H; Table S1) were made at the Southwestern Research Station, Arizona, USA (SWRS; 31°53′3′′N, 109°12′21′′W). In August and September of 2020, specimen collections and detailed observations of Gerris latiabdominis (16–19 mg; Fig. 9H; Table S2) were made at Gwanak Mountain, Korea (37°26′42′′N, 126°57′51′′E) and Seoul National University, Korea (37°28′57′′N, 126°96′04′′E), respectively. Each individual was weighed (GEM20 High Precision Digital Milligram Jewelry Scale, Smart Weigh, 0.001 g).

Videographic and photographic observations

We filmed four types of high speed and standard videos with Sony RX10-III at 959.04 frames per second (fps) and with Chronos 2.1-HD at 1000–4000 fps of individuals in acrylic containers (18 × 18 cm, filled with water). These standardized still-water conditions were maintained to enable direct, quantitative comparison of kinematic and microstructural variables between species. The strength of this approach is that it removed environmental variability and allowed the isolation of species-specific locomotor traits shaped by their respective ecological conditions.

• Type 1: directly from above (85 and 62 movies collected from six and seven individuals of R. distincta and G. latiabdominis, respectively).

• Type 2: from the side of various angles (below surface, surface level, and above surface) (249 and 87 movies of R. distincta and G. latiabdominis, respectively).

• Type 3: directly from below with light source positioned directly above the container (20 and 60 movies from two and five individuals of R. distincta and G. latiabdominis, respectively) to visualize the shadows on the bottom of the container; shadows correspond to dimples under legs on the water surface.

Two variables were extracted from the video types 2 and 3:

• Wetted midleg length (mm): maximal leg section in contact with the water surface in the middle of fast thrust strokes when midleg angle to body axis approximates 90°. The wetted midleg consisted of tarsus in R. distincta (6 individuals) and proximal tibia to tarsal tip in G. latiabdominis (6 individuals).

• Projected swimming fan area (mm²): fan surface area was estimated from six still frames taken from six type 2 clips of R. distincta, where the midleg tarsus was perpendicular to both the body axis and camera lens axis (Figure S3). Projected area was calculated assuming the fan forms a circular sector with radius equal to the average of six measured setal lengths. As the fan surface is tilted ~ 80° to the lens (rather than 90°), projections slightly underestimate true area. However, since tilt angles for individual frames were unavailable and the bias is minor, we used the projected area as our estimate.

Microscopic observations of leg microstructures

Using optical microscopy, we observed the morphology and behavior of the fan in specimens of R. distincta. Using Scanning Electron Microscopy (SEM), we visualized the hair structures on leg sections that interact with water (SI Part 3).

Contact angle measurements (SI part 4)

Contact angle (in degrees; °), height and width (mm), and respective shape index (height to width ratio) of small droplets (0.096 ± 0.032 mm) on the surface of ventral and dorsal microstructures of tarsomere 3 and tarsomere 2 of R. distincta and G. latiabdominis, respectively, and on the tarsal claw of R. distincta, were measured (with ImageJ 1.53t) in frames of three high-speed videos per condition (2000 fps; Chronos 2.1-HD Camera, Kron Technologies). The specimens, sprayed with water, were mounted on a micromanipulator (MM-3, Narishige, Japan) parallel to the camera (Video S4).

Kinematic profiles of a stroke

Detailed kinematic analyses of symmetric strokes by R. distincta and G. latiabdominis were restricted to data extracted with Tracker (https://physlets.org/tracker/) from selected videos: 21 and 12 strokes from six and six individuals for R. distincta and G. latiabdominis, respectively (Supplemental data 1; Supplemental data 2). All strokes were chosen based on strict inclusion criteria—specifically, individuals had to remain stationary before initiating a straight, uninterrupted thrust—to ensure consistent, high-quality data suitable for accurate 2D digitization and meaningful interspecific comparisons. Cartesian (x, y) coordinates of 10 and 9 points on the insect body for R. distincta and G. latiabdominis, respectively (midleg tips were not digitized in G. latiabdominis due to resolution issues) were digitized and subsequently smooth-splined using the “stats” package^32,33 (df = 5 and smoothing parameter = 0.5).

To compare the two species, we focused on five aspects (Fig. 9) of leg movements during a stroke and extracted the following kinematic variables for each frame, or two consecutive frames, through the thrust phase of a stroke:

• Midleg femur angle (degrees): the angle between the body axis and the femur at the coxae (Fig. 9B) was calculated at each frame. The coxal joint is where the major leg angular movement is performed in both species.

• Femur-tibia angle (degrees): the angle between the femur and the tibia at the femorotibial joint was calculated at each frame (Fig. 9C).

• Tibia-tarsus angle (degrees): the angle between the tibia and the tarsus at the tibiotarsal joint was calculated at each frame (Fig. 9D).

• Midleg angular velocity (degrees/s): calculated by dividing the between-frame difference in midleg femur angles by the latency between the two consecutive frames (i.e., 1/fps).

• Leg velocity (mm/s) (U): the linear velocity in horizontal plane of the midpoint of the wetted midleg length. It is calculated for each pair of two consecutive frames via dividing the distance traveled between by the latency by the two consecutive frames.

Three proxies of “effectiveness” of midleg application for thrust generation during a stroke were calculated (Fig. 9) using basic trigonometry and vector algebra:

• “Effectiveness” of leg velocity vector`s direction (Fig. 9E) (proportion; range 0–1): the proportion of the leg velocity vector (and of R. distincta`s fan protracted under the leg; green vector) employed along the direction parallel to the body movement (blue or violet vectors). Values closer to ‘1’ indicate “more effective” employment of legs on water surface because the backward leg velocity vector is near-parallel to the body axis line ((:{upalpha:}=0^circ:)) resulting in anteroposterior asymmetry of the dimple crucial for curvature force (i.e., surface tension) contribution to thrust⁶. Positive values indicate backward velocity vector (blue) that contributes to forward thrust, while negative values indicate forward vector (violet; when legs are dragged along body movement).

• “Effectiveness” of leg length`s use (Fig. 9F) (proportion; range 0–1): evaluation of the relative length of wetted leg projection (blue) on the line perpendicular to the body movement axis (relative to the actual wetted leg length marked green). In G. latiabdominis, it may be approximately viewed as the effective proportion of the total wetted midleg length (blue) that pushes the surface dimple directly backwards along the leg velocity vector parallel to the body movement direction (blue vector in Fig. 9E). Values closer to ‘1’ indicate “more effective” employment of the midleg length pushing the dimple backwards; they also indicate that the fan surface in R. distincta is approximately perpendicular to the body axis ((:{upbeta:}approx:90^circ:)), under the assumption that tarsus on the water surface lies approximately within the near-vertical plane with the surface of the R. distincta`s fan under water.

• “Effectiveness” of wetted leg orientation (Fig. 9G) (degrees): angle θ indicates the orientation of wetted midleg`s main axis (as well as the plane of the fan protracted under the leg, assuming fan surface`s plane includes the longitudinal axis of wetted leg) relative to the leg velocity vector. Angles closer to ‘90°’ indicate “more effective” employment of the full wetted midleg length in pushing the water surface dimple along the leg velocity vector and creating dimple asymmetry along the velocity vector. Under the assumption that longitudinal axis of wetted leg on the water surface approximately lies within the plane of the R. distincta fan`s surface protracted under water, θ values closer to 90° indicate that the angle between the fan surface and the fan movement direction is near perpendicular and hydrodynamic drag from the fan pushing backwards contributes to thrust.

We additionally extracted three variables from the body movements :

• Body velocity (mm/s): distance (mm) traveled by the body center (position derived from the average of head and abdomen tip positions) between consecutive frames divided by the latency between the two consecutive frames (1/fps).

• Body acceleration (mm/s²): rate of change of body velocity derived from each pair of consecutive body velocity values divided by the latency between the two frames.

• Net force (µN): body acceleration (mm/s²) multiplied by insect body mass (mg) and by 0.001 for unit conversion. It represents a horizontal vector of net thrust force during a stroke.

Kinematic characterization of a stroke

The following kinematic variables were extracted from 21 strokes of six individuals of R. distincta and 12 strokes of six individuals of G. latiabdominis (1 value per stroke):

• Distance traveled by wetted midleg (mm): sum of frame-by-frame distances of wetted midleg midpoint was measured along the actual trajectory of the midpoint during the thrust phase of a stroke.

• Midleg`s stroke amplitude (mm): direct straight-line distance from the initial (beginning of thrust stroke) to the final (end of thrust stroke; when midleg`s velocity vector is no longer opposite to the body velocity vector) positions of wetted midleg midpoint was measured.

• Average angular velocity (degrees/s): mean of all frame-by-frame midleg angular velocities calculated within the thrust phase of a stroke.

• Average midleg velocity (mm/s): mean of all frame-by-frame leg velocities calculated within the thrust phase of a stroke.

• Leg velocity at maximum body acceleration (mm/s): leg velocity that corresponds to the maximum body acceleration (i.e., maximum horizontal net force) within the thrust phase of a stroke.

• Maximal body acceleration (mm/s²): maximum value of body acceleration within the thrust phase of a stroke.

• Maximal net thrust force (µN): maximum value of body force within the thrust phase of a stroke.

• Final body velocity (mm/s): final value of body velocity (between the last two consecutive frames) in the thrust phase of a stroke.

• Midleg thrust duration (ms): latency from the initiation of midleg thrust movements to the moment of their disengagement.

Statistical analyses

All analyses were performed in R version 4.3.1³³. We used linear mixed-effects models (with “individual” as the random factor; “lme4” package³⁴; “lmerTest” package³⁵ to compare the effects of average leg velocity, stroke duration, and distance traveled by wetted midleg on the final body velocity between the two species (interaction with categorical variable “species”). However, these independent variables were correlated, and the statistical models would not allow proper evaluations of these effects. Therefore, we extracted principal components using functions fa.parallel and principal from the “psych” R package³⁶ from the pooled data for both species considering all seven kinematic variables: body velocity, maximum acceleration, average angular leg velocity, average leg velocity, midleg`s stroke amplitude, distance traveled by wetted midleg, and stroke duration. We focused on the variables with loading values > 0.75³⁷.