Parasite defensive limb movements enhance signal attraction in male little torrent frogs: insight into the evolution of multimodal signals

Many animals rely on complex sexual signals that target multiple senses to attract mates and repel rivals. These multimodal mating displays can however also attract unintended receivers, which can be an important driver of signal complexity. Despite being taxonomically widespread, we often lack insight into how multimodal signals evolve from unimodal signals and in particular what roles unintended eavesdroppers play. Here we assess whether the physical movements of parasite defense behavior increase the complexity and attractiveness of an acoustic sexual signal in the little torrent frog (Amolops torrentis). Calling males of this species often display limb movements in order to defend against blood-sucking parasites such as frog-biting midges that eavesdrop on their acoustic signal. Through mate choice tests we show that some of these midge-evoked movements influence female preference for acoustic signals. Our data suggest that midge-induced movements may be incorporated into a sexual display, targeting both hearing and vision in the intended receiver. Females may play an important role in incorporating these multiple components because they prefer signals which combine multiple programs. Our results thus help to understand the relationship between ecological and sexual selection pressure operating on signalers and how in turn this may influence multimodal signal evolution.

play. Here we assess whether the physical movements of parasite defense behavior Sexual selection can drive signal evolution through preferences for complex mating 40 displays [1][2][3] . Sexual signalers can either increase their attractiveness by enhancing the 41 complexity in a single sensory modality or by evolving displays that target multiple 42 sensory modalities 4, 5 . Such multimodal signaling can be highly complex, often 43 involving multiple underlying neuronal motor programs that need to be synchronized 44 in order to perform well [5][6][7][8] . The production and reception of multimodal signals is 45 often more costly in terms of energy loss or increased of predation and parasitism 46 when compared to unimodal signals 9 , and their evolution is therefore often explained 47 through functional benefits, such as cross-modal perception by receivers, which can 48 improve signal detection and discrimination, or enhance attention and memory time 7, 49 10, 11 . We know however far less how multimodal signals evolve from unimodal ones. 50 An important question remains whether and when multimodal signals evolve de novo, 51 or evolve through a process of co-option, by incorporating additional cues into a 52 unimodal mating display 7 . 53 Most species generate by-product cues during signaling. For instance, floating frogs 54 produce water ripples when calling from the water. These ripple cues have become 55 part of the sexual display, as their presence modulates receiver responses to their 56 acoustic signal components 12 . Multimodal signals can thus originate from cues 57 associated with primary signal production, either through a physical linkage (e.g. case 58 of call-induced water ripples) or through cues generated by other non-communicative 59 behaviors that have subsequently been integrated as part of a sexual display. Such 60 4 process of co-option has been proposed for many ritualized visual displays which are 61 predicted to have evolved from different intra-or interspecific activities such as 62 intention movements, protective and autonomic responses 13, 14 . For example, 63 comparative analyses on Anatidae (i.e. ducks) suggest that the precopulatory displays 64 of head-dipping seem to be derived from bathing behavior 15    Five observed limb movements are not only spontaneously generated, but also 126 induced by insects ( figure 2A). The passive visual movements were predominantly 127 evoked by some potential hematophagous parasites such as midges and sandflies 128 (movie S1, movie S2 and figure S2). Specifically, we identified Corethrella spp 129 midges and Phlebotomus flies, which prefer to feed on the blood of ectotherms 24, 28 .

130
As seen in figure 2A, the movements that were produced by parasite interactions had a  figure 2B). We also ran the same analyses but restricted to the two movements found 138 to be attractive to females (i.e., AW and HFL movements; see also below). As a result, 139 we found the same correlation between AW/HFL and the presence of parasites 140 (Pearson's correlation; N = 39; R = 0.737; P < 0.001).

141
Calling males show more parasite-evoked limb movements 142 We compared the limb movements between calling individuals (N = 39) and  Among the six types of limb movements, the TT and LS + FF were rarely induced by 145 parasitic insects (table S1). We thus only compared the difference between calling 146 males and silent males for the other four visual displays. We found calling males to  In little torrent frogs, we did not observe a potential visual signal related with body 201 displays. This species, however, performed diverse common limb displays.

359
We also filmed males that successfully attracted females to close-range and 360 interacted for about 10 min. Males with a female close by (less than 0.5 m) and those 361 without one (without female or other male within 1.5 m) were defined as the 362 close-range and long-range categories, respectively. The limb displays were classified 363 into seven types according to two published ethograms for anurans (see Results).

364
When midges or mosquitoes were on the body of frogs, they would shake their body 365 or move their limbs to repel those parasites (movie S1). Frogs also produced limb 366 movements towards parasites that were flying close to their body (movie S2). We with advertisement calls 32 . In addition, the LS and FF displays were not produced 390 during parasite interactions. We therefore did not include them when examining the 391 roles of parasite-evoked movements in female mate-choice.

392
The video stimuli were edited in Adobe Premiere Pro CS6. We firstly cut each clip 393 to 6 s and replicated them to generate a new video with a total length of 10 min, 394 respectively. In those videos, the display rate of each movement was within a natural 395 range. Next, we changed the audio channel of the video by replacing the original 396 recording with standard sound files. The standard sound files (with flowing water and 397 calls included) were produced according to an stimulus used in a previous study 51 . 398 The stimulus was synthesized based on the characteristics of thirteen males (average    447 We analyzed all data on male visual display and female mate-choice in R (v.3.5.3). 448 We used a Pearson correlation analysis to determine the relationship between the level 449 of parasite interference and the number of visual display. We carried out Wilcoxon 450 rank sum tests to evaluate the difference of parasite-induced visual displays (i.e.  Competing interests 465 We have no competing interests.

Supplementary 2
Movie S1. Video that frog produces defensive motions in order to repel midges.
Movie S2. Video that frog produces movement towards a flying parasite.
Supplementary 3 Table S1. The data of parasite-induced and spontaneous displays in each limb movement for calling males, silent males and males that have females nearby.