bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

1 Parameter tuning facilitates the evolution of diverse tunneling patterns in 2 3 Nobuaki Mizumoto1*, Paul M. Bardunias2, Stephen C. Pratt1 4 5 1 School of Life Sciences, Arizona State University, Tempe, AZ, USA 85287 6 2 Department of Biological Sciences, Florida Atlantic University, Boca Raton, FL 33431 7 * Correspondence: [email protected] 8 9 Abstract 10 The nest structures built by social are complex group-level patterns that emerge from 11 interactions among individuals following simple behavioral rules. The theory of complex systems 12 predicts that there is no simple one-to-one relationship between variations in collective patterns and 13 variation in individual behaviors; therefore, it is essential to know how actual behavior evolves to 14 change pattern formation. Here we demonstrate that the evolutionary divergence of tunneling 15 patterns is achieved by quantitative tuning of shared behavioral rules, rather than the acquisition of 16 novel behaviors. We compared tunnel formation between two closely related species, 17 tibialis and Heterotermes aureus, and found that H. aureus builds more highly branched tunnels than 18 R. tibialis. Our behavioral analysis and data-based modeling revealed that these species share the same 19 behavioral repertoire, but a quantitative difference in the probability of sidewall excavation leads to 20 diverse tunneling patterns. In contrast, we also found that Paraneotermes simplicicornis, which 21 evolved tunneling independently, possesses a distinct behavioral repertoire, but shows convergence of 22 branching patterns with R. tibialis. These results elucidate the complex relationship between individual 23 behavior and group-level patterns; in some cases, distinct behavioral rules can produce similar group- 24 level patterns, but in others, a common rule set can yield distinct patterns via parameter tuning. The 25 evolutionary process of collective behavior is flexible and much more complex than we can infer from 26 group-level patterns alone. 27 28 Introduction 29 The coordinated behavior of group-living often creates complex group-level patterns [1]. 30 Among these, nest structures built by social insects play an important role in their ecological success 31 by providing shelter and favorable microenvironments [2,3]. A wide variety of structures has evolved, 32 adapted to each species’ typical environment [4,5]. This leads to the fundamental question of what is 33 the behavioral mechanism underlying the evolution of diverse nest structures? In collective building, 34 group-level structures emerge from local interactions among individuals following simple behavioral 35 rules [1,6]. Thus, different collective outcomes may be obtained either by differentiated behavioral 36 rules or by regulation of a common set of rules to modify the interactions [7]. Theoretical studies have 37 supported the latter model; they predict that diverse nest structures can be explained by parameter 38 tuning, which is the quantitative modification of a single set of behavioral rules shared among species, 39 in ants [8,9], termites [10–12] and paper wasps [13,14]. However, because of the lack of comparative 40 studies, there is no empirical evidence for the sharing of behavioral rules across species, and thus the 41 key factor creating interspecific variation in patterns remains unknown. 42 Because evolutionary changes of nest structures occur through the modification of building 43 processes, it is essential to clarify the relationship between individual behaviors and group-level bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

44 patterns. Evolutionary developmental biology has treated similar challenges by elucidating the 45 mechanistic relationship between individual development and phenotypic change during evolution 46 [15,16]. Studies on the molecular basis for divergence and convergence of individual morphology 47 suggest that all scenarios are possible, where different forms may evolve by similar mechanisms [17] 48 and similar forms may evolve by different mechanisms [16]. Applying this perspective to nest 49 structures, which can be seen as the external morphology of a colony [18], social insects may not only 50 use parameter tuning mechanisms but also acquire differentiated behavioral rules during the evolution 51 of collective building. 52 In this study, we analyze the relationship between individual behavior and collective pattern in 53 the tunneling behavior of termites. Several termite species build tunnels to protect foragers from 54 desiccation and predators [19]. Termite species can vary in tunneling patterns, reflecting species- 55 specific foraging strategies and differences in the distribution of wood resources experienced by each 56 species [20–22]. Moreover, our phylogenetic analysis indicates that tunneling behavior has evolved 57 several times independently in termites (Figs. 1, S1). This provides an opportunity to explore the extent 58 to which behavioral rules are shared among species with different degrees of relatedness. 59

60 61 Figure 1. Simplified phylogeny of lower termites (modified from [23,24]) with information on tunneling 62 behavior. Ancestral states were reconstructed with maximum parsimony (detailed in SI text and Fig. S1). 63 Tunneling through the soil has evolved four times independently in Mastotermitidae, Hodotermitidae, 64 Paraneotermes, and . In this study, we used three species from the three underlined genera, 65 Paraneotermes (Kalotermitidae), Heterotermes, and Reticulitermes (Rhinotermitidae). 66 67 To trace the evolutionary changes of tunneling patterns and behaviors, we used three 68 subterranean termite species. Paraneotermes simplicicornis (Kalotermitidae) evolved tunneling 69 independently from Reticulitermes tibialis and Heterotermes aureus (Rhinotermitidae) (Figs. 1, S1). 70 We observed tunnel development at two different scales: the patterns of tunnel branching and the 71 behavior of each termite. We empirically demonstrate that there is no simple one-to-one relationship 72 between individual behaviors and group-level patterns. We find that R. tibialis and H. aureus build 73 tunnels with distinct branching patterns from each other by using shared behavioral repertoires, while 74 P. simplicicornis builds similar tunneling patterns to R. tibialis using a distinct behavioral repertoire. 75 We combine empirical observations and data-based simulations to show that different branching 76 patterns between R. tibialis and H. aureus result from parameter tuning of the same behavioral rules. bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

77 Thus, quantitative modification of shared behaviors can play an important role in the evolution of 78 diverse group-level patterns, without any change in the behavioral rule itself. 79 80 Results 81 Tunneling patterns 82 Our observation of tunnel development in a two-dimensional arena detected both convergence 83 and divergence of termite tunneling patterns (Fig. 2A). The most striking feature of group-level 84 patterns is the number of branches. When we counted the number of tunnel faces, we found that H. 85 aureus built a significantly larger number of branches than P. simplicicornis and R. tibialis, while there 86 was no significant difference between P. simplicicornis and R. tibialis (generalized linear model 2 87 [GLM]; likelihood- ratio test, χ 2 = 11.568, P = 0.00308; Tukey Contrasts: R. tib. - P. sim.: z = 0.376, 88 P = 0.925, H. aur. - P. sim.: z = 3.007, P = 0.0073, H. aur. - P. sim.: z = 2.611, P = 0.0245; Fig. 2B). 89 Branching events were mainly found near the tunnel entrance (Fig. 2A), where the length of the tunnel 90 before the branches is much shorter than that after the branches (Fig. 2C). 91

92 93 Figure 2. Interspecific comparison of termite tunneling patterns. (A) Typical tunneling patterns of each species. 94 The red lines in the simplified phylogeny above the photos indicate the independent evolution of tunneling. Red 95 circles indicate branching points and blue circles indicate the faces of the tunnels. (B) Comparison of the number 96 of tunnel faces among species. Different letters indicate significant differences (P < 0.05). (C) Comparison of 97 the tunnel length when divided into segments. Initial tunnel is a segment from the entrance to the first branch. 98 Secondary tunnel is a segment between two branches. Edge tunnel is the segment reaching the faces of the 99 tunnels. When a tunnel has no branch, it only contains an edge tunnel. 100 101 Individual digging behavior 102 Individual behavior during tunneling did not correspond directly to the group-level patterns; 103 instead, P. simplicicornis used a distinct transporting behavior unlike that of either R. tibialis or H. bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

104 aureus. In this behavior, each individual of P. simplicicornis excavated sand with its mandibles, 105 formed the sand into a ball with its legs, and then kicked the ball backward to the individual behind it 106 (Fig. 3A; Video S1). This behavior was observed only in this species (Fisher’s exact test, P < 0.0001; 107 Fig. 3B) and contrasts with the well-recognized behavior of rhinotermitid termites; these species 108 excavate and carry sand particles using their mandibles (Fig. 3A; Video S2). These different behaviors 109 were associated with different tactics for removing the sand; R. tibialis and H. aureus individually 110 carry sand out of the tunnel, while P. simplicicornis instead forms a bucket brigade of multiple 111 individuals. This was apparent in the fact that R. tibialis and H. aureus, but not P. simplicicornis, finish 112 transportation by compressing clumps of sand particles against the sidewall (Fisher’s exact test, P < 113 0.0001; Fig. 3B). P. simplicicornis just kicked sand balls into the tunnel passage, where they were 114 taken over by another individual. In addition, the order of individuals inside a tunnel was maintained 115 in P. simplicicornis, where the individual in the 1st-row at the tunnel face was less likely to move back 116 and change positions with 2nd-row individuals (one-way analysis of variance [ANOVA], F2 = 24.307, 117 P < 0.0001; Fig. 3C). Moreover, P. simplicicornis visited the tunnel face fewer times (ANOVA, F2 = 118 24.892, P < 0.0001; Fig. 3C), indicating that they transport a large amount of sand at once using the 119 bucket brigade. Finally, after visiting the tunnel face, 1st-row individuals of P. simplicicornis only 120 moved back a short distance as another individual can take over the sand ball. This trend was prominent 121 when the tunnel became longer, where P. simplicicornis typically moved back only about 10 mm 122 regardless of the length of the tunnels, while R. tibialis and H. aureus moved back increasingly long 123 distances as the tunnel lengthened (Fig. 3D). 124 We further investigated interactions among individuals to specify individual-level differences 125 that might account for the different tunnel patterns of R. tibialis and H. aureus. During excavation, 126 direct interaction between termites can occur in a clogged tunnel. We observed the behavior of 2nd- 127 row individuals when they found themselves immediately behind the 1st-row individuals who are 128 excavating at the tunnel face (Fig. 4A). We found that R. tibialis and H. aureus chose from the same 129 behavioral repertoire, either excavating the sidewall or waiting until the 1st-row individual has finished 130 excavation. However, the two species differed in the frequency of these behaviors (Fisher’s exact test, 131 P = 0.0035, Fig. 4A). The 2nd-row individuals in R. tibialis more often waited, while those in H. 132 aureus had a higher probability of beginning to excavate the sidewall (Fig. 4A). In contrast, P. 133 simplicicornis showed a distinct behavior not present in the other two species, where the 2nd-row 134 individual took over the transportation of the sand ball kicked back by the 1st-row individual (Fig. 3A, 135 4A). Thus, phylogenetically distinct species have different behavioral repertoires, while closely related 136 species share the same repertoire but quantitatively modify their use of it, indicating the existence of 137 parameter tuning. 138 bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

139 140 Fig. 3. Comparison of individual behaviors during tunneling. (A) Illustration of sand excavation behavior. After 141 excavating sand at the tunnel face, P. simplicicornis use their legs to form a ball of sand particles and kick it 142 back behind, where it is taken over by the 2nd-row individual. R. tibialis and H. aureus instead carry sand 143 particles with their mandibles, and the individual in the 1st-row often changes. (B) Interspecific comparison of 144 the probability to show kicking behavior and completing transportation. Completion of transportation is defined 145 as compressing or attaching sand particles, which are then not used by other individuals. (C) Interspecific 146 variation of the number of visits to the tunnel face and the number of position switches between 1st-row and 147 2nd-row individuals. (D) Comparison of the distance of backward movements after visiting the tunnel face. 148 Arrows indicate median values. 149 150 Simulations 151 We hypothesized that the observed quantitative difference in sidewall excavation between R. 152 tibialis and H. aureus is the mechanism of branching pattern variation. In fact, it has been reported that 153 such sidewall excavation widens the tunnel and can eventually result in a new branch [25]. To test our 154 hypothesis, we developed a cellular automaton model simplifying the tunneling process (Fig. S2). In 155 the model, each termite moves towards the end of the tunnel from the installation area as long as the 156 space in front is empty. When a termite reaches the face of the tunnel, it excavates the cell’s contents, 157 making it empty. After excavation, the termite moves back some distance to unload the sand particle. 158 If a termite arrives at the tunnel face to find another termite already excavating there, then it chooses bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

159 between waiting for the current excavator to finish or instead excavating the sidewall and thus starting 160 a new branch tunnel (Figs. 4B, S2). We used the species-specific probabilities of these behaviors in 161 the simulations, based on the experimental results for R. tibialis and H. aureus (Fig. 4AB). We also 162 simulated building by P. simplicicornis to predict if the same mechanism can explain the branching 163 patterns of this species. For them, we added the behavior that one termite can take over sand particles 164 from another termite. In the simulation, the side length of a single cell is 10 mm, and we observed the 165 development of tunnels until the longest path reached 100 mm. As in the experiments, we characterized 166 branching pattern by counting the number of tunnel faces. 167

168 169 Fig. 4. Parameter tuning mechanism for interspecific variation of tunneling patterns. (A) Interaction rules 170 between individuals within a clogged tunnel. In P. simplicicornis, a 2nd-row individual often takes over the 171 sand ball the 1st-row individual has kicked back. On the other hand, R. tibialis and H. aureus wait or excavate 172 side wall, where these two species are different in the frequency of these behaviors. (B) Behavioral rules 173 governed the simulated interactions for R. tibialis and H. aureus. If a termite finds an excavating individual in 174 front of it, it will choose an action of wait or excavate side wall depending on the probability obtained in the 175 experiments. (C) Comparison of the results of empirical experiments and simulations. Histograms indicate the 176 mean values for every 15 or 16 simulations (N = 1000), while red arrows indicate the mean value of empirical 177 experiments. 178 179 The model effectively reproduced the interspecific variation among rhinotermitid termites, where 180 R. tibialis built tunnels with less branching than H. aureus (Fig. 4C). However, the model 181 underestimated the branching rates of P. simplicicornis, suggesting that this species may have another 182 behavioral mechanism for branching in addition to sidewall excavation within a clogged tunnel. 183 Moreover, our model revealed that high local density causes branching in termite tunnels. Our 184 experiments showed that the branching of termite tunnels is concentrated near the beginning of tunnels 185 (Fig. 2D). The same pattern was reproduced by our simulations (Fig. S3). This is because, at the bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

186 beginning of the excavation, the area of the tunnels is smaller, which increases the local density of 187 termites and the probability of sidewall excavations. As the tunnels grow, the area increases and local 188 density declines, which results in lower branching rates in later stages of excavation. It is well known 189 that the group size and density affect the tunneling structures in many social insects [26–28]. But our 190 results indicate that even with the same group size, the change of the local density of individuals will 191 greatly affect pattern formation. 192 193 Discussion 194 Our comparative study revealed a complex relationship between behavioral mechanism and group- 195 level patterns. We found that two closely related species (R. tibialis and H. aureus) share behavioral 196 repertoires, but quantitative differences in the frequency of different actions result in divergent 197 branching patterns (Fig. 4). This result shows that parameter tuning of the same rule set plays an 198 important role in the evolution of collective building, and thus that a dramatic change of behavioral 199 repertoires is not required to produce diverse nest structures among species. In contrast, we also found 200 that two phylogenetically divergent species (P. simplicicornis and R. tibialis) possess different 201 behavioral repertories for collective excavation, but this does not necessarily yield different structures, 202 since both create tunnels with a similar branching pattern (Figs 2, 3). Thus, similarity of patterns need 203 not imply a shared behavioral algorithm. Altogether, we conclude that the evolutionary process of 204 collective behavior is much more complex than the transition of group-level patterns. This makes it 205 impossible to solve the inverse problem of inferring individual behavioral rules from the collective 206 structures that they produce. Our result emphasizes the importance of direct comparative studies of 207 behavioral mechanisms of self-organizing systems. 208 Termite phylogeny shows that tunneling behavior was present in the common ancestor of 209 Rhinotermitidae (Fig. 1). This suggests that parameter tuning of shared behavioral repertories explains 210 pattern diversification in this whole group. The Rhinotermitidae have a wide diversity of tunneling 211 patterns [19,20], which are often connected to different foraging strategies [21,29,30]. Optimal search 212 theory predicts a compromise between reducing detection errors and quickly exploring a wide area 213 [31]. In this sense, H. aureus engages in an intensive search by building highly branched tunnels, while 214 R. tibialis performs an extensive search by focusing on fewer tunnels. This may reflect their habitats, 215 where H. aureus is found in deserts with cactus resources that are small and relatively difficult to find, 216 while R. tibialis lives in pine forest with large wood resources. However, other factors including colony 217 size and traveling cost also determine search efficiency [32,33]. We found that the behavioral 218 mechanism underlying this adaptation is simple, where the tunnel geometry is sensitive to a single 219 behavioral parameter governing interactions; namely, a threshold for individuals in a clogged tunnel 220 to excavate a sidewall instead of waiting for access to the tunnel face. Thus, without the need for a 221 change in communication systems, selection acting on this parameter can easily result in the 222 evolutionary divergence of tunnel geometry, adapting it to the local environments experienced by each 223 species. This tunability of behavioral algorithm may have helped the Rhinotermitidae to spread to a 224 wide range of environments [34]. 225 On the other hand, the evolution of a differentiated behavioral rule in P. simplicicornis indicates 226 the importance of evolutionary contingency. This species’ behavioral repertoire appears to have been 227 shaped by the physiological and morphological traits of its family, Kalotermitidae. First, kalotermitids 228 move slower than rhinotermitids, possibly because of lower metabolic rates or shorter legs [35,36]. bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

229 During our observations, the maximum instantaneous moving speed of P. simplicicornis was 230 significantly slower than that of R. tibialis or H. aureus (Fig. S4). Kicking works well for slower 231 moving termites because it requires a shorter total movement distance to excavate a unit length of 232 tunnel (Fig. S4). Second, the body shape of kalotermitids is more elongated than that of rhinotermitids 233 (Fig. S5), which limits turning around inside narrow tunnels [37]. Because of this characteristic, P. 234 simplicicornis may do better with the kicking type of tunneling. Indeed, turning behavior, which often 235 involves transportation of sand particles for a longer distance, is less frequently observed in P. 236 simplicicornis (Fig. S5). Thus, the kicking type of tunneling is an adaptation to confined space for 237 species with lower mobility. 238 When a group of animals moves within a narrow and confined space, they face a problem of 239 high-density clogs which affect task performance and collective outcomes [27,37]. The bucket brigade 240 is one solution, because excavators do not need to pass each other [38]. In addition to P. simplicicornis, 241 there are a few observations of this behavior in social mole-rats [39] and army ants [40]. Another 242 mechanism, observed in fire ants, is individual idleness, which limits the number of excavators at the 243 face of tunnels and reduces the frequency of clogs [37]. The higher proportion of waiting behaviors by 244 R. tibialis is consistent with this idea. Thus, there are different clog control mechanisms behind the 245 convergence of reduced tunnel branching in P. simplicicornis and R. tibialis. Instead of reducing local 246 density, H. aureus exploits high-density clogs as a mechanism for building new branches, thus forming 247 highly branched tunneling patterns. A similar mechanism of high density creating a new branch has 248 also been proposed for ant nest construction [27]. Combined with previous studies, our results illustrate 249 that collective behavior in confined space is flexible within each taxon, suggesting that each 250 group’s behavior is an adaptive trait. 251 The central goal of evolutionary biology is to find the factors and mechanisms which are 252 responsible for the evolution of phenotypes. Two striking evolutionary phenomena are divergence of 253 phenotypes among closely related species and convergence between distinct related species. Studies 254 on morphological development of individual animals have found that there is not a one-to-one 255 relationship between morphology and its genetic determinants; different forms can evolve by altering 256 the expression of shared genes [17], and similar forms can evolve by using a different molecular basis 257 [16]. Under the superorganism analogy, the nest is the extended phenotype of an colony, formed 258 by the behavioral rules of colony members. Consistent with studies on morphological development, 259 our study reveals that diverse group-level patterns between species are produced from shared 260 behavioral rules, and also provides a novel example of convergent extended phenotypes based on 261 distinct behaviors. This challenges theoretical studies that often assume the same behavioral rules 262 across taxa based on the observation of a limited number of species. Direct comparative studies 263 promise a comprehensive view of the mechanisms of collective behavior and will give us an 264 understanding of the origin of coordination, as well as the general algorithms underlying it. 265 266 Materials and Methods 267 Termites 268 We used four colonies of Paraneotermes simplicicornis (Kalotermitdae) and five of 269 Heterotermes aureus (Rhinotermitidae) collected from cholla and mesquite desert in Gila and 270 Maricopa Counties, and five colonies of Reticulitermes tibialis (Rhinotermitidae) collected from a pine 271 forest in Pinal County, Arizona, USA. P. simplicicornis is the sole subterranean termite species in bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

272 Kalotermitidae [41], while all species in Heterotermes and Reticulitermes are counted as subterranean 273 termites. Colonies were maintained at 22 ℃ in plastic boxes with wood or cactus pieces and the soil 274 in which they were nesting in the field. 275 276 Macro-scale observation of tunneling patterns 277 To compare the branching patterns of tunnels, we prepared experimental setups for observing 278 two-dimensional tunneling patterns. These consisted of three layers. The middle layer, whose thickness 279 is adjusted to each species (1 mm for H. aureus and R. tibialis and 2 mm for P. simplicicornis) had a 280 round area filled with white sand (Marble White Sand, National Geographic, USA) moistened with 281 distilled water (10% by volume). At the edge of the round area was a teardrop-shaped area where 282 termites could be introduced (Fig. 2A). Sand particles were homogenized into 0.15 ~ 0.25 mm size 283 range using two screens with 60 and 100 mesh. The top layer had an opening only above the entry 284 area, which was covered by a glass plate. We used 20 termites for this experiment. Each individual 285 was used only once. After placing termites in the introduction area, we recorded tunnel development 286 for 24 hours. Snapshots were imported into ImageJ (US National Institutes of Health, Bethesda, MD, 287 USA) and measurements were taken by tracing the length of each branch after calibration. We defined 288 the beginning of the tunnel structure as a single point connected to the entry area. Tunnels greater than 289 one body length (6.2 mm, 4.4 mm and 3.9 mm for P. simplicicornis, R. tibialis, and H. aureus, 290 respectively) were counted as unique branches. 291 Overall, P. simplicicornis formed tunnels much more slowly than R. tibialis and H. aureus (Fig. 292 S6). To avoid an effect of environmental heterogeneity arising from the wall at the boundary [42], we 293 compared the structures of tunnels at the time when the first group in each species reached the wall 294 (14 hours in P. simplicicornis and 5 hours in R. tibialis and H. aureus; Fig. S6). We compared the 295 number of tunnel faces among species using a generalized linear model (GLM) with Poisson error and 296 a log-link function. The likelihood-ratio test was used to test for statistical significance of the 297 explanatory variable (type-II test). We pooled the data of three colonies for each species because we 298 did not find any significant colony variation (GLM, likelihood-ratio test, P > 0.12). In case of 299 significant effect of species, we ran Tukey’s post hoc test. 300 301 Micro-scale observation of digging behaviors 302 To compare micro-scale digging behaviors among the three species included in our study, we 303 prepared experimental arenas of two different sizes depending on species (small: H. aureus and R 304 tibialis; large: P. simplicicornis). The experimental arenas consisted of three layers; bottom and middle 305 layers made of acrylic board and a top layer made of glass plates. The middle layer was L shaped and 306 included an empty square area for introducing termites (small: 15 × 15 × 1 mm; large: 20 × 20 × 2 307 mm) and a narrow passage (small: 3 × 100 × 1 mm; large: 4 × 100 × 2 mm) filled with white sand (Fig. 308 S7A). 309 We selected ten termites haphazardly from available colonies for each trial of this experiment. 310 We used workers for H. aureus and R. tibialis; for P. simplicicornis we used pseudergates or nymphs 311 who play the role of the worker caste in Kalotermitidae, which have no true worker caste [43]. Each 312 P. simplicicornis group contained either all pseudergates or all nymphs. All termites were marked with 313 one dot on the head and two dots on the gaster (Racing Finish, Pactra, Testors, Rockford, IL, USA). 314 We used the marking on the head for tracking, and those on the gaster for individual identification. bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

315 After installing termites in the square area, we recorded their behaviors until they dug a tunnel 50 mm 316 long; as in the two-dimensional experiment, P. simplicicornis took longer to reach this milestone (Fig. 317 S6). A video camera was mounted above the arena covering the square area and first 50 mm of the 318 passage. We observed three groups from two colonies for each species; one replicate for P. 319 simplicicornis (colony A, rep 3) was censored at 24 hours after introduction of the termites, when the 320 tunnel had reached 47.60 mm in length. We used each individual only once. 321 All videos were split into 30 minute segments. Then we identified the segment in which the 322 termites started excavation. Starting with the immediately following segment, we observed their 323 behavior for 10 minutes every 60 minutes. During observations, we extracted the coordinates of each 324 termite’s head at a rate of 1 frame per second from each video using the video-tracking system 325 UMATracker [44]. We also measured the length of the tunnel at the beginning and the end of each 326 observation. 327 By analyzing the trajectories, we obtained the number of visits to the tunnel face by 1st-row 328 individuals and the number of changes in position between the 1st- and the 2nd-row individuals. We 329 estimated the mean numbers of these behaviors performed during the digging of a 1mm length of 330 tunnel. Then we compared the mean frequency of these behaviors among species using one-way 331 analysis of variance (ANOVA). We pooled the data of two different colonies for each species because 332 we did not find significant colony variations for any species (t-test, P > 0.10). 333 To form the tunnel, termites visit the tunnel face, excavate sand, and then transport sand particles 334 away from the tunnel face. Because of the narrow tunnel, only the 1st-row individuals can access the 335 tunnel face. Thus, we focused detailed analysis on the behavior of 1st-row individuals, and we 336 determined their behavioral repertories when the tunnel is longer than 40mm. We considered the 1st- 337 row individual to have visited the tunnel face when its position came within 1.5mm of the tunnel face 338 and then backed away more than 2mm (or 3mm for P. simplicicornis). We observed these visits to 339 check if the termites excavated sand, how they carried sand particles, and where and how they 340 deposited them. Next, we examined the interaction patterns among individuals by focusing on the 2nd- 341 row individuals who are found just behind a 1st-row individual visiting the tunnel face. We only 342 considered 2nd-row individuals that were within a minimum distance of the 1st-row termite. This 343 minimum distance was 6.5 mm, 4.5 mm and 4 mm for P. simplicicornis, R. tibialis, and H. aureus, 344 respectively (i.e., a little larger than body length for each species). The frequency of observed 345 behaviors was compared among species using Fisher’s exact test. 346 347 Individual-based model 348 We modeled a 2D discrete space, representing positions in tunnels. Each cell had two possible 349 states (empty and sand-filled), and termites were modeled as mobile agents, each one occupying a 350 single empty cell. All termites were initially placed in the introduction area, which can contain all 351 individuals. 352 Termites have five different states: moving forward (advancing), excavating, backing (with or 353 without loading) and waiting. Inside a tunnel, termites determine their behaviors depending on their 354 state and their interactions with other individuals. Each behavior takes one time step (Fig. S2). If 355 individuals don’t encounter others, advancing termites move forward as long as the cell in front is 356 empty. When the front is sand filled, advancing termites change to the excavating state; then they 357 excavate sand and change to the state of backing while loaded with sand. Backing termites move back bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

358 as long as the cell behind them is empty, until they have moved a given backing distance or left the 359 tunnel. As the backing distance increased proportionally to the tunnel length (Fig 3D), we determined 360 the backing distance by multiplying the tunnel length by a random number generated from a beta 361 distribution with parameters α and β (Table S1), which are variable among species and obtained by 362 fitting to observed transport distances (data are shown in Fig. S5). We used a beta distribution because 363 it can describe the bimodal shape observed in the data. After backing, individuals unload the sand and 364 enter the advancing state. Once unloaded, the sand is no longer available in simulations of R. tibialis 365 and H. aureus, but remains available in P. simplicicornis, where other individuals can pick it up in a 366 time step. 367 When there is an excavating individual in front of an advancing termite, it will enter either the 368 waiting or excavating state, with probabilities based on those observed for waiting and sidewall 369 excavation in the experiments (Fig. 4A). Then, waiting individuals will swap positions with the 370 individual in front once its state changes to backing; excavating individuals will excavate the sidewall 371 to create a new branch. When there is a backing individual in front of an advancing termite, the latter 372 changes to the backing state because of the confined space. Moreover, in P. simplicicornis, when an 373 advancing individual encounters a backing individual who is loaded with sand, the advancing termite 374 takes over the sand particles. After this, the backing individual changes to advancing state, while the 375 advancing individual changes to backing state. Similarly, individuals of P. simplicicornis can also 376 choose to take over a sand load when they encounter excavating individuals (Fig. S2). 377 In each trial, we modeled 20 individuals, as in the experiments. These 20 individuals act 378 sequentially in random order at each time step, except for swapping or taking over where we need to 379 compute the action of two individuals at the same time. The side length of a single cell is 10 mm, and 380 we observed tunnel development until the longest path reached 100 mm. As in the experiments, we 381 ran the simulation 15 times for R. tibialis and H. aureus and 16 times for P. simplicicornis. Then we 382 counted the number of tunnel faces and measured the mean values. We repeated this process 1000 383 times to estimate the expected range of the mean value. All calculations were performed using R 384 v.3.5.3 [45]. 385 386 Acknowledgments 387 We thank K. Baudier and C. Kwapich for helpful comments during analysis; T. Bourguignon for 388 information on termite phylogeny; M. De La Monja and A. Rizo for help during image analysis; 389 members of the Pratt laboratory and the Social Insect Research Group at ASU for helpful discussion. 390 N.M. is supported by a JSPS Overseas Research Fellowship. 391 392 References 393 1. Camazine S, Deneubourg J-L, Franks NR, Sneyd J, Theraulaz G, Bonabeau E. 2001 Self- 394 organization in Biological Systems. Princeton: NJ: Princeton University Press. 395 2. Hansell MH. 2005 Animal architecture. Oxford: Oxford University Press. See 396 http://library.wur.nl/WebQuery/clc/1748670. 397 3. Hughes D, Pierce N, Boomsma J. 2008 Social insect symbionts: evolution in homeostatic 398 fortresses. Trends Ecol. Evol. 23, 672–677. (doi:10.1016/j.tree.2008.07.011) 399 4. Tschinkel WR. 2015 The architecture of subterranean ant nests: beauty and mystery underfoot. 400 J. Bioeconomics 17, 271–291. (doi:10.1007/s10818-015-9203-6) 401 5. Perna A, Theraulaz G. 2017 When social behaviour is moulded in clay: on growth and form of 402 social insect nests. J. Exp. Biol. 220, 83–91. (doi:10.1242/jeb.143347) 403 6. Theraulaz G, Bonabeau E. 1995 Coordination in distributed building. Science. 269, 686–688. bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

404 (doi:10.1126/science.269.5224.686) 405 7. Pratt SC, Sumpter DJT. 2006 A tunable algorithm for collective decision-making. Proc. Natl. 406 Acad. Sci. U. S. A. 103, 15906–15910. (doi:10.1073/pnas.0604801103) 407 8. Khuong A, Theraulaz G, Jost C, Perna A, Gautrais J. 2011 A computational model of ant nest 408 morphogenesis. Adv. Artif. Life, ECAL 2011, Proc. Elev. Eur. Conf. Synth. Simul. Living Syst. , 409 404–411. 410 9. Franks NR, Gomez N, Goss S, Deneubourg J-L. 1991 The blind leading the blind in army ant 411 raid patterns: Testing a model of self-organization (Hymenoptera: Formicidae). J. Insect Behav. 412 4, 583–607. (doi:10.1007/BF01048072) 413 10. Mizumoto N, Kobayashi K, Matsuura K. 2015 Emergence of intercolonial variation in termite 414 shelter tube patterns and prediction of its underlying mechanism. R. Soc. Open Sci. 2, 150360. 415 (doi:10.1098/rsos.150360) 416 11. Ocko SA, Heyde A, Mahadevan L. 2019 Morphogenesis of termite mounds. Proc. Natl. Acad. 417 Sci. U. S. A. 116, 3379–3384. (doi:10.1073/pnas.1818759116) 418 12. Bonabeau E, Theraulaz G, Deneubourg J-L, Franks NR, Rafelsberger O, Joly J, Blanco S. 1998 419 A model for the emergence of pillars, walls and royal chambers in termite nests. Philos. Trans. 420 R. Soc. London B 353, 1561–1576. (doi:10.1098/rstb.1998.0310) 421 13. Karsai I, Pénzes Z. 1998 Nest shapes in paper wasps: can the variability of forms be deduced 422 from the same construction algorithm? Proc. R. Soc. London. Ser. B Biol. Sci. 265, 1261–1268. 423 (doi:10.1098/rspb.1998.0428) 424 14. Theraulaz G, Bonabeau E. 1995 Modelling the collective building of complex architectures in 425 social insects with lattice swarms. J. Theor. Biol. 177, 381–400. (doi:10.1006/jtbi.1995.0255) 426 15. Müller GB. 2007 Evo–devo: extending the evolutionary synthesis. Nat. Rev. Genet. 8, 943–949. 427 (doi:10.1038/nrg2219) 428 16. Rosenblum EB, Parent CE, Brandt EE. 2014 The molecular basis of phenotypic convergence. 429 Annu. Rev. Ecol. Evol. Syst. 45, 203–226. (doi:10.1146/annurev-ecolsys-120213-091851) 430 17. Carroll SB. 2008 Evo-Devo and an expanding evolutionary synthesis: a genetic theory of 431 morphological evolution. Cell 134, 25–36. (doi:10.1016/j.cell.2008.06.030) 432 18. Emerson EA. 1938 Termite nests: a study of the phylogeny of behavior. Ecol. Monogr. 8, 247– 433 284. 434 19. Su N-Y, Bardunias PM. 2005 Foraging behavior of subterranean termites (Isoptera: 435 Rhinotermitidae): food discovery and movement of termites within established galleries. In 436 Fifth International Conference on Urban Pests, Singapore, 11-13 July 2005, pp. 443–445. 437 International Conference on Urban Pests (ICUP). 438 20. Su N-Y, Stith B, Puche H, Bardunias PM. 2004 Characterization of tunneling geometry of 439 subterranean termites (lsoptera: Rhinotermitidae) by computer simulation. Sociobiology 44, 440 471–483. 441 21. Grace JK, Aihara-Sasaki M, Yates J. 2004 Differences in tunneling behavior of Coptotermes 442 vastator and Coptotermes formosanus (Isoptera: Rhinotermitidae). Sociobiology 43, 153–158. 443 22. Hapukotuwa NK, Grace JK. 2011 Preferences of Coptotermes formosanus Shiraki and 444 Coptotermes gestroi (Wasmann) (: Rhinotermitidae) among three commercial wood 445 species. Insects 2, 499–508. (doi:10.3390/insects2040499) 446 23. Krishna K. 1961 A generic revision and phylogenetic study of the family Kalotermitidae 447 (Isoptera). Bull. Am. Museum Nat. Hist. 122, 303–408. 448 24. Bourguignon T et al. 2015 The evolutionary history of termites as inferred from 66 449 mitochondrial genomes. Mol. Biol. Evol. 32, 406–421. (doi:10.1093/molbev/msu308) 450 25. Bardunias PM, Su N-Y. 2010 Queue size determines the width of tunnels in the formosan 451 subterranean termite (Isoptera: Rhinotermitidae). J. Insect Behav. 23, 189–204. 452 (doi:10.1007/s10905-010-9206-z) 453 26. Buhl J, Gautrais J, Deneubourg JL, Theraulaz G. 2004 Nest excavation in ants: Group size 454 effects on the size and structure of tunneling networks. Naturwissenschaften 91, 602–606. 455 (doi:10.1007/s00114-004-0577-x) 456 27. Toffin E, Dl Paolo D, Campo A, Detrain C, Deneubourg J-L. 2009 Shape transition during nest 457 digging in ants. Proc. Natl. Acad. Sci. U. S. A. 106, 18616–20. (doi:10.1073/pnas.0902685106) 458 28. Su N-Y, Lee S. 2009 Tunnel volume regulation and group size of subterranean termites 459 (Isoptera: Rhinotermitidae). Annu. Entomol. Soc. Am. 102, 1158–1164. 460 29. Hapukotuwa NK, Grace JK. 2012 Do tunnel patterns of Coptotermes formosanus and 461 Coptotermes gestroi (Blattodea: Rhinotermitidae) reflect different foraging strategies? bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

462 Sociobiology 59, 189–202. 463 30. Lee SH, Su N-Y. 2009 The influence of branching tunnels on subterranean termites’ foraging 464 efficiency: Considerations for simulations. Ecol. Inform. 4, 152–155. 465 (doi:10.1016/j.ecoinf.2009.04.003) 466 31. Bartumeus F, Campos D, Ryu WS, Lloret-Cabot R, Méndez V, Catalan J. 2016 Foraging success 467 under uncertainty: search tradeoffs and optimal space use. Ecol. Lett. 19, 1299–1313. 468 (doi:10.1111/ele.12660) 469 32. Lee S-H, Bardunias PM, Su N-Y. 2007 Optimal length distribution of termite tunnel branches 470 for efficient food search and resource transportation. BioSystems 471 (doi:10.1016/j.biosystems.2007.04.004) 472 33. Song H-S, Lee S-H. 2018 Simulation study to determine why only some termites are active 473 during tunneling activity. Entomol. Sci. 21, 185–192. (doi:10.1111/ens.12294) 474 34. Bourguignon T, Lo N, Šobotník J, Sillam-Dussès D, Roisin Y, Evans TA. 2016 Oceanic 475 dispersal, vicariance and human introduction shaped the modern distribution of the termites 476 Reticulitermes , Heterotermes and Coptotermes. Proc. R. Soc. B Biol. Sci. 283, 20160179. 477 (doi:10.1098/rspb.2016.0179) 478 35. Smith JL, Rust MK. 1993 Influence of temperature on tunneling, feeding rates, and oxygen 479 requirements of the western subterranean termite, Reticulitermes hesperus (Isoptera: 480 Rhinotermitidae). Sociobiology 21, 225–236. 481 36. Scheffrahn RH, Su N-Y. 2014 West Indian drywood termite , Cryptotermes brevis (Walker) 482 (Insecta: Isoptera: Kalotermitidae). Entomol. Nematol. Dep. Florida Coop. Extensive Serv. IFAS, 483 Univ. Florida Publ. , EENY079. 484 37. Aguilar J, Monaenkova D, Linevich V, Savoie W, Dutta B, Kuan H-S, Betterton MD, 485 Goodisman MAD, Goldman DI. 2018 Collective clog control: Optimizing traffic flow in 486 confined biological and robophysical excavation. Science. 361, 672–677. 487 (doi:10.1126/SCIENCE.AAN3891) 488 38. Ratnieks FLW, Anderson C. 1999 Task partitioning in insect societies. Insectes Soc. 46, 95–108. 489 (doi:10.1007/s000400050119) 490 39. Jarvis JUM, Sale JB. 1970 Burrowing and burrow patterns of East African mole-rats 491 Tachyoryctes, Heliophobius and Heterocephalus. J. Zool. 163, 451–479. (doi:10.1111/j.1469- 492 7998.1971.tb04544.x) 493 40. Weissflog A, Sternheim E, Dorow WHO, Berghoff S, Maschwitz U. 2000 How to study 494 subterranean army ants: a novel method for locating and monitoring field populations of the 495 South East Asian army ant Dorylus (Dichthadia) laevigatus Smith, 1857 (Formicidae, 496 Dorylinae) with observations on their ecology. Insectes Soc. 47, 317–324. 497 41. Light SF. 1937 Contributions to the biology and of Kalotermes (Paraneotermes) 498 simplicicornis Bank (Isoptera),. Univ. Calif. Publ. Entomol. 6, 423–464. 499 42. Lima JT, Costa-Leonardo AM. 2012 Tunnelling behaviour of the Asian subterranean termite in 500 heterogeneous soils: Presence of cues in the foraging area. Anim. Behav. 83, 1269–1278. 501 (doi:10.1016/j.anbehav.2012.02.020) 502 43. Noirot C, Pasteels JM. 1987 Ontogenetic development and evolution of the worker caste in 503 termites. Experientia 4, 851–860. 504 44. Yamanaka O, Takeuchi R. 2018 UMATracker: an intuitive image-based tracking platform. J. 505 Exp. Biol. 221, jeb182469. (doi:10.1242/jeb.182469) 506 45. R Core Team. 2017 R: A language and environment for statistical computing. 507 bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

508 Supplemental materials 509 SI text 510 Phylogenetical analysis 511 To reconstruct the evolutionary of tunneling behaviors of lower termites, we first mapped the 512 presence of tunneling behaviors on the phylogeny. We assumed that the multiple-piece or central 513 nesting termites do tunneling through the soil, while single-piece nesting termites do not [46]. 514 Although there are some observations of tunneling behaviors even in single-piece nesting termites 515 [47–51], the frequency of tunneling behaviors in the former is much higher than the later. Moreover, 516 note that the wood roach Cryptocercus, which is the sister group of all modern termite species, does 517 not dig in soil but excavates inside the gallery of well-rotten logs [52]. Because of the lacking of the 518 study, the information of Rhinotermitidae is limited. In this study, we follow the analysis in Inward et 519 al. 2007, to assume that only Prorhinotermes is classified as a single-piece nester [53]. Recent 520 molecular phylogeny can be found in Bourguignon et al. 2015 [24], but this does not include genus 521 Paraneotermes. Thus, based on the morphological phylogeny [23], we inserted Paraneotermes and 522 Kalotermes into the molecular phylogeny. We conducted the ancestral state estimates using a 523 maximum parsimony approach implemented in Mesquite v3.6 [54]. 524

525 526 Figure S1. Detailed phylogeny of lower termites (modified from [23,24]) with the information of tunneling 527 behaviors. The ancestral states were reconstructed with maximum parsimony. Tunneling through the soil has 528 evolved four times independently in Mastotermitidae, Hodotermitidae, Paraneotermes, and Rhinotermitidae. 529 bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

530 531 Figure S2 Model of termite collective excavation. The flowchart begins at the top with “State?”. Once it reaches

532 the endpoint, it goes back to the beginning, “State?”. Parameter values of pe, pw, and pt are in Table S1. 533

534 535 Figure S3. Comparison of the tunnel length for the simulations when divided into segments. Initial tunnel is 536 the segment from the entrance to the first branch. Secondary tunnel is a segment between two branches. Edge 537 tunnel is a segment reaching the face of a tunnel. When a tunnel has no branch, it only contains an edge tunnel. 538 See also Fig. 2D. 539 540 bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

541

542 Figure S4. Comparison of the maximum instantaneous moving speed and total moved distance during 543 observations. Maximum moving speed was measured for each individual separately, while moved distance was 544 measured for each group by summing up the distance traveled by all members in a group inside a tunnel. As the 545 observation was performed for 10 minutes every hour, we multiplied the distance by 6 to get the estimated 546 values. Bars indicate mean ± S.E. 547

548

549 Figure S5. Comparison of body shape and turning behavior inside a tunnel. Body length and head width were 550 measured for 30 inidviduals for each species. Bars indicate mean ± S.E. Turning behaviors were measured for 551 1st row individuals which leave the tunnel face after excavate and when the length of the tunnel is longer than 552 40mm. Logistic regression curve is shown for each species, where we tested the relationship between transport 553 distance and presence of turning behavior. 554 bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

555 556 Figure S6. The time until a group of termites reach a wall (two dimensional) or create a 50mm tunnel (one- 557 dimensional arena). Arrows indicate the timing we analyzed the tunneling patterns (two dimensional). 558

559 560 Figure S7. Experimental setup for behavioral observations with representative x-axis trajectories of each termite. 561 Different individuals are shown in a distinct color. Black arrows show examples of backward movements, where 562 R. tibialis and H. aureus often go back to the entrance of the tunnel, while P. simplicicornis mainly move only 563 within a limited range within the tunnel. 564 565 Table S1. Parameter values used in our simulations.

Species pe pw pt α β Paraneotermes simplicicornis 0.06 0.30 0.59 0.541 0.765 Reticulitermes tibialis 0.26 0.56 NA 0.567 0.155 Heterotermes aureus 0.5 0.4 NA 0.483 0.232 566 567 Reference for SI 568 46. Abe T. 1987 Evolution of life types in termites. In Evolution and Coadaptation in Biotic 569 Communities (eds S Kawano, J Connell, T Hidaka), pp. 125–148. University of Tokyo Press. 570 47. Castle GB. 1934 The dampwood termites of western United States, genus Zootermopsis 571 (formerly Termopsis). In Termite and termite control (ed CA Kofoid), pp. 273–310. Berkeley: 572 University of California Press. 573 48. Morgan FD. 1959 The ecology and external morphology of Stolotermes ruficeps Brauer 574 (Isoptera: Hodotermitidae). Trans. R. Soc. New Zeal. 86, 155–195. 575 49. Nkunika POY. 1988 The Bology and ecology of the dampwood termite, Porotermes adamson 576 (Froggati) (Isoptera: Termopsidae) in South Australia. University of Adelaide. See 577 http://content.ajarchive.org/cgi-bin/showfile.exe?CISOROOT=/0012-8789&CISOPTR=2337. 578 50. Waterhouse DF, Norris KR. 1993 Biological Control: Pacific prospects - Supplement 2. 579 Canberra: Australian Centre for International Agricultural Research. 580 (doi:10.1007/SpringerReference_92204) bioRxiv preprint doi: https://doi.org/10.1101/836346; this version posted November 9, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

581 51. Scheffrahn RH, Su N-Y, Cabrera B, Kern W. 2003 Cuban Subterranean Termite (proposed), 582 Florida Dampwood Termite (old unofficial name), Prorhinotermes simplex (Hagen) (Insecta: 583 Isoptera: Rhinotermitidae). Univ. Florida EENY 282, 1–3. 584 52. Bell WJ, Roth LM, Nalepa CA. 2007 Cockroaches Ecology, Behavior and Natural History. 585 53. Inward DJG, Vogler AP, Eggleton P. 2007 A comprehensive phylogenetic analysis of termites 586 (Isoptera) illuminates key aspects of their evolutionary biology. Mol. Phylogenet. Evol. 44, 953– 587 967. (doi:10.1016/j.ympev.2007.05.014) 588 54. Maddison WP, Maddison DR. 2018 Mesquite: a modular system for evolutionary analysis. 589 Version 3.6. 590