rao ‘Ant groups optimally amply the effec of rasierly informed inviduals: Nature Communications : Nature Publishing Group [NATURE COMMUNICATIONS | ARTICLE OPEN Ant groups optimally amplify the effect of transiently informed individuals Aviram Gelblum, tai Pinkoviezky, Ehud Fonio, Abhijit Ghosh, Nir Gov & Ofer Fel [Nature Communications 6, Article number: 7728 doi: 10.1038incomms®729 Received 21 April2015 Accepted 04 June 2015 Published 28 July 2015 Abstract, ‘To cooperatively transport a large load, It is Important that carvers conform in thelr efforts and align ther foroes. A downside of behavioural conformism is that t may decrease the group's responsiveness to exlernal information. Combining experiment and theory, we show how ants ‘optimize collective transport. On the single-ant scale, optimization stems from decision rues that balance indivaualty and compliance Macroscopicaly, these rules poise the system atthe transition between random walk and balistic motion where the calecve response tothe [Print steering ofa single informed ant ls maximized. We relate ths peak in response to the divergence of susceptibility ata phase transition. Our theoretical models predict thal the ant-load system can be transitioned through the criical point of this mesoscopic system by varying its size; we present experiments supporting these predictions. Our findings show that efficient group-level processes can arse from transient amplifcaion of Indhiduatbased knowledge. ‘Subject terms: Biological sciences Zoology Applied physics Introduction Collectively carying a large load requires a high degree of coordination and is rare outside humans and ants. One facet ofthis coordination isthe requirement to align forces such tha inefficient tug-of wars are avoided! Indeed, its known that group-lving animals increase thelr level of coordination as conformist group members algn thei actions wth thase of their neighbours?-+.5, A downside of behavioural cnformism i, however, that may increase the stabilly of maladaptive oehaviours and decrease the graup's responsiveness to external cues®-7:®, This, problem sharpens if information is Scarce or hel by a small number of individuals? "9,*7, 12,9. 4.15. The advantages and drawbacks of ‘behavioural conformism are accentuated in the case of cooperative load retrieval by ant teams® 18.17.18 the large degree of conformity, requires toensure efficient carying, may hinder the continuous responsiveness needed for navigation through a ragged terrain CCtical behaviour has been suggested as an organizing principle"? in several cllecive ological systems?°,21, The tansiion between order and clsorder is believed follow the group to exhibit behaviours tat are stable while maintaining flexibility. Cooperative transport by ants provides a nique opportunity to approach these ideas. A main reason fr this Is that, In ths case, the collective goal ofthe group is simple and wel understood: bringing the food to the nest, Ths allws for @ quantitative assessment of the relative contributions of efferent levels of persistence and responsiveness tothe colectve goal Here we combine experiment and theory to study the interplay between conformity and flextilty in groups of Paratrechinalongicomis ants as they cary a large food item towards thelr nest" P.longicomiss an invasive species wih worldwide cstribution that occupies a diversity of habitats from urban to rainforest22. These ants have @ ‘seasonal preference for proteins, and insects compose a main part of thelr det during the summer months? Sinoe these insects are often much larger than the size of worker ans (2.5-3mm), impressive dsplays of P.langicornis cooperative transport can be observed during this time? Here we describe field experiments of cooperative transpart by P. longcormis ants. By combining the analysis ofthe load motion with single-ant trajectory data, we find that while the combined force of the group determines the speed ofthe loa, itis insivcualinformed ants that ster the dection of movement, To understand how an indlvidual may influence the entre group, we present a microscoplealy realistic mode! that is formulated inthe language of single-ant decision rules and stands in quantitative agreement with the macroscopic characteristics ofthe load's ‘motion. The madel suggests that carying ans exhibit an intermediate level af social conformism which, onthe collective scale, alows a single individual to optimaly steer the oad, By mapping our model to a more abstract Ising model, we show how this optimalty is related to the divergence of susceptibilty ata crtical point. The mear-feld nature of his model implies that the gystem can be trankioned through ths crtical Point by varying load size, We provide experimental verification for th’ prediction, Results Experimental design and trajectory characteristics Ring-shaped food items were placed several metres away from a P.longicorns nest in he flld. Following discovery ofthe food and a recrutment phase, the ants commenced cooperative transport (Fig, 1a; Methods; Supplementary Note 1), We repeatedly flmed this behaviour over a distance of ~1m (ca, §00 ant lengths) and extracted the positions (Fig. tb) and angular orientations ofthe load from a total trajectory length of, ‘over 70m. We also measured the positions of all ants around the object as well asthe angles between the carrying nts’ body axis and the object radius (Fig, 1c; Methods; Supplementary Movie 1), We further lagged all events of ant attachment to and detachment from the load, Figure 1: Cooperative transport ‘tpt netre.com/ncomms/2018/150728ncommsB725"Ulincamms8729.nl anerao Ant groups optimally amply the effec of rasierly informed inviduals: Nature Commuricatons : Nature Publishing Group (a) Ateam of. fongicomis ants retieving fod tem ~360 times the mass ofa single art. (b) Sample nest-bound load Lajectres of ~1-m length each, ‘Scale bar, 10cm. (6) Snapshot ftom a tracked movie. Carrs are runerated in yelow and non-carers in wile. The recent trajectories offer ans are ‘marked by ciferent colours. The object's drection of movement is depicted by the blue arow. Scale ba, 2em. (f) The nual variably inthe number of ‘carers and thir distbuionarourd the load were used to extract mecian load speed as a futon of ether total ant number (orange line or eiference in ant ‘occupancy between the leading andthe trang edge (pik in). Te geen line depicts the efclency ofthe carving in terms of he faction of energy devoted toroaton, Ener bare are the $6, of @ dstbuton of medians calculated or 7,000 sample boottrappe tem the ta, NW6S7.B65 Frames Focusing on global features of the loa trajectory, we find that ile median speed increases linearly wth rumber of carrying ants (upto 18 attached ants), does not strongly depend on how these ants are distributed around the object (Fig. 16, orange and pink correspondingly: see ‘Supplementary Note 2). Furthermore, the relative amount of energy wasted on rotatng the object rather than translating it decreases wth number of carrying ants (Fig. 16, green). These properties indicate a high evel of conformism that enables all attached anis to efficiently align their efots. ‘Ths result stands in agreement with cooperative carrying in other ant species". n contrast, whe te trajectories generally head inthe corect rnest-bound direction they, nevertheless, exhibit substantial sinvosity, looping and detours (Fig. 1). Load storing Understanding load steering I facitated by the simple mechanical nature ofthe system, First, whatever the sources of information may be? the movement of the rigid load is determined by the sum of forces and torques applied by the carrying ants. Overall forces dictate linear load velocity and angular speed, both of which are experimentally measureable. Second, the forces that ants can exert on the oad are constrained by {heir anatomy: ants can either lit objects or apply pling forces that are, in general, aligned wit their body axis, while pushing is rare to absent? (Supplementary Note 3; Supplementary Fig. 1). Finaly the pling forces applied by a single ant are ofthe order of 0.1 mN (Supplementary Note 4) ‘There are many ways in which a glven net force may be dletributed between the individual ants. n one extreme ease al ans pull ith equal forces and load directionality s set by the angular distribution of attached ants along its edges, However, we find minimal correlation between ant angular distibution and direction of oad movement (Pearson Correlation: 0.104, test: P«10~"S, Ne6,591 frames from 47 experiments; see ‘Supplementary Note 5 and Supplementary Fig. 2a). This tug-of-war assumption is further incompatible withthe linear, rather than square rot, rise in the median load velocity withthe number of attached ants (Fig. 10) {tthe other end ofthe spectrum isthe case where, although all ants can apply a force, they are actualy enslaved by a single-informed ant that {guides the entire motion. Indeed, in most of our experiments, we could identify at east one ant that acted as a carter forthe fll duration ofthe retrieval, However, the distribution of correlations between the orientations ofthese persistenty attached ants wth the direction of motion ofthe cargo cannot be distinguished from that of non-persistent ants (est: P=0.8691, Npersistent14 ANS, Noon persistent 188 ants from ve ‘experiments; see Supplementary Note 5 and Supplementary Fig. 2). \We therefore conclude that the collective mavement does not arise neither from a wisdom-othe-crowels type averaging over all opinions nor by the continuous leadership of any single individual. Next, we aim at examining how the fotal force applied to the load is distributed among the carrying ants ‘Transient influence of newly attached ants, Freely moving ants are wel informed ofthe correct nest-bound direction (Supplementary Note 6; Supplementary Fig. 3a). When such ants attach to the oad they steer it so that it moves more accurately owards the nest. Quantilatively, we found that within the several seconds that fol their attachment, ans inject about 05 bits of cirectonal information into the system (see Fig. 2, Supplementary Note 6 and Supplementary Fig. 3b) ‘Ths causal effect implies that nevly attached ants adopt an influential role. Figure 2: Transient guidance. ‘tpt netre.com/ncomms/2018/150728ncommsB725"Ulincamms8729.nl anerao Ant groups optimally amply the effec of rasierly informed inviduals: Nature Commuricatons : Nature Publishing Group ae b ‘Atacnents Detachment Detacimor back 105r_9% 75¢ E10 120, 5 Boel 150" 20" wa a 8 a0" este beseae Flav impact direction 2 9g O3[ essing edge ‘og el Initial storing os 8B ho o aE Bor 3 Time (8), d Spee & onl § | 3 oz} | o 6 10 18 mea {@) The Information nthe angular spread of he loads rection of maton immediatly flowing the attachment ofa new ant at t20 (N= 184 attaches) Enors wore calculated fom the ertopy of artifaly generated histograms with added bincal nese.) A hal-plar histogram ofthe ares between the ‘llachmentidetachment pin of an ant ar the change in velocy lative impact erection) tha follows eterant events (N=252)(c) Relative impact rection asa function of time (hue line N=134 alachmonts) ae ference between distibutons oftime since attachment of ants in the leading and traling ‘0496 of the load (uruoise tine. The insets strate tis process fora sample newly attached ant (marked by a yellow cre). Load velocity (green arrow) and the acceleration caused by this art (tashed arow) are overlaid (@) An example ofa series of switches between steering ats along a trajectory. Overlad ‘colours mark rajectery segments where diferent aris steered the lad, Scale bar, 10cm, e) Mean magetude of velocity change caused by newly attached nls (donated by |dy| ony axis) 2 a function of number of ants already attached (N= 194 attachments). Error bars ar standard eer ofthe mean, ‘The peakin directional accuracy, evident in Fig. 2a, implies that carying ants are ess informed than neviy tached ants. The quick deterioration ‘of accuracy ater its intial rise shows that while newly attached ants have a transient positive influence, their directional nowledge rapidly decreases folowing the time of attachment (Fig. 2a). A tkely explanation to this effec is obstruction of ans’ antennae bythe lage load (these ants do not strongly rely on vision for ther navigation; see Supplementary Note 7 and Supplementary Fig. 4). The timescale ofa newly atlached ants intuence fs much shorter than the entre duration of he transport, which is qualitatively cifferent from more stable forms of influence observed in systems with distinct leadershio®-28:28, Finally, note thatthe high variation of directional accuracy in time (Fig. 28) cannot be explained by a ‘model in which carrying ans continuously reorient the load simiar to the way a single ant follows a pheromone tai, Microscopically, we estimate the eflect ofan individual carter by the change of load speed immediately fallowing her attachment or detachment. In ‘general, we find that ants at the leading edge of the object (Fig. 1c) tend to pull (evident by te recol ofthe load on detachment, Fig. 2b, red), hile those atthe back assist the motion by iting (see the non-specific direction change on detachment; Fig. 2b, green). Considering forces together with the ants’ angular locations on the load and their carying durations allows us to ilustrate the typical me course of a cari: on attachment, a new eartier tends to change the load's velocity towards her own puting direction (Fig 2, Blue). This influence lasts for several ig. 2a). The intial steering orks to bring the cartier to the leading edge ofthe load (Fig. 2c, turquoise line), where she remains for about 20s and continues puling. At seconds (blue line, Fig. 20) In agceement with the timescale ofthe ital improvement in load drectional accuracy longer times that carrier's angular postion becomes indstinguishable from that of any other ant (Fig, 2c, turquoise fine) and the forces she applies, doctease (Supplementary Note 8; Supplementary Fig. 5). In particular, the ant often ends up atthe taling edge of the oad from where she ‘cannot guide the motion, This provdes strong evidence forthe fac that all useful steering is provided by newly attached ants, Mutiptying the steering timescale of 5-208 (Fig, 22,¢) by he mean attachment rate (Supplementary Note ©), we estimate the number of mean concurrent steering ants at 0.35~1.4. Figure 2a iustrates changes between these steering ants along one trajectory and demonstrates how a high turnover can bridge the scale gap between that of seconds, relevant to the elfect of a single ant, and that of minutes, which characterizes the uration of the collective transport provess (see Supplementary Note 10). The ans are able to compensate forthe scary of information imposed by the small number of concurrent steering ants by maintaining high responsiveness to these ants, evan as the number of carters increases (Fig, 22) ‘tpt netre.com/ncomms/2018/150728ncommsB725"Ulincamms8729.nl anerao ‘Ant groups optimally amply the effec of rasierly informed inviduals: Nature Communications : Nature Publishing Group ‘Atiough our results demonstrate that, overtime, a large numberof ants steer the collective motion, we do not claim that all ants are equalin their susceptbilty lo guide the motion ofthe degree of intience that they exert. Figure 2¢ shows tha flowing attachment, some ants guide the motion for a longer duration than others. tis also possible that by detaching from the oad and then resttachingto it an ant may extend her influence even ‘further (examples for ths ae evident in Fig. 20). thas been previously shown thatthe first Formica schaufussi ants to locate food are crucial for ‘he transport process@®, Diferently, cooperative transport in P,longicomis relies on & larger number of transient steering ants and in ‘Supplementary Note 11, we flow the recrutment process inthis species and discus ils relevance tothe subsequent transport (also see ‘Supplementary Fig. 6). Finally, the carying ants’ susceptibity to influence (Fig. 28,0) seems to contrast the conformism’ that allows the ants to coordinate their forces and achieve high speeds (Fig. 14). Inthe next section, we presenta theoretical model that descibes the cooperative transport and use it to reconcle this seeming contradiction, ‘Theoretical mod ‘On the basis ofthe experimental properties outtned above, we constructed a theoretical model as specified in Fig. 3a, Supplementary Notes 12— 14, Supplementary Fig. 7 and the Methods section. In short the models based on the minimal assumption that cariers interact uniquely through local forces transmitted to them by the load’. Informed ants are assumed to ignore these forces and attempt to pul he load inthe corect nest bound direction. Our experimental data suggests thatthe information held by ants deteriorates afer they had been attached for a certain period (Fig, 22,0), For simplety, the model assumes that these ants become completely uninformed. Figure 3: Microscopie modal. a © 10 Sonlt Fos # os Bos gee 00 oa 4 6 6 0 Distance along wajctry (em) de, ee ic bo Bt b Bo es obese wD i 5 Numer of ans 5. e = © 025 2 Par J BF £ je Sor g i, Bor Z A & 00s wl . 04-02 0002 cea eam ‘Speed component (om =) Number of ants (@) Model sketch incusing the possible transitions for nominfome invidunt (bo) The four model parameters were set by fiting expetmenta data of ». The dstrbutlon ofthe odee's velety (jected on an ariray dretion) n pried of comtruous motion (N=55,090 frames). (e) Corlaton stance ‘uretion (N=17 tector) (6) Median speed (N=56,090 frames). (@) Mesan angular speed (N=56 020 frames). In each ofthese panels, the coloured nes represet the experimental data for ants transporting a lea in the absence of informed ants. The solid lack ines denote the results of cur model. Eor bars ine are the maxima and minima of eorelain functions produced by panning the data ito fou pars. Ear bars ind ane are the 8. ofa astnbution of ‘median called for 1,000 samples boostrapped from the dala, Non-informed ants either try to align thee pull with the force vector on the load (but align againt the local torque) or simply if the object to decrease fiction, An uninformed ant’s decison of which role to perform is random and is biased by her alignment wih the direction ofthe cen. ‘of-mass foree vector such that ans in the leading edge tend to pull (and ants in the rear tend to if). In this way, the roles of the ants depend on ‘heir location around the load with respect tothe direction of motion, An individuality parameter, Fin, determines the tendency of non-informed ‘ants fo randomly sich between puling and Htng. low indvidualty value indicates that an ant’ decision is governed by her algrment wth the Group while a high vake indicates random roke switching, irrespective ofthe pling force ofthe group. We denote the basal rate of this switching by Ke In adton, ants ean altach of detach trom the load. Finally, the objects speed and angular veloc were taken to be linearly correlated vith the total force and torque as applied by all ants (Supplementary Note 12). The mathematical form of our model is motivated by models of ‘tpt netre.com/ncomms/2018/150728ncommsB725"Ulincamms8729.nl anrao ‘Ant groups optimally amply the effec of rasierly informed inviduals: Nature Communications : Nature Publishing Group collective transport on the subcellular? and single-cell? levels. While the population of ants may have a dstrioution of internal properties, here we consider, fo simptcily, a population of identical ants (fr the behaviour of populations with non-uniform Fin: See Supplementary Note 18 and ‘Supplementary Figs 8-2); in particular, the forces applied by informed and uninformed ants are taken tobe equal \We begin by comparing the model wth the more elementary experimental condition from which informed ants are absent, This condition was ‘achieved by carefully transferring the cargo along vith he attached ants about 75cm away from is inal location to a nearby clean board. Note that the relocation itself does not signiicanty affect the transport behaviour (Supplementary Movie 2). Under these circumstances, the ants-cargo system exhibits a persistent random walk (Fig. 3b.c; Supplementary Note 16; Supplementary Fig. 10). The model has four free parameters (see Methods): two (Fin, Ke) ar@ associated withthe arts decision-making process, wile the other two are relate tothe mechanics ofthe specific ant-oad system. We aojusted these parameters (Methods; Supplementary Notes 19,17 and 18; Supplementary Figs 7 and 11) to ft a large: ‘number of experimentally measured features (Fig. b-2). The good agreement demonstrates that the observed behaviours can be reconciled with ‘@ madel thats based solely on mechanioal information transfer; that's, the new attached ant exerts its influence by causing a transient sturbance in the foal force vector to which the rest ofthe ants react, allthis wthout an acve signaling mechanism, Particular, any form of Influence must be implct and is similar, in his sense, tothe implct influence that characterizes an effective leadership". The sensitviy of the Calculation fo the chosen values of the free parameters is shown in Supplementary Note 19 and Supplementary Fig. 12 Balancing individuality and conformism In the simulation, we define the response ofthe system to an informed, newly attached ant (Methods; Supplementary Notes 16 and 20) as the distance thatthe load travels towards the nest in the characteristic time between two consecuthve attachments, Following Fig. 2a.c, we assume ‘hat an informed ant aisoriens, turing into an unformed anton a timescale of 10s, This simple model assumes a discrete (but stochastic) process of forgetting. In Supplomentary Note 15, we show that a more gradual forgelting process gives essentialy the same behaviour, even when combined witha vakie of Fjng that is not constant overall ants (Supplementary Figs 9 and 13) We fix three ofthe four free madel parameters and check for possible optimality in terms of system response as a function of Fng. We find that both complete conformity (small values of Fag) and strong individualism (large values of Fing) reduce the effectiveness of a nevdy atlached ant in steering the load (Fi. 4a). The filed value of Fing les between these two extremes and this suggests thatthe ants operate inthe transition region between strong and weak conformity, possibly to optimize thelr responsiveness to aliited influx of information (Fig. 42, uoper left inset). In addition, we fin that the working regime ofthe ans is such thatthe velociy distribution ofthe load les inthe transition region between unimodal (Qug of war or random wake) and bimodal (ersistent motion) behaviours (insets of Fig. 4a). awe scrap opti a Pig € Bos : Ab os : Iz Fane * G19 i Bas i be i io. | fas io "Gia 00 0) eo Numberct ans e fo 9 i 1s ms : ou i a0 3 wx) i eBEIr i "oe ae HO Yom fon dom Son oe (ab) Simulation data of th sponse ofthe abjoc oa sing attachment ofa single-knowledgoable ant asa function of ho inidualty parameter Fy (8) ‘tpt natre.comincomms/2018/150728ncommsB725"UlincammsB729.rinl srerao ‘Ant groups optimally amply the effec of rasierly informed inviduals: Nature Communications : Nature Publishing Group ofthe object's radi (0), Insts dict vloity component cstfbuton fr smal (range) and large (le) values of Fug a8 wel a its fied value (ink) Upper et inset depicts trajectories (al stating atthe yew dt, colour coded a before that ake Into account he coil aval of informed arts (scale bars represent 10m). The pink dot in b marks the radius of he experimental load. (€,) Exact Soin of the sing spin mae.) Normalized (Gmersioniss) systam response as afurcton ofthe ndvidually parameter Fipg, The blue curve marks the shore response toa newly attached ant an the re curve the mead suscepti, which dverges a he crea pir. ( Normatzed (dimensionless) shot-erm response to newly attached ant 36 function ofthe mean numberof as attached to the loa (a proxy for load size). Dated nes mark the erica tanstion pons. (e-g) Experimental vetcation.(e) Mean absolute curvature of tajctovies of objects of erent size (otal N=00). Syihetc and nar-syrthetc materia that were usd forthe ‘smalltom exhibited similar cuvatus (resans: syrthte: 8.84, ror-syiholc: 6.53, unpaired two-sample Hest. P=0,5285, N=d) and were therefor pool together. Thus, the effect presented inthe gue isa size effec that cannot be atnbuted to lose sttstance composition, Inet: mean cuvatre of simulates tracks of objects of ferent sizes (calculated on clesn board condtions). () Time spent a v>75% ransport speed fr one (aren) versus 2-4 (be) arts ‘carying @ smal load (atl N=20),(g) Top time to negotiate an obstacle (est; P<0,01) and (bttom) the maximal backwards ceplacement (est 0,000 towards a successful crossing ofa U-shaped bock (wich equred 5-om backtracking} fr two load sizes (taal N=) Scale bar, 10cm, ‘Our model suagests a correspondence between load size and the ants individuality (Supplementary Note 21). Namely, large loads correspond to ‘more conformist ants and vice versa. This behaviour naturally arises from the mean-field nature of this system (see below). We fixed Fng tos ‘ited value and simulated ring-shaped objects of different radi (such thatthe mass per ant remains constant o calculate the normalized response function to the attachment of a new ant. We find an optimal load size regime which is on the order of tom (Fig. 4b). This scale is compatible with natural prey and nest entrance dimensions. Finito-size criticality Figure 4a le reminiscent ofthe divergence of susceptibility near a second-order phase transition, and thereby suggests an analogy between Fing ‘and temperature, and between the response to @ nevly attached ant and suscepiibilly, To ascertain ths analogy, we constructed a simple Ising ‘model in which the group moves along one dimension (see Methods, Supplementary Notes 22-24 and Supplementary Fig, 14). This models anaicaly tractable and its equilbrium state describes the centre-of mass motion ofthe load, Note that affhough the model descrines a one- dimensional (1D) motion the spin connecinvty pattern has no spatial dimensionality (ll spins interact with each other over a complete graphy “The spins inthis sing model denote the ants roles: 1 for puller and ~1 for iter, while the extemal felt is analogous tothe force applied by an informed ant. Since a cater ants are attached to the same cargo, each of them senses the total force exerted by alothers an this makes the spin model ierenly mean feld. The mean-feld solution (Supplementary Equation (4), which fs an approximation fra finte-size system) reveals thal, similar othe extended model described above, the response of the spn system to a transient pul by an informed ant (Equations (2) and (3) peaks at FSing (FS pq*4.3 compared with the peak at Fing=4.25 in the extended mode) fr fxed N (Fig, 4), and at Ne for fixed Fing (Fig.40). The eiical value F%pq indicates the transition ofthis Ising model between ordored (Fiqa “The mean-field thermodynamic susceptibity fs, Fora mayo} Find > Fit @) 1 Fiat