Introduction

Predictable shifts in the ecological pressures that organisms face as they grow often shape the evolution of adaptive ontogenetic change in phenotype (Stearns 1992). In derived social insects, the colony is the adaptive organismal unit (Strassmann and Queller 2010), yet identifying how the ontogeny of this complex colony phenotype is shaped by ecology remains one of the major challenges of sociobiology (Oster and Wilson 1978; Wilson 1985; Tschinkel 1991; Tschinkel 2011). For such colonies, selection can act on a suite of additional social traits (Powell et al. 2020), adding more dimensions to colony ontogeny compared to ontogenetic change in non-social taxa. Here, we define colony ontogeny as the developmental trajectory of traits across the full range of colony growth seen in a species, from colony founding through to maximum reproductive size (derived from Wilson 1985; Tschinkel 1993). The study of colony ontogeny, also known as sociogenesis, is therefore concerned with understanding the evolution of adaptive change in traits explicitly associated with growth, contrasting more dynamic changes in phenotype or resource allocation within any particular phase of growth. Colony ontogeny, as a process of growth, is not explicitly defined by age, even though some trait changes will require more time to be attained. Indeed, previous work has shown that a colony’s position on its ontogenetic trajectory is best defined by an explicit growth trait like size, not colony age, and that reductions in size will reset or regress the ontogenetic state of other traits (Wilson 1983; Wilson 1985; Tschinkel 1993; Tschinkel 2011). Moreover, colony size and other traits undergoing ontogenetic change can be arrested or reset in nature by a range of abiotic and biotic pressures, such as damage from extreme weather or enemy attacks (e.g., brood and adult mortality without colony death associated with predatory grazing by army ants, e.g., LaPolla et al. 2002; Powell and Clark 2004; Powell 2011; Dejean et al. 2013). The decoupling of trait ontogeny from age is, therefore, likely to be widespread in nature. Nevertheless, across all contexts and challenges, the key question that remains is to what extent are the ontogenetic trajectories of colony traits, and the interactions between them, adapted to the ecological conditions a species faces?

Colony size and caste composition represent two major axes of the colony phenotype that typically undergo substantial ontogenetic change. Colony size is defined simply as the number of individuals in the colony, but it can span six orders of magnitude or more within the lifetime of a colony (Dornhaus et al. 2011). Consequently, it represents a common measure of overall growth in colony ontogeny studies (reviewed in Wilson 1985; Tschinkel 2011). Beyond serving as a measure of overall growth, ontogenetic change in colony size also alters the collective organization and actions of the colony (Dornhaus et al. 2011), and so interacts with many other ontogenetic changes in colony phenotype. Caste composition is one such trait, and it is defined here as the distribution of physical sizes and shapes of colony members. Composition can vary across species from moderate size differences to a number of distinct physical “castes” (Wills et al. 2018), and the presence of particular sizes and castes can change substantial with colony ontogeny (reviewed in Wilson 1985; Tschinkel 2011). Such ontogenetic shifts in caste composition fundamentally alter the range of functional capabilities present in the group (reviewed in Wills et al. 2018; Powell et al. 2020), and different sizes or castes can be employed individually and collectively, further increasing the colony’s full range of functional capabilities (reviewed in Anderson and Franks 2001; Dornhaus and Powell 2010; Wills et al. 2018).

Classic theory has proposed ways in which the ontogeny of colony size and caste composition may change adaptively under particular ecological contexts (Oster and Wilson 1978), but robust datasets to test these ideas are scarce (Tschinkel 2011). Indeed, most quantitative datasets used to both inform the theory (see Oster and Wilson 1978) and subsequently test it are from terrestrial soil-nesting ants (e.g., Tschinkel 1993; Tschinkel 1999; Kwapich and Tschinkel 2013). Consequently, we lack the data to address how colonies grow adaptively under many ecological contexts that differ substantially from those experienced by terrestrial soil-nesting colonies. Examining the contrasts between terrestrial and arboreal realms may be especially valuable for addressing this knowledge gap.

Perhaps the most fundamental ecological differences between terrestrial and arboreal realms are those associated with nesting conditions, and they are likely to exert contrasting selection pressures on the ontogeny of colony size and caste composition. Divergence in nesting ecology, and associated colony ontogeny, may be especially pronounced for ants. Whereas terrestrial ants can easily excavate nests of a variety of sizes and architectures in the soil as the colony grows (Mikheyev and Tschinkel 2004; Minter et al. 2012; Forti et al. 2017), the majority of arboreal ant species nest in pre-existing cavities in tree wood or under bark (e.g., Carroll 1979; Tanaka et al. 2009; Powell et al. 2011; Satoh et al. 2016; Novais et al. 2017). These pre-existing arboreal cavities are often the direct or indirect products of wood-boring beetle larvae, and the size and properties of the cavities can be relatively fixed by the original beetle occupant and the wood they are found in (Powell 2008; Dornhaus and Powell 2010; Calderón-Cortés et al. 2011; Novais et al. 2017). Without the capacity to extensively modify any individual nest, colonies grow by expanding into additional cavities, creating a modular colony structure of repeated nest units (Debout et al. 2007; Powell 2009; Powell and Dornhaus 2013). Moreover, arboreal ants rapidly and expansively colonize newly available cavities, while individual species often show strong preferences for specific cavity properties (e.g., Philpott and Foster 2005; Powell 2009; Powell et al. 2011; Powell and Dornhaus 2013; Jiménez-Soto and Philpott 2015; Powell et al. 2017; Powell et al. 2020; Priest et al. 2021). This pattern of cavity usage suggests that the number of occupied cavities is the ultimate limitation on growth, even though a colony may prefer expansion into larger volume cavities when there is a choice (e.g., Powell and Dornhaus 2013). The common demand for depletable cavity resources also generates widespread resource limitation, which is further compounded by cavity specialization, and a nesting environment with intense cavity usurpation pressure (Philpott and Foster 2005; Powell 2009; Powell et al. 2011; Jiménez-Soto and Philpott 2015; Priest et al. 2021). All considered, growth in arboreal ants across multiple, spatially constrained nesting units appears to generate a set of intense ecological pressures that are substantially divergent from those experienced by soil-nesting species. How might such divergent ecological contexts shape different adaptive ontogenetic trajectories for colony size and caste composition?

In the arboreal realm, the apparent coupling of changes in colony size to the acquisition of additional nest units may regulate the trajectory of identifiable growth phases. Existing theory predicts that the growth of perennial colonies should incorporate three distinct phases that are likely to influence the ontogeny of other colony traits (Oster and Wilson 1978): a concealed founding stage where the queen initially performs all social tasks, a more conspicuous and aggressive growth phase where only non-reproductive individual are added, and a reproductive phase producing sexual offspring that can mate and found their own new colonies. Once in the reproductive phase, which may span many years, the colony is then expected to cycle between bouts of non-reproductive growth and reproductive output each year (Oster and Wilson 1978; Tschinkel 1993; Smith and Tschinkel 2006; Murdock and Tschinkel 2015). During this cycling, the resource commitment to the reproductive bout, the associated stop to worker production, and season-specific worker mortality rates may also result in seasonal declines in overall colony size and biomass (Tschinkel 1988; Tschinkel 1993). Moreover, unexpectedly high worker mortality from chance abiotic or biotic pressures may reset colony ontogeny to an earlier phase. These expected phases in growth are further linked to the easy expansion of nesting space via soil excavation, with the physical, extended phenotype of the nest itself changing with colony size (e.g., Tschinkel 1987; Tschinkel 2004). Nevertheless, when colony growth is dependent on the acquisition of additional pre-existing nesting structures, as it is for arboreal cavity-nesting ants, the ontogenetic trajectory of colony size and growth phases may be more modular and constrained. While we know that larger arboreal ant colonies tend to occupy more cavities in some species (e.g., Powell 2009; Powell et al. 2017), quantitative understanding of how colony size and growth phase scale with the number and properties of arboreal nests is scarce (but see Hasegawa 1993; Powell 2009).

Interacting with ontogenetic changes in colony size, the nesting ecology of the arboreal realm may also favor more stable ontogenetic patterns of caste composition. Theory further predicts that both average worker size and proportional representation of larger castes will increase continuously with overall size and stabilize when the colony reaches reproductive maturity. The tension between limited resources and the ecological need to grow the workforce quickly is thought to have selected for this ontogenetic pattern of resource allocation that prioritizes worker number over size at smaller colony sizes (Oster and Wilson 1978; Porter and Tschinkel 1986; Wills et al. 2018). Similarly, delaying the production of functionally specialized castes may further prioritize worker number over compositional completeness (e.g., Oster and Wilson 1978; Wilson 1983; Tschinkel 1988). More specialized castes may also be more functionally valuable at larger colony sizes. For example, a distinct soldier caste may have little value until the colony is big enough to attract the most dangerous predators or competitors (Oster and Wilson 1978; Tschinkel 1988). However, when growing colonies experience frequent enemy attacks throughout life, selection may instead favor early and stable production of specialized castes (Oster and Wilson 1978). This hypothesis has not been tested directly, but the general scenario fits the specific ecological context of nesting in the arboreal realm. For many arboreal ant species, the most dangerous enemies are nest site competitors seeking to usurp occupied cavities (e.g., Hasegawa 1993; Powell 2009; Powell et al. 2017). Moreover, each occupied nest is of equal value to a usurper, so attacks against the single nest of a founding colony or one nest of a larger multi-nest colony are equally likely. Selection may then favor a more rapid transition to a stable compositional representation of large defenders or distinct soldiers, so that each nest has a similar likelihood of survival regardless of overall colony size.

To address how arboreal life shapes the ontogeny of colony size and caste composition, a tractable study system is needed where the various challenges of studying arboreal ants have been overcome. The turtle ants (genus Cephalotes) represent such a system (Fig. 1). This group is one of the great evolutionary success stories among arboreal ant lineages, diversifying to over 118 species throughout their New World range (de Andrade and Baroni Urbani 1999; Bolton 2020). Recent work has shown that members of this group occupy an exceptionally broad range of cavity-nesting niches (Powell et al. 2020). Turtle ants are also well known for their iconic caste system, including a specialized soldier caste in most species (de Andrade and Baroni Urbani 1999; Powell et al. 2020), the evolution of which is driven by their cavity-nesting ecology (Powell et al. 2020). Case studies have shown that turtle ant colonies follow the common growth pattern of arboreal ants, by expanding into additional cavities as available resources allow (Powell 2009; Powell and Dornhaus 2013). Workers, soldiers, and brood are then dynamically deployed across all nests occupied by the colony (Powell and Dornhaus 2013; Powell et al. 2017), with soldiers using their armored heads to physically barricade nest entrances against usurpation by other ant species (Fig. 1; Powell 2009; Powell and Dornhaus 2013; Powell et al. 2017; Powell et al. 2020). Moreover, these ants do not have the capacity to substantially increase the volume of each cavity, and the use of cavities in more modifiable softer wood results in competitors enlarging the nest entrance to circumvent the soldier blocking defense and usurp the cavity (Powell 2008; Powell 2009). The addition of new cavities then appears to be the primary process by which substantial new nesting space is added for growth. This combination of nesting ecology, colony structure, and caste composition allows existing theory on the ontogeny of colony size and composition to be confronted with data from a contrasting ecological context.

Fig. 1
figure 1

Nesting ecology of the turtle ant C. persimilis. a Workers exiting and returning to a nest via the small cavity entrance hole. b A soldier and worker caste of C. persimilis. c A soldier using the armored dorsal surface of its head to block a nest entrance. Note that the tight fit between soldier head area and entrance area allows the soldier’s head to be mechanically locked into position, providing a robust, immovable barricade against potential nest usurpers

Full size image

Here, we use a representative of the turtle ants, Cephalotes persimilis, to examine the process of colony ontogeny in the arboreal realm. More specifically, we examined whether this species conforms to existing theoretical predictions for the ontogenetic interaction of colony size and caste composition, and provide new foundational insight into how arboreal nesting ecology is associated with the ontogeny of these colony traits. We do this using a static snapshot approach to studying colony ontogeny (see “Methods” for details), collecting, censusing, and quantifying the ecology of full colonies that span founding phase to reproductive maturity in our focal species. The turtle ant C. persimilis is native to the low and highly accessible canopy of the cerrado biome of South America (de Andrade and Baroni Urbani 1999; Price et al. 2014), where it competes for nesting cavities within a diverse arboreal ant community (Fig. 1; Powell 2009; Powell et al. 2011; Camarota et al. 2016; Vasconcelos et al. 2017). Additionally, this representative species has a morphologically distinct disc-headed soldier, the most common soldier morphotype within the turtle ant lineage (Powell et al. 2020). We use our novel dataset to address the overarching hypothesis that arboreal cavity-nesting is associated with modular increases in colony size and transitions in growth phase, as well as early stability in caste composition. In tackling this hypothesis, we ask the following specific questions: (1) To what extent are increases in colony size and transitions in growth phase predictably coupled with expansion into additional cavities?; (2) Does the transition to colony reproduction alter the growth relationship between nest number and colony size?; (3) To what extent does the compositional representation of the soldier caste increase with increasing colony size, and is this relationship altered by colony reproduction?; (4) Do caste sizes and size distributions increase with colony size and stabilize beyond reproductive maturity?

Methods

Study site and focal species

Fieldwork and specimen collection were conducted at the reserve of Clube Caça e Pesca Itororó, Uberlândia, Brazil between 2005 and 2008. The habitat at this site is predominantly neotropical woodland savannah or “cerrado”, with a typical canopy height of 3–6 m and physiognomies spanning cerrado sensu stricto to cerrado denso (30–50% canopy cover; Oliveira-Filho and Ratter 2002). Most data collection was conducted across the wet and dry seasons of the study period, to incorporate any seasonal variation into the general quantification of the biology of the focal species. Nevertheless, some datasets were collected at specific times of year to capture reproductive output, as detailed below.

The focal species for this study was the turtle ant Cephalotes persimilis de Andrade (de Andrade and Baroni Urbani 1999) which is widespread throughout the cerrado biome of Brazil (de Andrade and Baroni Urbani 1999) and abundant in the study site. Like most Cephalotes species, C. persimilis has three morphologically distinct female castes: queen, soldier, and worker (following caste nomenclature of de Andrade and Baroni Urbani 1999). Moreover, C. persimilis is representative of species with a disc-headed soldier caste, the most common of four distinct soldier morphotypes in extant members of the genus Cephalotes (Powell et al. 2020). The close relative Cephalotes grandinosus was also found at the same study site but at lower abundance, and the two species are easily distinguished by the sculpturing on the head disc of the soldier caste (de Andrade and Baroni Urbani 1999).

Colony discovery

C. persimilis colonies were located initially by applying urine baits on all trees within 5 m by 50 m transects haphazardly positioned throughout the study site. Turtle ants feed naturally at patches of urine from arboreal mammals (Powell 2008) and recruit strongly to patches of both natural and bait urine (Powell 2008). Bait patches were created by applying baits to the primary tree forks of each censused tree using a spray bottle. Each tree was checked for C. persimilis recruitment activity 20–30 min after bait was applied. Preliminary studies had established that this observation window captures the peak recruitment response of turtle ants to a newly available bait. A total of 50 colonies were located in this way, providing a large random sample of C. persimilis colonies to assess the distribution of nest number across colonies within the population. The presence of a single resident colony on each tree was verified by subsequent aggression tests and colony collection (below).

On each tree with C. persimilis recruitment to bait, extensive baiting within the crown was then used on locate all nests and verify colony membership (following the procedure of Powell 2009). To locate nests, recruits were visually tracked from a bait back to the nest they originated from. For each nest identified in this way, a bait was subsequently placed immediately adjacent to the entrance, to redirect recruitment from that nest to the nearby bait and, thus, simplify the complex network of recruitment trails throughout the tree crown. As this stepwise procedure was followed, additional baits were placed in regions of the crown that had yet to show any ant activity, systematically providing recruitment opportunities to potential nests in all regions of the crown. Baits were also placed on connected tips of neighboring tree crowns, to allow for the discovery of nests from a colony that spanned multiple tree crowns, although this proved to be extremely rare. This iterative baiting, tracking, and re-baiting procedure was used until all regions of the crown had been baited exhaustively and no recruitment from new nests had been detected for an hour or more. During this nest discovery procedure, aggression tests among ants from different nests were used to assess colony membership. C. persimilis workers from different colonies fight vigorously when in close contact, so the lack of aggression among ants from different nests was used to indicate membership to the same colony (following Powell 2009; Powell and Dornhaus 2013; Powell 2016). Observations on natural interactions among workers recruiting from different nests were used to assess aggression where possible and supplemented with staged aggression tests in a small petri-dish arena as necessary. Subsequent colony collection and censusing verified that this nest discovery and colony membership procedure robustly identifies all nests of a focal colony, by universally recovering queenright colonies across a variety of colony sizes and number of nests (below). In all cases, initial discovery of C. persimilis recruits on a focal tree revealed a single colony with a variable number of nests that were typically limited to the crown of the focal tree only (below).

Measuring nest properties

From the 50 colonies for which all nests were located (above), a subset of 20 colonies was used to quantify nest entrance area, stem diameter at the entrance, and wood type across colonies. Nest-entrance area was measured by using a macro lens to photograph the entrance with an in-shot scale for image calibration (following Powell 2008). The scale was held in the same horizontal plane as the nest entrance, and the photograph was taken perpendicular to that plane. To maximize the accuracy of the in-shot calibration, the camera was set up with the minimum depth of field to ensure that crisp focus on the entrance and scale was only possible when they were on the same horizontal plane. Entrance area was then measured digitally using ImageJ (Schneider et al. 2012). For each entrance, the perimeter was traced manually using the polygon selection tool, and the pixel area was calculated using the measure function. The mean of three in-shot scale measurements was used to calculate pixel area in mm2 in each image, allowing the total pixel area of the entrance to be converted to mm2. Entrances were represented by a minimum of 10,000 pixels in the photographs, providing excellent precision. Stem circumference was measured with a fiberglass tailors tape measure (flexible, shape-fitting, but stretch resistant) at the midpoint of each entrance and the stem diameter was calculated from the circumference value. The wood type for each nesting cavity was categorized as either dead wood (DW), typically still attached to a live tree, or live wood (LW) on a live tree. Maximum soldier head area in the focal species, used to standardize the raw entrance data to head areas, was calculated using the formula for an ellipse and the maximum head width and head length measurements for the focal species (de Andrade and Baroni Urbani 1999; Powell 2016).

Colony collection and census

For a different subset of 20 colonies from the total 50 (above), nests were collected, opened, and censused in the laboratory. These data provided both detailed caste composition data for a variety of analyses and further verified the methods used for locating all nests of queenright colonies. This static, snapshot sampling of colonies has been a widely used and powerful approach to assessing the ontogenetic trajectory of colony traits in previous colony ontogeny studies (reviewed in Tschinkel 2011). It allows rapid assessment of trait relationships across many colonies and across the full range of colony size, which is the goal of the present study, while allowing those colonies to have experienced a natural range of challenges to their ontogeny and still be representative of any seasonal stages in the growth cycle (e.g., annual burst of reproductive output). Using this approach, any ontogenetic relationships identified between traits can be seen as characteristics of colony ontogeny in the species that are robust against the many challenges that growing colonies face. A stratified sample based on nest number was used to incorporate the full range of colony sizes, and to incorporate reproductive status. An initial batch of ten colonies was collected over the first 18 months of the study, to give a foundational understanding of caste composition and especially seasonal reproductive activity of colonies across the full range of colony sizes. A second batch of ten colonies was then collected within a few days of each other in mid-July, when the annual reproductive output of C. persimilis colonies is reaching maturity but none of the reproductives have yet to be released (Powell 2009). This second batch of colony collections ensured increased sampling of reproductive colonies, allowing contrasts of soldier and worker investment across non-reproductive and reproductive stages of colony growth within the overall dataset. These 20 colonies are the same colony collections used to address the relationship between colony size and reproductive output in Powell (2009), but here, we report more extensive and additional datasets from the colony censuses and use the data to address different questions.

To ensure accurate nest and colony population counts, nests were collected between 5 and 8 am only, when colonies are inactive and all colony members are inside a nest. Each entrance was sealed with flexible electrical tape prior to stem removal, and stems were cut well below the nearest proximal branching node to the nest entrance. Nesting cavities are in stem internodes and rarely extend past adjacent nodes, so this stem-collection procedure ensured that cavities remained intact and sealed during collection and for subsequent transport. In the laboratory, nests were chilled in a fridge to make the ants motionless and facilitate the censusing procedure. For each nest, the stem was split with a knife at one end along the long axis and carefully pulled apart, so that the internal nesting cavity was bisected and all contents could be gently brushed into a large sorting tray. For all colonies, counts were taken for all adult castes in each nest, defined as follows: workers, soldiers, queens (dealate gynes, presumed to be active female reproductives in the colony), alate gynes (presumed to be female reproductive offspring of the queen ready to disperse), and males. Additionally, presence or absence was recorded in each nest for the following brood stages: eggs, worker/soldier larvae, sexual larvae (gyne or male), worker pupae, soldier pupae, gyne pupae, and male pupae.

Caste measurements

For 12 of the 20 colonies used in the composition census (above), worker and soldier size distributions were also measured. A stratified sample of colony sizes was again used to ensure representation in the dataset of colonies from the full range of recorded colony sizes (counts of workers plus soldiers). For smaller colonies of 250 ants or less, all workers and soldiers were measured, while for larger colonies, a random subsample of approximately 250 ants was measured. Head width was measured for all workers and soldiers in these colony samples, with measurements standardized on the maximum head width behind the eyes (following de Andrade and Baroni Urbani 1999). Ants were also categorized as worker or soldier based on the absence versus complete formation of a head disc, respectively (following head disc definition in Powell et al. 2020).

Analyses

Nest number and nest properties

The relationship between median nest entrance size (response) and the number of nests a colony occupies (predictor) was analyzed with a linear least-squares regression fit, after log transformation of both variables to meet the assumptions of the analysis. Stem diameter data were summarized using a standard outlier boxplot analysis. The relationship between entrance size (response) and both stem diameter and wood type (predictors) was analyzed with a linear least-squares regression fit including both predictor variables. Both entrance size and stem diameter were log transformed to meet the assumptions of the test.

Nest number, colony size, and reproductive status

The relationship between colony size (response) and nest number (predictor) was analyzed using a linear least-squares regression fit, with both variables log transformed. To assess the influence of the reproductive phase on the relationship between colony size and nest number, a linear least-squares regression fit was first calculated for the 12 colonies that were not reproductive at the time of collection, and then recalculated for all 20 colonies, adding data for the 8 colonies that were reproductive at the time of collection. The colonies that were reproductive at the time of collection (all sampled within the month of July, when colonies have a mature reproductive brood) were naturally limited to the upper range of the full-colony size distribution represented by colonies that were sampled outside of the annual reproductive burst. Adding the data for reproductive colonies to the full distribution of colony sizes for non-reproductive colonies was then seen as the most appropriate approach for assessing the influence of reproduction of the overall relationship. Wald tests for differences in slope and intercepts were used to assess whether the addition of the reproductive colonies significantly altered the fits to the data. The log transformation of the variables did not violate the assumptions of the analysis and additionally allowed the degree of deviation from an isometric scaling relationship to be readily assessed. Thus, a slope not significantly different from one (isometry) indicates that the same proportional relationship is preserved over the full range of variable values. In contrast, a slope either significantly less than or greater than one (allometry) would indicate that there is a proportional decrease or increase in colony size with each additional nest, respectively. Wald tests for differences in slope were also used to assess whether linear least-squares regression fits deviated from isometry (slope of one). The analyses were run with colony size defined as the total number of workers and soldiers in the colony, and repeated with colony size defined as the total number of all adult ants (adding queens, alate gynes, and males to the colony totals).

The variation in nest populations between the different nests of a colony was assessed by calculating the median of the absolute deviations from the median (MAD). MAD is a standard metric of dispersion expressed in the original units of the data, similar to the more commonly used standard deviation but more appropriate for our count data with potentially small numbers of individuals per nest. Linear least-squares regression fits were then used to assess the relationship between MAD values (response) and nest number (predictor), as well as the relationship between MAD values (response) and colony size (predictor; defined as total number of workers and soldiers). Nest number was log transformed to meet the assumptions of the test. These analyses necessarily excluded the three incipient colonies with only one nest.

Colony size and caste composition

The relationship between soldier number (response) and worker number (predictor), to assess the pattern of caste composition across colonies, was analyzed using a linear least-squares regression fit with both variables log transformed. The log transformation of the variables did not violate the assumptions of the analysis and additionally allowed the degree of deviation from an isometric scaling relationship to be assessed. A slope not significantly different from one would, thus, indicate that the proportional relationship between soldier number and worker number is maintained across all colonies, indicating a stable compositional percentage of soldiers. In contrast, a slope either significantly less than or greater than one would indicate that there is a compositional decrease or increase in soldiers across colony sizes, respectively. The linear least-squares regression fit was first calculated for the 12 colonies that were not reproductive at the time of collection, and then recalculated for all 20 colonies. Wald tests for differences in slope and intercepts were then used to assess whether the addition of the reproductive colonies significantly altered the fits to the data. Wald tests for differences in slope were also used to assess whether fits deviated from isometric scaling (slope of one).

MAD was used to assess the variation among the nests of each colony in the compositional representation of soldiers, also known as the pattern of soldier deployment (Powell and Dornhaus 2013). A linear least-squares regression fit was then used to assess the relationship between MAD of percentage soldier deployment (response) and colony size (predictor; total number of workers and soldiers) across colonies. These analyses necessarily excluded the three incipient colonies with only one nest, as MAD of percentage soldier deployment among nests cannot be calculated when the colony has only one nest.

Colony size and caste size ontogeny

These analyses addressed the prediction that caste size increases with colony size but stabilizes in colonies that have reached reproductive maturity. ANOVA was used first to test whether caste sizes were larger in colonies beyond the colony size threshold of reproductive maturity. Head width was used as the focal metric of caste size, given the special importance of the head in caste function in turtle ants, and both maximum and median caste sizes were examined for soldier and worker castes. The categorization of whether colonies were above or below the threshold of reproductive colony size was based on the minimum reproductive colony size identified in other parts of the study (above). Linear least-squares regression of caste size (response) on colony size (predictor) was then used to test whether caste sizes were stable in those colonies that were above the colony size threshold of reproductive maturity. Only the colonies above the size threshold of reproductive maturity were included in these analyses, with stability in caste size interpreted as a slope not significantly different from zero. Both maximum and median caste sizes were examined for soldier and worker castes.

Results

Nest number and nest properties

For the random sample of 50 C. persimilis colonies located within the study site, nest number ranged from one to twenty. Figure 2 shows the distribution of the number of nests occupied by these colonies, highlighting the median of 4 nests per colony and a long tail on the upper end of the distribution. These results indicate that the typical C. persimilis colony occupies numerous nests, and can therefore be defined as obligately polydomous, with no clear limit on how many nests a colony may expand into.

Fig. 2
figure 2

An outlier boxplot showing the distribution of the number of nests per colony within the focal population of the turtle ant C. persimilis. The box encompasses the interquartile range (IQR), a line is drawn at the median, and the whiskers extend to the upper and lower quartiles (± 1.5 times the IQR). The mean is marked with the gray circle and outliers are marked by the smaller white circles

Full size image

A total of 91 nest entrances were measured across colonies, spanning entrance areas from 1.5 to 8.1 mm2 (Fig. 3a). Entrance usage is narrowly defined around a median of 2.2 mm2 (Fig. 3a), which represents considerable ecological specialization in entrance usage relative to the potential nesting resources in the system and those used by other turtle ant species (entrance area ranges from over 200 mm2 to just 1 mm2 across turtle ant species; (Powell 2008; Powell 2016; Powell et al. 2020). As reported previously, this tightly bound median entrance size is approximately equal to maximum soldier head area in this species (Fig. 3, right axis and dotted line; (Powell et al. 2020), allowing a single soldier to effectively block the typical entrance size used by the colony (Fig. 1C; Powell et al. 2020). Indeed, specialization on entrances that approximate one soldier head area is so pronounced that 98% of used entrances are below the threshold of two soldier head areas (Fig. 3a). Additionally, Fig. 3b shows that there is no significant relationship between the number of nests a colony occupies and the median nest entrance size (Solid line, F1,18 = 0.2, R2 = − 0.04, P = 0.7). Specialization on highly specific entrances sizes, which fit a single soldier head, is therefore maintained as colonies find and expend into additional nesting cavities. Importantly, the consistent pattern of a close 1:1 matching between head size and entrance hole included colonies with just one nest, revealing that founding queens and their initial nanitic offspring (below) use the same entrance sizes as established, larger colonies.

Fig. 3
figure 3

The area of nest entrances used by the turtle ant C. persimilis within the focal population. a A violin plot showing the distribution of entrance area for all C. persimilis nests located within the focal population (N = 91 nest entrances). Left axis shows the raw data in mm2. Right axis shows the entrance data represented in soldier head areas, by standardizing against the calculated maximum soldier head area in the species (2.4 mm2). This axis provides an estimate of how many soldiers it takes to block an entrance. The plot incorporates a kernel density plot of the data and a boxplot with a box encompassing the interquartile range (IQR), a line drawn at the median, and whiskers extending to the upper and lower quartiles (± 1.5 times the IQR). The dotted line shows the maximum head area of one soldier b A scatterplot showing the relationship between the number of nests occupied by a colony and the median entrance area of the occupied nests (N = 20 colonies). Solid line represents the linear least-squares regression fit to the data (F1,18 = 0.2, R2 = − 0.04, P = 0.7). The dark gray region mirrors the IQR of the population distribution of entrance area (see violin plot), the light gray region mirrors the minimum and maximum entrance areas in the population, and the dotted line shows the maximum head area of one soldier

Full size image

The stem sizes in which nests were located varied from 7 to 125 mm in diameter, but the interquartile range (middle 50% of data) of the dataset spanned a narrow range of stems only 12–31 mm in diameter, with a median of 18.5 mm. Typical stem usage is therefore relatively specialized, like entrance size, given the full range of stems used by the species and the total stem range found in the tropical canopy. Overall stem usage was also notably biased towards live wood (86% live vs. 14% dead), and the majority of the live stems containing scale insects far larger than the nest entrance. Nevertheless, neither stem diameter (P > 0.3) nor wood type (P > 0.9) were significant predictors of entrance area (F2,88 = 0.5358, R2 = − 0.01, P > 0.5). Specialized entrance area usage is then maintained regardless of the stem size.

Nest number, colony size, and reproductive status

Figure 4a shows that nest number was a significant positive predictor of colony size (total number of workers and soldiers), whether we consider only colonies that were not reproductive at the time of collection (F1,10 = 24.41, R2 = 0.68, P < 0.0006, slope = 1.6), or all colonies (F1,18 = 49.13, R2 = 0.72, P < 0.0001, slope = 1.3). Colony size ranged from 9 to 1007 ants, and we also see that reproduction is limited to colonies with approximately 200 ants (minimum reproductive size of 192 ants) and occupying five nests or more. Additionally, incorporating the reproductive colonies did not significantly alter the slope or intercept of the fit compared to non-reproductive colonies alone (Wald test for different slope, F1,18 = 2.33, P > 0.1; Wald test for different intercept, F1,18 = 0.53, P > 0.4). Moreover, the slope of the relationship for all data does not differ significantly from a slope of 1 (Wald test for slope different from 1, F1,18 = 2.97, P > 0.1), indicating that colony size increases are modular and scale isometrically with each additional nest.

Fig. 4
figure 4

Number of nests, colony size, and nest populations in the turtle ant C. persimilis. a Scatterplot showing the relationship between the number of nests occupied by a colony and colony size, defined as the total number of workers and soldiers in all nests occupied by the colony. Both axes are plotted on a log scale. White circles represent colonies that were not reproductive when collected, while gray circles represent colonies that were reproductive when collected. The dashed line shows the linear least-squares regression fit to the data for non-reproductive colonies (F1,10 = 38.93, R2 = 0.78, P < 0.0001), while the solid line shows the linear least-squares regression fit to the data for all colonies (F1,18 = 43.88, R2 = 0.69, P < 0.0001). b The nest populations (workers plus soldiers) for each nest occupied by a colony. Nest populations are plotted on a logged axis. Gray lines connect all the nests from each colony, and colonies are ordered by overall colony size

Full size image

When incorporating all adult ants (workers, soldiers, queen, and all adult reproductive offspring) into the colony size totals, we again saw that adding reproductive colonies did not significantly change the slope or intercept of the fit between colony size and nest number compared to non-reproductive colonies alone (Fit for non-reproductive colonies, F1,10 = 24.67, R2 = 0.68, P < 0.0006, slope = 1.6; Fit for all colonies, F1,18 = 51.49, R2 = 0.73, P < 0.0001, slope = 1.3; Wald test for different slope, F1,18 = 2.05, P > 0.1; Wald test for different intercept, F1,18 = 0.48, P > 0.4). Moreover, the slope of the relationship between colony size on nest number for all data also did not differ significantly from isometric scaling (Wald test for slope different from 1, F1,18 = 2.97, P > 0.1). These data, thus, indicate that annual reproductive investment does not change total colony size enough to significantly alter the modular and isometrically scaling relationship between colony size and the number of nests the colony occupies.

In contrast to the consistent positive relationship between nest number and overall colony size, nest populations were highly variable across the nests of each colony (Fig. 4b). Indeed, there was no significant relationship between the variation (MAD) in nest populations and nest number across colonies (F1,15 = 1.36, R2 = 0.02, P = 0.2), or between the variation (MAD) in nest populations and colony size (workers and soldiers) across colonies (F1,15 = 0.006, R2 = − 0.06, P = 0.9). These non-significant patterns indicate that variation in the populations of individual nests does exist but does not increase with colony size, such that larger colonies are not the product of the ants creating larger individual nest populations. Expansion into additional cavities then remains the major predictor of colony size.

Colony size and caste composition

Across the 128 nests from the 20 colonies that were censused, all nest contained workers while only five lacked brood and one lacked a soldier. Notably, at least one soldier was present in all three colonies that occupied a single nest and had a total colony size of 14 ants or less. Nineteen of 20 colonies had a single queen (dealate gyne, presumed to be the active female reproductive in the colony). One colony had two dealate gynes, but the colony was also reproductive at the time of collection. Consequently, we cannot rule out the possibility that at least one of these individuals was a female offspring of the colony that had lost its wings and was retained. Regardless, these findings establish that C. persimilis colonies are typically monogynous, and therefore that queen number is not a variable trait that might influence the ontogeny of other traits in this species. The reliable recovery of queenright colonies also reaffirms the robustness of our colony collection technique.

Worker number was a significant positive predictor of soldier number across colonies, whether considering only colonies that were not reproductive when collected (F1,10 = 217.7, R2 = 0.95, P < 0.0001, slope = 1.0; Fig. 5) or both non-reproductive and reproductive colonies (F1,18 = 239.1, R2 = 0.93, P < 0.0001, slope = 1.1; Fig. 5). Incorporating reproductive colonies did not change the slope (Wald test for different slope, F1,18 = 1.0, P = 0.3) or the intercept (Wald test for different intercept, F1,18 = 0.2, P = 0.7) of the relationship. Moreover, the relationship between worker and soldier number across all colonies scaled isometrically (Wald test for slope significantly different from 1, F1,18 = 1.5, P = 0.2; Fig. 5). Using the fit to all data, the stable percentage composition of soldiers across colonies was approximately 19%. These results indicate that colonies have stable proportional representation of worker and soldier castes regardless of colony size and reproductive phase, and including colonies with a single nest and less than 20 ants.

Fig. 5
figure 5

The relationship between the number of workers and soldiers in colonies of the turtle ant C. persimilis. Both axes are plotted on a log scale. White circles represent colonies that were not reproductive when collected, while gray circles represent colonies that were reproductive when collected. The dashed line shows the linear least-squares regression fit to the data for non-reproductive colonies (F1,10 = 217.7, R2 = 0.95, P < 0.0001), while the solid line shows the linear least-squares regression fit to the data for all colonies (F1,18 = 239.1, R2 = 0.93, P < 0.0001). The dotted line represents the isometric scaling relationship (slope = 1) assuming the same intercept as the linear least-squares regression fit to all data

Full size image

Contrasting the highly stable soldier investment at the colony level, the percentage soldier deployment among the individual nests of a colony was highly variable, ranging from no soldiers present to 58% soldiers across all colonies (Mean 21%, SD ± 10%). Moreover, the variation (MAD) in percentage soldier deployment across the nests of each colony was not predicted by colony size (F1,15 = 217.7, R2 = − 0.05, P = 0.6). These finding then indicate that percentage soldier deployment across the nests of a colony varies substantially around the stable overall colony-level percentage representation of soldiers, and that this variation in deployment is not predicted by colony size.

Colony size and caste size ontogeny

Maximum caste size (head width) was significantly larger for colonies that had passed the threshold of reproductive colony size (Fig. 6a), considering both soldier (F1,10 = 64.6, P = 0.00001) and worker castes (F1,10 = 26.9, P = 0.0004). The same significant differences between colonies of reproductive and non-reproductive size held when considering median caste size (Fig. 6b) for the soldier (F1,10 = 28.3, P = 0.0003) and workers castes (F1,10 = 25.7, P = 0.0005).

Fig. 6
figure 6

The relationship between colony size and caste size in colonies of the turtle ant of C. persimilis. In both figures, circles represent soldiers, squares represent workers, the vertical dotted line represents minimum reproductive colony size (N = 192 ants), white data points represent colony sizes below this reproductive threshold, and gray data point represent colonies above it. a Scatterplot of the relationship between colony size and maximum head width in members of the soldier and worker castes. Solid line represents the linear least-squares fit to the soldier data for colonies larger than minimum reproductive size (gray circles; F1,6 = 2.8, R2 = 0.20, P = 0.1), while the dashed line represents the linear least-squares fit to the worker data for colonies larger than the minimum reproductive size (gray squares; F1,6 = 9.8, R2 = 0.56, P = 0.02, slope = 0.0002). b Scatterplot of the relationship between colony size and median head width in members of the soldier and worker castes. Solid line represents the linear least-squares fit to the soldier data for colonies larger than the minimum reproductive size (gray circles; F1,6 = 2.0, R2 = 0.12, P = 0.2), while the dashed line represents the linear least-squares fit to the worker data for colonies larger than the minimum reproductive size (gray squares; F1,6 = 36.9, R2 = 0.84, P = 0.0009, slope = 0.0002)

Full size image

When examining only those colonies larger than the threshold for reproductive maturity, colony size was not a significant predictor of soldier size (Fig. 6a; F1,6 = 2.8, R2 = 0.20, P = 0.1), indicating that maximum soldier size was stable across colonies in this reproductive category. Colony size was a significant positive predictor of maximum worker head size for reproductive colonies, but with a shallow slope close to zero (Fig. 6a; F1,6 = 9.8, R2 = 0.56, P = 0.02, slope = 0.0002), again indicating largely stabilized maximum caste size in reproductively capable colonies. Notably, for the colonies with a single nest, maximum soldier and worker sizes were approximately 25% smaller than maximum caste sizes in reproductively mature colonies, consistent with the production of nanitic soldiers and workers in founding C. persimilis colonies. Additionally, the smallest maximum soldier size recorded for a non-reproductive colony was less than the typical maximum worker size in reproductive colonies, highlighting the substantial shift in relative caste sizes with colony size ontogeny.

Closely mirroring the relationships for maximum caste sizes, colony size was again not a significant predictor of median soldier size when considering only colonies of reproductive size (F1,6 = 2.0, R2 = 0.12, P = 0.2). While median worker head width was significantly predicted by colony size in reproductive colonies, the relationship was again characterized by a shallow slope close to zero (F1,6 = 36.9, R2 = 0.84, P = 0.0009, slope = 0.0002). These findings complement the pattern seen for caste maximums, establishing that both the upper limit and central tendency of caste sizes initially increase across smaller colonies and then stabilize in reproductively mature colonies.

Discussion

Here, we have addressed the overarching hypothesis that arboreal cavity-nesting is associated with modular increases in colony size and transitions in growth phase, as well as early stability in caste composition. We have done this by examining the ontogeny of colony size and caste composition in colonies of the arboreal turtle ant C. persimilis, and by contrasting these traits against theory informed largely by soil-nesting ants. We have seen that our focal species obligately expands into additional nesting cavities with specialized entrance and cavity properties as it grows (obligate polydomy). Moreover, colony size scales isometrically with each additional nest unit occupied by colonies, reproduction is limited to a colony size that requires five nests or more, and the reproductive phase does not significantly change the relationship between colony size and nest number. We also saw no indication that larger colonies get large in part by having larger individual nest populations, further underscoring that growth is primarily dictated by the addition of modular nest units. In strong contrast to typical expectations for caste composition, we saw that soldiers are produced immediately, when a newly founded colony occupies just a single nest. Moreover, proportional soldier representation scales isometrically, at approximately 19% of the colony, across all colony sizes and spanning growth and reproductive phases. Yet this stable caste investment pattern at the colony level is contrasted by highly variable deployment of castes across nests. Finally, ant size increases steadily for both castes in early colony growth and then stabilizes at reproductive maturity, as predicted by theory. Nevertheless, the transition to reproductive maturity and the production of stable caste sizes happens early in colony ontogeny, at only 200 ants. Broadly, these findings show that colony size and caste composition in our focal species are intimately connected to the context of multi-nest arboreal ecology, with key aspects of colony growth and caste production contrary to typical theoretical and empirical expectations. The implications of these results for our understanding of how ecology shapes the evolution of colony ontogeny are discussed below.

The pattern of cavity selection seen in our focal species indicates that colonies maintain the same specialized nesting preferences during colony ontogeny, likely minimizing enemy threats throughout the colony’s life. Specialization on entrance sizes that tightly fit the head disc of one soldier is typical in turtle ants (Powell et al. 2020) and it maximizes defensive performance (Powell 2009; Powell et al. 2017). Moreover, some degree of nest entrance specialization is common in the larger arboreal ant communities that turtle ants are members of (Powell et al. 2011; Jiménez-Soto and Philpott 2015; Camarota et al. 2020; Priest et al. 2021), and direct competitive interactions over shared nesting resources can be intense (Powell 2009; Camarota et al. 2020). Thus, ecological specialization yields functional benefits during defense and narrows the range of potential nest competitors a colony faces. Here, we add the new insight that the species-level entrance specialization of C. persimilis is also maintained as colonies grow and expand into additional cavities (Fig. 3). Moreover, entrance specialization is coupled with stem size specialization and a strong preference for live wood in this species. Stem specialization further narrows the occupiable regions of tree crowns, and the use of live wood appears to facilitate the farming of scale insects and stable access to the food supply they provide. All considered, these data identify a consistent multi-nest, defensible nesting niche maintained throughout colony ontogeny.

The isometric scaling of colony size with number of nests provides new insights into the potential constraints of modular colony growth in the arboreal realm. We know that pre-existing cavities are often a limited nesting resource for arboreal ants and that the availability of cavities with suitable properties can further compound resource limitation and influence colony expansion (Powell 2009; Powell et al. 2011; Powell and Dornhaus 2013; Jiménez-Soto and Philpott 2015; Powell et al. 2017). In a very different obligate ant-plant system, colony size also scales with the size of the host tree (Marting et al. 2018). Here, we add a new ontogenetic perspective on the resource constraints faced by arboreal ants. We have seen that colony size increases across two orders of magnitude require expansion from one to twenty nests, with an isometric relationship between the colony size increase achieved from each additional nest. Moreover, reproduction appears to require five nests, but does not significantly alter the growth relationship between colony size and nest number. The simplest explanation for the regularity in colony size increases with each additional nest is that all cavities are very similar in size. Some combination of limited natural variation in the original beetle damage, limited excavation capabilities of the ants in hard wood, and the use of cavities already close to internode perimeters may explain this similarity. This interpretation that any variation in cavity size has limited importance for the overall trajectory of colony growth is supported by the variation in nest populations within a colony (Fig. 4b) having no significant relationship to colony size. A more complex, but not mutually exclusive possibility is that colonies seek additional cavities once the populations of existing nests reach a size that represents too high of a risk. This balance between resource constraint and risk distribution is a general challenge for understanding why multiple-nest strategies evolve in social organisms (reviewed in Debout et al. 2007; Robinson 2014). Turtle ant colonies expand into newly available cavities quickly in the field and laboratory, and prefer larger cavity volumes when available (Powell 2009; Powell and Dornhaus 2013). Yet the expansion process is also risk averse with respect to less-defensible entrance sizes in more dangerous enemy contexts (Powell et al. 2017). Some balance between the limitations of nest properties and risk-averse strategies is, therefore, likely and worthy or future investigation. Regardless, the isometric pattern of colony growth with increasing nest number stands as a robust component of colony ontogeny in the focal species.

Shifting focus from colony size to caste composition, the immediate and stable percentage representation of soldiers across all colony sizes suggests each nest unit faces similar high-level threats regardless of colony size. Theory and previous empirical studies on ground nesting ants show that investment in a specialized major or soldier caste increases later in colony growth (e.g., Wilson 1983; Porter and Tschinkel 1985; Gibson 1989; Tschinkel 1993), when interspecific enemy pressures are more varied and intense (Oster and Wilson 1978). Yet the same theory also proposed that it may benefit small colonies under frequent attack to invest early in soldier production, albeit without empirical support in ants at the time. Here, we show exactly this pattern of immediate investment in soldiers and a stable percentage investment throughout life. These results suggest not just frequent attacks early in colony life, but also the potential for sustained pressure by the same or similar enemies throughout colony life. This interpretation is consistent with our nesting ecology data, which shows that the first nest is no different in defensive properties (approximate 1:1 fit with a soldier head) to the subsequent nests acquired across the full range of colony sizes (Figs. 3b, 4a). Early production of soldiers has also been documented in American and Asian species of Colobopsis (Walker and Stamps 1986; Hasegawa 1993), arboreal cavity-nesting ants distantly related to turtle ants but with functionally similar soldier morphology. Moreover, Colobopsis impressus also showed stable percentage brood investment in soldiers across colony sizes (Walker and Stamps 1986). This immediate soldier production in Colobopsis and stable brood representation in one species was also interpreted as an adaptive response to a high level of usurpation risk throughout colony life (Walker and Stamps 1986; Hasegawa 1993), although the scaling of caste composition and the link to quantitative nesting ecology data was not part of these studies. While the extent of early soldier production across arboreal taxa is unclear, these comparisons suggest the convergent evolution of a similar caste composition strategy under similar nesting conditions. Our data further show that each new nest adds a similar number of individuals while maintaining the same overall soldier representation, and that neither relationship is changed significantly by reproduction (Figs. 4a, 5). Combined, this evidence of growth via repeated, well-defended colony units further indicates that consistently high usurpation pressure is driving the evolution of colony ontogeny in the arboreal realm. More extensive datasets on the strength and pervasiveness of nest usurpation pressure across colony sizes, and the associated nest turnover dynamics, will therefore be important for future studies to tackle.

Contrasting the striking stability in soldier representation at the colony level, our data also identify highly variable soldier deployment among nests of a colony, consistent with rapidly shifting threats at individual nests. If environmental pressures shift more rapidly than the development time of new individuals, then selection is expected to favor a stable pattern of caste production that is appropriate for meeting the full range of environmental conditions (Oster and Wilson 1978; Walker and Stamps 1986). Selection should then also favor the dynamic usage or deployment of the existing members of a caste to meet rapidly changing environmental pressures (Powell and Dornhaus 2013). Concordantly, we have seen that the consistent 19% representation of soldiers at the colony level is contrasted against a 0–58% representation of soldiers across nests, and that the variation in percentage soldier deployment across the nests of each colony was not predicted by colony size. Recent work in a different turtle ant species has shown that soldier deployment dynamically adjusts to nest availability, nest quality, and level of threat in the environment (Powell and Dornhaus 2013; Powell et al. 2017). Our results, combined with previous insights into soldier deployment, then add that a highly stable pattern of colony-level investment can be complemented by more dynamic usage of available caste members at any particular time. The dynamic deployment process is nested within the larger issue of social organization or division of labor within complex societies (Robson and Traniello 2016; Gordon 2019). Yet our findings highlight how colony-level ontogenetic stability in caste composition contrast the dynamic use of colony members, and how both are consistent with the overarching ecological pressures faced by the colony. These links between ecology, colony ontogeny, and dynamic organization are rarely made, but doing so promises more comprehensive understanding of the evolution of the complex colony phenotype.

In the final layer of our analysis, we did see increasing caste size with colony size, as predicted by theory, but this happened especially early in colony growth. Members of all castes are expected to increase in size with increasing colony size, with this process started later in colony growth for specialized castes like soldiers (e.g., Oster and Wilson 1978; Wilson 1983; Tschinkel 1988; Wills et al. 2018). Additionally, the energetic and ergonomic demands of having large, specialized castes within the colony has been seen as a constraint that limits their evolution to species with larger mature colony sizes (e.g., Oster and Wilson 1978; Wheeler 1991; Fjerdingstad and Crozier 2006). Yet this expectation has been challenged by comparative analyses (Fjerdingstad and Crozier 2006). In this study, the immediate production of a specialized soldiers in the first cohort of small individuals, or “nanitics”, was followed by maximum caste sizes reached at a reproductively mature colony size of only two hundred ants. These data then provide direct ontogenetic evidence that it is necessary to reevaluate the expectation that specialized castes are constrained to species with large colonies, and then limited to late development within such species. Here, we have seen a variety of evidence that soldier production and representation are likely the product of intense selection pressures associated with the nesting ecology of the arboreal realm. Ecology then drives the evolution of specialized castes, and perceived energetic or ergonomic constraints on caste evolution do not apply when selection in a specific environment is strong enough.

Broadly, our analyses of the ontogeny of colony size and caste composition in an arboreal cavity-nesting turtle ant have highlighted key patterns that contrast classic theory and previous empirical data. These differences appear to be directly shaped by obligately nesting in multiple, interconnected arboreal cavities, which is common among arboreal ants. Most notably, increases in colony size are modular via the acquisition of additional cavities with specialized properties, production of soldiers is immediate and proportionally stable across all colony sizes, and worker and soldier sizes increase rapidly and stabilize early in colony growth. Additionally, none of these ontogenetic patterns are meaningfully altered by the transition to the reproductive phase of colony growth. More robust and complete collection of colony ontogeny data are critical for addressing a wide variety of open questions in the study of the ecology and evolution of social taxa (Tschinkel 1991; Tschinkel 2011). Here, we have shown that colony ontogeny data collected in a less-studied nesting context has yielded new insights into the general question of how ecology shapes the evolution of colony size and caste composition. Integrating contrasting ecological contexts into future studies of colony ontogeny will likely further extend our empirical and theoretical understanding of how complex colony phenotypes evolve.