Biological characterization of the obligate symbiosis between Acropyga sauteri Forel (Hymenoptera: Formicidae) and Eumyrmococcus smithii Silvestri (Hemiptera: Pseudococcidae: Rhizoecinae) on Okinawa Island, southern Japan

The ant Acropyga sauteri Forel has an obligate, mutualistic symbiosis with a mealybug, Eumyrmococcus smithii Silvestri, on Okinawa Island, southern Japan. The mealybugs live inside ant nests nearly all their lives, and the ants depend on them for food. Alate foundress queens carry mealybugs during their nuptial flights, using them to establish new colonies at new sites. However, important aspects of the symbiosis have not yet been elucidated. The present study characterizes the basic biology of the symbiosis and describes for the first time the morphologies of all growth stages of E. smithii. Our study suggests that E. smithii has only one nymphal stage, followed by a female pupal stage or male prepupal stage. Intensive sampling of ant nests across seasons showed that A. sauteri prefers nest sites 5–20 cm underground. Acropyga sauteri produced reproductive stages mainly in mid‐March or early April, and numbers of both ant workers and mealybugs increased from spring to summer. Experimental determination of colony identity with a method using nestmate recognition by ants suggested that each ant colony rarely has a perimeter greater than 30 cm, that the ants are monogynous, and that different ant colonies are densely aggregated along the root system of a plant, adjacent to each other but not interflowing. Both symbiotic partners were vulnerable to attacks by several common subaerial ant species following physical disturbance to their nests.


Introduction
Mutualistic interactions between ant and hemipteran species are not uncommon (e.g. Way 1963;Buckley 1987;Hö lldobler and Wilson 1990;Gullan and Kosztarab 1997). Hemipterans provide honeydew for ants in exchange for protection against parasites and life cycle of E. smithii, the spatial distribution and structure of nests of A. sauteri, the withinnest distribution of both partners, and the seasonal changes in age structure for both partners. In addition, we investigated other ant species occurring around the nests of A. sauteri and their effects on the symbiosis.

Study site
The study site (26u109N, 127u459E, 65 m in altitude) was in Sashiki Town, the southern part of Okinawa Island, which is a grassland dominated by a perennial grass species, Miscanthus sinensis Anderss, but has a mixture of several grass and shrub species. The study site was on a hillside surrounded by secondary forests and sugarcane fields. A plot (2563 m) was established among dense growth of M. sinensis (average plant height: ca 2.0 m).

Ants and mealybugs
The body length of an A. sauteri worker is 2.0-2.5 mm (Yamane et al. 1999). The body of the adult female of E. smithii is white in colour, 1.7 mm in length (Silvestri 1927), and has a dilated cephalothorax and an abruptly narrowing abdomen with a dorsally curled tip (Silvestri 1927;Williams 1970Williams , 1978Williams , 1998). The mealybugs were observed feeding on the roots of a species of bamboo (Uye 1933), Saccharum officinarum L. (Takahashi 1934;Williams 1970;Terayama 1988) and Imperata sp. (Takahashi 1934). At the study site, the mealybugs commonly feed on the roots of M. sinensis.

Examination of the growth stages of Eumyrmococcus smithii
To observe the morphology of all growth stages of E. smithii, we collected 262 individuals by sampling cubic clods from the ground surface to a depth of 0-15 cm in randomly chosen quadrats in the study plot in April 2000. All collected individuals were prepared on microscope slides by the method of Takagi (1970) with Kawai's (1980) improvement. For each individual, we measured the width and length of the body and antennae, the width of the spiracle, the length of the hind legs (trochanter, femur, tibia, tarsus, and claw) and body setae, and the width and length of the labium (if present). In individuals with anal rings, we also measured the width of the base of the abdominal segment, the width of the anal ring, and the lengths of anal lobe setae and anal ring setae.
Examination with magnification showed that some of the specimens had the exoskeletons of subsequent stages within their bodies, which indicated that they were just about to moult. Based on the morphology of the mealybugs measured in this study and the morphology of the conspecifics and congenerics described by Williams (1970Williams ( , 1978Williams ( , 1998, we attempted to identify the growth stages of these mealybugs.

Spatial distribution of both partners
To clarify the spatial distribution of nests of A. sauteri and the within-nest distribution of A. sauteri and E. smithii, we mapped ant nests in the field by digging out clods from the ground (ranging from 0 to 30 cm in depth) and checking for the presence of both partners in the nest in each clod. We removed 184 cubic clods (56565 cm) in a randomly selected 55625 cm area (ranging from 0 to 20 cm in depth) in August 2001, and 212 cubic clods (56565 cm) in three randomly selected 45645 cm areas and a 45675 cm area (ranging from 0 to 30 cm depth) in June 2002 ( Figure 1a; Table I). We recorded the locations of the quadrats, and all the cubic clods were mapped on the grid in each quadrat (Figure 1a). Each clod was transported to the laboratory in a plastic bag. All ants and mealybugs in each cubic clod were then collected. Only the August 2001 samples were used for analysis of the spatial distribution of the nests in relation to the depth, but all the samples were used for the other analyses.
Determination of colony identity and spatial range of a single colony Ant nest tunnels ran along the root systems of host plants, crossing reticulately, and the colonies were very densely distributed. It was consequently extremely difficult to recognize the boundaries of neighbouring colonies of A. sauteri by finding an uninhabited area between two ant aggregations situated in neighbouring clods. We used ant nestmate recognition behaviour (shown by ant workers in many ant species) to determine the colony identity of two ant worker aggregations that were haphazardly selected. Workers of most ant species are known to show some antagonistic behaviour against non-nestmate conspecifics (Hö lldobler and Wilson 1990). If antagonistic behaviour was observed when a worker of one ant aggregation was experimentally introduced into another aggregation, we considered that the two ant aggregations belonged to different colonies. Conversely, if receptive behaviour was observed, we considered the two aggregations to be parts of the same colony. We tested the validity of this method for A. sauteri by examining whether A. sauteri workers showed distinct antagonistic behaviour toward a worker experimentally introduced from a different colony and receptive behaviour toward a worker from the same colony. We removed 18 cubic clods (10610610 cm) in 12 randomly located quadrats (10610 cm, ranging from 0 to 20 cm in depth) in January 2002 ( Figure 1b; Table I). In the laboratory, we collected eight aggregations from the clods. Each ant aggregation was derived from a single cavity or tunnel. For each aggregation, five to seven workers were placed in a plastic cup (60 cm 3 ) covered with black paper ( Figure 2b) and rested until they resumed normal behaviour. Then we arranged 14 inter-aggregational experimental matches ( Figure 2c). In each match, we introduced an ant worker from one cup (aggregation) into another cup (aggregation) and observed the behaviour stimulated by the introduced worker ( Figure 2d). The interactions between the ants were classified as follows: ants from one aggregation may have been (1) bitten, (2) chased, (3) groomed with mouthparts, or (4) ignored by ants from another aggregation. The behaviour of the former two was regarded as ''antagonistic'', whereas the behaviour of the latter two was regarded as ''acceptive''. After the workers from different aggregations had contacted each other two or three times, we removed the introduced worker and replaced it in its original cup ( Figure 2e). We never used an introduced worker for more than one trial. For a match, these trials of introduction (Figure 2c-e) were repeated at least seven times. For all matches, all trials for a match were consistently categorized as a single type of behaviour, as either antagonistic or acceptive. We therefore concluded that this method was effective for the determination of colony identity between ant aggregations derived from two locations.
Using this method (Figure 2c-e) on the ant samples collected in June 2002 (mentioned above; Table I), we checked the colony identity of neighbouring ant aggregations to At each sampling event we randomly chose a ground surface area for sampling clods, from which individuals of Acropyga sauteri and its symbiont Eumyrmococcus smithii were collected. The dates of sampling events and the numbers and sizes of areas and cubic clods are listed in Table I. determine the range of a single colony in the clods. For the experiment of colony identity, we collected 29, 32, and 34 ant aggregations from cubic clods in the three different study areas. They were selected in such a way that the distances (D) between two clods containing the subjective two aggregations varied from 5 to 30 cm within each selected area ( Figure 2a). In addition, we set up matches between aggregations from different study areas so that the distances between the two aggregations were much more than 30 cm. A total of 96 experimental matches of two aggregations with the use of 95 ant aggregations were conducted. Trials (Figure 2c-e) were repeated five times for each match between two particular ant aggregations.

Seasonal changes in age structures for both partners
We examined the seasonal changes in age structure for A. sauteri and E. smithii using the samples taken in August 2001 and January and June 2002 (Table I). In addition, we collected extra ants and mealybugs in 12 cubic clods (15615615 cm) in 12 randomly positioned quadrats (15615 cm, ranging from 0 to 15 cm in depth) in October 2001, in three cubic clods (15615615 cm) in three randomly positioned quadrats (15615 cm, ranging from 0 to 15 cm in depth) in March 2002, and in five cubic clods (10610610 or 15615615 cm) in two randomly positioned quadrats (30630, 50650 cm, ranging from 0 to 15 cm in depth) in April 2002 (Table I).
Several colonies of A. sauteri often aggregate densely in a small space, such as the root system of a host plant (see Results). In this case, it was extremely difficult to specify the spatial range of a single colony of A. sauteri by recognizing the intervals between ant colonies. Therefore, when multiple colonies might possibly be sampled from certain clods in the vicinity, we estimated the average numbers of ants and mealybugs at different growth stages for the colonies in the clods by dividing the total numbers of ants and mealybugs at different growth stages by the total number of ant queens present in the clods. The validity of this estimation was based on the assumption that A. sauteri forms monogynous colonies, which seemed to be a safe assumption for the following reasons. First, Uye (1933) observed Table I. Protocol for sampling cubic clods in the field in order to determine the seasonal changes of age structures of both partners, the spatial distribution of both partners, and the colony identity of ant aggregations (we used the samples at all sampling events for the seasonal changes, those at two events for the spatial distribution, and those at two events for the colony identity).

Date
Size of an area for sampling cubic clods (cm) a The samples were used for specifying spatial distribution of both partners. b The samples were used for determining the colony identity of ant aggregations. c The samples contained several queens in one cubic clod. Therefore we estimated the average numbers of ants and mealybugs per colony by dividing the total number of queens into the total numbers of ants and mealybugs.
colony foundation by solitary queens in this species, and the number of solitary queens drastically increases just after the putative nuptial flight season (see Results). Second, antagonistic behaviour between workers, which would indicate contact between nonnestmates, were frequently observed when multiple queens were found in a small space (see Results). Moreover, antagonistic behaviour was always observed between workers derived from two distant points (see Results). Third, throughout our study, multiple queens in the same nests and intimate interaction between queens were never observed. Figure 2. Schematic illustration of the method used to determine colony identity. (a) Distance (D) between two particular ant aggregations was defined as the distance between the centre of the two clods containing the two aggregations; (b) five workers were placed in a plastic cup covered with black paper; (c) an ant worker was introduced into another cup (of recipient workers); (d) the contact behaviour of recipient and introduced workers was observed; (e) the trial ended after contact had occurred two or three times, after which the introduced worker was returned to the original cup. In a match between two aggregations, the method described from (c) to (e) was repeated five times.
Ants and mealybugs in clods that were more than 30 cm away from the nearest clod containing a queen were omitted from the analysis because they may not have been a representative fraction of a colony.
We statistically compared the total numbers of individuals of A. sauteri and E. smithii by one-way analysis of variance. We could not statistically analyse the numbers of alate females and males of A. sauteri because the sample size was insufficient and because the data were skewed. Instead, we presented the distribution of variables for each colony (sometimes variables estimated from multiple colonies).

Other ant species
We censused ants species other than A. sauteri in the study plot during every period of clod sampling. We recorded the nest locations of the other ant species, observed their behaviour with special reference to their responses to A. sauteri and E. smithii, and collected specimens for species identification. We identified the ants using keys provided by Yamane et al. (1999).

Features of Eumyrmococcus smithii
Mealybugs were morphologically allocated into one of six growth stages by microscopic examination. One stage (39 individuals) was considered to represent adult females, and some of these mealybugs contained eggs. Adult females could additionally be distinguished from the other five stages on the basis of their anal characteristics (e.g. width of anal ring, anal ring setae, and anal lobe setae). Two individuals were determined to be adult males on the basis of their genitalia and the absence of a labium (Appendix; Figure 11). Three stages were characterized by pupa-like morphologies, which were similar to those described by Williams (1998) for the prepupae and pupae of congeneric species. One of the stages (19 individuals) was identified as the female pupal stage (Appendix; Figure 12a), because the mealybugs contained the exoskeletons of female adults (Figure 3a). Similar to patterns observed in congeneric species (Williams 1998), female pupae at this stage had a labium and longer antennae than observed in the other two pupa-like stages. Of the two remaining pupa-like stages, the antennae of one group of mealybugs was shorter than those of the other, and three of these mealybugs (of a total of seven individuals) contained male adult exoskeletons (Figure 3b). They were thus identified as male pupae (Appendix; Figure 12c). Two (of a total of 15) individuals in the remaining pupa-like group contained the exoskeletons of male pupae, and this stage was therefore considered to represent the male prepupa (Appendix; Figure 12b).
The remaining growth stage (180 individuals) was identified as nymph. One sampled individual had died during emergence from an egg and was not yet sclerotized. Although details of its morphology were not clear because of the manner of death, strong similarities in spiracle width and width of the anal ring between this individual and the remaining 179 nymphs were observed. Consequently, all 180 nymphal individuals were categorized as first-instar nymphs (Appendix; Figure 13). The variations in body size and morphological characteristics, including setal characteristic, observed within this group were continuous and unimodal. Two of these nymphs contained the exoskeletons of female pupae, and 10 contained exoskeletons of male prepupae, which confirmed that this first instar was followed by a male prepupal stage or a female pupal stage (Figure 4). The morphology of the first-instar nymphs, female and male pupae, male prepupa and male adult are described in the Appendix.

Spatial distribution of nests of Acropyga sauteri
Acropyga sauteri workers were distributed at a depth of 0-38 cm in August 2001; no worker was found at depths greater than 40 cm. The numbers of clods with more than five workers varied significantly among depths (x 2 510.837, P,0.05, chi-square test of independence) and was lowest at soil depths of 0-5 cm ( Figure 5). All three queens were collected at 10-20 cm depth. The shortest distance between queen-containing clods was 5 cm.   The average number of ant workers in clods with E. smithii was significantly higher than the number of ant workers in clods without E. smithii (Figure 6; August; U51110.00, P,0.0001, June; U51420.00, P,0.0001, Mann-Whitney U test). No individual of E. smithii was found in clods in which A. sauteri was absent.

Colony identity and spatial range of a single colony
In matches between two ant aggregations, done for the purposes of colony identification, recipient and introduced workers consistently showed similar responses in all trials; all displayed either antagonistic or acceptive behaviour. When the distance between two clods containing ant aggregations ranged from 5 to 30 cm, the frequency of acceptive behaviour was 38-47% ( Figure 7). The greatest distance between two clods containing ant aggregations with workers displaying acceptive behaviour was 25 cm. When the distance between two clods was more than 30 cm, antagonistic responses were observed in every trial in all matches (Figure 7). Only one queen was found within this presumed spatial range of a single colony (except in the case of new alate foundress queens).

Within-nest distribution of both partners
Ant nests had tunnels 1-2 mm in diameter and two types of chambers, both ca 4-7 mm in width and height. One chamber type was penetrated by small plant roots, on which adults and nymphs of E. smithii were often observed to be settled. The other chamber type did not contain any plant roots. In these chambers, A. sauteri and E. smithii individuals at various growth stages were often huddled in clusters. Clusters consisted mainly of E. smithii at pupal or prepupal stages and eggs or brood of A. sauteri. The average number of ants in all stages (alate female, alate male, worker, pupa, and brood) per colony was significantly higher in October 2001 than in January, March, and April 2002 (Figure 9; P,0.05, one-way ANOVA followed by Tukey-Kramer test). Average numbers were lowest in January 2002 (Figure 9).  There were seasonal changes in the female adult: nymph ratio corresponding to changes in the total number of mealybugs. The ratio was relatively low as the total number of mealybugs per colony increased from March to June, whereas it was relatively high when the total number of mealybugs per colony decreased from August to January (Figure 10

Other ant species
At least nine ant species other than A. sauteri were found in the soil around nests of A. sauteri (Table II). Of the nine species, Camponotus kaguya Terayama, Monomorium sechellense Emery, Oligomyrmex hannya Terayama, Paratrechina sp., and Tetramorium biacarinatum (Nylander) had nests around those of A. sauteri (Table II). At least 10 nests of any of the five species were observed at soil depths of 0-30 cm in August 2001 and June 2002 (Table II). Seven nests were observed at 0-5 cm in depth (Table II).
Pheidole noda F. Smith and T. biacarinatum were often observed invading nests of A. sauteri through holes that were accidentally made during sampling activities. These ants carried individuals of both A. sauteri and E. smithii out in their mandibles. Except when disturbed nests of A. sauteri were available, ants of P. noda were always observed Figure 10. Seasonal changes in the average numbers (with SE) of individuals of Eumyrmococcus smithii per colony and the age structure (percentage of components). Numerals above the bars indicate the sample size (presumed number of ant colonies). ''Pupa'' here includes pupae of both sexes and male prepupa, which were difficult to discriminate when not on slides.
foraging on the ground and were not found in the soil. Although nests of T. biacarinatum were found in the soil in the study plot, nests of P. noda were not (Table II).

Growth stages of Eumyrmococcus smithii
The present study is the first to describe the female pupal stage of E. smithii. Within the family Pseudococcidae, the female adult stage normally follows three nymphal stages (Williams 1998). However, Williams (1988) described a female pupal stage for Xenococcus and argued that the pupal stage could be regarded as the third instar. Moreover, Williams (1998) found similar female pupal stages in females of at least five Eumyrmococcus species. Our study has shown that the morphology of the female pupa of E. smithii is very similar to that of congeneric species (Williams 1998). All mealybug species in intimate symbioses with Acropyga ants have thus far been found to have a non-feeding pupal stage, and thus our identification of a female pupal stage of E. smithii reinforces the association (pointed out by Williams 1998) between the formation of the female pupal stage and symbiosis with Acropyga ants.
We identified the morphology of first-instar nymphs for the first time in Eumyrmococcus by scrutinizing the morphology of many individuals of E. smithii. Moreover, our study suggests that E. smithii has only one nymphal stage, which is followed by a pupal stage in females and by a prepupal stage in males (Figure 4). Williams (1998) argued that two nymphal stages would be expected for other species of Eumyrmococcus, and he reported that second-instar nymphs were available for 11 species, but described only that of E. taylori in detail and those of six other species very briefly. Williams (1998) also reported that no firstinstar nymphs were available, except for a single individual still within its egg membrane. Our results suggest that the individual inside its egg membrane and the ''second-instar'' nymphs described by Williams (1998) are the same instar. Rearing experiments will be necessary to more precisely describe the life cycle of E. smithii.

Nest structure of Acropyga sauteri
Our observations show that nests develop along the root systems of plants on which the mealybugs feed. As a consequence, the spatial borders between two neighbouring colonies cannot be simply mapped in a horizontal dimension. Using nestmate recognition methods to determine colony identity, we showed that colony boundaries often existed between two ant aggregations separated by only ca 5 cm and that the greatest distance between two aggregations that were judged to belong to the same colony was 25 cm. In addition, when the distance between two ant aggregations compared was greater than 30 cm, the aggregations were always found to belong to different colonies. These results suggest that a nest is restricted to a space of 0-30 cm, that an individual colony of A. sauteri may be densely concentrated in a small space around the root systems of plants of M. sinensis, and that colony boundaries can run diagonally, horizontally, and vertically through the root system.
Mapping ant distributions showed that the shortest distance observed between two queens was only 5 cm. This might cause problems with the determination of whether A. sauteri is monogynous or polygynous. As previously mentioned, however, our observation that many colonies were concentrated in a small space and the fact that multiple queens (except for new alate foundress queens) were not found within a single colony suggest that A. sauteri is monogynous. This suggestion is supported by Uye's (1933) observation that A. sauteri colonies are founded by solitary queens. Without our tests of colony identity, the field situation could have been interpreted as a widely distributed polygynous nest.

Seasonal changes in age structures for both partners
On the basis of the following two observations, we inferred that the nuptial flight of A. sauteri mainly occurs from mid-March to early April on Okinawa Island. First, crossseasonal comparisons of the numbers of reproductives in colonies that produced at least one reproductive showed that the number of reproductives per colony was rather high in March. Second, solitary queens unattended by workers were only observed in April. These queens were presumed to be solitarily founding their nests immediately after their nuptial flight. This inference agrees with Terayama's (1988) observation of nuptial flights for this species in mid-March on Tokunoshima Island, near Okinawa Island. However, we found alate females and males in seasons other than March. Currently, we do not have sufficient information to judge whether these reproductives perform nuptial flights in seasons other than spring or whether they survive inside the nests until the following year's nuptial flight. Uye's (1933) observation of nuptial flights of A. sauteri in December in Oita Prefecture (to the north of Okinawa Island) supports the former possibility, although he regarded the observed phenomenon as an abnormal event.
The observed seasonal trends in numbers and age structures of the partners suggest that E. smithii reproduces mainly from April until August and that the total number of mealybug individuals decreases from October to March. These seasonal trends may be associated with seasonal fluctuations in temperature. Temperatures are usually under 20uC from November or December to March or April on Okinawa Island. The decrease in mealybug numbers may be due to a low temperature, which is likely to negatively affect production activities (such as photosynthesis) by host plants. The seasonal trends in ant numbers also may be indirectly affected by temperature fluctuations. This is suggested by the fact that there was a lag period between fluctuations in ant numbers and fluctuations in mealybug numbers and by the reasonable inference that colony growth of A. sauteri depends on the reproduction of its symbiont, E. smithii. Delabie and Fowler (1993) reported that the population size of an Acropyga ant in symbiosis with a mealybug species showed significant lagged correlations with temperature, rainfall, and leaf flush. They proposed that the reason for the lagged correlations was the indirect dependence of the ant population on plants, via feeding on plant roots by the mutualistic mealybugs.

Other ant species
The observed frequent invasion of damaged nests by common ant species such as P. noda and T. biacarinatum suggests that nests of A. sauteri were normally well defended from ant attacks because of their almost completely subterranean lifestyle in certain kinds of soils. For nest sites, A. sauteri prefers rather hard clay layers lying underneath soft surface soil. By contrast, the other ant species located their nests mainly in the surface soil, usually at soil depths of 0-5 cm, which suggests that they may have difficulties digging in hard clay. The clay layer preferred by A. sauteri may function as a barrier to invasion by potential predators, such as the two ant species observed in this study.