Phylogenetic evidence for a single, ancestral origin of a ‘true’ worker caste in termites


Graham J.Thompson School of Tropical Biology, James Cook University, Townsville, Queensland 4811, Australia. Tel.: (61 7) 4781 5582, fax: (61 7) 4725 1570. e-mail:


Phylogenetic analysis based on sequence variation in mitochondrial large-subunit rRNA and cytochrome oxidase II genes was used to investigate the evolutionary relationships among termite families. Maximum likelihood and parsimony analyses of a combined nucleotide data set yield a single well-supported topology, which is: (((((Termitidae, Rhinotermitidae), Serritermitidae), Kalotermitidae), (Hodotermitidae, Termopsidae)), Mastotermitidae). Although some aspects of this topology are consistent with previous schemes, overall it differs from any published. Optimization of ‘true’ workers onto the tree suggests that this caste originated once, early in the history of the lineage and has been lost secondarily twice. This scenario differs from the more widely accepted notion that workers are derived and of polyphyletic origin and that extant pseudergates, or ‘false’ workers, are their developmentally unspecialized ancestor caste. Worker gains and losses covary directly in number and direction with shifts in ‘ecological life type’. A test for correlated evolution which takes phylogenetic structure into account indicates that this pattern is of biological significance and suggests that the variable occurrence of a worker caste in termites has ecological determinants, apparently linked to differences in feeding and nesting habits.


Termites comprise an insect order of over 2300 species consisting of seven taxonomic families ( Pearce & Waite, 1994). This classification stems from earlier works by Emerson (1955), Snyder (1949) and Grassé (1949) who used morphological, anatomical, and to a lesser extent, behavioural and ecological characteristics to distinguish among what were previously identified as the principal lineages ( Holmgren, 1911, 1912). Relationships among these lineages have long been hypothesized (e.g. Hare, 1937; Emerson, 1941; Krishna, 1970) but the application of phylogeny to developing and testing models of termite social evolution has been minimal.

All extant species of termite have a reproductive division of labour into a small number of reproductives and a large number of sterile or subfertile workers and soldiers (i.e. eusociality sensuWilson, 1971). This eusocial uniformity has rendered it more difficult to deduce the likely evolutionary transitions that preceded the presumedly graduated appearance of this trait ( Thorne, 1997). By contrast, evolutionary ‘stepping-stone’ type scenarios, though not essential ( Michener, 1985), are more readily formulated for the socially polyphyletic Hymenoptera. Here, eusocial groups are separated from solitary species by numerous species with intermediate forms of social organization ( Wilson, 1971).

Despite their typological uniformity, there is a growing appreciation that termites do present a broad-ranging eusociocline. This is because of a number of comparative studies that have highlighted differences in life history characteristics underpinning termite eusociality (e.g. Abe, 1987; Noirot, 1989; Lenz, 1994; Shellman-Reeve, 1997; Keller, 1998; Myles, 1999) and more directly, the result of a conceptual shift that considers ‘eusociality’ in general not as an homogenous category but rather as continuously varying in proportion of offspring that forego reproduction ( Sherman et al., 1995 ). In termites, such variation is most conspicuously manifest through two well-differentiated subfertile castes. The soldier caste is ubiquitous throughout the order (only the derived Apicotermitinae genus Anoplotermes have lost soldiers secondarily; Sands, 1972), but ‘true’ workers are present only in the families Mastotermitidae, Hodotermitidae, Serritermitidae, Rhinotermitidae and Termitidae ( Abe, 1987; Noirot & Pasteels, 1987).

True workers (hereafter workers) were defined by Noirot & Pasteels (1987) as ‘individuals diverging early and irreversibly from the imaginal development, with a morphology typical of their caste but largely of larval appearance, and taking part in most of the social tasks’. Thus, workers are not necessarily sterile but have reduced reproductive potential brought about through a combination of behavioural, morphological and developmental characteristics. Worker-less species have instead a developmentally unspecialized worker-like caste called a pseudergate (‘false’ workers; after Noirot & Pasteels, 1987; Grassé & Noirot, 1947), which more readily differentiates into reproductive form.

Two alternate hypotheses have emerged to account for the evolutionary history of the worker caste. The traditional view is that the early and irreversible development of workers is a derived feature of termite eusociality that evolved on a least three separate occasions ( Noirot & Pasteels, 1987, 1988). By contrast, Watson & Sewell (1981, 1985) have suggested that worker origins may have preceded or accompanied the origin of termite eusociality, implying that workers are therefore an ancient feature of termite eusocial systems. Whether workers are ancestral or, alternatively, derived and of polyphyletic origin has not yet been firmly established and these two alternative hypotheses could lead to different scenarios for the evolution of termite eusociality.

Moreover, the occurrence of workers is known to have clear ecological correlates. Specifically, workers are invariably present in species with ‘intermediate’ or ‘separate’ life types that exhibit central-place nesting with distant foraging locations and absent in ‘one-piece’ life types that consume solely the wood in which they nest (after Abe, 1987). Abe (1991) and Higashi et al. (1991) used this observation, together with a polyphyletic view of worker origins, to hypothesize that workers evolved from pseudergates through ecologically mediated selection favouring worker-based altruism. It has, however, never been established whether the correlation between life type and worker presence evident among extant taxa is the result of an evolutionarily significant adaptive association between these two traits.

Thus, to clarify patterns and correlates to worker evolution, a family level phylogeny is clearly desirable. To this end, recent work has begun to build on the traditional interpretation of the interfamilial relationships within the Isoptera ( Kambhampati & Eggleton, 2000) and between this and the related orthopteroid orders ( Vawter, 1991; DeSalle et al., 1992 ; Thorne & Carpenter, 1992; Kambhampati, 1995; Maekawa et al., 1999 ; Lo et al., 2000 ). Although certain relationships have become firmly established (e.g. Termitidae and Rhinotermitidae are sister groups), others remain disputed or untested (e.g. Is Mastotermitidae plus Kalotermitidae the sister group to all other termites?) with no single family level hypothesis having yet gained widespread acceptance. In this paper, we re-examine the phylogenetic relationships among all seven families within the order in light of new information obtained from two mitochondrial genes, cytochrome oxidase II (COII) and large ribosomal subunit RNA (l-rRNA). In addition to inferring the best tree from these data, we test previous phylogenetic schemes and use the overall best estimate to test scenarios concerning the evolutionary history of a worker caste. Specifically, we examine the number and pattern of origins and losses of the termite worker caste and provide a test for association between life type and workers by comparing their phylogenetic distributions against that expected under the null hypothesis that both characters evolved independently.

Materials and methods

Sequence data and alignment

Partial sequence data from l-rRNA and COII genes used in this study were obtained from representative species from all seven families as indicated in Table 1. We included sequence data from Periplanata americana (Blattidea), Tenodera angustipennis (Mantodea), and Locusta migratoria (Orthoptera) as outgroups to our analyses on the basis that these taxa form a sister-group assemblage to the termites ( Maekawa et al., 1999 ). Data not obtained from previously published sources as indicated in Table 1 was generated manually for the remainder of taxa following procedures analagous to those described in Lo et al. (2000) .

Table 1.  List of taxa included in this study with collection site and GenBank accession numbers. Thumbnail image of

Primary alignments of the protein-encoding sequence were performed using C LUSTALX ( Higgins et al., 1996 ), then verified manually against existing insect alignments ( Liu & Beckenbach, 1992). Ribosomal RNA gene sequence was aligned in relation to the secondary structure proposals of Locusta migratoria ( Uhlenbusch et al., 1987 ) and Drosophila yakuba ( Clary & Wolstenholm, 1985), with gaps inserted in a manner that preserved nucleotide complementarity in hydrogen-bonded stem regions ( Kjer, 1995).

Phylogenetic analysis

All maximum likelihood (ML) analyses of nucleotide data were preformed using PAUP*, version 4.0b2a (PPC) ( Swofford, 1998) and M OLPHY, version 2.3 ( Adachi & Hasegawa, 1996b), applying the Hasegawa-Kishino-Yano (HKY85) model of nucleotide substitution ( Hasegawa et al., 1985 ). The probablility rate-matrix of this model allows for unequal base frequencies and different rates for transitions and transversions; molecular calibrations shown to be applicable to our data through inference from the hierarchical hypothesis testing framework of M ODELT EST, version 2.0 ( Posada & Crandall, 1998). ML analyses of translated nucleotide sequence were performed using P UZZLE, version 4.0 ( Strimmer & von Haesler, 1997) and employed the mtREV24 model of amino acid substitution ( Adachi & Hasegawa, 1996a).

Maximum parsimony (MP) analyses of both nucleotide and protein data were performed using the heuristic algorithm of PAUP* with tree bisection-reconnection branch swapping and 10 replicates of random taxon addition. In order to improve congruence among characters found to be evolving at different rates, we applied an a priori differential weighting strategy to nucleotide sequences that consisted of downweighting transitions relative to transversions by a factor of 4 : 1.

Testing alternative phylogenetic hypotheses

In addition to inferring the most likely tree from our data, we also tested this tree against previous phylogenetic schemes. These include criteria-based estimates with complete ( Donovan et al., 2000 ; Kambhampati & Eggleton, 2000) or incomplete ( Vawter, 1991; Thorne & Carpenter, 1992; DeSalle, 1994; Kambhampati et al., 1996 ; Miller, 1997) family level resolution, as well as morphology-based classifications with complete ( Ahmad, 1950; Krishna, 1970; Roonwal, 1975; Noirot & Noirot-Timothée, 1977; Abe, 1987; Myles, 1988) or incomplete ( Hare, 1937; Noirot & Pasteels, 1988; Abe, 1991; Noirot, 1995a, b) family level resolution. To evaluate the extent to which our molecular data support alternative hypotheses represented by these studies, we estimated the variance of the difference in likelihoods between topologies derived under searches constrained to uphold the relationships reported therein (as implemented under the ‘backbone constraints’ option in PAUP*) and those derived from searches without any such constraints. The calculation of likelihood differences and the associated tests for significance were performed using the formulae of Kishino & Hasegawa (1989), as implemented in N UCML and T OTALML of M OLPHY. As an additional criterion for evaluating the relative merit of alternate topologies, we calculated the bootstrap probability of different trees using the RELL method ( Kishino et al., 1990 ; Hasegawa & Kishino, 1994) with a sample size of 104, also implemented in N UCML and T OTALML of M OLPHY.

Trait mapping and comparative analyses

To identify the number and temporal sequence of origins and losses of a worker caste we reconstructed these events over the focal tree under parsimony criteria ( Swofford & Maddison, 1987). To test for association between the occurrence of a worker caste and alternate life types we applied a Maddison’s (1990) concentrated-changes test for correlated character evolution using one-piece (1+) and intermediate or separate (1−) life types and worker-presence (W+) and worker-absence (W−) as the independent and dependent binomial variables, respectively. We grouped intermediate and separate life types into a single category as both have uncoupled feeding from nesting sites (they differ only in the extent of this uncoupling; Abe, 1987). Cockroaches, the outgroup to termites, were coded as ‘worker-absent’ and ‘one-piece’ (after Seelinger & Seelinger, 1983; Nalepa, 1984). All trait mapping and comparative analyses were performed using M ACC LADE, version 3.07 ( Maddison & Maddison, 1992).


Sequence data and alignment

All sequences used in this study are available on EMBL/GenBank databases as indicated in Table 1. In total, 612 bp of the COII gene (including indels and corresponding to positions 3098–3707 of the locust mitochondrial genome; Flook et al., 1995 ) were aligned for all 21 taxa listed in Tables 1, and 884 bp of the l-rRNA gene (including indels and corresponding to positions 12 922–13 731 of this same reference sequence) were aligned for 12 of these species. Regions of the l-rRNA alignment in which primary homology could not reliably be inferred were excluded prior to our analyses (in total 130 bp; Supplementary material).

The nucleotide compositions of the data partitions are summarized in Table 2, and compared with the g1 statistics for all partitions. For analysis, the l-rRNA alignment is considered in its entirety as a single partition, whereas the COII alignment is considered in its entirety and also as three equal-sized partitions corresponding to codon position. Each class of data thus defined shows a significant AT-bias compared to a null distribution assuming equal base proportions (χ2-test, P < 0.0001). As AT-rich genomes are common in insects ( Liu & Beckenbach, 1992; Flook et al., 1995 ), this pattern suggests that sequence divergence is often accompanied by high A-T mutation pressure either nonadaptively ( Crozier & Crozier, 1993; Jermiin et al., 1994 ; 1995) or less probably resulting from selection for the incorparation of amino acids encoded by AT-rich codon families ( Wolstenholme & Clary, 1985). In termites, the AT-bias is particularly evident at third codon positions.

Table 2.  Nucleotide composition and g1 statistics as a function of genetic partitions used in phylogenetic analyses. Thumbnail image of

Moreover, the relatively nondegenerate second codon positions have the lowest proportion of transversions, whereas third codon positions have the highest proportion; a pattern expected if substitutional constraints are operating at the amino acid level ( Moritz et al., 1987 ). The RNA-specifying gene also shows a transversion bias, although the average proportion of transversions in each partition is significantly lower than that expected if partitions had reached a point of total saturation as a result of multiple substitutions at single sites ( Table 3). However, because functional constraints will restrict the total number of possible substitutions, the expected values presented should be considered as theoretical maximums.

Table 3.  Observed proportion of transversions and their 95% confidence intervals for each data partition with corresponding expected values at saturation. Thumbnail image of

Tree-length distributions generated for each data partition were significantly skewed towards those of shorter length (g1-kurtosis statistic, P < 0.01; Hillis & Huelsenbeck, 1992), indicating substantial hierarchical structure and by inference, phylogenetic signal. The signal is stronger within the l-rRNA data (g1=0.546) than for the COII data (g1=0.394). As each partition appears to remain above mutational saturation and is shown to be phylogenetically informative, we did not exclude any data partition prior to phylogenetic analysis.

Phylogenetic analysis: translated COII data

Tests for compositional heterogeneity among whole sequences within the overall COII alignments revealed significant variation between the nucleotide compositions of eight of the 21 taxa (χ2, P < 0.05; implemented in P UZZLE). The nucleotide composition of these eight species were outside the range of two standard errors expected from random sampling and therefore violated the assumption of stationarity ( Strimmer & von Haesler, 1997). Compositional heterogeneity was corrected by re-coding this alignment into amino acid sequence (204 amino acids) according to the invertebrate mitochondrial code ( Wolstenholme & Clary, 1985) prior to phylogenetic analysis.

Figure 1 shows the quartet puzzling tree derived from the translated COII data for all 21 taxa. Branches not supported by at least 50% of puzzling replicates ( Strimmer & von Haeseler, 1996) were collapsed into polytomies. The MP analysis of this data set returned 70 equally parsimonious trees; a strict consensus of these trees produced a multifurcating tree consistent with that depicted in Fig. 1. Despite the low level of phylogenetic resolution obtained from this single-gene analysis, these data were able to identify the Kalotermitidae, the Termitidae and the Termitidae + Rhinotermitidae  + Serritermitidae as monophyletic groups. Although the Rhinotermitidae are presented here as polyphyletic, there is little support for this relationship. The likelihood score for this tree was not significantly different from one in which the Rhinotermitidae were constrained to form a monophyletic group, exclusive of the Serritermitidae, within the Serritermitidae + Rhinotermitidae + Termitidae complex ( Table 4). Although we do not reject the monophyly of Rhinotermitidae, we thus note that a conclusive branching order among termite families derived from COII amino acid data remains elusive at this point.

Figure . 1.

Phylogenetic relationships among 18 termite taxa plus the cockroach, mantid and locust based on ML analysis of translated COII sequence, applying the mtREV24 model as implemented in PUZZLE. ML support values (above branches) were obtained by 1000 puzzling replicates ( Strimmer & von Haeseler, 1996), of which values below 50% were collapsed to form polytomies. MP analysis of this same alignment returned 70 equally parsimonious trees (length 501, CI=0.617, RI=0.537); a strict consensus of which was consistent with the resolved portions of the ML tree presented. Corresponding MP support values (below branches) are superimposed on the ML topology and were obtained through 1000 bootstrap replicates ( Felsenstein, 1985a). Abbreviated family designations are as in Table 1.

Table 4.  Paired-sites test ( Kishino & Hasegawa, 1989) comparing the difference in likelihood between the unconstrained topology presented in Fig. 1 (in which Rhinotermitidae is polyphyletic) and the highest-likelihood topology in which the Rhinotermitidae were constrained to be monophyletic. Thumbnail image of

Phylogenetic analysis: combined nucleotide data

Prior to the concatenation of COII and l-rRNA nucleotide data sets, we evaluated the presence and/or extent of phylogenetic incongruence between them. This was done by reciprocal paired-sites tests ( Kishino & Hasegawa, 1989) of single-gene ML trees (as implemented in PAUP*). These tests indicated that individual data sets were not mutually incongruent and consequently we combined the two nucleotide data sets into a single alignment prior to subsequent analysis. No sequences in this alignment were significantly heterogenous with respect to nucelotide composition.

Internal branch nodes of the most likely tree derived for the combined nucleotide data set (1366 bp) appear relatively short, compared with terminal branch lengths, but ML bootstrap support values remain uniformly high across the full range of divergences ( Fig. 2). The tree shows the Isoptera as a monophyletic group separate from the outgroup taxa; the monotypic Mastotermitidae (Mastotermes darwiniensis) is placed as the earliest branching lineage within the order. The remaining families are collectively the sister group to the Mastotermitidae and are arranged into separate phyletic lines; the first consists of the Hodotermitidae and Termopsidae and the second leads to the Termitidae, with the Kalotermitidae, Serritermitidae and Rhinotermitidae separating from it. The topology presented in Fig. 2 is identical to that returned under our differentially weighed MP analysis [this tree had a length of 5396 steps, a consistency index ( Kluge & Farris, 1969) of 0.523 and a retention index ( Farris, 1989) of 0.351; not shown] which showed mixed levels of boostrap support, tending to be lower among the deeper divergences ( Fig. 2). The topology presented is consistent with the local branching arrangements presented by various authors ( Noirot, 1995a, b; Kambhampati et al., 1996 ; Miller, 1997). However, considered in its entirety, this topology differs from any previously published arrangement that has included all seven families.

Figure . 2.

Most likely tree (HKY85 model, α/β=2.491, ln L=−10705.68 ± 203.47) of 12 taxa representing all termite families derived from a combined analysis of l-rRNA and COII nucleotide data, totalling 1366 bp. The local bootstrap probabilities (in percentage) are given above each internal branch. A differentially weighed MP analysis which consisted of downweighting transitions with respect to transversions by a factor of 4 : 1, returned an identical topology; bootstrap probabilities ( Felsenstein, 1985a) are given below each branch. Lengths of horizontal lines are proportional to the estimated number of nucleotide substitutions. Family abbreviations are as in Table 1.

Testing alternative phylogenetic hypotheses

Previous hypotheses at variance with Fig. 2 are presented in Table 5 along with their relative likelihood and boostrap scores. These alternatives differ mainly in the relationship of (1) Serritermitidae to Termitidae (2) Kalotermitidae to Mastotermitidae and (3) Termopsidae to Hodotermitidae. The conditional likelihoods calculated in relation to the combined nucleotide data set favour the unconstrained tree ( Fig. 2). By comparison, the highest-likelihood topologies supporting the relationships represented in previously published alternatives –Kambhampati & Eggleton (2000), Krishna (1970), Ahmad (1950), Ampion & Quennedy (1981), Vawter (1991), Noirot & Noirot-Timothée (1977), Roonwal (1975), Myles (1988), and Abe (1987) – have significantly lower likelihoods than the most likely tree ( Table 5). The highest-likelihood topologies supporting the relationships presented by Donovan et al. (2000) and Thorne & Carpenter (1992), however, are not significantly less likely than the most likely tree ( Table 5). Yet, as the latter two alternatives lack high bootstrap support (at 15 and 6%, respectively), they remain less favourable hypotheses than that of the present study (with 75% bootstrap support).

Table 5.  Summary of trees describing evolutionary relationships among termite families (abbreviations as in Table 1). The natural likelihood value (ln L) is given for the most likely topology, that of the present study, whereas the difference from this value and its standard error (SE) is given for inferior topologies. Significant P-values indicate that the corresponding phylogenetic hypothesis can be rejected by the standard criterion Δ ln L/SE > 1.96 ( Kishino & Hasegawa, 1989), and BP indicates the boostrap probability of support calculated by the RELL method ( Kishino et al., 1990 ) for each topology. Number of evolutionary events refers to the required number of true worker caste origins and losses over each topology as inferred from parsimony mapping. Thumbnail image of

We tested the generality of the above conclusions by re-calculating the likelihood differences and bootstrap probabilities in reference to each genetic partition (as defined above) within our combined whole-gene data set. Summing partition-specific likelihoods (implemented in T OTALML of MOLPHY) allows for any interpartition heterogeneity ( Table 2) to be more effectively calibrated into the substitution model ( Adachi & Hasegawa, 1996b). When the totality of likelihoods are summed across all four partitions, the current phylogeny is favoured (BP=81.8%) and the remaining alternatives are ranked similarly. Thus, the overall evidence from analyses carried out in this work suggests that the current phylogeny ( Fig. 2) is the best estimate of the true tree from the given data currently available, and that the alternative phylogenies tested are incorrect in one or more of their arrangements. Two of the rejected hypotheses were also based on molecular sequence data – namely, Vawter (1991) and Kambhampati & Eggleton (2000). Differences between these and the present molecular analysis may stem in part from difficulties in inferring organismal trees from single-gene trees (cf. Vawter, 1991; Kambhampati & Eggleton (2000)) or low taxon sampling within the Isoptera (cf. Vawter, 1991).

Trait mapping and comparative analyses

Figure 3 shows the focal tree optimized for worker-presence and worker-absence. Ancestral character state reconstructions were identical under both ACCTRAN and DELTRAN optimization rules ( Swofford & Maddison, 1987). The most parsimonious explanation to account for the current distribution of workers is a single origin early in the history of the lineage, followed by two secondary losses. Specifically, under this scenario workers originated in the internodal branch segment leading to the ingroup node (i.e. between Blattoidea and the termites) and were lost in branch segments leading to the Kalotermitidae and the Termopsidae ( Fig. 3). A third loss occurs in the rhinotermitid genus Prorhinotermes (where workers are known to be absent; Roisin, 1988) but is not considered further in the present analysis except to say that when the Rhinotermitidae as a whole are coded as worker-absent, it does not change the monophyletic origin result.

Figure . 3.

Cladogram of termite families plus the cockroach family Blattidae showing the phylogenetic distribution of a worker caste (worker present=solid lines; worker absent=empty lines) as determined through parsimony-based optimization. The reconstruction of ancestral character states suggest that the worker caste evolved once early in the history of the lineage and was secondarily lost on two separate occasions. Specifically, workers evolved in the branch leading to the ingroup node (i.e. between Blattidae and modern termites) and were lost in the branches leading to the Kalotermitidae and the Termopsidae. This figure also serves to demonstrate the distribution of the alternate life types, with intermediate or separate types (=solid lines) evolving from a one-piece-like ancestor early in the lineages history. One-piece types sensu stricto (=empty lines) appear later in branches leading to the Kalotermitidae and the Termopsidae. Correlated transformations between alternate states of this character and workers supports the hypothesis that the variable occurrence of a worker caste in termites has an ecological basis.

Figure 3 equally serves to show the phylogenetic distribution of alternate life types. In this case, ancestral character state reconstruction suggests that intermediate or separate types evolved from a one-piece-like life type ancestor in the internodal branch segment leading to the ingroup node. A one-piece life type sensu stricto ( Abe, 1987) then evolved in branches leading to the Kalotermitidae and the Termopsidae. The probability of observing the two worker losses on branches leading to the one-piece Kalotermitidae and Termopsidae is small (P= 0.006; concentrated-change test of Maddison (1990)) under the null hypothesis that the observed gains (n=1) and losses (n=2) of workers are randomly distributed with respect to life type. Character transformations involving the loss of a worker caste are significantly more concentrated than expected by chance on those branches of the clade in which there is a transformation to one-piece life types. This result implies that the two characters have undergone correlated evolutionary change throughout the history of the order.


Branching order in termite evolution

This study is the first to employ a multigene data set and maximum likelihood test criterion toward resolving the evolutionary relationships among the principle lineages within the Isoptera. A combined-data phylogenetic analysis yielded a single topology under both ML and differentially weighed MP criteria ( Fig. 2), and although the hypothesized relationships inferred from this topology agree with certain previously proposed local arrangements ( Hare, 1937; Noirot, 1995a, b; Kambhampati et al., 1996 ; Miller, 1997), the topology differs from any published studies that have included all seven families.

These results support the long standing morphology-based hypothesis that the relictual and monotypic family Mastotermitidae is the most basal lineage within the Isoptera and is the sister taxon to all other living termites ( Krishna, 1970). The families Hodotermitidae and Termopsidae are found to be a closely related sister clade separate from the Kalotermitidae + Serritermitidae + Rhinotermitidae + Termitidae clade. The family Termopsidae is sometimes considered a subfamily of the Hodotermitidae ( Krishna, 1970), as originally proposed by Snyder (1949), but its subsequent re-classification ( Grassé, 1949) to family rank has been generally accepted ( Pearce & Waite, 1994; but see Emerson, 1965). Most of the alternative topologies tested consider these two families to be closely related through an immediate common ancestor, or through a single intermediate ancestor, with our results favouring the former arrangement.

The Kalotermitidae are found to be the most primitive family among the Kalotermitidae + Serritermitidae + Rhinotermitidae + Termitidae clade as suggested by Noirot (1995a). This family was not shown to be closely related to a basal Mastotermitidae as has traditionally been understood ( Ahmad, 1950; Krishna, 1970). More basal placements of Kalotermitidae variously represented among the alternatives, are generally not supported by these data. The family Serritermitidae was placed as immediately basel to the Rhinotermitidae + Termitidae clade and our analyses supports the elevation of this monotypic taxon (Seritermes serrifer), previously included as a subfamily of Rhinotermitidae, to full family status ( Emerson, 1965).

These results confirm the widely accepted notion that the Rhinotermitidae and Termitidae are sister lineages that occupy the most apical branch of the isopteran tree. The Termitidae are referred to as the ‘higher termites’ primarily because they harbour only bacteria in their hindguts; the ‘lower termites’ (all families excluding Termitidae) have cellulolytic protozoa as well as bacteria in their hindguts ( Krishna, 1970). We suggest, however, that because evolutionary taxonomy should represent genealogical relationships (i.e. monophyletic clades) rather than shared attributes and because the lower termites have long been recognized as a paraphyletic assemblage, that the terms ‘higher’ and ‘lower’ be abandoned in reference to termite systematics. The subjectivity inherent within these terms impart no explicit phylogenetic meaning and can be misleading if used as synonyms for ‘primitive’ and ‘advanced’ eusociality as in the Hymenoptera ( Kukuk, 1994).

A single origin of workers

The phylogenetic distribution of the worker caste obtained by reconstructing ancestral character states over the family level tree suggests a single origin of workers early in the history of the lineage, followed by two secondary losses ( Fig. 3). This scenario, which posits a temporal proximity between the origin of termite eusociality itself and the origin of a worker caste, is at variance to the notion that the worker caste evolved later in the lineage’s history and on three separate occasions ( Noirot & Pasteels, 1987; Myles & Nutting, 1988; Noirot & Pasteels, 1988; Abe, 1991; Higashi et al., 1991 ; Roisin, 1994; Thorne, 1997). Only one of the alternative topologies ( Vawter, 1991; Table 5) tested would require three separate origins of the worker caste with no losses, but this topology is rejected by our analyses and therefore is not a valid template from which to infer character state transformations.

Conclusions regarding worker origins are based upon the formal reconstruction of ancestral character states over a family level phylogeny shown to be the most likely among various alternatives ( Table 5). However, this interpretation of worker origins is robust against alternate phylogenetic hypotheses. For example, reconstructing the ancestral character states over the tree assumed in Noirot & Pasteels (1988; analysis not shown), yields a monophyletic origin early in termite history as the most parsimonious explanation. Thus, this work provides empirical evidence that suggests that the early and irreversible differentiation characteristic of a worker caste may be an ancestral feature of termite social systems and not a derived, polyphyletic one. A single origin or workers is consistent with a single origin of the other well-differentiated caste, the soldier. It is widely accepted that the soldier caste is monophyletic and that it evolved early in the lineage’s history ( Hare, 1937; Wilson, 1971; Noirot & Pasteels, 1987) (NB: soldiers are normally sterile, but can become reproductive in Zootermopsis, Termopsidae; Myles, 1986).

The hypothesis that the pseudergate caste represents the developmentally unspecialized ancestor of the worker ( Miller, 1969; Noirot, 1985b; Noirot & Pasteels, 1988; Abe, 1991; Higashi et al., 1991 ; Roisin, 1994) is also challenged. Pseudergates are present, to the exclusion of workers, among extant kalotermitid and termopsid taxa ( Noirot, 1985b) as well as in Prorhinotermes ( Roisin, 1988). The temporal sequence of character state transformations inferred from this study suggests that modern pseudergates are phylogenetically derived with respect to workers and therefore do not represent an intermediate stage from which workers differentiated. Whether workers evolved from an extinct pseudergate-like ancestor in the distant past remains a possible scenario, but it is difficult to test because of the unavailability of these ancestors.

Ecological correlates to worker evolution

Abe (1987) noted that workers occur exclusively in those species that feed outside the nest and because food-nest separation is expected to increase nest stability, he suggested that worker evolution is promoted through ecologically mediated inclusive fitness benefits related to nest stability ( Abe, 1991). Specifically, there is a critical level of nest stability above which workers will be favoured over developmentally plastic pseudergates ( Higashi et al., 1991 ). Small, short-lived, unstable societies facing environmental uncertainty are expected to favour temporary ‘workers’ that are relatively flexible in the roles they can adopt. A central prediction from this hypothesis is that nest stability (measured here as a function of life type) and workers are adaptively associated. It has not been established, however, whether the association between life type and workers that is evident among extant taxa has arisen through correlated evolution.

Standard correlative tests for association are not appropriate because of the statistical nonindependence of taxa with shared phylogenetic history ( Felsenstein, 1985b; Maddison, 1990; Harvey & Pagel, 1991). Therefore, correlative measures which take phylogenetic structure into account are desirable and accordingly we apply Maddison’s (1990) concentrated-changes test for correlated character evolution. The results indicate that character transformations involving the loss of a worker caste ( Fig. 3) are significantly more concentrated than expected by chance on those branches of the tree reconstructed to be of one-piece life type. This suggests that the association between life type and workers observed among extant taxa is of biological significance and supports existing hypotheses which suggest that changes in life type are sufficient to account for the varied phylogenetic occurrence of workers ( Abe, 1991; Higashi et al., 1991 ).

The causal association between colony life history and worker polymorphism inferred here is indicative of other correlates to termite social complexity. The social characteristics of the Kalotermitidae and Termopsidae (i.e. small colony sizes, no elaborate nest architecture, living entirely within a single piece of wood) intuitively suggests that they are ‘primitively’ eusocial ( Noirot, 1985a, b). However, the present study suggests that these characteristics may be secondarily derived. The variation in social complexity of termite societies varies among lineages and does not necessarily correspond to phylogenetic position. This realization has been advanced previously for termites ( Watson & Sewell, 1985; Shellman-Reeve, 1997) and has been discussed at greater length for other social forms (e.g. Danforth et al., 1999 ).

Towards the continued study of termite eusocial breeding systems

The lack of presocial intermediates among living termites removes one of the options for studying the evolutionary processes underlying termite eusociality. The intractability of reconstructing termite social ancestries is, at least in part, an artefact of the blanket application of the term ‘eusocial’ to all living species. Social characteristics do vary among lineages and the study of termite social evolution becomes operational by simply measuring how these characteristics vary and covary over time. This can be done by avoiding a too-narrow view of eusociality and adopting alternate criteria for quantifying differences in the degree of eusociality ( Keller & Perrin, 1995; Sherman et al., 1995 ).

Preliminary applications of this broad sense view have demonstrated that termite social systems can be modelled along a ‘eusocial continuum’ based on differences in the reproductive potential of individuals within the colony ( Shellman-Reeve, 1997). Although the application of this approach remains controversial ( Crespi & Yanega, 1995; Costa & Fitzgerald, 1996; Reeve et al., 1996 ; Wcislo, 1997), a first indication of evolutionary trends between simple and complex states of eusociality can be achieved through mapping of life history characters directly associated with eusociality (e.g. Bourke, 1999) and which reflect differences in the reproductive potential of individuals in much the same way as indices of reproductive skew are intended to do. For example, termite life history characteristics such as colony size and longevity ( Lenz, 1994), queen physogastry ( Keller, 1998) or neotenic reproductive potential ( Myles, 1999) could be optimized for clades showing variation in these characteristics. A further aspect would be the extended phenotypes ( Dawkins, 1982; Stone & Cook, 1988; Crespi & Worobey, 1998) of social complexity such as nest architecture ( Emerson, 1938) or food preference ( Abe, 1987). These characters may then be separately optimized on phylogenetic trees and the states of the interior nodes examined. In this manner, separate stages or grades in termite eusocial history could be characterized by the possession of different suites of traits.

We anticipate that from the phylogenetic framework presented here, more explicit termite phylogenies which incorporate newly established intrafamilial relationships ( Miura et al., 1998 ; Thompson et al., 2001 ) will then serve as a template against which any number of more detailed or clade-specific hypotheses concerning other aspects of their social diversification could be developed or tested.


The authors wish to thank M.A.D. Goodisman, T. Evans, L. Atkinson, B.L. Thorne, L. Keller and an anonymous reviewer for constructive comments on earlier drafts of this manuscript; B.J. Crespi, J. Shellman-Reeve, and D. Rowell for useful discussion; the Australian Research Council for research grants (RHC) and scholarship support (GJT), as well as the Natural Sciences and Engineering Research Council of Canada (GJT) and Harley Rose (NL) for additional financial support.

Supplementary material

The following material is available from

Alignment and secondary structure of l-rRNA sequences for 10 termite species and two outgroup taxa (Cockroach, Locust). Secondary structural features were used to facilitate a manual optimization of the primary alignment. To this end, the secondary structure proposal for the locust was obtained from Uhlenbusch et al. (1987) and initial alignments included that of D. yakuba ( Clary & Wolstenholm, 1985). Inferred secondary structures of termite DNA that varied from these models were examined more closely using compensatory base changes as evidence for the validity of stem structures ( Kjer, 1995). Brackets [] indicate hypothesized long-range interactions, and parentheses () hairpin stem-loops. For certain regions, structural homology among sequences could not be reliably inferred and were therefore excluded prior to our analyses (130 bp, as indicated). Bases in a stem that are likely to be paired are underlined, while bulges, loops, and regions where this could not be determined unequivocally, are not. Base positions are marked with an ^ symbol below the sequence.