Five species in the African lovebird genus Agapornis are the only parrots, other than Monk Parakeets (Myiopsitta monachus), that construct nests. Four species (A. personata, A. fischeri, A. lilianae, and A. nigrigenis) build domed nests within cavities, and a fifth (A. roseicollis) builds a cup-shaped nest within a cavity. The other members of the genus have nesting behavior that is more typical of other parrots: A. cana and A. taranta nest in cavities that are lined with nesting material, and A. pullaria excavates burrows in arboreal ant or termite nests. To reconstruct the evolution of nest-building behavior in Agapornis, I sequenced a 622-bp portion of the cytochrome-b gene (mtDNA) and used the sequence data to build a phylogenetic tree. The phylogeny shows that the divergence between the nest-building species and cana, taranta, and pullaria occurred early in the evolution of the genus. The nest builders form a monophyletic clade, and the small amount of sequence divergence between personata, fischeri, lilianae, and nigrigenis indicates that they probably should be considered subspecies of a single species. A reconstruction of the evolution of nest-building behavior on the phylogeny indicates that the construction of a domed nest is derived from the habit of lining the nest, because the nesting material is used to build progressively more complex nest structures. Within Agapornis, nest building is associated with colonial breeding. The construction of a nest within a cavity may allow breeding pairs to modify and use cavities that otherwise might be unsuitable. This would, in turn, give pairs added flexibility in nest-site choice, thereby facilitating colonial breeding. Received 5 May 1997, accepted 11 November 1997.

Department of Ecology and Evolutionary Biology, Princeton University, Princeton, New Jersey 08544, USA CAVITY NESTING has evolved multiple times among birds (Collias and Collias 1984). Be- cause cavity nests are safe and well protected, elaborate nest-building behavior is not predict- ed to evolve in cavity-nesting lineages (Collias and Collias 1984). Parrots present some excep- tions to this pattern: tree hollows probably are the primitive nest type of parrots (Eberhard 1997), yet two genera construct complex nests. Monk Parakeets (Myiopsitta rnonachus), which are native to South America, build domed stick nests, and some members of the African genus Agapornis construct domed nests within cavi- ties (Forshaw 1989). A survey of the nesting be- havior of extant parrots suggests two hypoth- eses for the evolution of nest-building behavior in the Psittaciformes: (1) nest building evolved from the habit of lining the cavity with nest ma- terial; and (2) nest building evolved via nest adoption. The genus Agapornis is an interesting group for studying the evolution of nest-building be- Present address: Smithsonian Tropical Research Institute, Unit 0948, APO AA 34002-0948. E-mail: eberharj@naos.si.edu havior in parrots because its nine species in- clude examples of most types of nesting behav- ior observed across the parrot family. Members of the genus Agapornis are small short-tailed parrots that are native to African forests and savannas. Two species (A. cana and A. taranta) nest in tree holes, and one (A. pullaria) uses cav- ities excavated in arboreal ant or termite nests (Forshaw 1989). In all three of these species, the nest cavity is lined with material (e.g. seed husks, small pieces of bark, grass or leaves) that the female carries tucked in her body feath- ers (Forshaw 1989). The nesting habits of a fourth species, A. swinderniana, are poorly known, but it may nest in termitaria as well (Forshaw 1989). Agapornis roseicollis females carry nesting material (strips of bark, leaves, or grass) tucked in their rump feathers and build cup-shaped nests within cavities. Females of the remaining four species (A. fischeri, A. per- sonata, A. lilianae, and A. nigrigenis) carry nest- ing material (long stalks and strips of bark) in their beaks and build bulky, domed nests with- in cavities (Forshaw 1989). The nesting material is woven together, and the resulting structure retains its shape even if removed from the cav- ity (Dilger 1960, Vriends 1978). TABLE 1. Taxonomic characters used by Moreau (1948: table 5) in classifying species in the genus Agapornis. Characters in italics are shared characters. Character Group A n A. roseicollis Group B b Natal down White Red Red Distinct juvenal plumage Present Present Absent White skin around eye Absent Absent Present Sexual dimorphism Present Absent Absent Black bar on central tail feathers Present Absent Absent Method of carrying nest material In feathers In feathers In beak Form of nest Pad Cup Dome A. cana, A. pullaria, A. taranta. b A. fischeri, A. personata, A. lilianae, A. nigrigenis. The first comprehensive attempt to under- stand the evolution of the genus Agapornis was made by Moreau (1948), who summarized what was known about the distribution, ecol- ogy, morphology, and behavior of lovebirds. Using all of this information, he proposed a classification of the genus. Moreau's (1948) grouping of Agapornis is based on seven mor- phological and behavioral characters (see Table 1). The designation of some as "primitive" and others as "advanced" results from the selection of Loriculus as the most closely related genus, because members of the latter genus also carry nesting material in their feathers and are sex- ually dimorphic. A. roseicollis appears to be in- termediate with respect to the "primitive" (Group A) and "advanced" (Group B) species, because it shares three characters with each group. Placing it as a phylogenetic intermedi- ate would indicate that nest-construction be- havior evolved from the habit of lining the nest, reflecting a gradual elaboration of nest build- ing. However, Moreau (1948:236) suggests that A. roseicollis and the Group B species are dif- ferent lineages: "... it is curious that of all the Group B birds, A. nigrigenis, the member that is geographically nearest to A. roseicollis, should be most unlike it. This suggests that, although A. roseicollis is intermediate in characters, Group B evolved independently." Most of our knowledge of the behavior of lovebirds is a result of detailed observations of captive individuals made by Dilger (1960, 1961). His studies included all of the species in Agapornis except for A. swinderniana, and they provide descriptions of breeding and social be- havior. Based on his behavioral observations, Dilger (1960, 1961) characterized cana, taranta, and pullaria as "primitive," and fischeri, person- ata, lilianae, and nigrigenis as the most "highly evolved." The behavior of the remaining spe- cies, roseicollis, appeared intermediate, leading Dilger to conclude that it arose from the "prim- itive" species and is ancestral to the "highly evolved" species (Dilger 1960, 1961). Dilger concurred with Neunzig (1926) and Hampe (1957) that fischeri, personata, lilianae, and nigri- genis probably are subspecies of one species. He also suggested that taranta and pullaria are more closely related to each other than to other species in the genus, and that they are the clos- est relatives of cana, which is endemic to Mad- agascar. Although he generally agreed with Moreau's arrangement, Dilger (1960:650) sug- gested that Group B species are most closely re- lated to A. roseicollis and probably were "de- rived from a roseicollis-like ancestor" In order to determine the relationships among Agapornis species, and to reconstruct the evolutionary history of nest building in the genus to test the nest-lining hypothesis, I used mtDNA sequence data to infer a phylogeny for the group. The choice of mtDNA data for phy- Iogeny reconstruction was made for two main reasons: (1) morphological data have proven to be of limited taxonomic value in parrots (Smith 1975, Forshaw 1989); and (2) some of the char- acters that have been used to define the "prim- itive" and "advanced" species of Agapornis (Moreau 1948; Dilger 1960, 1961) are associated with nest building and would introduce an in- appropriate bias (de Queiroz 1996) to this test of the nest-lining hypothesis. METHODS Blood feathers were used as sources of DNA for most polymerase chain reactions (PCR) and sequenc- ing reactions; DNA from one of the A. personata sam- ples was extracted from buffered blood. Feather sam- TABLE 2. Feather samples used as sources of DNA for PCR and sequencing. Source Band or ID no. Agapornis cana J. Landvater 17L-R-96-53-ALBS J. Landvater 17L-R-96-54-ALBS A. taranta San Diego Zoo 391108-7EWR-3 San Diego Zoo 495050-SDZ15 A. pullaria San Diego Zoo 393091-1GT-846 San Diego Zoo 393105-817 A. roseicollis San Francisco Zoo 197-288912 Dickerson Park Zoo -- A. fischeri Dallas Zoo DZST 053 Dickerson Park Zoo 3608 A. personata Fort Wayne Zoo -- Dickerson Park Zoo 3507 D. Emlen/K. Bright -- A. nigrigenis San Diego Zoo 391505-0127 San Diego Zoo 391513-002 A. lilianae Brookfield Zoo 24152 Brookfield Zoo 25100 Loriculus galgulus San Diego Zoo CEM 50 San Diego Zoo CEM 51 pies from eight of the nine species of Agapornis and one species of Loriculus (the Blue-crowned Hanging- Parrot [Loriculus galgulus]) were taken from captive birds in zoos in the United States and, in one case, from an aviculturist's collection. The A. personata blood sample was taken from a pet bird. A list of samples used in this study is provided in Table 2. DNA sequences were obtained from at least two in- dividuals of each species in order to check for intra- specific sequence variation. Within-species variation might be expected in the white eye ring taxa, because they are known to hybridize in captivity (Moreau 1948, Vriends 1978). In most cases, these individuals were unrelated, and where possible from different zoos; the only exception was A. cana, for which the samples were from full sibs (and therefore expected to be identical due to the maternal inheritance of mtDNA). The feather samples from each individual were stored in separate bags or envelopes at room temperature. All extractions were done within two months of sample collection. I was unable to locate any zoos or aviculturists in the United States, Eu- rope, or Africa who keep A. swinderniana. Because the extraction of ancient DNA from museum skins and the designing of primers necessary for ampli- fying such DNA were beyond the scope of this study, A. swinderniana was not included in the analyses. DNA extraction from blood feathers was done ac- cording to a modified hair lysis buffer extraction protocol (P. Wade pers. comm.). A small (<2 mg) piece of the feather tip was immersed in 400 p,L of a hair lysis buffer (0.9% Tween 20, 10 mM Tris pH 8.0, 50 p,g/mL Proteinase K, 35 mM DTT) and incubated overnight at 56C. Following the addition of 500 of 5% Chelex, the samples were incubated at 95 C for 30 min, cooled to room temperature, and then mi- crofuged for one minute at maximum speed. Finally, 350 p,L of the supernatant were transferred to a new tube and stored at -20C. The supernatant was used in amplification reactions without dilution or further purification. For all sets of extractions (usually three to six samples each), two negative extraction controls also were performed. Extraction of DNA from the single buffered blood sample was done by first in- cubating the sample with SSC, TNE, and Protease K, followed by a standard phenol/chloroform extrac- tion and dialysis. PCR amplification (Saiki et al. 1988) of a portion of the cytochrome-b gene (mtDNA) was done using primers L14841 (Kocher et al. 1989) and CB3-H (Pal- umbi et al. 1991). Amplifications were done using 1 p,L of extraction supernatant as template in 10 p,L to- tal-volume PCRs that contained a buffer (10 mM Tris/HC1, 50 mM KC1, 2.5 mM MgC12, pH 8.3), each dNTP at 0.2 mM, each primer at 0.2 p,M, and 0.5 units of Thermus aquaticus polymerase. The first cycle of the PCR reaction consisted of denaturation for 3 min at 94 C, annealing for 2 min at 50C, and extension for 3 min at 72 C. This was followed by 35 cycles of de- naturation for 30 s at 94C, annealing for 1 min at 50 C, and extension for 1.5 min at 72C. The reaction was completed with a single cycle of denaturation for 30 s at 94C, annealing for 1 min at 50 C, and exten- sion for 4 min at 72C. With each set of PCR reac- tions, a negative control was included to permit the detection of any contamination of reagents. Electro- phoresis of PCR products in a 0.4% agarose mini-gel (TAE buffer), followed by staining in ethidium bro- mide (EtBr), was done to check for amplification. The PCR product was then diluted (up to a 1:10 dilution depending on the concentration of product as esti- mated by the EtBr staining) and used as template for sequencing reactions. Sequencing of most samples was done using thef- mol cycle sequencing kit (Promega), using end-la- beled 32p, according to the manufacturer's instruc- tions. Both strands of the fragment between L14841 and H15149 were sequenced for an A. personata sam- ple and found to match without any discrepancies. For all of the other samples, the 622-bp fragment was sequenced in two sections: approximately half of the region was sequenced using primer H15149 (Kocher et al. 1989), and the other half with primer CB3-H. The A. cana samples were sequenced by the auto- matic-sequencing facility at Princeton University. Se- quences were easily aligned by eye to each other, and to the Night Parrot (Geopsittacus occidentalis) se- quence published in Leeton et al. (1994); no inser- tions or deletions were found. Maximum-parsimony tree searches were done us- ing test version 4.0.0d59 of PAUP* (provided by D. L. Swofford). Maximum-parsimony trees were found using the exhaustive search option, and bootstrap and jackknife resampling with random branch ad- dition were implemented on all trees (2,000 repli- cates) to assess the resolving power of the data. Trees were rooted with both the Loriculus galgulus and the Geopsittacus occidentalis sequences (Leeton et al. 1994). Loriculus is thought to be a sister genus to Aga- pornis (Dilger 1960); Geopsittacus is Australian and a member of a different tribe (Forshaw 1989) and thus is more distantly related. I conducted maximum-par- simony searches with equal weighting of all base po- sitions, and also with codon weights of 3:10:1 (lst: 2nd:3rd position weights, reflecting the relative number of variable sites at each position). These weightings were calculated from the numbers of base changes occurring at each codon position in the data set. Tree statistics (treelength, and consistency and retention indices) for parsimony trees were cal- culated using MacClade (Maddison and Maddison 1992), with uninformative characters excluded. Be- cause of the difficulties with calculating treelength for trees that contain polytomies of uncertain reso- lution ("soft" polytomies), as did the trees in this study, it should be noted that the treelengths are not exact and are likely to be underestimates (Maddison and Maddison 1992). PAUP* also was used to com- pare alternative tree topologies by performing Kish- ino-Hasegawa tests (Kishino and Hasegawa 1989). Maximum-likelihood trees were reconstructed us- ing the quartet-puzzling method (Strimmer and von Haeseler 1996) implemented in the PUZZLE pro- gram (Strimmer and von Haeseler 1997), with 1,000 puzzling steps. I used the Hasegawa-Kishino-Yano (1985; HKY) substitution model with the transition/ transversion parameter and nucleotide frequencies estimated from the data (exact parameter estimation option). Trees were found assuming uniform substi- tution rates and Gamma distributed rates. The reli- ability values generated by PUZZLE reflect the num- ber of times a given group is reconstructed during puzzling steps and are highly correlated with boot- strap values (Strimmer and von Haeseler 1996). I calculated pairwise distances among sequences using: (1) the uncorrected percentage sequence di- vergence, with all positions given equal weight and no substitution model assumed; and (2) the maxi- mum-likelihood distance, which was found using program DNAML in PHYLIP 3.5 (Felsenstein 1993). The latter distances were calculated to estimate di- vergence times using the crane calibration of Kra- jewski and King (1996); a rate calibration is not avail- able for parrots. To test whether the Agapornis data were consistent with a molecular clock (i.e. whether use of the above calibration was valid), a maximum- likelihood tree search was done using PAUP* (equal- ly weighted data, HKY substitution model, transi- tion/transversion ratio = 2), with and without en- forcing a molecular clock. I then used a likelihood- ratio test (Felsenstein 1988, 1993) to compare the like- lihood scores of the two trees; no statistical differ- ence between the likelihood scores would indicate that enforcement of a molecular clock is compatible with the data. RESULTS The sequences of the eight Agapornis species and of Loriculus galgulus are deposited in GenBank under accession numbers AF001324 to AF001332. Of the 622 bases included in the cytochrome-b fragment that was sequenced, 169 positions (27%) were variable, and 96 (15%) were parsimony-informative. Of these variable sites, 123 (73%) were located at third-codonpo- sitions; 33 (19%) and 13 (8%) occurred at the first and second positions, respectively. Be- cause there was no intraspecific sequence vari- ation, I used only a single sample per species for tree building. Uncorrected sequence diver- gence between Agapornis species ranged from 0.5% (nigrigenis-lilianae and personata-fischeri) to 12.7% (nigrigenis-cana; see Table 3), and dif- ferences between Agapornis and L. galgulus ranged from 13.3% (L. galgulus-A. pullaria) to 15.3% (L. galgulus-A. taranta). Pairwise numbers of transitions (Ti) and transversions (Tv) showed a bias in favor of transitional changes. When all codon positions were considered, Ti:Tv ratios within the Lori- culus/Agapornis data set ranged from 6.00 (A. personata-A. lilianae) down to 1.14 (L. galgulus- A. roseicollis). If Geopsittacus was included in the comparisons, ratios were as low as 0.93 (Geop- sittacus-A. roseicollis). This indicates that the deeper comparisons among these taxa will be affected by saturation due to multiple hits. Plots of transitions and transversions show that for this dataset, first position Ti and Tv increase linearly with increasing overall sequence di- vergence, without showing signs of saturation (Fig. 1A). However, at the third position, Ti changes reach saturation for comparisons of TABLE 3. Pairwise distances between taxa over a 622-bp fragment of the cytochrome-b gene. Above the di- agonal, uncorrected divergence; below the diagonal, maximum-likelihood distances calculated using the DNAML program in PHYLIP 3.5 (Felsenstein 1993). 1 2 3 4 5 6 7 8 9 10 1 L. galgulus -- 0.1367 0.1334 0.1527 0.1383 0.1383 0.1447 0.1350 0.1399 0.1672 2 A. personata 0.1864 -- 0.1061 0.1061 0.0096 0.0048 0.1579 0.0113 0.1206 0.1752 3 A. pullaria 0.1796 0.1313 -- 0.0981 0.1109 0.1077 0.1061 0.1093 0.1141 0.1672 4 A. taranta 0.1946 0.1463 0.1070 -- 0.1141 0.1061 0.1109 0.1125 0.1206 0.1704 5 A. nigrigenis 0.1917 0.0098 0.1366 0.1517 -- 0.0113 0.0627 0.0048 0.1270 0.1801 6 A. fischeri 0.1880 0.0048 0.1329 0.1480 0.0115 -- 0.0595 0.0129 0.1238 0.1768 7 A. roseicollis 0.1876 0.0606 0.1324 0.1475 0.0660 0.0623 -- 0.0611 0.1238 0.1736 8 A. lilianae 0.1934 0.0115 0.1383 0.1534 0.0048 0.0131 0.0676 -- 0.1254 0.1768 9 A. cana 0.1739 0.1475 0.1408 0.1558 0.1528 0.1491 0.1486 0.1545 -- 0.1865 10 Geopsittacus 0.1995 0.2367 0.2297 0.2449 0.2420 0.2383 0.2379 0.2437 0.2242 -- taxa whose sequences differ by more than. ap- proximately 10% (Fig. lB). Therefore, down- weighting third positions for parsimony anal- yses (as described in the Methods), or using a likelihood model that incorporates rate hetero- geneity, should better reflect the phylogenetic information content of changes at the different codon positions. The disproportionately high number of changes occurring at the third po- sition indicates that the region amplified and sequenced probably is in the cytochrome-b cod- ing region (as intended) and is not a nuclear pseudogene (Arctander 1995). This is further supported by the fact that there were no inser- tions or deletions in the sequences relative to other known sequences (e.g. Leeton et al. 1994), and that when translated to an amino-acid se- quence, no stop or nonsense codons were found. An exhaustive parsimony search found four equally parsimonious trees, and a strict con- sensus of these was identical to the bootstrap tree shown in Figure 2A, except that taranta and pullaria formed a clade. Parsimony analysis of the weighted data produced a tree of similar to- pology but of improved resolution and better bootstrap support (Fig. 2B). Maximum-likeli- hood trees, reconstructed assuming either uni- form or Gamma distributed rates, had the same topologies but differed in their likelihood scores (uniform rate: In L = -2324.94; Gamma distributed rate: in L = -2258.31). The trees differed only slightly from the parsimony tree in Figure 2A in their grouping of the white eye ring group: personata andfischeri were paired in the likelihood tree (66 and 54% reliability val- ues for uniform and Gamma distributed rates, respectively) and placed as a sister group to the nigrigenis-lilianae clade found in the parsimony tree (98 and 85% reliability values). Support for the cana-taranta clade was better (87% reli- ability value) when uniform rates were as- sumed than when a Gamma distribution was used (50% reliability value), but support for the roseicollis and the white eye ring group nodes was 100% under both likelihood models. In all of the tree searches that were done, the best trees grouped the white eye ring forms as a clade, with roseicollis at its base. Bootstrap and jackknife analyses, as well as quartet puz- zling, showed that the white eye ring clade and the position of roseicollis are well supported us- ing both parsimony and likelihood algorithms, with bootstrap/jackknife and reliability values of 99 to 100%. Also consistent was the close as- sociation between nigrigenis and lilianae, which was supported with bootstrap/jackknife and reliability values of at least 85%. All trees had cana, taranta, and pullaria at their bases but dif- fered in their placements relative to each other Rooting trees with Loriculus and Geopsittacus produced nearly the same topology within Agapornis. Using equally weighted data, the use of Loriculus as an outgroup provided better res- olution within the Agapornis clade, but the in- clusion of Geopsittacus improved the resolution when weighted data were used in the analysis. This is not surprising, because sequence differ- ences between Geopsittacus and the other spe- cies used in this analysis range from 16.7 to 18.7% (Table 3), and so third-codon position changes are likely to be affected by saturation (see Fig. 1). The basal position of cana, which is endemic to Madagascar, suggests that it diverged from its congeners early in the history of the genus. A molecular "clock" calibration for cranes (Krajewski and King 1996), which is based on a 20 15 10 5 0 0 [] [] [] [] 0.05 [] [] [] 0.1 0.15 [] [] 0.2 b 10- 0 , 0 0.05 0.1 0.15 0.2 Sequence divergence Fc. 1. Numbers of transition (Ti) and transver- sion (Tv) substitutions plotted against total sequence divergence for all species pairs included in Table 3. Shown are plots of Ti and Tv occurring at codon first positions (A) and third positions (B). taranta and cana-pullaria distances is 0.1483, so their distance from a common ancestor is esti- mated to be 0.0742. According to the crane cal- ibration, this indicates that cana began diverg- ing from its mainland relatives approximately 4.4 to 10.6 million years ago (MYA). All of the trees generated from the sequence data are consistent with Dilger's (1960, 1961) grouping and with the characters used by Mo- reau (1948) to divide the genus into two groups. The phylogenies lend some support to Moreau's suggestion that roseicollis belongs to a lineage that is separate from the Group B spe- cies, but they also show that roseicollis is most closely related to the white eye ring forms, and that they all shared a relatively recent common ancestor. All trees place roseicollis and the white eye ring forms in a monophyletic clade, with ro- seicollis ancestral to the others. This segment of the tree also is the most robust, as indicated by the consistently high bootstrap values. When nest types are mapped onto the tree, the construction of a cup nest appears on the tree with roseicollis, immediately preceding the origin of domed nest building (Fig. 2). Use of nest lining is ancestral in the group, and bur- rowing in arboreal termite nests is derived from nesting in a lined cavity. Alternative branching patterns, in which roseicollis was moved to different sections of a maximum-like- lihood tree, were compared to the optimal tree using Kishino-Hasegawa tests. Forcing roseicol- lis to fall within the white eye ring clade re- suited in a log-likelihood score that was signif- icantly worse than that of the optimal tree (Kishino-Hasegawa test, P = 0.0043). Placing roseicollis in the taranta/pullaria clade also was a significantly unlikely arrangement given the data (Kishino-Hasegawa test, P = 0.0001). maximum-likelihood distances, was used to es- timate the date of cana's divergence from its closest relatives, taranta and pullaria. The com- parison of likelihood scores for trees built with and without a molecular clock (In L = -2328.65 and In L = -2324.98, respectively) shows that they did not differ significantly (likelihood-ra- tio test, 0.5 < P < 0.9), indicating that a molec- ular clock holds for these data. According to the crane calibration, the maximum rate of cyto- chrome-b sequence divergence (using maxi- mum-likelihood distances; Table 3) is 0.7 to 1.7% per million years. The mean of the cana- DISCUSSION The phylogenetic hypothesis presented here is consistent with the arrangement proposed by Dilger (1960). The white eye ring forms (per- sonata, fischeri, lilianae, and nigrigenis) and ro- seicollis clearly compose a monophyletic clade. Agapornis roseicollis is at the base of the clade, and thus shared a common ancestor with the other four species. The percentage of sequence divergence between personata, fischeri, lilianae, and nigrigenis is much lower (range 0.5 to 1.1%) than the divergence between other species in 100/100 85/86rA. nigrigenis 100/99 nigrigenis D 93/67 A. [.A. lilianae A. li#anae D lOO/lOO lOO/lOO A. fisched A. fischori D A. roseicollis s$/6 A. roseicolls C A. taranta 69/66 r A. taranta LH A. pullaria t-----A. pullada AT A. cana A. cana LH Lorulus Lodculus LH Geopsittacus Geopsittacus FIG. 2. Maximum-parsimony bootstrap trees found using PAUP* (test version 4.0.0d59, provided by D. L. Swofford). Numbers above branches are bootstrap values (2,000 replicates), followed by jackknife values (2,000 replicates). For each tree, the treelength (see text), consistency index (CI), and retention index (RI) are given. All tree statistics were calculated with uninformative characters excluded. (A) Maximum-parsimony tree found using equally weighted data. Treelength = 173, CI = 0.72, and RI = 0.75. (B) Maximum-parsimony tree found using weighted data (see text). Treelength = 328, CI = 0.67, and RI = 0.70. Letters next to species names indicate the type of nest used or built by that taxon: LH (lined hole), C (cup nest), D (domed nest), or AT (burrow in arboreal termitarium). the genus (range 6.0 to 12.7%), indicating that these four species should be considered sub- species of A. personata. This supports the grouping advocated by Dilger (1960). Moreau (1948) also recognized the close relationship among the white eye ring forms, but he did not recommend combining them into one species. The slightly closer relationship between ni- grigenis and lilianae was first noted by Moreau (1948), and this relationship is supported by the mtDNA data presented here. As Moreau (1948) pointed out, the separation between the nigrigenis / lilianae and personata / fischeri clades probably was caused by mountain building at the head of Lake Nyasa. The northern end of the personata/fischeri division coincides with the location of the Rift Valley, which may be re- sponsible for their separation; however, at the southern end of the division, no major geo- graphic barriers separate the members within the two species pairs (Moreau 1948). Moreau (1948) suggested that the separation is main- tained by regions of "miyombo" woodland, and, in the case of nigrigenis and lilianae, by a region of high elevation. The mtDNA phylog- enies also show a close relationship between taranta and pullaria, which was noted by both Moreau (1948) and Dilger (1960). The basal position of cana indicates that it probably became isolated on Madagascar rel- atively early in the course of diversification of the genus Agapornis. However, the amount of sequence divergence between cana and its clos- est mainland relatives (taranta and pullaria) suggests that colonization of Madagascar oc- curred after the island's separation from main- land Africa. According to a molecular clock cal- ibration for cranes (Krajewski and King 1996), cana is estimated to have been diverging from other members of its genus for about 5 to 10 million years. Madagascar completed its sepa- ration from Africa some 121 MYA (Rabinowitz et al. 1983), long before cana began to diverge from other members of its genus. This indicates that this species did not diverge due to vicari- ance, but colonized an already insular Mada- gascar and thereafter was isolated from its con- geners. When the nest-type character is mapped onto the phylogeny (Fig. 2B), a reconstruction of the evolution of nest building in Agapornis supports the nest-lining hypothesis (see Intro- duction). In the genus, the habit of lining the nest cavity is ancestral to the use of nesting ma- terial for construction of a nest. Also, the four species that build domed nests are the most re- cently derived in the group and shared a com- mon ancestor with roseicollis, which builds a simpler cup nest. This is consistent with the nest-lining hypothesis, which proposes that the construction of a domed nest evolved as the nest material initially used to line the nest cav- ity was used to construct progressively more complex nests. Another possibility that is equally parsimonious given these data is that cup nests and domed nests evolved separately. Alternative topologies that would lend less support to the nest-lining hypothesis are not likely given the sequence data. For example, placing roseicollis within the white eye ring group, which would not reflect the gradual elaboration of nest building predicted by the nest-lining hypothesis, is significantly unlike- ly. If this arrangement were correct, it would suggest that the cup nest built by roseicollis is derived from the domed nest. Putting roseicollis in the cana / taranta / pullaria clade, which would imply two clearly independent origins of nest building and again would fail to reflect the gradual elaboration of nest building, also is in- consistent with the sequence data. It should be noted that within Agapornis, there is a correlation between nest adoption and nest construction: both roseicollis and lili- anae frequently use the nests of colonial weaver birds (Forshaw 1989), and fischeri and personata appear to use other birds' nests (Moreau 1948). This is consistent with the nest-adoption hy- pothesis, but the data more strongly support the nest-lining hypothesis by indicating a pro- gression from nest-lining behavior toward more elaborate building behavior. The ob- served correlation between nest construction and nest adoption in some Agapornis could oc- cur if a bird that constructs a nest might be more likely to recognize and accept the nest of other species as a breeding site. The construction of a nest within a cavity may be advantageous because it allows birds to modify cavities that might otherwise be un- suitable for breeding (Eberhard 1997). Obser- vations of captive A. personata indicate that nest-construction behavior is flexible, and that a function of nest building is the modification of cavities (Vriends 1978). Alternatively, love- birds would be expected to build a nest that was essentially the same, regardless of the size or shape of the cavity. Vriends (1978) found that if the nest-box opening is too large, birds sometimes pile nesting material up against it to make the entrance smaller. He also noted that the extent of nest construction depends on the size of the cavity--if space is limited, the dome may be omitted. Vriends also observed that fe- males will add extra material to their nests if light leaks into the nest chamber, or if there are any sharp projections in the nest interior. An important difference between the Agapor- nis species that construct nests and those that do not is breeding density. Those that build nests (i.e. the white eye ring forms and rosei- collis) are colonial whereas the others are sol- itary breeders. In captivity, breeding pairs of cana and taranta are extremely territorial and must be kept in separate cages (Vriends 1978, Erhart 1983). Gregarious breeding may have been facilitated by the evolution of nest build- ing, because the ability to construct a nest could give breeding pairs flexibility in their choice of nesting sites (Eberhard 1997). An as- sociation between increasingly complex nest construction and gregarious nesting also has been shown in a phylogenetic analysis of the evolution of nest construction in swallows (Winkler and Sheldon 1993). In the case of swallows, Winkler and Sheldon suggest that the transition from cup nests to domed nests permitted higher breeding densities by reduc- ing forced extrapair copulation attempts. In Agapornis, the shift from sexual dimorphism (in the "primitive" species) to sexual monomor- phism (in roseicollis and the white eye ring forms) may be a result of the shift to colonial breeding (Dilger 1960), perhaps because of nonsexual social selection on plumage involved in signaling (which is likely to occur more fre- quently in a colony) by both sexes (West-Eber- hard 1983). A complete assessment of the adaptive value of nest building and colonial breeding clearly is an ecological question that involves knowl- edge about the habitat in which these species live (and have evolved), the availability of nest cavities, and quantification of the costs and benefits associated with gregarious breeding. Field studies of Agapornis are lacking and are necessary to evaluate the hypotheses proposed here regarding the adaptive value and evolu- tion of nest-building behavior in the genus. In particular, data on the availability and distri- bution of nest cavities, and the lovebirds' pat- tern of occupation of available cavities, could be used to test the hypothesis that the construc- tion of nests within cavities facilitates gregari- ous breeding. The phylogenetic data presented here sup- port a historical reconstruction of the evolution of nest-building behavior in which the con- struction of a nest within a cavity is derived from the habit of lining the nest. Several of the differences traditionally used to distinguish the nest-building species from the "primitive" species probably are linked with the change in nesting behavior. One of these is the method of carrying nesting material--the change from carrying material in the feathers to carrying it in the beak permitted transport of larger and/or heavier items. Plumage changes may have fol- lowed changes in nest-building behavior, be- cause the switch from dimorphism to mono- morphism probably is linked to the shift from solitary to gregarious breeding. It is likely that gregarious breeding itself was facilitated by the ability to construct nest structures within cav- ities, because the ability to modify unsuitable cavities would give breeding pairs increased flexibility in nest placement. ACKNOWLEDGMENTS I thank H. Hollocher for providing me with a lab in which to work; she and D. Kutzler patiently taught me PCR and sequencing techniques. P. Wade provid- ed an easy and reliable feather-extraction protocol; J. Rowden and K. Petren kindly gave me the primers that were used for PCR and sequencing. Feather sam- ples were provided by a number of people: C. Fal- zone (Brookfield Zoo), M. Mitchell (Dickerson Park Zoo), J. Landvater and K. Raby (San Francisco Zoo), D. Rimlinger (San Diego Zoo), C. Siegel (Dallas Zoo), M. Weldon (Fort Wayne Zoo), and D. Emlen and K. Bright; A. Lyles helped me to locate zoos with Aga- pornis and Loriculus in their collections. While work- ing on this project, I was supported by Princeton University; funding for the laboratory work was pro- vided by the Association for Women in Science Educational Foundation and Princeton University. The manuscript benefitted from the comments of P. Grant, H. Hollocher, A. Dobson, H. Horn, M. Bjvrklund, and R. Gray. LITERATURE CITED ARCTANDER, P. 1995. Comparison of a mitochondri- al gene and a corresponding nuclear pseudo- gene. Proceedings of the Royal Society of Lon- don Series B 262:13-19. COLL1AS, N. E., AND E. C. COLLIAS. 1984. Nest build- ing and bird behavior. Princeton University Press, Princeton, New Jersey. DE QUEIROZ, K. 1996. Including the characters of in- terest during tree reconstruction and the prob- lems of circularity and bias in studies of char- acter evolution. American Naturalist 148:700- 708. DILGER, W. C. 1960. The comparative ethology of the African parrot genus Agapornis. Zeitschrift ffir Tierpsychologie 17:649-685. D1LGER W. C. 1961. The behavior of lovebirds. Sci- entific American 206:88-98. EBERHARD, J. R. 1997. The evolution of nest-building and breeding behavior in parrots. Ph.D. disser- tation, Princeton University, Princeton, New Jer- sey. ERHART, R. R. 1983. Breeding lovebirds in cage or aviary. Watchbird 10(2):18-21. FELSENSTE1N, J. 1988. Phylogenies from molecular sequences: Inference and reliability. Annual Re- view of Genetics 22:521-565. FELSENSTE1N, J. 1993. PHYLIP (Phylogenetic Infer- ence Package) 3.5c. Distributed by the author. University of Washington, Seattle. FORSHAW, J. M. 1989. Parrots of the world, 3rd ed. Landsdowne Editions, Sydney, Australia. HAMPE, H. 1957. Die Unzertrennlichen. Pfungstadt, Germany. HASEGAWA, M., H. K1SH1NO, AND t. YANO. 1985. Dating the human-ape splitting by a molecular clock of mitochondrial DNA. Journal of Molec- ular Evolution 22:160-174. K1SH1NO, H., AND g. HASEGAWA. 1989. Evaluation of the maximum likelihood estimate of the evo- lutionary tree topologies from DNA sequence data, and the branching order in Hominoidea. Journal of Molecular Evolution 29:170-179. KOCHER, T. D., W. K. THOMAS, A. MEYER, S. V. ED- WARDS, S. PBO, AND E X. V1LLABLANCA. 1989. Dynamics of mitochondrial DNA evolution in animals: Amplification and sequencing with conserved primers. Proceedings of the National Academy of Sciences USA 86:6196-6200. KRAJEWSKI, C., AND D. G. KING. 1996. Molecular di- vergence and phylogeny: Rates and patterns of cytochrome b evolution in cranes. Molecular Bi- ology and Evolution 13:21-30. LEETON, P. R. J., L. CHRISTIDIS, M. WESTERMAN, AND W. E. BOLES. 1994. Molecular phylogenetic af- finities of the Night Parrot (Geopsittacus occiden- talis) and the Ground Parrot (Pezoporus wallicus). Auk 111:833-843. MADD1SON, W. P., AND D. R. MADD1SON. 1992. MacClade: Analysis of phylogeny and character evolution, version 3.0. Sinauer Associates, Sun- derland, Massachusetts. MOREAU, R. E. 1948. Aspects of evolution in the par- rot genus Agapornis. Ibis 90:206-239. NEUNZ1G, R. 1926. Zur Systematik und Biologie der Gattung Agapornis. Verhandlungen der Orni- thologischen Gesellschaft in Bayern 17:112. PALUMB1, S., A. MARTIN, S. ROMANO, W. O. MCMIL- LAN, L. STICE, AND G. GRABOWSKI. 1991. The simple fool's guide to PCR, v. 2.0. University of Hawaii, Honolulu. RAB1NOWITZ, P. D., M. E COFFIN, AND D. FALVEY. 1983. The separation of Madagascar and Africa. Science 220:67-69. SAIKI, R. K., D. H. GELFAND, S. STOFFEL, R. HIGUCHI, G. T. HORN, K. B. MULLIS, AND H. A. ERLICH. 1988. Primer-directed enzymatic amplification of DNA with a thermostable DNA polymerase. Science 239:487-491. S4IT4, G. A. 1975. Systematics of parrots. Ibis 117: 18-68. STRIMMER, K., AND A. VON HAESELER. 1996. Quartet puzzling: A quartet maximum-likelihood meth- od for reconstructing tree topologies. Molecular Biology and Evolution 13:964-969. STRIMMER, K., AND A. VON HAESELER. 1997. PUZ- ZLE: Maximum likelihood analysis for nucleo- tide and amino acid alignments. Zoologisches Institut, Universitaet Muenchen, Muenchen, Germany. VRIENDS, M. M. 1978. Encyclopedia of lovebirds and other dwarf parrots. TFH Publications, Neptune City, New Jersey. WEST-EBERHARD, M. J. 1983. Sexual selection, social competition, and speciation. Quarterly Review of Biology 58:155-183. WINKLER, D. W., AND E H. SHELDON. 1993. Evolu- tion of nest construction in swallows (Hirundin- idae): A molecular phylogenetic perspective. Proceedings of the National Academy of Sci- ences USA 90:5705-5707. Associate Editor: A. J. Baker