Genetic structure of lions (Panthera leo L.) in the Selous Game Reserve: implications for the evolution of sociality


G. Spong, Department of Animal Ecology, Evolutionary Biology Centre, Uppsala University, Norbyvägen 18D, 752 36 Uppsala, Sweden. Tel.: +46 18 471 26 34; fax: +46 18 471 64 84; e-mail:
Present address: Department of Zoology, Cambridge University, Downing Street, Cambridge, CB2 3EJ, UK.


Abstract We use 14 microsatellites to examine the genetic structure of a lion (Panthera leo L.) population in southern Tanzania. Heterozygosity levels were high (0.75 ± 0.08). Relatedness estimates showed that prides often had close relatives in neighbouring prides, whereas few relatives were found in prides not sharing a border. The drop-off in relatedness with distance was highly significant. Female pridemates exhibited a higher mean relatedness (0.26 ± 0.07) to one another than did pride males (0.11 ± 0.07). Mean relatedness among females was significantly higher in small prides than in large ones. Prides exhibited significant inbreeding avoidance (FIL: −0.11). Mating did not detectably differ from random across prides (FIT: −0.02 ns). In addition to being recognizable behavioural and demographic units, prides were statistically distinct genetic units (FLT: 0.07). Some neighbouring prides grouped together both geographically and genetically, forming ‘superprides’ in the population (FZT = 0.05). Thus, although individual prides were genetically distinct, there was an important genetic structure above the level of social groups.


Social systems can have profound effects on genetic structure (e.g. Sugg et al., 1996; Storz, 1999). For example, patterns of mating and dispersal are of major importance in determining the distribution of genetic variation in a population (Chesser, 1991; Dobson et al., 1997; Dieckmann et al., 1999). Dispersal is generally thought to have evolved to minimize inbreeding and/or competition among relatives. However, dispersal, relatedness patterns and the cost of inbreeding are likely to have evolved in concert (Gandon, 1999; Perrin & Mazalov, 1999). Although the formation of social groups may lead to inbreeding, it may also result in strong differentiation between groups. For example, groups may exhibit high levels of genetic relatedness and comparably little variation because of coancestry rather than inbreeding (Sugg et al., 1996).

However, it has recently been suggested that the importance of kin for the formation of social groups has been exaggerated and that the structure of relatedness within many groups is merely a consequence of direct benefits derived from natal philopatry (e.g. Lambin et al., 2001; Clutton-Brock, 2002). For example, if individuals are able to cooperate better with familiar individuals or if dispersal costs are high, clustering of related individuals could result. Furthermore, new theoretical models (Queller, 1992, 1994; Taylor & Frank, 1996) suggest that competition between relatives can counteract kin selection, as reported by West et al. (2001).

African lions, Panthera leo L., are large, social carnivores (Schaller, 1972). They exhibit a number of cooperative behaviours, including group hunting, territory defence, and communal rearing of cubs, all of which might affect the costs and benefits attending decisions about group formation (Packer & Ruttan, 1988; Packer et al., 1991; Scheel, 1993; Pusey & Packer, 1994; Heinsohn & Packer, 1995). Prides of related females and their dependent young occupy a permanent territory, vigorously defended against intruders (Packer & Pusey, 1983; Mccomb et al., 1994). Males fight their way into prides by ousting the resident males and are generally unrelated to the females they join (Pusey & Packer, 1987; Packer et al., 1991). The relative importance of these behaviours in terms of group formation has been debated. For example, analyses of cooperative hunting in lions have met with considerable difficulties, mostly because of the complexity of correctly estimating the costs and benefits of various types of hunting methods and prey items (Caraco & Wolf, 1975; Mangel & Clark, 1986; Creel & Creel, 1995; but cf. Packer et al., 1990).

In a study of the Serengeti lions, Gilbert et al. (1991) used multilocus genotypes based on minisatellites. This analysis showed that pride females are always closely related, that male coalitions sometimes are composed of nonrelatives (but only in small coalitions), and that mating partners generally are unrelated (see also Packer et al., 1991). However, minisatellites can be difficult to interpret, because different loci may overlap in size (Lewin, 1989; Bruford & Wayne, 1993). For example, the use of minisatellites in studies of mating patterns and group structure requires empirical calibration of band sharing among individuals of ‘known’ pedigree, e.g. from behavioural observations. Furthermore, a minisatellite band does not necessarily correspond to a neutral allele, as is the case for microsatellites. Lastly, the large number of bands on a typical minisatellite gel may also bias the interpretation process. While minisatellites were the best markers available to Gilbert et al. (1991), they are less than ideal for analyses of genetic structure and mating patterns.

In this paper, we therefore use more powerful microsatellite markers in further pursuit of these questions, using F-statistics and estimates of relatedness, and consider the implications our results have for the evolution of lion sociality. The complexities of group dynamics in lions provide an excellent opportunity to analyse the genetic consequences of sociality. In addition to differences in genetic markers employed, our study population differs from the Serengeti population studied by Packer et al. In our study population in the Selous Game Reserve no pride held more than seven females, markedly lower than the maximum pride size of 18 recorded in the Serengeti (Pusey & Packer, 1987). Furthermore, while many earlier lion studies have been conducted in relatively open habitats (e.g. Serengeti in Tanzania, see Pusey & Packer, 1987; Etosha in Namibia; see Stander, 1992), our study population primarily occupied woodland, a more typical habitat for lions. Habitat differences may affect demographic parameters, genetic properties, and the social system by, for example, altering the costs and benefits of cooperative hunting (Packer & Ruttan, 1988; Caro, 1994). We investigate whether behavioural units in the population (prides) are also distinct genetic units, and describe the effects of dispersal on the genetic structure of the population. Relatedness plays a pivotal role in most theories of the evolution of sociality. Detailed information under a variety of ecological conditions is therefore of great importance, particularly in a species like the lion, the only social member of an otherwise solitary family.

The new findings presented in this paper extend our knowledge of genetic patterns both within and between social groups. The contrasting discoveries of a genetic structure above the level of the social group and the lack of relatedness within some groups have important ramifications for how we view the evolution of social behaviour in some species.


Field work and study area

All data were collected from a free-ranging lion population in the northern sector of the Selous Game Reserve, in south-eastern Tanzania (lat 7°35′S, long 38°15′E). The study site covered about 1000 km2 and consisted mostly of wooded savannah, miombo and Combretum thickets. We collected data between June 1993 and February 1999. Individual lions were identified mainly from whisker spot patterns (Pennycuick & Rudnai, 1970) aided by the use of a picture library, but natural scars, colour, and other attributes were also used. A total of 141 lions in 16 prides were identified during the study. Pride membership was inferred from association patterns of identified individuals. In no instance was pride membership inferred via shared associations with a third individual. We made all observations from inside a vehicle at distances from 10 to 200 m. For females fitted with radiocollars (Telonics MOD-500, Telonics, Mesa, AZ, USA), anaesthesia was induced by 200 mg Telazol and 100 mg Rompun administered by a Telinject (Vario 3V) CO2 rifle (Telinject, Saugus, CA, USA). Rompun was reversed by 17 mg Yohimbine after the collar had been fitted or removed, usually within 45 min. Lions were usually mobile without visible sign of anaesthesia within 2 h after i.v. injection of Yohimbine.

Tissue collection and molecular methods

We collected tissue samples in the form of biopsy. For this purpose, a CapChur CO2 pistol with custom-made aluminium darts was used (Karesh et al., 1987). Lions usually reacted to the impact of the dart by startling, often growling and moving away a few metres. The lions' reactions were sufficiently mild that up to five animals could be sampled in one session. Because lions can be hunted in most parts of Selous, some are wary of humans and cars, and this method did not allow sampling of shy lions that were impossible to approach to within 25 m. We refrained from sampling cubs less than 1 year old, as the risk of serious injury was considered too high. Tissue was stored at ambient temperatures in 95% ethanol containing 100 mm ethylenediaminetetraacetic acid in the field for up to 4 months. In the laboratory, samples were stored at −20 °C. DNA was extracted using a standard phenol/chloroform protocol (Maniatis et al., 1982), dissolved in water, and stored at −20 °C. Three fluorescently labelled primers (dyes HEX, TET, and FAM; PE Applied Biosystems, Foster City, CA, USA) were run in 10 μL multiplex reactions under the following conditions: 25 ng target DNA, 2.5 nmol dNTP, 4 pmol of each primer, 0.4 units of Taq polymerase, 1× PE buffer (including 15 mm MgCl2), at 94 °C for 3 min, 25× (94 °C for 30 s, 52 °C for 45 s, 72 °C for 1 min), followed by 72 °C for 10 min. A few samples that failed to amplify under these conditions were run at a lower annealing temperature (>49 °C) for up to 35 cycles. A 2-μL polymerase chain reaction product was then loaded with 6 μL formamide and 0.03 μL GS500 TAMRA size standard and run on a ABI Prism 310™ Genetic Analyser (PE Applied Biosystems). Primers used were Fca001, Fca008, Fca026, Fca031, Fca045, Fca077, Fca126, Fca272, Fca275, Fca391, Fca506, Fca567, Fca628, and F115 (Menotti-Raymond et al., 1999).

Statistical methods and software

Levels of observed and expected heterozygosity were calculated using the software BOTTLENECK (Cornuet & Luikart, 1996). Hardy–Weinberg equilibrium tests were performed using GENEPOP (Raymond & Rousset, 1995).

Pairwise estimates of relatedness (r) were calculated using the computer program DELRIOUS (Stone & Björklund, 2001) in which the algorithm described by Lynch & Ritland (1999) is implemented. The formulas prescribed assume random mating. This assumption is unfulfilled if molecular markers are obtained from populations exhibiting social structure. Therefore, to account for lion pride structure, we used computer simulation. In the simulation, a single adult breeding male and female were selected pseudorandomly from each pride and coalition, respectively, and used to create a ‘virtual population’. Virtual population allele frequencies were calculated across all loci (repeated 1000 times) and the mean virtual population allele frequencies were determined and used as effective breeding population allele frequencies. The virtual population thus consisted of individuals chosen pseudorandomly among breeding individuals within prides. This procedure entails that the initial virtual population was chosen nonrandomly over the entire population, while reconstructing the genepool that might have existed.

F -statistics were calculated using ARLEQUIN ( Schneider et al., 2000 ). We allowed for the specifics of lion social behaviour by setting the pride as the lowest population structure above the individual level. This caused FST , FIS , and FIT , to be transformed into FLT , FIL and FIT , respectively (I = individual, L = lineage/pride, and T = total population). This adjustment allowed us to investigate the genetic structure at three levels; the entire study population, prides, and individuals. The study population was itself an artificial subset of a larger continuous population that extended beyond our study area. We detected a secondary structure of the population, and this was analysed by defining a fourth level in the population (denoted Z = superpride to avoid confusion with FST ). We used F -statistics in preference over Rho-statistics, which work less well for samples scored for less than 20 loci ( Gaggiotti et al., 1999 ). Means are followed by standard deviations throughout the paper, unless otherwise indicated.


Pride structure

We obtained samples from 14 of 16 prides in the study area (Fig. 1 and Table 1). Pride size averaged 5.6 (range 2–9) adults, including 1–4 males. The average number of males was 2.4 ± 1.0 (Table 1). One of the prides comprised only one female associating with three males. She was included in the analyses because she maintained an exclusive home range in an area where no other lions could be found. We obtained genetic data from 69% of the adults in the 14 prides included in the analysis and from 35% of the cubs, a total of 70 individuals.

Figure 1.

Schematic map showing the relative position of each pride. Each circle corresponds roughly with the centre of each pride's territory. Superpride clusters denoted by A, B, C, and D, respectively. Territories were overlapping and changing with time. Prides shortcut N and Mbuyuni (superpride D) therefore occasionally came into contact, contrary to what the Figure indicates. Prides in the superpride ‘C’ were, however, never located in the same area. The line shows the Rufiji river.

Table 1.  Pride composition in early 1999. Numbers within parentheses indicate number of individuals included in molecular analyses.
PrideNo. of femalesNo. of malesNo. of cubsTotal no. of adults
  • *

     Two subsequent tenure holders.

  •  These two males held tenure in both prides.

Sand River5 (4)2  5 (4)7
Tagalala2 (2)3 (3)5
Selous Mdogo3 (1)25
Selous Grave7 (7)1 (1)  9 (1)8
BehoBeho3 (2)2  25
Beho Mdogo1 (1)34
Shortcut5 (4)4 (3)  4 (1)9
Shortcut North2 (2)?2
Manze4 (3)04
Nzerakera5 (3)4 (1 + 4)*11 (6)9
Mbuyuni3 (3)  1 (1)3
Siwando 22 (2)1 (1)  23
Siwando3 (3)2 (2)  55
Mzizimia3 (2)2 (2)  4 (3)5
Mean ± SD3.4 ± 1.62.4 ± 1.0  5.3 ± 3.25.3 ± 2.1

Heterozygosity levels

At the individual level, the mean number of loci scored was 13.3 ± 1.6 (range 5–14). For 53 individuals, we had data from all 14 loci. Average heterozygosity was 0.75 ± 0.08 (range 0.58–0.89). Three loci were found to significantly deviate from Hardy–Weinberg equilibrium, but after Bonferroni correction, this significance evaporated (Table 2). For each locus, over 85% of the individuals were scored, but, despite repeated amplification attempts, allelic dropout remained a problem for a few individuals, as discovered by a complete absence of alleles at some loci. These were mostly samples from shy lions that were sampled at larger than optimal distance, from which only hair could be obtained from the biopsy dart.

Table 2.  Summary statistics of the 14 loci used in the analysis.
LocusNo. of individualsNo. of allelesHexp (IAM) Hobs
  • *

     Significantly deviating from HW. No locus deviates significantly from HW after Bonferroni correction.

Fca00866  60.560.72
Fca02666  90.690.84*
Fca03170  90.680.70
Fca04566  40.410.67*
Fca27260  70.610.63
Fca27570  70.600.79*
Fca39170  80.640.75
Fca56761  60.560.70
Fca62870  50.480.58
Mean ± SD66.9 ± 3.5  8.9 ± 3.60.65 ± 0.120.75 ± 0.08

Relatedness estimates

Analyses of relatedness revealed that r-values declined significantly with distance between individuals (Mantel test, n = 2415, P < 0.001) (Fig. 2). Relatedness estimates >0.25 were in 167 of 209 cases between individuals within the same pride or individuals in proximate prides. Consequently, only 32 large relatedness estimates were obtained between individuals in prides further apart. An adult male was involved in just over half of these (n = 18). The remaining relatedness estimates >0.25 were either between offspring (n = 6), between adult females (n = 6), or between adult females and offspring (n = 2). Mating between close relatives was observed in only one pride, where the two individuals mating had an estimated r = 0.59 ± 0.04 (jackknife value ± SE).

Figure 2.

Pairwise relatedness estimates for all individuals plotted against number of prides apart. Line simple linear fit. The dropoff in relatedness with distance is highly significant (Mantel test P  < 0.001). The 95% confidence intervals were generated by bootstrap resampling (5000 iterations) to account for the fact that each lion contributes to more than one pairwise estimate (see also Spong & Creel, 2001 ).

Pride females were on average related to each other at a level equivalent to that of half-sibs (Table 3, n = 34 females, 12 prides). Mean r was correlated negatively with the number of adult females in a pride: smaller prides showed greater mean r than did larger prides (Spearman r = −0.58, P < 0.05, n = 12). However, some prides exhibited very low, or zero, within pride relatedness (Table 3).

Table 3.  Mean relatedness estimates (by DELRIOUS) among different groups within each pride.
  • Numbers within parentheses indicate number of individuals included in the genetic analyses.

  • *

    Coalition males held tenure in both prides simultaneously.

  • Pride of single female.

Tagalala  0.59 (2)  0.08 (3)  0.22 (5)
Sand River −0.06 (2)0.06 (6)  0.13 (8)
Selous Md.
SelousGrave  0.07 (3)  −0.14 (4)0.17 (5)  0.09 (8)  0.15 (6)
Shortcut  0.01 (2)  0.26 (3) −0.04 (5) −0.01 (3) −0.05 (4)
BehoBeho  0.52 (2)
Beho Md.  (1)
Manze  0.11 (3) −0.08* (4) −0.19 (7)
Nzerakera  0.27 (3) −0.08* (4) −0.07 (7)0.12 (6)  0.11 (9) −0.00 (10)
Shortcut N  0.14 (2)
Mbuyuni  0.08 (2)  0.09 (4)
Siwando 2  0.49 (2)  −  0.18 (3)
Siwando  0.34 (3)  0.18 (2)  0.12 (5)
Mzizimia  0.55 (2)  0.18 (2) −0.05 (4)0.15 (3)  0.22 (5)  0.08 (5)
Mean ± SD  0.26 ± 0.23  0.09 ± 0.14 −0.03 ± 0.140.13 ± 0.05  0.11 ± 0.07  0.05 ± 0.09

Pride males exhibited a lower mean r, than female pride mates (Table 3). All but one coalition included one pair of full- or half-sibs. The third member present in two three-member coalitions was unrelated to the other two members. A four-member coalition included two males that exhibited r = 0.12, whereas the other two males were unrelated to any other member. Pride males were in general unrelated to the pride females. Offspring mean r to females, males, and to other offspring, were all below that of first order cousins (i.e. 0.125, Table 3).

Population structure

Global F-statistics showed significantly negative FIL-values, FIT-values not significantly different from zero, and significantly positive FLT-values (Table 4). When we removed the genetical component of the dispersing sex (males), this pattern was reinforced. That is, the absolute values of both FIL and FLT increased, while FIT remained undetectably different from zero.

Table 4. F -statistics describing the social structure of the population. The first row lists F -statistics when including all pride members. The second includes adult females and dependent young only, and the last row includes adult females only.
  • * *

    P  < 0.01,

  • * * *

    P  < 0.001 based on randomization (1023 permutations).

All pride members (n = 70) −0.11*** −0.02 ns0.07***0.02 ns
Excluding males (n = 55) −0.12*** −0.02 ns0.07***0.02 ns
Adult females only (n = 34) −0.20*** −0.01 ns0.11***0.05**

Pairwise FLT values among prides showed an insignificant increase with increasing numbers of intervening prides (Fig. 3). Mantel test statistics for the correlation either including all pride members or only females and young, yield the same value (Mantel, P = 0.12). With adult females only, the correlation with distance was almost significant (Mantel, P = 0.06). Pride and territory sizes varied greatly, so the number of intervening prides, rather than geographical distances, provided more meaningful values of the degree of separation among prides. The absolute values of these pairwise comparisons did not increase significantly when the male component was removed (F < 1.04, ns; for all three pairwise comparisons), nor did the slopes change significantly (t < 0.79, ns; for all three pairwise comparisons).

Figure 3.

Pairwise FLT /(1 −  FLT ) for all defined demographic groups (all pride members, females and cubs, and adult females only) plotted against number of prides apart. Lines simple linear fit. Regression lines were not significantly different ( F  < 1.04, ns; for all three pairwise comparisons). Slopes of regression lines not significantly different ( t  < 0.79, ns; for all three pairwise comparisons).

When we examined patterns of similarity among prides, we found distinct sets of prides, ‘superprides’ (Fig. 1 and Table 4). However, the pairwise pride FLT-comparisons did not necessarily concur on low values for all dyads in a cluster, so defining the number of clusters of prides and members of each cluster became difficult (Table 5). To our knowledge, there is no general method for analysing these kinds of spatially distributed molecular data where the dispersal pattern and social system of the species can be included. We therefore approached the problem from the opposite direction and asked the question: if prides can be clustered into distinct sets, do prides within groups share any common trait (e.g. spatial proximity)? We first investigated if the correlation of FLT with distance had any effect on the grouping pattern. As genetic relatedness decreases with increasing distance, it is conceivable that any grouping of prides is simply a reflection of this phenomenon. To control this we used residuals from the regression of FLT on distance, thereby making the pairwise FLT-values independent of distance. This imparted no change to the grouping pattern of prides; each superpride still contained the same prides. The statistical procedure ‘K-means clustering’ groups prides into a specified number of superprides, minimizing within superpride variability while maximizing between superpride variability. Essentially an analysis of variance in reverse, the most significant grouping pattern is then presented. Using K-means, three prides (of 11) in the suggested clusters were not neighbouring, and the obtained FZT-value was highly significant. The suggested superprides comprised two (superpride A), four (superpride B), three (superpride C), and two (superpride D) prides, respectively (Fig. 1). It should be noted that the highest FZT-value was obtained when defining 13 clusters (the maximum allowed number), as all but one cluster then contained only one pride. FZT then was almost identical to FLT and, as expected, assumed nearly the same value. Removing single prides from superprides yielded similar results. Testing for alternative pride constellations of proximate prides all gave insignificant FZT-values.

Table 5.  Pairwise pride FLT -values.
  • *

    P  < 0.05. TA, Tagalala; SR, Sand River; SM, Selous Mdogo; SG, Selous Grave; SH, Shortcut; BB, BehoBeho; BM, Beho Mdogo; MA, Manze; NZ, Nzerakera; SN, Shortcut N; MB, Mbuyuni; S2, Siwando 2; SW, Siwando; MZ, Mzizimia. No value remains significant after Bonferroni correction.

SR  0.02            
SM  0.13 −0.02           
SG  0.07  0.03 −0.07          
SH  0.07 −0.03 −0.13 −0.04         
BB  0.08  0.08  0.04  0.09  0.16*        
BM  0.10  0.15 −0.03  0.06  0.090.26       
MA  0.11*  0.10* −0.17  0.02 −0.010.18*0.14*      
NZ  0.15*  0.08 −0.02  0.07*  0.010.16*0.14*0.02     
SN  0.11  0.00  0.01  0.06    
MB −0.02  0.06  0.07  0.09  0.080.17*0.210.13*0.08 −0.02   
S2  0.17*  0.14 −0.03  0.04  0.17*0.26*  
SW  0.17*  0.10*  0.19*  0.09*  0.080.22*0.27*0.12*0.12* 
MZ  0.11  0.15  0.01  0.17*  0.19*0.20*0.320.15*0.16*  0.24*


Our basic results confirm expectations derived from behavioural descriptions of lion social organization, as well as previous molecular findings (i.e. Packer et al., 1991), but with some important exceptions and extensions. Some prides exhibit very low among female relatedness and some prides show significant kinship ties forming ‘superprides’. The differences in the two populations might be attributed to the higher resolution obtained with microsatellites, but contrasting ecological conditions may also play a role.

Relatedness estimates

As expected, relatedness between any chosen pair of lions in the population drops of with distance. At distances of more than one or two prides away, it was uncommon for any lion to have close relatives. This supports previous findings of short male dispersal distances in this population (Spong & Creel, 2001).

All females in a pride normally breed, and mean relatedness estimates among offspring will therefore decline from that of their mothers, assuming that there is more than one father. Accordingly, the highest mean among offspring relatedness estimate was found in a pride held by a single male (Selous Grave; Table 3). However, even for the pride with a single male, mean among offspring relatedness was surprisingly low, suggesting that females mate with males outside the pride.

Previous studies in the Serengeti have shown that coalitions comprised of more than three males were always composed of relatives, whereas smaller coalitions sometimes included unrelated individuals (Packer et al., 1991). In our study population, all four coalitions sampled included at least one related pair, and a four-member coalition included two males that were unrelated to any other coalition member. Although our sample of male coalitions is small, it nonetheless detected greater genetic diversity among coalition members than has been found previously. Apparently, relatedness among males plays a weaker role in coalition formation for males in Selous than in Serengeti. Perhaps sport hunting of males in Selous (Creel & Creel, 1997) affects the turnover of adult males and thus reduces the size of male sibships.

Population structure

Female lions are strongly philopatric (Selous: personal observation; Serengeti: Pusey & Packer, 1987). This explains the high among female relatedness in many prides and the divergence of prides from one another. Female philopatry and male dispersal can also be inferred from the fact that the global FLTs increased when the male component of the prides' genetic composition was removed from the analysis. Coancestry thus explains why pride females maintain high levels of genetic relatedness in the absence of inbreeding. Accordingly, global FIL-values were negative, and more highly negative values were obtained when the male contribution to the prides was excluded. The increase in pairwise FLTs with distance were insignificant for all demographic groups, and when only females were included the pattern remained insignificant, but only marginally so (Mantel, P = 0.06). The inclusion of more prides probably would have rendered significant results. While, the number of prides included was not sufficient to detect a significant correlation of pairwise FST values to distance, it was sufficient to provide significant global F-statistics. Furthermore, the significant correlation of pairwise relatedness estimates to distance strongly suggests that the direction of the slopes of the lines in Fig. 3 are correct. The observed superpride structuring could, in theory, be created by either a high proportion of male dispersal events occurring over short dispersal distances or by pride fissions yielding groups of prides with a common predecessor. The superpride C (Fig. 1), which included nonadjacent prides may have arisen from male dispersal in the past. In Selous, male dispersal distances have been shown to be short (Spong & Creel, 2001), and repeated nonrandom dispersal of male coalitions into groups of relatives probably play an important role in creating the observed pattern (e.g. Keane et al., 1996). Especially considering that male dispersal is a more frequent demographic event than pride fissions are.

Whatever the cause, the existence of geographically distinct superprides raises interesting possibilities for the evolution of sociality. Perhaps most importantly, kin selection effects can operate between members of neighbouring prides. Earlier studies of the Serengeti lion population have shown that, although the degree of interpride relatedness remains high for the first few years after a fission, it rapidly dwindles after 10–15 years (Packer et al., 1991). Our data show that genetic associations may exist between groups of prides. Even if fissioned prides eventually diverge genetically, this does not preclude the high genetic relatedness between some prides playing a vital role in social evolution. During the first years after fission, interpride tolerance between such prides may allow the establishment of new prides, and relatedness among adjacent prides would decrease the costs of showing tolerance.

Our data show that patterns of relatedness among groups can be strong, as in some other social carnivores. Examples include dwarf mongooses, Helogale parvula (Keane et al., 1996); wild dogs, Lycaon pictus (Creel & Creel, in press); and white-nosed coatis, Nasua narica (Gompper et al., 1998). When such groups are proximate, kin selection may potentially affect behavioural interactions among individuals in different groups (e.g. in territorial contests). The evolutionary impact of social organization above the level of the social group will be an interesting topic for future research.

Seemingly contrasting to the above arguments is our discovery that not all prides are composed of related females (Table 3). Two prides in our study (Sand Rivers and Shortcut) had mean among female r close to zero. Recently the relative importance of indirect vs. direct fitness benefits of group living has been highlighted (Clutton-Brock, 2002). If, for example, individuals simply prefer to group with familiar individuals, kinship ties among pride females follow, and may not be contingent on the inclusive fitness benefits that derive from living with relatives. For example, in meerkats (Suricata suricatta), Clutton-Brock et al. (2000) failed to find any correlation between relatedness and investment in costly group activities. In lion prides all females usually breed, and direct fitness benefits of group living continually accrue for all pride members. Lion prides would thus seem an ideal setting for group formation based solely on mutualistic and direct benefits. But, free-ranging female lions have never been observed to pair with unfamiliar females. The potential indirect fitness benefits deriving from grouping with relatives may be an evolutionary force strong enough to preclude the evolution of nonkin groups in female lions. The two prides with low among female relatedness are probably the result of persistent matrilines, resulting in genetic divergence. If familiarity is used as the criterion for amicable behaviour among pride females, low among female relatedness is expected in some prides. It is important to realize that lack of relatedness between group members, or lack of correlation between relatedness and cooperative behaviours, does not per se mean that kin selection is unimportant for the evolution of group living in lions.


For permission to conduct research in Tanzania, we are grateful to Tanzania Wildlife Division, Serengeti Wildlife Research Institute, and the Commission of Science and Technology. We thank Staffan Ulfstrand and Peter Waser for valuable comments on earlier versions of the manuscript. This study was supported by the Royal Swedish Academy of Sciences, the Swedish International Developmental Agency, the Lennander and the Helge Ax:son Johnson's foundation, and Frankfurt Zoological Society (project 1112/90), Magnus Bergwall's foundation and the Swedish Natural Science Council. Logistical help during fieldwork was kindly provided by S. Liljeholm; T. Semya and W. Minja at the Tanzania Wildlife Division; S. Arsene, M. Wiener at Selous Safari Company; and S. Heep at Baobab Village.