2.  General methods

Rhesus macaques have the most extensive geographical range of any non-human primate (Southwick et al. 1996). This species originating from Asia lives in habitats varying from semi-desert, temperate forest to tropical woodland and swamp, from sea level to 3000m (Seth & Seth 1986). Each social group occupies a home range, which is not defended, but is likely to overlap between two neighbouring groups. The home range varies in size between 0.37 and 16 km² (Soutwick et al. 1996). In the wild, the group size varies between 10 to 125 individuals, close to humans they have been observed in groups up to 240 individuals (Seth & Seth 1986, Southwick et al. 1996). Groups are composed either of many small matrilines (definition below) or of a few large matrilines (Melnick & Kidd 1983). After a certain rise in number a large group is likely to fission, e.g., by splitting along matrilines and then pushing the lowest-ranking female with her descendants to the periphery of the group. This subgroup finally splits up and starts a group on its own (Chepko-Sade & Sade 1979, Melnick & Kidd 1983).

Rhesus monkeys live in multi-male, multi-female groups. The females are philopatric, i.e., they stay in their natal group throughout their life and interact with both kin and non-kin (Gouzoules & Gouzoules 1987). As a consequence, females are closely associated with their maternal kin which is called a female-bonded society (Wrangham 1980). Maternal kinship is established via matrilines, with a matriline defined asall descendants by a founder female (Melnick & Pearl 1987). In contrast, male macaques are the dispersing sex (Colvin 1986, Melnick & Pearl 1987). They emigrate from their natal group around puberty to join other social groups (Lindburg 1969, Colvin 1986). This happens especially during the mating season which Lindburg (1969) named mating season mobility. Most males change groups every few years. This behaviour has been interpreted as inbreeding avoidance (Melnick et al. 1984, Pusey 1987, Pusey & Wolf 1996). Although male rhesus were found to join groups into which their elder brothers had immigrated (Meikle & Vessey 1981), males mainly interact with non-kin once they emigrated.

Adult males are dominant to adult females, and females, in addition, are clustered in a strict matrilineal hierarchy. Mothers are dominant to their daughters, and daughters socially inherit [page 10↓]the dominance rank of their mother (Koyama 1967, Berman 1980). The rank order among daughters is inverse to their birth order (Datta 1988). This has three consequences: (i) all daughters rank directly below their mothers in the hierarchy, (ii) all maternal half-siblings occupy adjacent dominanceranks within a troop (Chapais & Schulman 1980), and (iii) the eldest daughter is the lowest ranking among her sisters. High-ranking females are often preferred as social partners (Seyfarth 1980), suggesting that dominance rank is also influencing the social structure among females. Macaques are capable of detecting their own individual dominance relation towards others, but likewise they are capable to detect the rank relation between two other monkeys (reviewed by Cheney & Seyfarth 1990) which is influencing the likelihood of interventions in conflicts among others (see chapter 6).

Females reach adulthood around 3.5 years of age, males around 4.5 years of age (Bercovitch & Berard 1993, van Hooff 1988). However, there is probably substantial inter-individual variation in the ages at which the full development is attained (Pereira & Altmann 1985, Bercovitch & Goy 1990).

Rhesus monkeys exhibit a promiscuous mating system, males and females mate with several sexual partners. Mating with multiple males has been interpreted as a female reproductive tactic designed to obscure paternity, decrease the risk of infanticide and increase the chance of parental care (Hrdy 1981, Stacey 1982). Female rhesus macaques are found to mate with an average of three males per conception cycle and they tend to conceive on the first cycle of a mating season (Bercovitch 1997). Sexually receptive females form temporary mating bonds (consortships) with males (Carpenter 1942, Lindburg 1971, Chapais 1983). Aggressive competition and male coalitions over access to fertile females are rare in rhesus macaques (reviewed in Bercovitch 1992), but males suffer more wounds during the mating season (Bercovitch 1997). High-ranking males sire more offspring than low-ranking males (Bercovitch & Nürnberg 1996), but females prefer to mate with low-ranking (Manson 1995a) and novel males (Manson 1995a, Bercovitch 1991, 1997, Berard 1999).

Cayo Santiago is a 15.2-hectare island situated approximately 1 km off the south-east coast of Puerto Rico (18°09' N, 65°44' W). The climate is subtropical, with average rainfall of 163 cm per annum and mean daily temperatures ranging from 23.8 °C to 27.1 °C (Kessler & Berard 1989). Vegetation on the island ranges from sparsely wooded areas to dense undergrowth, with periodically waterlogged mangrove areas and exposed cliffs.

The island is inhabited by a free-ranging colony of approximately 800 rhesus macaques, which are distributed over the island in several social groups. All individuals are descendants of the founding population of 409 individuals trapped in India in 1938 (more details in Carpenter 1942, Rawlins & Kessler 1986). No individuals have been added on Cayo Santiago since then except through births, but genetic analyses suggest that the island population is not closely inbred (Widdig et al. 2001). Historically seen, this study site is/was important as studies on primates have been restricted to captivity, before studies on wild primates became first successful in the seventies. The advantage of studying kinship in this population is that census records have been kept continuously since 1956, therefore detailed maternal genealogies and male migration histories are known for all individuals. Day of birth and day of death (or removal) are also available for all individuals. In addition, paternity analyses were started with analysing the birth cohort 1989 from group S.

This population lives under so called semi-free conditions and monkeys are habituated to human observers. The animals are provisioned with commercial high protein biscuits distributed once daily in the morning at the three food dispensers located in corrals. Nevertheless, the macaques forage extensively, with an estimated 50% of feeding time spent on natural sources, e.g., foliage, fruits, and insects (Marriot et al. 1986, Marriot 1988). The ingestion of soil is thought to aid against enteric parasites (Knezevich 1998). Available water is supplemented via the collection of rainfall in cisterns and piping of water to drinking basins. Although the primary sources of mortality are starvation and injury (Berard 1990), the lack of natural predators on the island requires an artificial control of the population size. In the past, entire social groups were removed, but since 1996 a proportion of randomly selected 2-year-olds from each social troop has been chosen annually with respect to family membership. Intervention is limited to an annual trapping period (January to March), during which the 1-year-olds are assigned identification codes and blood samples are obtained for paternity analyses.

Like a typical macaque group, the study troop, group R, consists of a stable core of maternally related females with their offspring and adult males who change in number, especially around the mating season where immigration and emigration is at a maximum (Hill 1986). Group R had been formed as the result of a fission event in 1985, whilst group BB subsequently fissioned from R in the period January-February 1996 (Kazem pers. comm.), with the partition being complete prior to the collection of the behavioural data presented here. In the [page 12↓]census provided by the Caribbean Primate Research Center (CPRC), group BB has been considered as a separate group from R since June 1996. Group BB was smaller than group R and mainly formed by the lower-ranking females from R and their offspring. During the behavioural observation period in 1997, group R had a total of 126 core members i.e., 91 natal females and 35 natal males (not involving adult males which are non-natal, see below), but the behavioural analyses were restricted to interactions among adult females and their female descendants (N=91). In addition, the number of adult non-natal males changed between 21 and 31 during the behavioural study due to migration. All males (natal and non-natal) were excluded from the behavioural analyses. After migration males interact more with non-kin than with kin, so for the majority of their adult life they mainly interact with non-kin. However, it could be that males also learn to discriminate paternal kin from non-kin in their natal group. In order to compare interaction between kin (both maternal and paternal) and non-kin, adult females are the appropriate sex to study.

Two data sets will be distinguished in this thesis. The first data set will be used in chapter 3 (reproductive skew) where paternity analyses for all infants born in group R and BB between 1993 and 1998 (N=263) will be presented without additional behavioural observations. The second data set will be used in chapter 4-6 where paternity analyses on all females of group R present in 1997 (N=91) will be presented in combination with behavioural observations. The latter data set, the study group will now be introduced in detail (see also Appendix 1 on the study group).

Age of the group subjects (N=91)

Age (date of birth) was known for all group members from the demographic data base of the CPRC. Since age classification differs across studies and inter-individual variation is known (see above), females in this study have been generally categorised as “adults” when they were born in 1994 or before. In other words, they were at least 3 years old during the behavioural study in 1997, because such a female has already given birth to an infant. Further definitions of age classes used in this study are given in Table 2.1.

Table 2.2.1: Age categories among the group females (N=91)

The rhesus macaque breeds on a seasonal basis (Lindburg 1971, Drickamer 1974, Bercovitch 1997) with an inter-birth interval of approximately one year (Rawlins & Kessler 1986). The mating season at Cayo Santiago usually lasts from May to October, following the birth season from November to April, with a peak of births in January to February. Thus, individuals can be classified into discrete a birth cohort, although individuals born within the same birth season may be born on the same day or up to a maximum of 6 months apart. Individuals born within the same birth season are defined as peers, whereas individuals born into different cohorts are non-peers. Both age categories were distinguished throughout the study, but in some analyses the exact age differences (in years) are used as a refinement of the non-peer category which includes multiple ages (all age differences except for age mates). More information on age and birth cohorts are given in the Appendix 1 (Study group).

Kinship of group subjects (N=91)

Maternal kinship for all subjects was known up to several generations backwards using the demographic database of the CPRC which was started in 1956. During the behavioural study, group R was comprised of three matrilines, two smaller (D07: N=11, CS: N=15) and a large one (262: N=65, considering only females). Because the focal group was composed of descendants of only three matrilines, a narrow definition of „maternal relatedness“ was adopted which included only close family members. A family consisted of the eldest surviving daughter (or granddaughter, if the daughter was deceased) of a matrilineal founder and her offspring (if n≥3 offspring) which is comparable to the definition of a “matriline” used by Dunbar (1984) to distinguish close family relatives from more distant kin in large matrilines. Therefore, two elderly maternal half-sisters with multiple offspring could have their own family, even though both were descended from the same founding matriarch.

Paternal kinship was defined containing all paternal half-siblings and their descendants. Sirehood was revealed via paternity analyses (see below) and paternal kinship could not be reconstructed for more than 2 generations backwards. However, this seems to be sufficient, as in female-philopatric species adult females are likely to be distant kin, whereas adult males (i.e., potential sires) are likely to be unrelated. For example, de Ruiter & Geffen (1998) found for long-tailed macaques, Macaca fascicularis, that the average degree of relatedness among dyads of adult females was much higher than for dyads of adult males (adult female r=0.14 vs. adult males r=-0.10). It was therefore assumed that sires are mainly unrelated.

Female rhesus bear only a single offspring in more than 99% of cases (Rawlins & Kessler 1986) suggesting that twins are extremely rare to be found. Assuming that females indeed tend to prefer novel males (Manson 1995a, Bercovitch 1991, 1997, Berard 1999) females are [page 14↓]expected to reproduce with different males rather than always with the same male in consecutive years. This would result in full-siblings to be rarely found, as infants born to the same mother would be maternal half-siblings of different ages (i.e., non-peers). Assuming male reproductive skew within the birth season considered, offspring born to different mothers, but sired by the same male, would be paternal half-siblings of the same age (i.e., peers) or often in close age proximity.

To investigate the effect of kinship and age proximity on female social relationships (see chapter 4), analyses are restricted to maternal half sisters (same mother, paternally unrelated, coefficient of relatedness r = 0.25), paternal half sisters (different family, same father, r ~ 0.25), and non-kin (different family and different father, r ~ 0.0). For testing the effect of the exact age differences on female social relationships, analyses are restricted to dyads that were either zero (i.e., peers), one, two or three years apart in age.

To investigate the effect of the degree of relatedness on female social relationships (see chapter 5), the analyses were extended to include a wider range of relatedness for both maternal and paternal kinship. In addition to maternal, paternal half-siblings and non-kin used in chapter 4, the following kin categories will be added in chapter 5: mother-daughters (r = 0.5), maternal grandmother-granddaughters (r = 0.25), maternal aunt-nieces (r = 0.125), maternal grandaunt-grandnieces (r = 0.0625), maternal cousins (r = 0.0625) and paternal aunt-nieces (r = 0.125). More information about maternal kinship of the study group is given in the Appendix 1 (Study group).

Dominance rank of group subjects (N=91)

Dominance ranks were determined based upon the outcome of dyadic agonistic interactions (ad libitum or continuous data, see below). First, matrilines themselves can be arranged in a matrilineal hierarchy which is fairly stable over time. Females from the matrilines D07 were highest ranking, followed by matriline CS, and than the large matriline of 262 was lowest ranking. The only exception was the mother-daughter pair 897-J12 from matriline CS, which both ranked lowest within group R, being on the periphery of the group.

Second, females themselves were arranged in an individual hierarchy. All mothers were found to be dominant over their daughters and all maternal half-sisters could also be ranked as predicted from their birth order with the exceptions of the two sister pairs (33C-X34 and X91-T95).

Kawai (1958) distinguished between dependent rank where individuals still dependent upon support of a third individual in order to achieve their basic rank where they have already acquired their adult rank. However, including all 91 females of the study group in the [page 15↓]hierarchy could bias data if younger females were still dependent in rank upon their mothers. In order to control for this potential bias, hierarchy was restricted to adult females with basic ranks (N=49). The hierarchy of the matrilines is given in the Appendix 1 (Study group).

From the study group R, 34 out of all 49 adult females were chosen as focal subjects. These females were consistently monitored in respect to their social interactions with other group females (see below). Regarding age, all focal females were adults. All females born between 1994 and 1988 and still present during the behavioural study in 1997 were chosen as focal females regardless of determined paternity. This decision was due to two reasons: First, for adult females born before 1988 it was less likely to establish paternity since the probability that their actual sire was genotyped decreased for elder females. Even though all males on the island were genotyped, systematically blood collection for paternity analyses did not start before 1992 (Nürnberg, pers. comm.). However, for the very old females with undetected paternity the existence of a paternal half-sibling within the study group in 1997 could be excluded (see paternity analysis below). Second, females born in 1995 or later were still dependent in their behaviour upon their mothers and therefore less suitable as focal subjects. All 34 females had 1 to 5 peers and 71 to 82 non-peers. Regarding kinship, all 34 females had a maternal half-sister (ranging from 1 to 6) and 22 out of the 34 focal females had a paternal half-sister (ranging from 1 to 5). Fifteen of these 22 focal females had a paternal half-sister being a peer (range between 1 to 2) and nineteen of these 22 females had a paternal half-sister being a non-peer (range between 1 to 4). Due to trapping (and perhaps early death) individuals lost part of paternal half-siblings which they had at the time of their birth (compare chapter 3 showing that the average number of paternal half-siblings is larger compared to maternal half-siblings). All females had female non-kin in the group, their number varied between 1 to 5 being peers and 64 to 82 non-kin non-peers across focal females. Regarding rank, focal females came from all three matrilines and were of high, medium or low maternal rank. Focal females are marked in bold in Appendix 1 (Study group).

Field work started on the 3 ^rd of May in 1997. After 3 weeks all core members of the group plus most adult males could be distinguished individually. Infants could be identified later, since they were still depending on the mother at the start of the study and could therefore be assigned to their mothers when they went back to suckle. At the same time all observed [page 16↓]behaviour was catalogued and the focal protocol (a written checksheet) was designed. All behavioural data were taken between May 28^th and December 23^rd of 1997, six days a week between 7 a.m. and 3 p.m. (weekdays) or 7 a.m. and 4.30 p.m. (weekend and holiday), respectively, due to the boat schedule given by the CPRC. Even though there is evidence of seasonal variance in behaviour (e.g. Wilson & Boelkins 1970, D'Amato et al. 1982), most data were taken during the mating season in order to avoid disturbance during the trapping period which finished around the end of March 1997 and started again in January 1998. During the mating season, females were monitored for signs of oestrus (coloration and oedema of the sexual skin, presence of vaginal plug, obvious consort behaviour), because social behaviour may change over the course of these cycles. Whenever the focal female entered the feeding corral, this would be noted in the focal protocol, since increased interactions (especially regarding competition) were expected.

Observations were scheduled in order to achieve an equal distribution per individual over the day, since activity could dependent on the time of the day. The sequence of animals observed was determined by randomly selecting a focal from those remaining to be sampled during that particular time block. No individual was observed more than once per day. A focal protocol was interrupted when the focal animal was out of view and continued if this break was less than 5min. Otherwise the protocol was discarded. Census was checked once per observation day including data on group membership, transfers, births and deaths. Where animals appeared to be transferring in or out of the group, their position was continuously checked, until final state was reached. These records were contributed to the census of the CPRC.

A repetition of a behaviour was scored as a new bout if (i) more than 10 seconds had elapsed between occurrences, (ii) at least one partner had switched to a mutually exclusive activity (e.g., from grooming to aggression) or (iii) at least one partner moved out of the 2m radius. The only exception were aggressive events in which a number of different agonistic patterns occur in quick succession. Unless one or both opponents had switched to a mutually exclusive behaviour those bouts were treated as one, but only the most intense kind of aggression was used for the analyses. Acts between mother and their dependent infants were scored as interactions only, when the act was specifically directed towards or initiated by the infant.

Behavioural data are either states or events (Martin & Bateson 1986). State data are counted as point time sampling (see below). Within a 20min focal protocol, state data were collected [page 17↓]at min 4, 8, 12, 16 and 20. These data include information on the activity of the focal female and all neighbours in spatial proximity (≤ 5 m) at this second (see below).

Event data are sequence data which were either counted as continuous data or ad libitum data (see below). Sequence data were collected on (i) affiliation interactions such as grooming, friendly approach, co-feeding, co-drinking, (ii) agonistic interactions such as agonistic approach, non-physical and physical aggression, and (iii) interventions in conflicts (coalition formation)including both a co-operative and competitive interaction (see below) and (iv) other interactions. All sequences of behaviour included the identity of the initiator and the recipient of a certain interaction which (i) involved always an interaction between a focal female with one of the remaining potential female social partner (N=90) ignoring interactions between the focal and adult males (see continuous data) or (ii) involved an interaction between any group members others than the focal female (see ad libitum data). Behaviour patterns were scored either as affiliative or agonistic interactions or in other contexts. Focal protocols included always the activity of the recipient before the interaction (context) and the response of the recipient following the interaction. All interactions were scored as bouts. Grooming, in addition, was timed as duration (measured in seconds) with one bout lasting at least 5 seconds.

When a third party intervened in an ongoing dyadic conflict (coalition formation), the following information was collected whenever possible: (i) the initiator and recipient of the original dyadic conflict, including kind and context of aggression, response of the recipient, (ii) the sequence of intervening individuals, (iii) the kind of aggression by the intervener/s, (iv) whether the intervener/s supported the aggressor or the victim of the original dyadic conflict, (v) the distance of the supporter towards the dyadic conflict, (vi) whether or not the supporter approached the scene and (vii) the response by the target of support. One possibility to assess the number of potential support by a given female (opportunities of support, see below), all non-silent agonistic interactions were counted ad libitum to determine how often a maternal kin, paternal kin or non-kin of the female considered was involved in a dyadic conflict. A non-silent agonistic interaction was defined as either non-silent aggression including vocal threat, lunge, charge, chase, bite, attack or/and non-silent responses of any aggression such as fleeing and screaming. Silent agonistic interactions as open mouth threat, head-bobbing or displacement were not counted as opportunities of support as they could have been occurred out of sight to potential interveners (e.g., in dense vegetation).

Description of behaviours used in this study are given in the Appendix 2 (Ethogram).

Point time sampling data

Two kinds of point time sampling data were collected. First, data on spatial proximity (hereafter: proximity) were based upon point samples taken every 4 min during each focal protocol (Altmann 1974). Here, all neighbours within the 5m radius were identified with their exact distance (in m) towards the focal female including their activity at this time. Where no animal was within 5m of the focal female, the distance (in m) to and identity of the nearest visible individual including their activity were recorded instead, but not used in the analyses. If two neighbours were interacting or in contact they were recorded as social partners. As a second type of point time sampling data, the current activity state of the focal was also noted. (see also Appendix 2 Ethogram).

Continuous data

During focal sampling, all behavioural interactions involving the focal female and any of the remaining potential female social partners (N=90) were noted. The behaviour patterns were continuously recorded in the sequence in which they occurred (Altmann 1974).

Ad libitum data

Ad libitum data (Altmann 1974) are additional information collected in a focal protocol with interactions not involving the focal animal (non-focal data). Two kinds of ad libitum data were recorded in their sequence whenever they occurred. First, all dyadic and polyadic (coalition formation) agonistic encounters among group members were noted if the identities of the original dyadic conflict were observed. Ad libitum data are expected to be biased towards non-silent agonistic interactions or to conflicts involving individuals which consistently share spatial proximity with the focal female. These data were also used to construct the hierarchy among both males and females. Second, all grooming bouts observed between any group members were recorded in the way described before, but here the duration of a grooming bout is missing.

General analyses

In total, 645 hours focal sampling of nearly 1000 hours collected were used for analysing interactions of the 34 focal females, with a mean of 1130 min (ranging between 980 to 1180 min) per focal female. All behavioural data collected have later been entered into a complex data base designed in Microsoft Access ©. To answer specific questions, data have been re-[page 19↓]filtered from Access into a two dimensional table which were transferred to Excel ©. In Excel, the data were arranged in a 34 x 91 matrices (focal female x potential female social partner) to calculate either (i) rates per hour, or (ii) mean duration for each of the 34 females with respect to all potential female social partners within the group. As a general procedure, for each of the focal subjects the observed frequency of a given behaviour per social partner was divided by the observation time, so data on single behaviours are rates per hour.

Analyses on affiliation and aggression

Analyses were confined to three affiliative dyadic interactions (spatial proximity, grooming and approach) and two agonistic dyadic interactions (physical and non-physical aggression) because the remaining interactions (co-feeding, co-drinking, passes) were not suitable or did not occur frequently. Dyadic scores for social partners who were related to the focal animal, but who were neither maternal nor paternal half-siblings of the focal female, were excluded from analysis on paternal kin discrimination (chapter 4). Here, the crucial comparison was between non-kin on the one hand, and close kin of identical genetic relatedness (r=0.25), but of different co-parental gender, on the other. In other analyses, dyadic scores for social partners who were maternally or paternally related were calculated in order to compare different degree of relatedness of maternal and paternal kin (chapter 5).

Analyses on coalition formation

In chapter 6 data on coalition formation were analysed. A coalition is formed when one individual intervenes in an ongoing conflict between two opponents in order to support one of them. Since support in favour of one party is simultaneously targeting the other party, coalitions are triadic interactions involving a supporter, a recipient and a target (see chapter 6 for more details). Whenever the sequence of intervention by multiple supporters (polyadic support) into the original dyadic conflict was observed, this event was split into single triads each including a target, a recipient and a supporter (cf. Widdig et al. 2000).

Different analyses on coalition formation are all based on interventions observed over the 8 months study period, either collected ad libitum or in a focal protocol. I assume that these data collection did not bias the sample for the following reasons. Coalitions are relatively long lasting and often noisy events, especially as rhesus macaques tend to redirect aggression they received to other individuals which often produces a series of conspicuous aggressive events (pers. observation). Thus, together with the fact that individuals could be recognised over long distances, it is unlikely that a substantial number of dyadic conflicts followed by coalition [page 20↓]formation were missed (cf. Altmann 1974). As coalitions are based on ad libitum observations data are not rates per hour.

Testing the kin selection theory, all coalitions were included where a focal female (N=34) was intervening either in favour of or against a maternal half-sibling, a paternal half-siblings or a non-kin. This implies that the third party in this coalition could be any other individual of group R including adult males. Two different procedures were used to determine whether focal females intervened more on behalf of a particular kin and age category (i.e., maternal half-siblings being non-peers) than expected. The number of observed interventions was either divided (i) by the number of potential partners available in a particular kin and age category (hereafter: availability) or (ii) by the number of opportunities to intervene on behalf of particular kin and age categories (hereafter: opportunities) (cf. Silk et al. 2002). The number of available partners was based on the number of individuals of a particular kin and age categories present during the study period. The number of opportunities to intervene on behalf of a potential recipient were derived from the number of non-silent dyadic conflicts (see above) in which this potential recipient was involved.

Testing reciprocity or interchange, all coalitions were included where an adult female (N=49) was intervening in favour of or against another adult female due to the limitations in the program used for matrix correlation (see below). The third party in this coalition could be any other individual of group R including adult males.

Testing co-operation, all coalitions were included where an adult female (N=49) was intervening either in favour of a maternal half-sibling, a paternal half-siblings or a non-kin. Again, the third party in this coalition could be any other individual of group R including adult males.

The unit for analyses were either the individual or a dyad units (for the latter see Matrix tests below).

Individual units

However, for most analyses the behaviour of a given focal female, as an individual’s behavioural strategy, was the focus of interest. All interactions in a focal female protocol involved the focal female either as the initiator or the recipient of this particular interaction with any other group female. Females differed in the number of partners in each kin and age classes. For each focal female, her proximity index towards any given social partner was calculated. Then the mean proximity was taken over all female group members meeting a [page 21↓]specific kin and age category with respect to the focal female considered. Analogously, mean indices were calculated for each behaviour. Mean rates per hour for all behaviours (except coalition formation see above) were then compared between kin and age categories by means of paired t-tests applied to the focal females.

If the assumptions of parametric tests were not met (e.g., unequal variance between the categories, data not normally distributed) non-parametric tests such as the Wilcoxon-test were used. In the case of small sample size, the exact instead of the asympotic p-value was calculated for non-parametric tests (Mundry & Fischer 1998). Exact p-values are indicated as P_e, while asympotic p-values are given as P.

In order to control the probability of a type I error over multiple statistical tests, the Dunn-Šidák method was adopted for each behaviour (Sokal & Rohlf 1995). The Dunn-Šidákwas favoured over the more common Bonferroni correction because the latter is more conservative. Only P-values less than or equal to the corrected significance level (P’) indicate a significant test result which will be marked in bold.

In contrast, in the next chapter (chapter 3) no behavioural data will be analysed, instead results of the paternity data in comparison with demographic data of potential sires are to be presented.

Dyadic units

As a second approach, dyadic units were analysed using the matrix correlation method (Hemelrijk 1990a,b, de Vries 1993) which was restricted to chapter 6 (coalition formation). A relationship between N individuals of a group is described as a NxN Matrix where each cell represents a dyad. Three kinds of questions were analysed: (i) reciprocity for the same behaviour among dyads (Hemelrijk 1990a), e.g. is female A grooming female B as often as female B grooms female A, (ii) interchange for different behaviours among dyads (Hemelrijk 1990a), e.g. is female A grooming female B as often as female B supports female A and (iii) the correlation between a behaviour and a dyadic attribute (such as rank distance, age distance and degree of relatedness) among dyads, e.g., is an increase in the grooming frequency between dyads associated with an increase in the degree of relatedness between these dyads. The advantages of the matrix correlation method compared to analysis using individual units is that dyadic interdependencies were taken into account. A limitation of this method is that it cannot handle multiple factors (Vervaecke et al. 2000).

Three different matrix tests were used. Firstly, to investigate the correlation between a behaviour (such as grooming, support given or support against) and a dyadic attribute (such as rank distance, age distance or degree of relatedness) on the dyadic levelthe Mantel R-test [page 22↓]was applied. Secondly, to investigate reciprocity or interchange on the individual level the
K _r -test was used. Finally, (i) to investigate the correlation between a behaviour and the degree of relatedness on the individual level while controlling for rank distance in a third matrix or (ii) to investigate reciprocity or interchange on the individual level while controlling for a dyadic attribute (e.g., maternal sibship) in a third matrix, the partial K _r -test was applied (cf. Hemelrijk 1990a,b, de Vries 1993). The Mantel R-test, in contrast to the Mantel Z-test, controls for outliers (cf. Hemelrijk 1990a,b) and investigates whether the sequence of preference among all dyads is correlating between the two matrices. The K_r-test, which also controls for outliers, investigates e.g., whether the sequence of preferred partners groomed by individual A is associated with the sequence of partners from whom individual A is being groomed. The partial K_r-test is testing the same as the K_r-test, but additionally controls for a third dyadic relation which is also described by a NxN matrix. The strength of correlation is measured as a Spearman’s rank correlation coefficient in the case of the Mantel R-test and as a row-wise Kendall correlation coefficient in the case of the K_r-test or partial K_r-test (cf. de Vries 1993). Matrix correlations were undertaken using the program MatMan by Noldus (licence to Jürgen Streich, IZW). Matrix tests were restricted to females categorised as adults (N=49) using focal and ad libitum data on grooming and support given to a recipient or support against a target. This limitation was necessary, because the program provides only 50 cells.

The significance criterion for all tests was set at alpha=0.05. All statistic tests were two-tailed and taken from Sokal & Rohlf (1995). All these analyses were performed with the SPSS 10.0 package. In order to distinguish degrees of relatedness (r) from the Pearson product moment correlation coefficient (r) and the Spearman’s rank correlation coefficient (r), the former is referred to as r_p and the latter as r_s.

Single locus versus multi-locus approach

DNA typing to identify kinship has mainly been directed at DNA of tandemly repetitive structures, such as (CA)n, (CAG)n, (GATA)n sequence motifs. They are also referred to as short tandem repeats (STRs). Depending upon the length of the repeat motif and the number of copies tandemly arranged at a single locus, repetitive DNA has been classified by Tautz (1993) as either microsatellites (1-6 bp with 10-100 copies) or minisatellites (9-100 bp with 10-1000 copies). In most cases micro- and minisatellites are characterised by an extreme hight of polymorphism (Luty et al. 1990). Polymorphisms are due to mutations such as insertion or [page 23↓]deletion of one or more repeat elements during meiosis resulting in changes of the length of the repeat block. In order to be useful for kinship testing, mutation rate has to be small enough to avoid changes of allele sizes during meiosis to still detect allele sharing between close relatives. Most micro- and minisatellites fulfil this criterion (average mutation rate estimated 10–3 per locus, Edwards et al. 1992). Microsatellites are generally preserved in closely related species (Moore et al. 1991). This allows cross-species amplification of orthologous DNA loci (Kayser et al. 1996). A second approach to kinship analysis uses multi-locus probes. These probes hybridise to multiple sites within the genome which produces a unique pattern of DNA for each individual.

Both, single and multi-locus approaches have advantages and disadvantages which vary depending upon the characteristics of the study species and the population under study (cf. Nürnberg et al. 1998). Paternity assessment by the multi-locus approach was found to have several limitations over microsatellites (ibid.). First, this method requires large amount of DNA, whereas for microsatellites small amount of even degraded DNA are sufficient. Microsatellites also promote non-invasive techniques where samples can be drawn from hair roots or faeces instead of blood (e.g., Launhardt 1998, Smith 2000). Second, data are difficult to compare between gels. Therefore a direct comparison on a single gel including the infant considered, its mother and all the putative candidates (usually 2-4) needs to be available.

At this point two problems related to the present study should be emphasised. The first point of interest is related to the fact that primates in general have a long life span (more than 30 years for rhesus macaques, van Hooff 1988). Paternity might be difficult to assess among the older individuals since it is less likely that their sires have been typed or even sampled. For that reason, it is difficult if not impossible to reconstruct complete patrilineal lineages. Therefore, the main focus of the paternity analyses was restricted to younger individuals. The second point which needs to be considered is the high number of potential sire (i) within the group (up to 62 males in the data set) and (ii) from outside the group because it can not be assumed that sirehood is restricted to resident males. Berard et al. (1993) have demonstrated that a proportion of infants was sired by males from neighbour groups. But since Cayo Sanitago is a close colony (no individuals have been added since 1938 except through births) and because the whole population has been systematically sampled (starting in 1992) it is likely that all potential sires of infants considered were actually genotyped.

Given that (i) the number of microsatellites used is high enough and (ii) that the loci used are polymorphic enough, the best method of choice seems to be a dual approach. In a first step, STR markers are used to reduce the number of potential sires, ideally to N=1. If the number [page 24↓]of potential sires could not been reduced to one sire after testing 15 DNA markers, as a second step, a DNA fingerprinting was undertaken (see below). For the discussion of this combined approach see Nürnberg et al. (1998).

DNA extraction

DNA was available from blood samples which were taken during annual trapping on Cayo Santiago. All samples required for this study were already available as part of the DFG-project of Peter Nürnberg (Nu 50/3-2) which aimed at genotyping the whole population. Genomic DNA was prepared from leukocytes using the Genomix DNA extraction kit (Labortechnik Fröbel GmbH, Lindau, Germany).

Paternity assessment via microsatellites

Primers for the STRs were chosen randomly from the human genome map and their applicability for rhesus macaques was tested on unrelated individuals. Screening of STR markers has earlier been described by Kayser et al. (1996). A total of 14 microsatellites were found to be polymorphic for rhesus macaques with a mean heterozygosity rate of 0.74 (Table 2.2). They comprise di-, tri-, tetranucleotide repeats. STR markers are as variable as in humans with the mean number of alleles of 8.6 and a slight tendency of shorter alleles in rhesus compared to humans (see Table 2.2).

Table 2.2.2: Microsatellites used for paternity analyses

Abbreviations are as follows: bp is the number of base pairs, numbers in square brackets are human allele length ranges, N_alleles is the number of alleles found on that locus, H is the observed heterozygosity, N_typed is the number of individuals typed on that locus.

In addition to the STR markers listed in Table 2.2, the highly polymorphic DQB1locus from the major histocompatibility complex (MHC) of rhesus monkeyswas analysed as described by Sauermann et al. (1996) and used for paternity determination. Briefly, exon 2 of Mamu-DQB1 was amplified using a single primer pair, followed by digestion of the PCR products with up to ten different restriction endonucleases.

Application of microsatellite markers was favoured by the simple use of the polymerase chain reaction (PCR) has especially promoted microsatellite loci. Unique PCR primers flanking the STR locus of interest were designed. With the amplification of this particular stretch the length of the alleles could identified using gel electrophoresis. The main steps of paternity analysis using microsatellites are summarised in Box 2.1.

Box 2.1: Procedure of paternity analysis using a single locus.

Amplification of the locis via PCR

PCR was performed in a total volume of 10 µl using a thermal cycler (Trioblock). The reaction mix contained 20-100 ng genomic DNA, 10x buffer (PE), 20 mM dNTP's (Boehringer, Mannheim, Germany), 10 pMol of each Primer (for- and backward), 5 unit Taq Polymerase (Applied Biosystems, except for D8S601 Quiagen), Aqua bidest., 10% Triton, 25 mM MgCl₂. The exact amount of each substance was optimised for each primer pair and is given in Table 2.3. Initial denaturation varied from 2-15 min at 92°C followed by 30-35 cycles of (i) 30-60 s at 94°C, (ii) 30-60s at the annealing temperature (see Table 2.3) and (iii) 90-120 s at 72°C with 10 min at 72°C after the last cycle (elongation). One PCR primer (forward) was labelled by fluorescin for allele size determination. Each PCR included 2 control animals of known allele sizes (positive controls).

[page 26↓]

Table 2.2.3: Master-mix for PCR amplification of the 14 microsatellites

Abbreviations are as follows: T_a is the annealing temperature. Note each master-mix was 10 µl per sample, the amount of primers in this table is just given for one of the two.

Determination of the allele size using an automatic sequencer (A.L.F.)

Allele sizes were determined using an automatic sequencer (A.L.F., Pharmacia). PCR products were transferred on a 6% polyacrylamide gel and fluorescence labelled fragments were compared against standards in order to determine both alleles of an individual. At first samples were denatured at 95°C. Depending upon the length of the PCR products gels ran at 900V, 45mA, 45W, 45°C between 2 and 4 hours. Peaks of the microsatellites and standards were analysed with a service tool (fragment manager of A.L.F.). More details on chemicals used are given in Voltz (1999).

Identifying the sire using the program FIND SIRE

As an individual inherits one allele by its mother (which was genotyped and therefore known by the investigator), the second allele, if the infant is heterozygous at this locus, must come from its father. All homozygous genotypes were at least genotyped twice. If homozygosity was not confirmed, at least two identical results had to be obtained before the genotype was entered in the data base.

Any male 1250 days older than a given infant, and residing on the island 200 days prior to the infant’s birth, was considered a potential sire in the paternity analyses. After the use of a maximum of 15 markers most potential sires could be excluded, except from 1 male. The genotypes were transferred into a data base and genotypes of the infant-mother pair were compared against all potential sires using the FIND SIRE program written by Michael Krawczak (Kiel, Germany).

Solving individual cases

Here the point should be made, that paternity for each individual offspring (see paternity data set below) had to be established for every single offspring. In other words, genotyping mother-offspring pairs and all potential sires was only the first step. Any discrepancy between mothers and their offspring had to be re-analysed in order to exclude typing errors. This also revealed systematic typing errors which may have had an influence on the whole data set. Some mutation events could be verified.

Criterions of established paternity

Following Krawczak (1999) Log10-likelihood ratios (LR) for paternity vs. unrelatedness were calculated, using the 15 markers for which nearly all animals were genotyped. Paternity was regarded as established when a male fitted into two criterions: (i) he had to reach a LR in favour of paternity that was larger than two (corresponding to a standardised paternity probability of 99%) and (ii) his LR had to be at least one unit larger than the LR of any other male.

Some potential sires have not have been successfully genotyped on all 15 markers and which marker has not been genotyped also varied among potential sires. The LR calculation for a given infant and his potential sire A was based on all marker available for both, but LR calculation for the same infant and his potential sire B may have been based on other markers depending upon which they both shared. However, because the two paternity criterions listed above are very strict, it is unlikely that this was influencing the decision in favour of a male being the actual sire (Krawczak, pers. comm.).

Paternity assessment via DNA fingerprinting

If no unequivocal sire could be determined after the use of all 15 DNA markers, DNA fingerprinting was applied to detect paternity. For example, two males had reached a log-likelihood ratio larger than two, but the differences was smaller than one unit. It could have been that these two males were actually father and son themselves as assessed via [page 28↓]microsatellites, but the only way to prove which of the two male sired the considered infant, a comparison of the males DNA fingerprint with whose of the infant-mother dyad was needed.

The methodology of DNA fingerprinting using oligonucleotide probes has been described in detail by Epplen (1992). In this study three synthetic oligonucleotide probes (GATA)₄, (CA)₈ and (GTG)₅ were used to produce a specific band pattern per individual. Infant-mother dyads were compared on the same gel with the sires in question in order to avoid problems derived from comparing gels (Nürnberg et al. 1998). DNA samples were digested with the restriction endonuclease Hinf I under standard conditions and DNA fragments were separated on 30 cm 0.7% agarose gels in TAE buffer for up to 48 hours at 1V/cm. ³²P-labeled probes were hybridised to DNA in the dried gels, and exposed to Kodak XAR-5 film over night (cf. Berard et al. 1993, 1994). The DNA fingerprint for each infant were then compared to those of its mother. Bands which the infant did not share with its mother (non-maternal bands) had to be present in the actual sire. All offspring bands of necessarily paternal origin were marked and compared directly to the corresponding mother-infant dyad. A male was regarded as the sire, if he had no mismatches in all three probes for all non-maternal bands, but all other males had at least two mismatches. This techniques was conducted by Ingrid Barth. Band sharing was analysed by visual insepection by myself.

Paternity data sets

Paternity analyses have been part of a co-operation aimed at the typing of the whole population of Cayo Santiago. Paternity assessment in groups R and M was undertaken by a team headed by Peter Nürnberg (Charité, Humboldt-University of Berlin), in the remaining groups were analysed by a team headed by Jörg Schmidtke (Medizinische Hochschule, Hannover). This thesis uses two kinds of paternity data sets: (i) all individuals born in group R or its sister group BB between 1993 and 1998 and (ii) all females in group R alive at the time of behavioural study in 1997. While the first data set is only used in chapter 3 (skew), paternity data of females, present in group R in 1997, are used in all remaining chapters 4-6. More information on the paternity results are given in the Appendix 3 (Paternity results).