1. Ecology
  2. Evolutionary Biology
Download icon

Social groups buffer maternal loss in mountain gorillas

  1. Robin E Morrison  Is a corresponding author
  2. Winnie Eckardt
  3. Fernando Colchero
  4. Veronica Vecellio
  5. Tara S Stoinski
  1. Dian Fossey Gorilla Fund, Rwanda
  2. Centre for Research in Animal Behaviour, University of Exeter, United Kingdom
  3. Department of Mathematics and Computer Science, University of Southern Denmark, Denmark
  4. Interdisciplinary Center on Population Dynamics, University of Southern Denmark, Denmark
Research Article
  • Cited 0
  • Views 809
  • Annotations
Cite this article as: eLife 2021;10:e62939 doi: 10.7554/eLife.62939

Abstract

Mothers are crucial for mammals’ survival before nutritional independence, but many social mammals reside with their mothers long after. In these species the social adversity caused by maternal loss later in life can dramatically reduce fitness. However, in some human populations these negative consequences can be overcome by care from other group members. We investigated the consequences of maternal loss in mountain gorillas and found no discernible fitness costs to maternal loss through survival, age at first birth, or survival of first offspring through infancy. Social network analysis revealed that relationships with other group members, particularly dominant males and those close in age, strengthened following maternal loss. In contrast to most social mammals, where maternal loss causes considerable social adversity, in mountain gorillas, as in certain human populations, this may be buffered by relationships within cohesive social groups, breaking the link between maternal loss, increased social adversity, and decreased fitness.

eLife digest

Most mammals depend entirely upon their mothers when they are born. In these species, losing a mother at a young age has dramatic consequences for survival. In cases where orphaned individuals do reach adulthood, they often suffer negative effects, like reduced reproductive success or lower social status. But this is not the case for humans. If a child loses their mother; relatives, friends and the wider community can take over. This does not tend to happen in nature. Even our closest relatives, chimpanzees, are much less likely to survive if their mothers die before they reach adolescence.

Although orphan survival is not the norm for mammals, humans may not be entirely unique. Mountain gorillas also live in stable family groups, usually with a dominant male and one or more females who care for their offspring for between 8 and 15 years. It is possible that gorillas may also be able to provide community support to orphans, which could buffer the costs of losing a mother, just as it does in humans.

To answer this question, Morrison et al. examined 53 years of data collected by the Dian Fossey Gorilla Fund to assess the effects of maternal loss in mountain gorillas. The analysis examined survival, reproduction and changes in social relationships. This revealed that, like humans, young gorillas that lose their mothers are not at a greater risk of dying. There is also no clear long-term effect on their ability to reproduce. In fact, gorillas who lost their mothers ended up with stronger social relationships, especially with the dominant male of the group and young gorillas around the same age. It seems that gorilla social groups, like human families, provide support to young group members that lose their mothers.

These findings suggest that the human ability to care for others in times of need may not be unique. It is possible that the tendency to care for orphaned young has its origins in our evolutionary past. Understanding this in more depth could provide clues into the social mechanisms that help to overcome early life adversity, and have a positive impact on future health and survival.

Introduction

Maternal loss, along with a number of other indicators of early-life adversity, is one of the strongest predictors of lifespan in humans and other social mammals (Snyder-Mackler et al., 2020). In mammals, mothers are vital for the survival of young offspring, providing nutrition, thermoregulation, and protection (Clutton-Brock, 1991). In some species, particularly social species with slow life histories (Mitani et al., 2013), mothers continue to provide benefits to their co-residing offspring throughout immaturity and even into adulthood (Surbeck et al., 2019; Surbeck et al., 2011; Andres et al., 2013; Stanton et al., 2020). The active support of mothers can increase the rank of their offspring (Strauss et al., 2020; East et al., 2009; Maestripieri and Mateo, 2009; Lea et al., 2014) and improve their integration in the group (Tung et al., 2016), both of which are linked with greater survival (Snyder-Mackler et al., 2020; Archie et al., 2014). Maternal presence can also influence nutrition at these later stages of development by buffering against feeding competition (Samuni et al., 2020), providing access to valuable ecological knowledge (Stanton et al., 2020; Foley et al., 2008; Brent et al., 2015) or increasing opportunities for the social learning of complex feeding techniques (Lonsdorf et al., 2004; Estienne et al., 2019).

In social mammals maternal loss can therefore reduce the fitness of offspring across a broad age range through long-term effects on their social environment, negatively influencing their social integration and social status throughout their lives (Snyder-Mackler et al., 2020; Strauss et al., 2020; Tung et al., 2016). Multiple studies have now confirmed effects on survival for individuals orphaned well past the period of nutritional dependency, with these negative changes to their social environment, often termed social adversity, posited to be the key mechanism by which this occurs (Andres et al., 2013; Stanton et al., 2020; Tung et al., 2016; Watts et al., 2009; Foster et al., 2012). In social mammals, the social environment can have extreme consequences for health, fitness, and lifespan, mediated through pathways such as chronic stress, immune function, or environmental exposure (Snyder-Mackler et al., 2020).

The impact of maternal loss varies based on the loss in benefits relative to individuals with mothers present. In species with sex-biased dispersal, the consequences of maternal loss can differ between the sexes due to longer periods for potential investment in non-dispersing offspring (Clutton-Brock, 1991; Fairbanks, 2009; Altmann and Alberts, 2005; Greenwood, 1980). In female-philopatric red deer (Cervus elaphus), maternal loss increases mortality for males and females, but this effect is only detectable in males under 2 years of age, whilst for females, maternal benefits continue throughout their lives (Andres et al., 2013). In male-philopatric chimpanzees (Pan troglodytes), males that suffer maternal loss before reaching 15 years of age have lower survival, whilst females only show reduced survival from maternal loss under the age of 10 (Nakamura et al., 2014; Stanton et al., 2020). In killer whales (Orcinus orca), where neither sex disperses, maternal loss when offspring are over 30 years old reduces survival for both sexes, although considerably more so for males (Foster et al., 2012). However, maternal loss between 15 and 30 years appears to reduce male but not female survival. This is thought to be due to higher maternal investment in males which mate outside the group and whose offspring therefore do not increase within-group feeding competition.

As a result of the numerous benefits mothers can provide, it is not surprising that maternal loss not only influences offspring survival but can also impact other components of their offspring’s fitness, such as reproduction and the survival of grand-offspring. Male bonobos (Pan paniscus) residing in groups with their mothers sire three times the number of offspring (Surbeck et al., 2019), whilst maternal loss before weaning negatively affects antler development in male red deer – a trait found to correlate with reproductive success (Andres et al., 2013). In chimpanzees, females mature faster, first give birth younger (Walker et al., 2018) and enter the dominance hierarchy higher (Foerster et al., 2016) if their mothers are present which is expected to considerably increase their lifetime reproductive success. In savannah baboons (Papio cynocephalus), if mothers had themselves suffered maternal loss in the first 4 years of their life, their offspring had 48% higher mortality throughout the first 4 years of their life, suggesting an intergenerational effect of maternal loss driven by lifelong developmental constraints (Zipple et al., 2019). Maternal loss in social mammals with extended maternal care can therefore have long-term fitness consequences mediated through multiple pathways that detrimentally affect survival and reproduction.

Due to the extended periods of mother–offspring co-residence and the important social support mothers can provide, highly social species often have the most to lose from maternal loss. However, social groups also provide the potential for support from other group members following maternal loss. Both kin and non-kin group members have been suggested to compensate for the loss of close kin to varying extents (Hamilton et al., 1982; Engh et al., 2006; Goldenberg and Wittemyer, 2017; Reddy and Mitani, 2019) and the strengthening of relationships with remaining group members may buffer against changing social environments (Firth et al., 2017). In chacma baboons (Papio ursinus), social support from group members is thought to alleviate the stress of losing a close relative (Engh et al., 2006). Similar social support has been suggested in African savannah elephants (Loxodonta africana) which associate more with group members of a similar age and siblings in response to maternal loss (Goldenberg and Wittemyer, 2017). However, these orphaned elephants interact less with matriarchs which may decrease their access to key knowledge and high-quality resource patches. In chimpanzees, older siblings can ‘adopt’ younger siblings after maternal loss, increasing their social contact and showing heightened vigilance in dangerous situations (Reddy and Mitani, 2019; Hobaiter et al., 2014). But despite these compensatory social behaviours, the negative consequences of maternal loss post-weaning are well documented in all three genera (Stanton et al., 2020; Tung et al., 2016; Samuni et al., 2020; Nakamura et al., 2014; Goldenberg and Wittemyer, 2017; Goldenberg and Wittemyer, 2018).

Humans (Homo sapiens) are a rare example of a group-living mammal in which compensatory social behaviours have been suggested to have the capacity to consistently overcome these negative consequences of maternal loss. In a meta-analysis of historic and contemporary human populations, the death of a mother was associated with increased child mortality in all 28 populations studied (Sear and Mace, 2008). However, this effect appeared to decline substantially with age, disappearing for children that suffered maternal loss over 2 years of age in 5 of the 11 populations in which it was investigated (Sear and Mace, 2008). This reduced mortality was thought to be due to the care provided by other kin, particularly after weaning, suggesting that social buffering from other group members can overcome the negative effect of maternal loss on survival in certain circumstances. Whilst the effects of care from specific kin members varied across populations, at least one kin member significantly impacted child survival in all studies (Sear and Mace, 2008), and there is evidence for the importance of maternal grandmothers (Lahdenperä et al., 2004; Sear et al., 2000) and fathers (Hurtado and Hill, 1992; Hill and Hurtado, 2017) in particular. In killer whales, maternal grandmothers, especially those that are post-reproductive, are also known to improve grand-offspring survival (Nattrass et al., 2019). Whilst the specific effect of killer whale grandmothers on orphan survival has not been investigated, this finding suggests that killer whales may represent a further species in which care from other kin has the capacity to overcome the effects of maternal loss. In humans, there is also evidence for the benefits of care provided by non-kin such as step-mothers (Andersson et al., 1996; Campbell and Lee, 2002) and through the modern practices of non-kin adoption (Bentley and Mace, 2009).

Mountain gorillas (Gorilla beringei beringei) show extended maternal care with offspring remaining in their natal groups at least until sexual maturity and approximately half remaining beyond sexual maturity (48% of females [Robbins et al., 2009a] and 55% of males [Stoinski et al., 2009a]). Females that disperse from their natal group tend to do so earlier (mean age of 7.9 years [Robbins et al., 2009a]) than males (mean age of 15.3 years [Stoinski et al., 2009a]) and therefore have a shorter period of potential maternal investment. The complexity of gorilla social structure with numerous types of differentiated social relationship both within and among groups (Morrison et al., 2019; Mirville et al., 2018; Morrison et al., 2020a; Morrison et al., 2020b) suggests that detrimental long-term effects on individual gorillas’ social environments could have particularly negative fitness consequences. However, these stable, cohesive, social groups also have the potential to provide a social buffer to the negative consequences of maternal loss. Mountain gorilla groups either contain a single adult male (approximately 64% of groups) or multiple adult males (approximately 36% of groups) (Gray, 2010), at least one adult female, and their offspring (Robbins, 1995). Single male groups are polygynous whilst multimale groups have high reproductive skew towards the dominant male who sires the majority of offspring (Nsubuga et al., 2008; Stoinski, 2009b; Bradley et al., 2005). Infants (<4 years of age) are nutritionally dependent on their mothers until being weaned at a mean of 3.3 years (Eckardt et al., 2016) and are reliant on their mothers for thermoregulation and transport, being carried for prolonged periods (Breuer et al., 2009). Juveniles (4–6 years old) are nutritionally independent but remain in close proximity to their mothers the majority of the time (Breuer et al., 2009). Dominant males’ primary form of care is through protection from out-group males and potential predators (Harcourt and Greenberg, 2001). But they also show high levels of affiliative behaviour towards infants, grooming and resting in contact with them, with no evidence that they discriminate between infants based on paternity (Rosenbaum et al., 2018; Rosenbaum et al., 2015).

In this study, we use the long-term demographic records of the Dian Fossey Gorilla Fund’s Karisoke Research Center collected over 53 years (1967–2019) to (A) quantify the effects of maternal loss on multiple fitness measures: survival, female age at first birth, female survival of first offspring through infancy, and male dominance; and (B) investigate the social responses of group members to maternal loss by immature gorillas. We hypothesize that as demonstrated in chimpanzees (Stanton et al., 2020), gorillas may face greater fitness costs if they suffer maternal loss at an earlier age and that males may face greater costs from maternal loss than females due to their longer periods of mother–offspring co-residence. Alternatively, as observed in many human populations, the cohesive, stable social groups of mountain gorillas may enable social buffering from group members to compensate for the social costs of maternal loss with minimal fitness consequences to maternal loss. In particular, dominant males may take on crucial roles in buffering the social adversity faced by maternal orphans (hereafter, orphans), as past research has demonstrated the strong bond between dominant males and young orphans who may regularly share a nest at night (Robbins et al., 2005; Gatesire et al., 2016).

Results

Effect of maternal loss on survival

To determine the effect of maternal loss on survival, we carried out a Cox-proportional hazards analysis separating individuals into four orphan classes based on their age when their mother died: (a) infants (2–4 years old), (b) juveniles (4–6 years old), (c) subadults (6–8 years old), and (d) non-orphans (>8 years old) if their mothers died after they had reached maturity. Due to the small sample sizes for the juvenile and subadult classes, analysis was also run with these two classes merged. We found no significant differences in survival between all orphan classes and the non-orphan class for both sexes irrespective of using three or four classes (Table 1, Figure 1, Figure 1—figure supplement 1,Supplementary file 1 -Table 1). Bayesian survival trajectory analysis (Colchero and Clark, 2012; Colchero et al., 2012) showed similar results, whereby the model with highest support for both sexes was the null model without orphan classes as covariates (Table 2).

Figure 1 with 1 supplement see all
Survivorship curves for each of the four maternal loss categories for each sex.

Plots show the proportion of surviving (a) females (n = 98) and (b) males (n = 102) that suffered maternal loss as infants, juveniles, and subadults compared to non-orphans that did not suffer maternal loss under the age of 8 years. Figure 1—figure supplement 1 shows survivorship curves of each orphan age and sex class plotted separately against non-orphans for further clarity. Grey dashed line indicates the age from which orphan and non-orphan survival are modelled separately.

Table 1
Cox-proportional hazards models showing the effects of the four maternal loss categories: infants, juveniles, subadults, and non-orphans, for each sex (98 females and 102 males), on survival.

All results are relative to the non-orphan class.

FemalesMales
Age-classEst ± SEpEst ± SEp
Infants0.73 ± 0.5920.218−0.34 ± 0.5510.540
Juveniles1.22 ± 1.1700.2980.05 ± 1.0600.959
Subadults1.77 ± 1.2400.1520.59 ± 1.1100.591
Table 2
Deviance information criterion (DIC) for the three models tested.

(a) No covariates (i.e. null model where all individuals have the same hazard rate); (b) proportional hazards (where mortality differs proportionally between orphan classes); (c) covariates modifying all Siler mortality parameters (where each orphan class has a different age-specific mortality), for each sex (98 females and 102 males). The delta DIC shows the difference in DIC from the model with lowest DIC.

FemalesMales
ModelDICΔ DICDICΔ DIC
No covariates371.460486.080
Prop. hazards374.162.71486.440.36
All mortality parameters377.045.59499.4613.38

Effect of maternal loss on dispersal

We examined the effect of maternal loss on the likelihood of females dispersing before giving birth to their first offspring using a binomial generalized linear model (n = 51). Female orphans were not significantly more likely to disperse from their natal group prior to first birth than female non-orphans (Table 3). However, there was a close to significant increase in the likelihood of dispersal for females that lost their mothers as juveniles or subadults. 37.5% of non-orphan females (n = 32) dispersed prior to their first birth (mean dispersal age ± SD: 7.96 ± 1.55 years) compared to 54.5% (n = 11) of infant orphans (dispersal age: 7.75 ± 0.60 years) and 75.0% (n = 8) of juvenile and subadult orphans (dispersal age: 8.21 ± 2.36 years). We examined the effect of maternal loss on male dispersal based on whether a male had dispersed from their natal group prior to the age of 16 years (the median age of male dispersal). Only three males that reached the age of 16 had lost their mothers as infants, but all three remained in their natal group. Due to this small sample size this was not examined statistically. However, a binomial generalized linear model demonstrated that juvenile and subadult orphan males were significantly more likely to disperse before reaching 16 years of age (84.6%, n = 13) than non-orphan males (37.5%, n = 40, Table 3).

Table 3
The influence of age at maternal loss (infant or juvenile/subadult (J/SA)) relative to non-orphans on a female’s decision to disperse from their natal group prior to their first birth and a male’s decision to disperse prior to the age of 16, modelled using binomial generalized linear models.
Females (n = 51)Males (n = 53)
Est ± SETPEst ± SETp
Intercept−0.511 ± 0.365−1.3990.162−0.511 ± 0.327−1.5640.118
Infant0.693 ± 0.7070.9800.327---
J/SA1.609 ± 0.8941.7990.0722.216 ± 0.8352.6530.008

Effect of maternal loss on female reproduction

Using a generalized linear model, we found that maternal loss had no significant effect on the age at which females first gave birth (Supplementary file 1 - Table 2, n = 53). The mean age at first birth (± SD) for non-orphans was 10.24 ± 1.61 years compared to 9.72 ± 0.73 years for those orphaned as infants and 9.67 ± 1.82 years for those orphaned as juveniles or subadults. After accounting for age at first birth and dispersal, there was also no evidence that maternal loss influenced whether a female’s first offspring survived infancy (Supplementary file 1 - Table 3, n = 50, binomial generalized linear model). 51.5% of non-orphan females’ first-born offspring (n = 33) survived infancy compared to 60% (n = 10) of first-born offspring of those orphaned as infants and 57.1% (n = 7) of first-born offspring of those orphaned as juveniles or subadults.

Effect of maternal loss on male dominance attainment

The mean age (± SD) at which a male first became the dominant male of a group was 17.88 (± 2.56) years. The oldest that a male first reached dominance was 22.99 years, with all males that had not become dominant by this age, never reaching dominance. We therefore compared the proportion of males over the age of 23 that had attained dominance in each orphan class. 52% of non-orphan males had become the dominant male of a group for at least 6 months (n = 21) by the age of 23 compared to all three infant-orphaned males and no juvenile- or subadult-orphaned males (n = 5). This finding suggested maternal loss as a juvenile or subadult male could limit a gorilla’s ability to become dominant. However, this could also be purely an artefact of the small sample sizes. We therefore examined the dominance status of the seven juvenile- or subadult-orphaned males that were over the age of 16 but had not yet reached 23 years by the end of the study period. 71% (n = 7) of these males had already become dominant despite their younger age, indicating the capacity for males orphaned in any age category to become dominant later in life.

Changes in network position following maternal loss

The social responses of group members to 21 incidents of maternal loss were investigated (Supplementary file 1 - Table 4). Focal data collected daily in each group were used to construct social networks based on (a) 2 m proximity and (b) affiliative contact (resting, playing, or feeding in physical contact and grooming, but excluding physical aggression) for the 6 months leading up to a maternal loss incident and the 6 months immediately after a maternal loss incident. In the 6 months prior to maternal loss, orphans had spent a mean of 13% (± 8, n = 31) of their time while monitored in affiliative contact with their mother and 39% (± 16, n = 31) of their time within proximity (<2 m) of their mother, who was predominantly their closest social partner (Table 4). After maternal loss, orphan’s affiliative contact with other group members increased on average by 28% and the proportion of their time spent within 2 m of other group members increased on average by 42%. Overall, this resulted in a net decrease in orphan’s affiliative contact of 31% and a net increase in orphan’s proximity to others of 11% following maternal loss.

Table 4
The percentage of gorilla orphans for which their mother was their closest social partner prior to maternal loss based on affiliative contact and proximity within 2 m.
ContactProximity
Infants (n = 9)89%78%
Juveniles (n = 14)86%79%
Subadults (n = 8)25%75%

To investigate how these changes influenced the social network position of orphans we compared orphan’s change in binary degree (number of connections), weighted degree (strength of connections), and eigenvector centrality (how connected they were to other well-connected individuals) with that of non-orphans within the same networks (n = 136). These networks included only individuals that were present both before and after the maternal loss incident (excluding the mothers of orphans) to enable direct comparison. Using generalized additive mixed models (GAMMs) with node-level permutations of orphan status we found that within proximity-based networks, orphans’ eigenvector centrality (Est = 0.169 ± 0.037, t = 4.594, p < 0.001, Pnull < 0.001) and weighted degree (Est = 0.075 ± 0.032, t = 2.337, p=0.021, Pnull = 0.024) increased significantly more after maternal loss than non-orphans within the same pair of networks. This led to orphans and non-orphans having similar weighted degree and centrality values following maternal loss despite orphans losing a key social partner (Figure 2, Supplementary file 1 - Table 5). However, within contact-based networks orphans did not show these same gains, with no significant change in eigenvector centrality (Est = −0.034 ± 0.057, t = −0.593, p = 0.554, Pnull = 0.598) or weighted degree (Est = 0.031 ± 0.047, t = 0.656, p = 0.513, Pnull = 0.442) relative to non-orphans (Figure 2—figure supplement 1, Supplementary file 1 - Table 5). Binary degree did not increase to a greater extent in orphans than non-orphans in either contact or proximity-based networks (Supplementary file 1 - Table 5).

Figure 2 with 1 supplement see all
The change in proximity-based social network position (eigenvector centrality, weighted degree, and binary degree) for orphaned (n = 28) and non-orphaned immature gorillas (n = 108) within the same group in the 6 months before an incident of maternal loss and the 6 months after.

Weighted degree and binary degree values are calculated as proportions of the greatest value observed within their specific network to enable comparison across multiple networks (multiple incidents of maternal loss). Networks include only individuals that were present both before and after maternal loss and therefore exclude the mothers of orphans. Blue lines and shading indicate mean and 95% confidence interval. Figure 2—figure supplement 1 shows the change in affiliative contact-based social network position. Source data available in file: ‘Figure 2—source data 1’.

Figure 2—source data 1

Changes in the network position of orphaned and non-orphaned immature gorillas following an incident of maternal loss.

https://cdn.elifesciences.org/articles/62939/elife-62939-fig2-data1-v1.xlsx

Relationship changes following maternal loss

GAMMs were also used to investigate changes in individual pairwise relationships pre- and post-maternal loss. Our first pair of models included all pairwise relationships involving an immature gorilla (both orphans and non-orphans). They demonstrated that affiliative contact with dominant males, subordinate males, and subadult females increased to a greater extent for orphans than other immature gorillas within the same group during the same time period (n = 3486, Figure 3; Supplementary file 1-Table 6). Based on proximity, orphan’s relationships with dominant males, subordinate males, adult females, subadult females, and juveniles strengthened more than those between other immature gorillas and the same age-sex classes of group members (n = 3486, Figure 3; Supplementary file 1 Table 6). Our second pair of models examined only pairwise relationships involving an orphan (n = 755) to provide more detailed information on how orphans relationships changed. The extent to which an orphan’s affiliative contact with and proximity to other group members increased following maternal loss did not differ depending on the orphan’s sex (Table 5). The increase in proximity with other group members following maternal loss was smaller for older orphans but this difference was not significant for affiliative contact (Table 5). This suggests that social support after maternal loss through proximity with other group members is lower for older orphans who may already have been less reliant on their mothers. Age-mates (those within 2 years age of the orphan) showed a greater increase in proximity after maternal loss relative to other group members. However, this was not the case for affiliative contact (Table 5). The change in relationship strength between maternal siblings (hereafter, siblings) after maternal loss depended on the age-sex class of the sibling (Figure 3—figure supplement 1). Subordinate adult males and subadult females had more affiliative contact with younger siblings following maternal loss but siblings in all other age-sex classes did not (Table 5). Both forms of social support (affiliative contact and proximity) showed the greatest increase from dominant males (Figure 3, Table 5). For affiliative contact this was significantly greater than all other age-sex classes, but for proximity the increase was only significantly greater than that of subordinate adult males.

Figure 3 with 2 supplements see all
The change in relationship strength (SRI) between immature gorillas and other group members in both affiliative contact and proximity, between the 6 months prior to an incident of maternal loss and the 6 months post-maternal loss (n relationships = 3592, orphaned immature gorillas = 31, non-orphaned immature gorillas = 51).

Black points show values for orphans, and grey points show values for immature gorillas within the same group that did not suffer maternal loss. Error bars indicate the standard error. Dashed red line indicates no change in relationship strength. NS indicates no significant difference between orphan and non-orphan changes in relationship strength, * indicates significance at <0.05, ** indicates significance at <0.01, and *** indicates significance at <0.001. Exact p-values (L to R), contact: <0.001, 0.001, 0.997, 0.280, 0.122, 0.021, 0.941, 0.019; proximity: 0.006, 0.009, 0.004, 0.265, 0.680, <0.001, <0.001, 0.328. Source data available in file: ‘Figure 3—source data 1’. Figure 3—figure supplement 1 shows the change in relationship strength between orphans and group members that were their maternal siblings (dark blue) and those that were not (light blue) after maternal loss based on (A) affiliative contact and (B) proximity. Figure 3—figure supplement 2 shows the change in relationship strength between orphans and dominant and subordinate adult male group members by kinship after maternal loss based on (A) affiliative contact and (B) proximity.

Figure 3—source data 1

Changes in dyadic relationship strengths following an incident of maternal loss within a gorilla group.

https://cdn.elifesciences.org/articles/62939/elife-62939-fig3-data1-v1.xlsx
Table 5
GAMMs predicting the change in dyadic relationship strength (SRI values for affiliative contact and proximity) between orphans (O) and other group members (GM) following maternal loss (n = 755).
Affiliative contactProximity
Est ± SEZpEst ± SEZp
Intercept0.028 ± 0.0055.508<0.0010.058 ± 0.0183.1800.002
Orphan
Age (years)−0.001 ± 0.000−1.6400.101−0.004 ± 0.002−2.0330.042
Sex (male)0.000 ± 0.002−0.0340.973−0.003 ± 0.006−0.5030.615
Group member
Age-mate (within 2 years)−0.001 ± 0.002−0.6230.5340.020 ± 0.0072.8850.004
Maternal sibling−0.014 ± 0.011−1.3050.1920.010 ± 0.0370.2840.777
Group member age/sex class (relative to a dominant male)
Adult male (subordinate)−0.023 ± 0.005−4.911<0.001−0.033 ± 0.017−1.9730.049
Adult female−0.024 ± 0.005−5.083<0.001−0.016 ± 0.017−0.9590.338
Blackback male−0.023 ± 0.005−4.731<0.001−0.032 ± 0.018−1.7820.075
Subadult male−0.026 ± 0.005−4.801<0.001−0.021 ± 0.020−1.0380.299
Subadult female−0.018 ± 0.005−3.2050.001−0.003 ± 0.020−0.1510.880
Juvenile−0.021 ± 0.005−3.930<0.001−0.002 ± 0.019−0.1240.902
Infant−0.021 ± 0.005−4.224<0.001−0.006 ± 0.018−0.3340.738
Age/sex class–sibling interaction (relative to a non-sibling dominant male)
Adult male (sub.) sibling0.041 ± 0.0133.2100.0010.052 ± 0.0431.2310.219
Adult female sibling0.015 ± 0.0141.0160.310−0.048 ± 0.048−0.9860.324
Blackback male sibling0.004 ± 0.0150.2430.808−0.044 ± 0.049−0.8960.371
Subadult male sibling0.001 ± 0.0170.0420.966−0.107 ± 0.057−1.8760.061
Subadult female sibling0.058 ± 0.0173.3870.0010.011 ± 0.0580.1970.844
Juvenile sibling−0.020 ± 0.021−0.9550.340−0.118 ± 0.072−1.6430.101
Infant sibling−0.004 ± 0.015−0.2820.778−0.096 ± 0.052−1.8500.065
  1. Smooth term: s(mean focal scans) Contact: F = 6.955, p=0.008; Proximity: F = 7.288, p<0.001.

Table 5—source data 1

Model: relationship change ~ age_O + sex_O + age mate + age sex class_GM * sibling + s(MeanDen), random = ~(Group|ID_O) + (1|ID_GM).

https://cdn.elifesciences.org/articles/62939/elife-62939-table5-data1-v1.xlsx

Changes in the relationship with adult males following maternal loss

For 67% of orphans of known paternity (n = 18), the dominant male at the time of maternal loss was their genetic father. Paternity did not influence the social support provided by adult males after maternal loss (contact: z = 1.130, p=0.262; proximity: z = −0.552, p = 0.583) but adult male maternal siblings increased both their affiliative contact and proximity more than non-siblings (contact: z = 3.807, p < 0.001; proximity: z = 2.237, p = 0.028). However, the increased social support from adult male maternal siblings relative to non-siblings was largely driven by an effect in subordinate males, whilst social support from dominant males did not differ greatly by kin relationship (Supplementary file 1 - Table 7, Figure 3—figure supplement 2).

Discussion

In contrast to many social mammals with extended periods of mother–offspring co-residence (Andres et al., 2013; Stanton et al., 2020; Foster et al., 2012), in mountain gorillas, we found no evidence for higher mortality in immature offspring of either sex following maternal loss. Whilst our sample size of 59 orphans between the ages of 2 and 8 years may limit our ability to detect relatively weak effects on survival, significant effects have been found in other species with comparable sample sizes (Stanton et al., 2020). This suggests that either maternal loss after offspring have reached 2 years of age does not reduce survival in mountain gorillas, or that the reduction in survival is considerably weaker than that observed in other species and therefore undetectable within our sample.

The effect of maternal loss on dispersal decisions differed depending on the age at which maternal loss occurred. Whilst no effect was found for those orphaned as infants, juvenile- or subadult-orphaned gorillas were more than twice as likely to disperse than non-orphans (although for females this effect was not quite significant). This suggests that there may be some benefits to co-residence with mothers for both sexes that do not directly translate to survival, but could influence lifetime reproductive success, like the faster maturation and higher dominance in female chimpanzees (Walker et al., 2018; Foerster et al., 2016) or greater mating opportunities in male bonobos (Surbeck et al., 2019). Alternatively, potential benefits may not be specific to the mother–offspring bond. Instead, the loss of a key social partner close to dispersal age may reduce the social benefits of remaining in a group. Without this strong social bond in the natal group, dispersal to a new group may lead to fewer social costs relative to remaining, whilst also providing considerable benefits for inbreeding avoidance (Vigilant et al., 2015). This might better explain the age-dependent influence of maternal loss, as by the time those suffering maternal loss as infants approach an age at which dispersal could occur, they may have compensated for the loss of such a key social partner through strengthening their relationships with other group members. Our analyses also suggest that the strengthening of relationships with group members post-maternal loss may be greatest for younger individuals, which could further reduce the likelihood that they disperse later in life.

We found no significant effect of maternal loss on the elements of female reproduction we investigated. Female orphans gave birth slightly younger, but despite this, their first offspring was marginally more likely to survive infancy. Whilst neither effect was significant, the direction of the effect, with earlier first birth, is consistent with the stress acceleration hypothesis: that those suffering from early caregiving adversity may show accelerated development as an adaptive strategy to low parental care (Ellis, 2004; Callaghan and Tottenham, 2016). Female dispersal has also not been found to cause any reproductive delays in mountain gorillas, suggesting that the potentially higher likelihood of those suffering maternal loss as juveniles or subadults to disperse is unlikely to reduce their reproductive capacity (Robbins et al., 2009b).

Male reproduction was harder to assess due to limited paternity data and later sexual maturity. However, as dominant males usually sire the majority of offspring, even within multi-male groups (Nsubuga et al., 2008; Stoinski, 2009b; Bradley et al., 2005), analyses of male dominance status provided some insights into the reproductive success of male orphans. Although only around half of males ever reached dominant male status, some males in both orphan age categories were able to do so, demonstrating that this is possible in the absence of maternal support. Small sample sizes limited the extent of analysis possible, but maternal loss appeared to have differing effects depending on the age of maternal loss, with infant-orphaned males more likely and juvenile- or subadult-orphaned males less likely to become the dominant male of a group relative to males that did not suffer maternal loss. This mirrors the effect of maternal loss on dispersal decisions. Modelling has suggested that males that disperse suffer a 50% reduction in lifetime reproductive success (Robbins and Robbins, 2005), as they lose the potential for mating opportunities as a subordinate male within their natal group or for eventually taking over as dominant male of that group and must instead attempt to attract females to form a group of their own (Stoinski et al., 2009a). It is therefore likely that males suffering maternal loss as juveniles or subadults may suffer from reduced siring opportunities throughout their lives due to their increased likelihood of dispersal. As discussed above, it is possible that the greater strengthening of relationships between group members and those orphaned at an earlier age may reduce the likelihood of these infant-orphaned gorillas dispersing relative to those suffering maternal loss as juveniles or subadults, ultimately influencing male orphans’ subsequent reproductive success.

Overall, these findings suggest there are no strong negative fitness consequences for maternal loss in female mountain gorillas over 2 years of age, although we cannot rule out longer-term effects on lifetime reproduction. In males, maternal loss after 2 years does not appear to influence survival but maternal loss as juveniles or subadults (4–8 years) could lower their future reproductive success. This lack of effect on survival reflects patterns found in human populations with natural fertility and mortality, where the negative effect of maternal loss on survival declines substantially with age and in many cases disappears entirely over the age of 2 (Sear and Mace, 2008). However, it is contrary to research in most other social mammals with long periods of mother–offspring co-residency (Snyder-Mackler et al., 2020; Andres et al., 2013; Stanton et al., 2020; East et al., 2009; Tung et al., 2016; Watts et al., 2009; Nakamura et al., 2014; Goldenberg and Wittemyer, 2017; Goldenberg and Wittemyer, 2018). In these species, one of the major mechanisms posited to link maternal loss with increased mortality is through the social adversity caused by mother absence. Specifically, mother absence is found to reduce social integration (Tung et al., 2016; Archie et al., 2014), competitivity (Surbeck et al., 2019; Samuni et al., 2020), dominance rank (Strauss et al., 2020; East et al., 2009; Maestripieri and Mateo, 2009; Lea et al., 2014), and opportunities for social learning (Stanton et al., 2020; Foley et al., 2008; Brent et al., 2015; Estienne et al., 2019). But the lack of increased mortality in gorillas and certain human populations suggests that mother absence does not always lead to social adversity in mammals with extended maternal care.

When immature mountain gorillas suffered maternal loss, we found that their social relationships with other group members estimated through proximity increased in strength (weighted degree), and their social integration within the group (eigenvector centrality) increased considerably. There was no increase in binary degree suggesting that orphans strengthen existing social bonds rather than forming new ones following maternal loss. The increase in orphan eigenvector centrality was especially large (Figure 2), which may be explained by the strengthening of the orphan-dominant male bond in particular, with dominant males typically being extremely well connected within the group network. The increase in proximity-based relationship strength with other group members was actually great enough to outweigh the loss of their mother, although this was not the case for relationships based on contact which did not change significantly more than non-orphans overall. In contrast to fission–fusion social systems such as in chimpanzees, where group composition regularly changes, mountain gorilla orphans remain in stable cohesive social groups in which their social integration is not dependent on their mother – as shown by the strengthening of relationships following maternal loss. Their increased social integration within the larger cohesive group immediately following maternal loss suggests that mountain gorilla orphans are unlikely to have fewer opportunities for social learning as they are regularly in close proximity to multiple group members. Gorillas also do not appear to require the complex feeding techniques such as nut-cracking or termite-fishing, for which close contact with mothers may be most beneficial (Lonsdorf et al., 2004; Estienne et al., 2019). Instead, common group-membership may be sufficient for learning the complex foraging behaviours observed in gorillas (Watts, 1984; Byrne et al., 2001) as groups travel and feed as a cohesive unit.

The greatest increases in relationship strength following maternal loss were found with the dominant male of the group, regardless of whether or not he was the genetic father. This suggests that unlike in elephants where maternal loss leads to weaker relationships with dominant group members, gorillas that suffer maternal loss instead have increased access to the most dominant member of their group. Whilst the cohesion of gorilla social groups already limits the extent to which orphans are likely to suffer from reduced access to resources or knowledge, the strengthening of social relationships, particularly with the dominant male, is likely to further buffer any potential reduction to their competitivity or future dominance rank within the group. Other cohesive social groups in which the effects of maternal loss have been studied have primarily been matriarchal, with groups led by older females, e.g. elephants (Goldenberg and Wittemyer, 2017; Goldenberg and Wittemyer, 2018), or where strong female dominance hierarchies are inherited, e.g. spotted hyenas (East et al., 2009; Watts et al., 2009) and cercopithecine species such as the savannah baboon (Archie et al., 2014). In contrast, males are the dominant sex in gorillas (Wright et al., 2019). Female dominance hierarchies are relatively weak, and the common dispersal of both sexes means maternal support is unlikely to provide considerable benefits for dominance to offspring of either sex. Orphaned gorillas are therefore unlikely to suffer costs in this regard. However, for males that remain in their natal group it is possible that stronger relationships with the dominant male, such as those developed post-maternal loss, could aid dominance acquisition. For males that lose their mothers at particularly early ages and show some of the strongest relationships with the dominant male as a result, it is possible that this could ultimately improve their chances of inheriting dominance of their natal group (Stoinski et al., 2009a; Robbins and Robbins, 2005). Although, due to the rarity of such events (three cases in 53 years) there may never be a large enough sample size to thoroughly investigate such a hypothesis.

The lack of paternity discrimination in the support provided by adult males after maternal loss is consistent with previous findings on paternal care in mountain gorillas, where the highest ranking males sire the majority of offspring and provide the most care, regardless of paternity (Rosenbaum et al., 2018; Rosenbaum et al., 2015). As observed in chimpanzees (Reddy and Mitani, 2019), support from some maternal siblings also appears to occur, with relationships between subordinate adult males and subadult females, and their younger siblings strengthening following maternal loss. In these two classes, caring for siblings may have additional benefits to those of inclusive fitness (Riedman, 1982). Care of infants by adult male mountain gorillas has been found to be linked with dramatically higher reproductive success. Males in the top tertile for showing affiliative behaviour towards infants were found to sire 5.5 times more infants than those in the bottom tertile, even after accounting for rank (Rosenbaum et al., 2018). This suggests that females may prefer males that demonstrate more caring behaviour towards infants, and that subordinate male kin that increase their care of younger maternal siblings following maternal loss may additionally increase their own reproductive success. Pre-reproductive female siblings, particularly subadult females, may also benefit through developing parental experience that may improve their own reproductive success (Riedman, 1982).

Our analyses of social relationship changes used undirected data on pairwise proximity and affiliative contact. Whilst both measures of affiliation indicate at a minimum, high tolerance for the other individual, we cannot determine who was responsible for initiating the relationship changes. Future research could investigate whether it is the orphans themselves that are strengthening their relationships by approaching other group members more often or whether it is the other group members responding to the maternal loss faced by these young gorillas. Our study also only investigates the immediate social response to maternal loss in the 6 months after the mother dies or leaves the group. Whilst this might be expected to be the period where individuals face the greatest social costs from maternal loss, our analyses cannot tell us whether the social buffering we detected during this period is maintained in the longer term. However, previous observations have suggested that the close associations that young gorillas develop following maternal loss continue into subadulthood (Robbins et al., 2005). It is therefore highly likely that the dramatic short-term changes we have detected are part of a longer-term social response to maternal loss with the potential to overcome the long-term fitness costs of maternal loss.

In many human populations, care from other family members is believed to buffer the negative consequences of maternal loss, but the identity of these carers can vary greatly between populations. In rural Gambia, elder sisters and maternal grandmothers increased offspring survival but not fathers, other grandparents, or brothers (Sear et al., 2003). In contrast, in the Ache of Paraguay the loss of fathers significantly impacted offspring survival but the loss of grandparents or adult siblings did not (Hill and Hurtado, 2017). Support from family members in rearing offspring is thought to be a human universal but the composition of those families and the specific family members involved in cooperative care appears to be flexible and responsive to ecological conditions (Sear and Mace, 2008). In killer whales, as observed in many human populations, grandmothers have been found to influence offspring survival and this additional support could in part contribute to the lack of reduced survival for younger female killer whales after maternal loss (Foster et al., 2012; Nattrass et al., 2019). In mountain gorillas, grandparents and their grand-offspring are rarely in the same social group, as females often transfer between groups multiple times within their lives (Robbins et al., 2009a). Instead, fathers (or dominant males with a high likelihood of paternity), siblings, and group members close in age seem to play a key role in buffering the detrimental effects of maternal loss. What appears to set humans, gorillas, and possibly killer whales apart from other species with extended maternal care where high fitness costs to maternal loss are observed is the potential for cooperative care from within the social group.

Conclusion

We found that immature mountain gorillas do not appear to face increased social adversity or a detectable reduction in fitness following maternal loss. It is not yet possible to demonstrate the direct link between the strengthening of relationships with other group members after maternal loss and the absence of fitness costs to maternal loss. However, our analyses show that at least in the short-term, a key mechanism by which maternal loss is hypothesized to lead to reduced survival and fitness in other social species – social adversity - does not apply in mountain gorillas over the age of 2 years. The social support provided by other group members within mountain gorillas’ cohesive social groups, particularly from dominant males, siblings, and those close in age, appears to buffer against the negative consequences of maternal loss. In mountain gorillas, like humans (Sear and Mace, 2008; Lahdenperä et al., 2004; Sear et al., 2000; Hurtado and Hill, 1992; Andersson et al., 1996; Campbell and Lee, 2002; Bentley and Mace, 2009) social support appears to come from a number of group or family members. This could provide a buffer to the loss of any single relationship, even one as important as the mother–offspring relationship, once an individual can be nutritionally independent. In the absence of nepotistic matriarchal dominance hierarchies and when social buffering is possible due to cohesive strongly bonded social groups, it may matter less who is providing care as long as care is provided.

Materials and methods

Demographic data

Request a detailed protocol

Mountain gorillas in the Volcanoes National Park, Rwanda, have been monitored almost continuously by the Dian Fossey Gorilla Fund’s Karisoke Research Center since 1967. Habituated mountain gorilla groups are monitored daily by field teams who collect data on demography, behaviour, ranging, and health. From 1967 to 2015 (inclusive), 59 (28 males, 31 females) out of the total 200 immature mountain gorillas (102 males, 98 females) that reached the age of at least 2 years suffered maternal loss between the ages of 2 and 8 through the death or permanent transfer of their mother.

Gorillas were classified as infants up to 4 years of age, as juveniles from 4 to 6 years of age and as subadults from 6 to 8 years of age (Breuer et al., 2009). From 8 years of age females were classified as adults. Males were classified as blackbacks from 8 to 12 years. From 12 years, males were classified as either subordinate adult males or dominant adult males from their dominance hierarchy. Male dominance hierarchies were based on displacements and avoidances using the Elo-rating method (Albers and de Vries, 2001; Neumann et al., 2011). Dominance hierarchies were calculated using the R package EloRating, version 0.43 (Neumann and Lars, 2014) as described by Wright et al., 2019. Only one adult male was classified as dominant in each group at a given time. Dominant males were those with the highest dominance status unless they were the only adult male in the group in which case they were automatically classified as dominant.

Survival

Request a detailed protocol

The youngest infant to survive maternal loss was a 2.45-year-old female. Our data set included only one infant younger than this that suffered maternal loss at 0.67 years and died after 1 day. Three infants aged 1.91, 2.42, and 2.52 became separated from their mothers after a suspected poacher encounter. During this separation they travelled with a small number of group members not including their mothers. The 1.91-year-old died after 6 days of separation. The 2.42-year-old died after 9 days of separation. The 2.52-year-old survived until they were reunited with their mother and the larger group 18 days after the initial separation. These infants were not considered as orphans in the data set as their mothers did not permanently transfer or die, but in combination with those that did, they suggest that infant mountain gorillas cannot survive independently from their mothers under the age of at least 2. We therefore investigated the effect of maternal loss after the age of 2.

To determine the effect of maternal loss on survival, we carried out a Cox-proportional hazards analysis separating individuals over the age of 2, based on four general age classes of maternal loss: (a) infants, (b) juveniles, (c) subadults, and (d) non-orphans, where mothers did not die or leave the group before the individual reached 8 years of age. First, we ran a Cox-proportional hazards model for each sex (males: n = 102, females: n = 98) with a time-varying covariate for the age at maternal loss and using the four classes as covariates. Due to the small sample size for the subadult class, we merged this with the juvenile class into a single juvenile/subadult class and ran a new set of Cox-proportional hazards models on these new classes. In all cases we truncated the analysis to start at age 2.

In order to verify our results and to account for the uncertainty in some of the dates of birth, we ran a Bayesian survival trajectory analysis (Colchero and Clark, 2012; Colchero et al., 2012) for each sex truncated at the age of maternal loss for all orphans, and at 2 years of age for non-orphans. We used orphan status as a binary covariate (orphan vs non-orphan) and, using the Siler, 1979 mortality model for the baseline mortality, we tested three models: (a) no covariates (i.e. null model where all individuals have the same hazard rate); (b) proportional hazards (where mortality differs proportionally between orphan classes); (c) covariates modifying all Siler mortality parameters (where each orphan class has a different age-specific mortality). We used deviance information criterion for model fit and selection (Spiegelhalter et al., 2002; Celeux et al., 2006). Model (b) is equivalent to the Cox-proportional hazards model. However, these tests facilitate further exploration of the hypotheses on the effect of maternal loss on mortality, namely that there is no effect (model a) or that the entire age-specific trajectory of mortality changes for each category.

Female dispersal and reproduction

Request a detailed protocol

Between 1967 and 2019, 66 females gave birth to what was known to be their first offspring. For 53 of these females, their age could be accurately estimated within a 90-day period. For these individuals, we extracted their age at first birth, whether they dispersed from their natal group prior to first birth and whether they had suffered maternal loss when immature, from the long-term database. Maternal loss was investigated with the orphan age classes described above with juvenile and subadult classes merged due to small sample sizes. We investigated the effect of maternal loss on the decision of females to disperse from their natal group prior to their first birth using a binomial generalized linear model. We investigated the effect of both maternal loss and dispersal prior to first birth on age at first birth using a generalized linear model with a Gaussian distribution. Due to the positive skew of age at first birth, we used the square root of age at first birth minus 8 (the earliest recorded age at first birth) as the response variable. We checked Q–Q plots to verify the normal distribution of residuals and ran Levene tests using the ‘rstatix’ package to check for heteroskedasticity. Finally, we examined survival of each female’s first offspring through infancy (1: survived to 4 years, 0: died before reaching 4 years) according to age at first birth, dispersal prior to first birth (1: yes, 0: no), and maternal loss, with merged juvenile and subadult classes as above, using a binomial generalized linear model. Multicollinearity of all models with multiple variables was checked using variance inflation factors in the ‘car’ R package.

Male dispersal and dominance

Request a detailed protocol

Between 1967 and 2019, 56 males whose age could be accurately estimated within a 90-day period reached the median age of dispersal (16 years). We used a binomial generalized linear model to predict whether each of these males dispersed from their natal group prior to this age according to orphan age classes (as above). To investigate the effect of maternal loss on dominance, we gave males that became the dominant male of a stable group for at least six consecutive months a dominance score of 1. This included adult males of single-male groups and the most dominant male of multi-male groups based on Elo-ratings. Males that never reached dominance or only transiently (for <6 months) received a score of 0. We recorded the age at which a male first became dominant for those for which this could be accurately estimated within a 90-day period. This represented the age at which they first successfully attracted and retained a female to join their group, the age at which they split from their natal group with at least one adult female to form a new group, or the age at which their Elo-rating surpassed that of all other adult males in their group, if those groups remained independent and did not disintegrate within 6 months of that date. The mean age (± standard deviation) at which a male reached dominance was 17.88 ± 2.56 years. The median was 17.29 years. The oldest age at which a male first became dominant was 22.99 years. Therefore, to investigate the influence of maternal loss on dominance status we analysed only males that had survived and remained in the study population until at least 23 years (n = 40). We did not attempt to statistically examine the effects of maternal loss on dominance status due to small sample sizes.

Social network analysis

Request a detailed protocol

Habituated gorilla groups were monitored for up to 4 hrs daily and all gorillas were individually identified by physical characteristics. Behavioural data were collected on each group member via 50 min focal sampling during which the researcher would typically be within 10–20 m of the focal individual. Researchers systematically worked their way through a randomly ordered list of all individuals in the group. If an individual could not be observed (e.g. obscured by dense vegetation), the researcher moved on to the next individual on the list and returned to them the subsequent day. During focal sampling, a focal scan was completed every 10 min which recorded all gorillas within 2 m of the focal individual and all gorillas in physical contact with the focal individual. This data was therefore in the form of frequencies (the number of focal scans during which individuals were associating) rather than durations. Affiliative contact included resting, playing, or feeding in contact and grooming, and excluded physical aggression. This focal sampling approach limited the extent of potential sampling bias due to individual-level differences in observation propensity, with all group members systematically observed and an extremely low likelihood of individuals within 2 m or in physical contact of the focal individual being missed at these close proximities.

The social response of group members to an incident of maternal loss was investigated in 31 of the 59 total cases – those that suffered maternal loss after 2003 for which adequate social behaviour data was available (more than 12 focal scans of the individual were recorded in the 6 months prior to maternal loss and the 6 months after maternal loss, Supplementary file 1 - Table 3). For each case of maternal loss, two types of weighted social network were constructed based on (a) 2 m proximity and (b) affiliative contact. For each type, edge values of the networks were calculated using the Simple Ratio Index (SRI) (Whitehead, 2008) with edges between a pair of individuals calculated as the proportion of focal scans of either individual during which the pair was recorded as associating. These values represented an estimate of the proportion of time two individuals were either within 2 m of each other or in physical contact. For example, in the contact network a value of 1 would indicate that the two individuals were in physical contact every time a focal scan of either individual was conducted, whilst 0 would indicate that they were never observed in physical contact during a focal scan. Both network types were constructed for two time periods for each case of maternal loss: pre-maternal loss (the 6 months leading up to maternal loss) and post-maternal loss (the 6 months immediately after maternal loss). Social networks were constructed using all focal scans during these time periods. Only gorillas for which more than 12 focal scans were available in both of the 6 month periods (pre- and post-maternal loss) were included in the networks (Farine and Whitehead, 2015). This meant that only the social relationships with group members that were present both pre- and post-maternal loss were analysed, except for the mother–offspring relationships pre-maternal loss which were extracted separately. The mean (± SD) number of focal scans used to estimate edge values pre-maternal loss was 144.81 ± 90.10 and post-maternal loss was 149.16 ± 88.56.

Changes in network position following maternal loss

Request a detailed protocol

Binary degree, weighted degree, and eigenvector centrality in the pre- and post-maternal loss networks (excluding orphan’s mothers) were calculated using the ‘igraph’ package for both network types (contact and proximity). To enable comparison across networks of the same type, binary degree and weighted degree metrics were calculated as a proportion of the maximum value for an individual within the network (as is already the case for eigenvector centrality). This meant that for all networks, each metric could have a maximum value of 1 and a minimum value of 0. Metrics were extracted for all immature gorillas that were between the ages of 2 and 8 years on the date that an immature individual within their group suffered maternal loss. We then calculated the change in these network metrics between time periods for all immature gorillas in each set of group networks and used GAMMs in the ‘gamm4’ R package to determine whether orphan’s network metrics changed differently to those of non-orphans within the same group during the same time period. To ensure significant changes were not driven by unusually high or low initial values, the deviance of the network metric in the initial network from the mean value for immatures in the network was calculated. Orphan status, age, and the deviance of the initial value from the group mean were included in the model as fixed factors. The number of focal scans per individual across both time periods was included as a smoothing factor to account for any potential differences driven by sampling effort. The specific set of group networks (the pair of networks for each incident of maternal loss) was included as a random effect to account for differences in network composition (Supplementary file 1 - Table 5).

Due to the non-independence of network metrics, null models generated through node-level permutations were used to assess the significance of orphan status on the change in network metric. Permutations were run by swapping orphan status between immature gorillas (orphans and non-orphans) within the same set of paired group networks (same maternal loss incident). Orphans that suffered maternal loss in groups with no other non-orphaned immature gorillas were excluded from the analyses, leaving a sample of 19 incidents of maternal loss, 28 orphans, and 108 non-orphans (Supplementary file 1 - Table 3). The mean age ± SD of orphans was 5.12 ± 1.49 years and for non-orphans was 4.71 ± 1.71 years. 10,000 sets of node-permutations were generated by permuting orphan status between immature gorillas in the same network and extracting node labels every 200th permutation. The same GAMMs were then run on all 10,000 sets of node permutations to produce a null distribution of t-values for the effect of orphan status. Pnull was calculated using a two-tailed approach. For observed t-values greater than the median of the null distribution, Pnull was calculated as:

2×numberofnullt-valuesgreaterthanobservedt-valuetotalnullt-values

For observed t-values lower than the median of the null distribution, Pnull was calculated as:

2×numberofnullt-valueslowerthanobservedt-valuetotalnullt-values

Relationship changes following maternal loss

Request a detailed protocol

We extracted SRI edge values (representing the strength of relationship between a pair of gorillas) between the orphan and all other group members pre- and post-maternal loss. We also extracted SRI edge values between all immature gorillas (aged 2–8 years) within the same group that had not suffered maternal loss and all other group members, for the same time periods. We then calculated the change in relationship strength between an immature gorilla (both orphans and non-orphans) and other group members following an incident of maternal loss within the group as the change in SRI value between periods (SRI post-maternal loss – SRI pre-maternal loss) in both network types (contact and proximity). GAMMs were used to predict this change for both affiliative contact and proximity which enabled the non-independence of relationships involving the same individual to be accounted for through random effect structures and sampling variation to be accounted for using a smoothing term. We ran an initial set of GAMMs on the change in SRI values for relationships involving the orphans or other immature gorillas within the same group that had not suffered maternal loss, excluding mother–offspring relationships. This was to verify that changes in the relationships of orphans with other group members were significantly different to those observed for other immature gorillas present within the same group during that period. These GAMMs included the identity of the immature gorilla as a random factor, nested within the specific maternal loss incident (group and time period), nested within the group. The identity of the other group member was included as a further random factor (Supplementary file 1 - Table 6). The mean number of focal scans used to estimate the SRI value across both time periods was included as a smoothing term in the model to account for any biases from sampling intensity. The age and sex of the immature gorilla were included as fixed factors, along with the age-sex class of the other group member, whether the immature gorilla suffered maternal loss during this period (1: Yes, 0: No), and the interaction between maternal loss and the age-sex class of the other group member. Non-orphan relationships where the other group member was an orphan were excluded from the analysis.

We then ran GAMMs on only the relationships involving orphans to investigate in more detail how these changed following maternal loss. The identity of the orphan was included as a random factor, nested within the group. The identity of the other group member was included as a further random factor (Table 5). The mean number of focal scans used to estimate the SRI value across both time periods was again included as a smoothing term. The age and sex of the orphan were included as fixed factors to predict the social response of group members, as well as whether those group members were maternal siblings (1: Yes, 0: No) and whether those group members were age-mates (1: <2 years age difference with the orphan, 0: ≥2 years age difference). This 2-year cut-off was chosen to be consistent with the width of age categories, such that age-mates would be within the same age class during large proportions of their immature life. The age-sex class of the other group member at the date of maternal loss and the interaction between this and maternal sibling status were also included as fixed factors for predicting the change in relationship.

Adult male relationship changes following maternal loss

Request a detailed protocol

Paternity was known for 18 of the 31 orphans for which social data was available, from a previous study (Vigilant et al., 2015). To investigate the influence of paternity, we ran an additional set of GAMMs to predict the change in relationship (both contact and proximity) between adult males and orphans following maternal loss (n = 121). The identity of the orphan and the identity of the adult male were included as random factors (Supplementary file 1 - Table 7). The mean number of focal scans was included as a smoothing term. Orphan age, orphan sex, male dominance, and the binary kinship variables: maternal sibling and paternity were included as fixed factors. The interaction between dominance and sibling status was also included in the model, but the interaction between dominance and paternity could not be, due to possible issues of multicollinearity (variance inflation factors > 3).

Data availability

Demographic data available within the Primate Life Histories Database at https://datadryad.org/stash/dataset/doi:10.5061/dryad.v28t5 Social data available as source data files.

The following previously published data sets were used
    1. Bronikowski
    2. Anne M
    (2017) Dryad Digital Repository
    Data from: Female and male life tables for seven wild primate species.
    https://doi.org/10.5061/dryad.v28t5

References

  1. Report
    1. Campbell C
    2. Lee JZ
    (2002)
    When Husbands and Parents Die: Widowhood and Orphanhood in Late Imperial Liaoning, 1789-1909, When Dad Died Individ Fam
    Coping with Distress Past Soc.
  2. Conference
    1. Gatesire T
    2. Stoinski TS
    3. Caillaud D
    4. Ndagijimana F
    5. Eckardt W
    (2016)
    Going alone? Social bonds of gorilla orphan
    International Primatological Society Congress.

Decision letter

  1. Josh Firth
    Reviewing Editor; Edward Gray Institute, United Kingdom
  2. Christian Rutz
    Senior Editor; University of St Andrews, United Kingdom

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

We were very pleased to see that the previous round of recommendations and suggestions were taken onboard and responded to in full. We believe the manuscript reads very clearly, and the methods are well explained throughout. As expressed in previous decision letters, we are highly impressed by this interesting work and believe it will be a valuable contribution to the literature that will be enjoyed by eLife's broad readership.

Decision letter after peer review:

Thank you for submitting your article "Social groups buffer maternal loss in mountain gorillas" for consideration by eLife. Your article has been reviewed by Christian Rutz as the Senior Editor, a Guest Editor, and three reviewers. The reviewers have opted to remain anonymous.

The reviewers have had the opportunity to discuss their reviews with one another, and the Guest Editor has drafted this decision letter to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, we are asking editors to quickly process manuscripts, like yours, that they judge can potentially stand as eLife papers without additional data. Thus, the revisions requested below are aimed at: (a) addressing clarity and presentation of the manuscript; and (b) addressing further considerations about the current analyses and arising conclusions.

Summary:

This manuscript provides an in-depth assessment of the consequences of maternal loss for wild mountain gorillas, and its rigorous and detailed approach was appreciated by all reviewers. However, some concerns were raised by the reviewers in relation to the reported analyses, and the wording of certain text passages.

Essential revisions:

We judge that the manuscript would benefit from addressing the reviewers' comments point-by-point, and have therefore decided to append their full reports below. In relation to some of the comments detailed below, we would like to highlight the following three points:

1) The use of paired t-tests for the examination of social changes is a bit confusing. Specifically, we are unsure how these properly control for repeated sampling of individuals when considering changes in dyadic association scores. Also, the changes reported here appear to just assess how orphan behaviour altered between the pre- and the post-maternal loss period, but how do we know that such changes aren't expected anyway over time? Ideally, these analyses would directly compare orphaned to non-orphaned individuals. Note the comments from reviewer #1 about these paired t-tests that also need to be addressed.

2) Animal social networks (either dyadic association scores or individual-level metrics) are strongly influenced by observation. But it is not clear how these analyses controlled for, or considered, how individual-level differences in observation propensity might alter inferences of their social network position/association scores, and what this might mean for the arising conclusions. We think the manuscript would benefit from including a much more detailed report of how individuals differed in their likelihood of being observed in relation to the social behaviour considered here, and how this may shape the arising networks, and whether controlling for these effects might alter the results/conclusions.

3) We are unsure why only weighted degree was used as a network metric; reviewer #3 also raises a point about this metric. We suggest that this could be addressed through also considering mean average non-zero dyadic bond scores as a metric. Furthermore, you may want to re-run the network metric-based analyses, but with other measures of social behaviour (e.g., eigenvector centrality) to get an idea of how wider indirect centrality within the network may relate to maternal loss.

Reviewer #1:

This is an interesting paper about how the loss of a mother influences individuals future development, or rather the fact that it might not due to social connections with other group members. The paper examines a very comprehensive set of question, and on the whole I found it very well written, though I have some quibbles about how certain ideas are presented/supported. The dataset presented seems impressive. The methods used generally seem appropriate and robust, though I'd like to see some extra justifications for some of the approaches taken during network analysis. For more info, see detailed comments:

Introduction

I think an example of how the presence of the mother affects fitness in later life would be nice here, in order to lend weight to the statement about maternal loss having affects throughout an individual's lifetime. The third paragraph contains examples, and currently feels repetitive in terms of the points raised.

Similarly, I think "these negative changes" could be expanded on in the preceding sentences with an example of negative changes to the social environment.

I am not sure how convinced I am that the human example here adds to the authors' point. As mentioned above, I would prefer to see more general examples of these ideas/effects in social mammals (or animals in general). If the authors' really want to keep a human example, I'd like to see significantly more detail about the circumstances in which this result was obtained.

It feels slightly odd to discuss sex specific consequences before the general consequences.

As above, I am slightly unconvinced by the need/relevance to relate these results back to humans. This might be a result of me not having an anthropology background of course. Related to this, I feel this is making a rather large assumption about the familiarity a reader might have with this literature. Some small details about what these populations are might help convince that they are relevant to build the argument in the same paragraph as killer whales. There are a few other instances of this in the Discussion too, but I'll avoid repeating myself.

Any directional predictions going into the study? There are quite a few variables under investigation, and it would be nice to link them more strongly to the ideas articulated in the Introduction. Is social buffering more likely in certain group compositions of age/sex?

Results: The term age mates initially confused me a little, I thought it was a typo in the table. Perhaps "cohort" or similar accepted term?

Discussion: I feel you could just say "non-breeder" or "virgin" here.

Materials and methods:

I am also not quite sure about the use of simple paired t-tests to address a rather complex question, with many potential confounding variables. While I am all for using simple stats where possible, it seems like there would be more going on (age difference of a dyad, group size, age of orphan, year of study) that should be controlled for/investigated. Given the detailed models that follow, perhaps some clarity about why this is not necessary for the question currently being addressed would be useful?

Additionally, is there a citation for this permutation approach over permuting the networks themselves?

This suggests multiple models, but it seems only one model is described. I think I need to see these models/model formula laid out in a table for clarity. I'd also like to see a citation for the use of the random effects structure to control network independence.

Reviewer #2:

This is a novel, very well written and well-researched project, making use of a fantastic long-term dataset. It is interesting to a wide readership and of high scientific value. I do not really have any major concern but a few queries and suggestions in order improve the clarity in places.

List of slightly more substantial comments/queries:

1) Results – throughout – can you please include sample sizes when stats values are given?

2) Results – can you please also give us an idea of how much time they spent in proximity and contact? It’s good to see the stats but it would be nice to get a feel for who much time we are talking about (and how big the actual change is).

3) Discussion – I don't follow this argument – can you explain the reduced siring opportunities? Also, can you explain why infant-orphans are doing BETTER than older age orphans? This seems counter-intuitive and I am wondering why this would be the case?

4) Can you discuss a little who initiated the contact and proximity? Would that be on the initiative of the orphan or of the other group members?

5) Can you provide any thoughts on what happens after the 6 months? You are concluding that the strengthening of bonds for 6 months after the loss of maternal care is responsible for the lack of adverse effects – but do you have any evidence or indication that this strengthening is extending over a longer time period? I understand that you chose 6 months in order avoid other confounding variables in the analysis, such as changes to group composition – but in the discussion it reads like these bonds are strengthen “forever”. Some indicators that this indeed the case would be good to see if this generalization is warranted.

6) Materials and methods – a few more details about the data feeding into the networks would be good – which behaviours were included as affiliative behaviours? And how was the SRI calculated, ie did you use frequencies or durations of time in contact? More details on how focal sampling was carried out (length of focals and frequency of data recording) would be useful.

7) I would also like to see more details on number of groups, group sizes and compositions used in this study (maybe as supplementary material); all of this can go in supplementary material – but would be nice to have. Also, information on observation times for each social group and how stable they were would be helpful.

Reviewer #3:

The manuscript "Social groups buffer maternal loss in mountain gorillas" examines the potential consequences of maternal loss in mountain gorillas. The topic of early life adversity and consequences of maternal care that extends beyond weaning is of growing interest to researchers including behavioral ecologists and anthropologists. The results presented in the manuscript are an interesting and important contribution to that literature. Whereas most studies have found negative fitness consequences associated with maternal loss, the results reported here indicate that gorillas who experience early maternal loss do not face negative consequences in terms of survival, maturation (age at first birth), or an indicator of reproductive success (first offspring survival). Furthermore, rather than speculating the authors follow-up on social buffering as a potential explanation for this somewhat unexpected (given outcomes observed in other social species) result using social network analyses. My comments on the manuscript primarily concern organization and clarity to help the reader, particularly the reader of a broader journal such as eLife.

1) I realize that the structure of an eLife article has the methods at the end, but as currently presented most of the results are impossible to interpret without a lot of flipping back and forth searching for information. For example, the Results start out by briefly stating that survival was examined using a cox proportional hazards model, which was very helpful for interpreting the results. However, in the next section there is reference to model results without any information about what type of model it is. Later, there are t statistics and p values with no explanations and no indication they came from permutation-based tests.

2) Some of the abruptness of the transition from introduction to results might be helped by stating clearer predictions that help set up the outcomes you tested. I found the descriptions and expectations concerning the fitness outcomes clearer than the social network outcomes and suggests more information be given about the networks before the results of the analyses are presented.

3) I also have some questions concerning the social network analysis. One additional analysis that might be interesting is looking at binary degree along with weighted degree. High weighted degree can result from few strong connections or many weak connections and presenting both weighted and binary degree might indicate one strategy (find a strong buddy) versus another (cast a wide net). Furthermore, what was the variation in group size in these data? Did you take variation in group size into account when calculating network metrics. The maximum weighted degree of an individual in a group with 5 individuals is much lower than the maximum weighted degree of an individual in a group with 15 individuals.

4) Results: Regarding the results for dispersal – any reason to believe group size will influence likelihood of dispersal?

5) Materials and methods: Social buffering of maternal loss results: can you clarify whether the edge to mom included when weighted degree was calculated?

6) Table 5:

- How was age included in the GAMM models? Age in years?

- This table could be presented more clearly. Sometimes the bold is used to describe the two variables under it (Maternal orphan and Group member) and other times the bold itself is a variable with multiple levels under it. Maybe just write out "Orphan age" "Orphan Sex" and not include the extra bold rows?

- Should the age/sex class – sibling interaction results be relative to the dominant male who is or isn't a sibling?

- Where are the results of the smoothed term?

7) Figure 1: Curious about why each age category has its own line. Since maternal loss is a time varying covariate, shouldn't the survival probability of individuals who lost their mother at 7 (for example) be the same as non-orphans until age 7? Apologies if what I described above is the actually the case. It is tricky to tell where the dashed lines start.

8) Materials and methods: Can you explain why sampling variation warranted a smooth? Not necessarily questioning your decision – I find it interesting and looking for more information about it!

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for revising and resubmitting your article "Social groups buffer maternal loss in mountain gorillas" for consideration by eLife. Your article and response have been evaluated by a Guest Editor and Christian Rutz as the Senior Editor.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, we are asking Editors to quickly process manuscripts, like yours, that they judge can potentially stand as eLife papers without additional data. Thus, the revisions requested below are aimed at (a) addressing clarity and presentation of the manuscript, and (b) addressing further considerations about the current analyses and arising conclusions.

As outlined in the first decision letter, this manuscript provides an "in-depth assessment of the consequences of maternal loss for wild mountain gorillas" and was found to be a "rigorous and detailed examination" that was appreciated by the Editors and all the reviewers. During the first assessment by the reviewers and Editors, some potential problems with the wording of particular parts of the text were raised, and this revision appears to address all of these in full. However, the first assessment also raised some issues with the reported analyses, and unfortunately, there is still some lack-of-clarity and some issues that need to be resolved. To outline these issues in as clear a way as possible, the below text takes a 3-step approach for each issue: (A) it revisits the initial comment, (B) provides the authors' response (and associated manuscript text), and finally (C) describes why a problem still exists and what needs to be addressed.

1) Differences in social changes

1A) Revisiting the issue: "The use of paired t-tests for the examination of social changes is a bit confusing here. Specifically -on my part- I'm unsure how these properly control for repeated sampling of individuals when considering changes in dyadic association scores? Also, the changes reported here appear to just assess how orphan behaviour altered between the pre- and the post- maternal loss period, but how do we know that such changes aren't expected anyway over time? Ideally this analysis would directly compared orphaned to non-orphaned individuals. Note the comments from reviewer 1 about these paired t-tests need to be addressed too"

1B) Author response: "Yes, we entirely agree and have changed these analyses to instead compare whether network metrics (binary degree, weighted degree and eigenvector centrality) change differently in orphans relative to other immature gorillas within the same group during the same time period using node-based permutations constrained to swap only between individuals within the same incident of maternal loss (based on the same pair of networks). Described in the Materials and methods and Results."

1C) Current Issue:

While we appreciate that the authors have taken the first comment onboard and reconsidered the approach, there are two remaining problems with the current approach. Namely:

1Ci) In the Results, the results of these tests are reported. But, when reporting these results, it is important to not entirely focus on the p-value generated from the null models, but to also report the full results of the standard statistical test as well (i.e., the estimate, the SE, the t value, and the standard p-value). The p-value from the null model is simply used to 'double-check' the results of the standard statistical test, and shouldn't be fully relied upon for conclusions like this. For example, the first result reported here is "(t=1.721,p=0.029)" which isn't clear to the reader, as a t-value this low wouldn't generally result in a p-value this low. Instead, the text needs to report the actual observed statistical test results from the standard test (including the real p-value) and then also include a Pnull value (the null value generated from the node permutations). If the primary standard statistical test shows no significant difference (as may be the case here), then this primary result should be relied upon for drawing conclusions (not the null model p-value, as this is mainly just for double-checking tests, rather than drawing direct conclusions from them in this particular way).

Also, as a small side-point, the method used here for actually calculating the node permutation p-value is quite a strange one. Specifically, it is stated in the revised text that "two-tailed p-values were calculated as two times the proportion of permutation t-values greater or smaller than the t-value from the real data set". But this is problematic and doesn't make clear sense? What is actually being done here?

1Cii) The text goes on to state that "within contact-based networks orphans did not show these same gains, with no significant increase in binary degree (t=0.099, p=0.730), weighted degree (t=1.110, p=0.397) or eigenvector centrality (t=0.129, p=0.741) relative to non-orphans". But it isn't clear where these results are actually derived from. We are assuming that they come from a subset of tests the authors refer to in their response, i.e., 'node-based permutations constrained to swap only between individuals within the same incident of maternal loss'. But, if the swaps are constrained between individuals who experience the same incident of maternal loss, and you subsequently try to look at the comparable difference between incident of maternal loss, then the tests will never find a difference because this is already controlled for earlier in the pipeline? Some clarity about what is actually going on here needs to be added to the manuscript (and also some thought about whether such tests are appropriate for testing the questions at hand).

2) Social networks and observations

2A) Revisiting the issue: The first assessment of the paper flagged up that animal social networks are strongly influenced by observation, and requested some analytical consideration of this in the revisions.

2B) Author Response: The authors responded by providing more details about the data collection methods, which is very much appreciated. However, the revision doesn't include any attempts to actually control for/integrate consideration of differences in observation number. Instead, the response states that the Simple Ratio Index includes this (stating "variation in the number of focal scans per individual which is accounted for within the simple ratio index" and "we did not believe that these extra data would bias the estimate of the specific relationships based on SRI (as these account for the total number of observations)".

2C) Current Issue: Unfortunately, the SRI does not control for differences in observation number in the way that the authors lay out here. While the SRI takes some of the difference in observation into account during the calculation of the dyadic values, it in no way fully accounts for this; it just uses it for the calculation. For instance, even in simulations where all individuals have the same social phenotype, those individuals which are observed more will have higher centrality (e.g., higher degree) than those which are observed less. This is simply a product of how these metrics are calculated and how these networks emerge. Therefore, the previous comments from the Editors and the reviewers that the analyses should directly consider 'observation number' still stands.

3) Outstanding analytical issues

3A/B) Revisiting the issue: In the reviewer comments, a few analytical (smaller) issues were pointed out, and in most cases the authors have addressed these fully and clearly. However, a couple of key points remain. Particularly, controlling for age and network differences within the models themselves, and also using 'change in metric' as the response.

3C) Current Issue: The revision didn't make any attempt to actually integrate age of individuals into the models, nor did it try to consider network groups, but the reviewers made solid points that these would be good to consider. We believe these should be integrated into the paper as supplementary analyses. Finally, the switch to considering “change in metric” as the response is certainly useful in lots of ways, but have the authors' considered problems with “regression to the mean” in this sense? Specifically, we can automatically expect this metric to be altered significantly by the initial value, whereby those which initially have a large SN metric value are likely to (just by chance) have large -ve change values, and whereby those which initially have a small SN metric value are likely to (just by chance) have large +ve change values. Have the authors assessed this as a driver/contributor to the results here? (Note this consideration relates to the above point about considering things like age, and network group, which might alter these initial values and thus affect the change values.)

https://doi.org/10.7554/eLife.62939.sa1

Author response

Essential revisions:

We judge that the manuscript would benefit from addressing the reviewers' comments point-by-point, and have therefore decided to append their full reports below. In relation to some of the comments detailed below, we would like to highlight the following three points:

1) The use of paired t-tests for the examination of social changes is a bit confusing. Specifically, we are unsure how these properly control for repeated sampling of individuals when considering changes in dyadic association scores. Also, the changes reported here appear to just assess how orphan behaviour altered between the pre- and the post-maternal loss period, but how do we know that such changes aren't expected anyway over time? Ideally, these analyses would directly compare orphaned to non-orphaned individuals. Note the comments from reviewer #1 about these paired t-tests that also need to be addressed.

Yes, we entirely agree and have changed these analyses to instead compare whether network metrics (binary degree, weighted degree and eigenvector centrality) change differently in orphans relative to other immature gorillas within the same group during the same time period using node-based permutations constrained to swap only between individuals within the same incident of maternal loss (based on the same pair of networks). Described in the Materials and methods and Results.

2) Animal social networks (either dyadic association scores or individual-level metrics) are strongly influenced by observation. But it is not clear how these analyses controlled for, or considered, how individual-level differences in observation propensity might alter inferences of their social network position/association scores, and what this might mean for the arising conclusions. We think the manuscript would benefit from including a much more detailed report of how individuals differed in their likelihood of being observed in relation to the social behaviour considered here, and how this may shape the arising networks, and whether controlling for these effects might alter the results/conclusions.

We have added further details of the sampling methods to the manuscript. Data were collected on fully habituated mountain gorilla groups using focal sampling. Under this approach researchers systematically worked their way through a randomly ordered list of all individuals in the group. If an individual could not be observed, the researchers moved on to the next individual on the list and returned to them the subsequent day. During focals, researchers recorded all individuals within 2m and within physical contact of the focal individual every 10 minutes for an hour. This meant there was no bias in certain individuals being missed that were within 2m or in physical contact of the focal individual as with these small distances and extended periods of observation it is extremely unlikely any individuals at this close proximity were missed. We therefore do not believe there are any biases in the likelihood of certain individuals in a group being observed when participating in the social behaviour considered here, except the variation in the number of focal scans per individual which is accounted for within the simple ratio index.

During the 17 years of social data used in these analyses, some additional projects took place. For example, one project examining the relationships of dominant males and infants. This meant that additional focal data were available for certain individuals and led to some variation in the number of focals per individual within a group. However, we did not believe that these extra data would bias the estimate of the specific relationships based on SRI (as these account for the total number of observations) but would likely just lead to more accurate estimates of these relationships relative to others in the group. We therefore kept these extra focal data within the larger dataset but included the number of focals of either individual involved in a pairwise relationship as a smoothing term in the GAMMs in case there was some potential non-linear bias introduced by this sample size variation that we hadn’t thought of.

3) We are unsure why only weighted degree was used as a network metric; reviewer #3 also raises a point about this metric. We suggest that this could be addressed through also considering mean average non-zero dyadic bond scores as a metric. Furthermore, you may want to re-run the network metric-based analyses, but with other measures of social behaviour (e.g., eigenvector centrality) to get an idea of how wider indirect centrality within the network may relate to maternal loss.

Thanks so much for these suggestions, we think they add really useful information on the overall network changes faced by orphans. We have rerun the analyses looking at the change in binary degree, weighted degree and eigenvector centrality. We used mixed models to account for differences between networks and permuted only between individuals within the same network (as discussed for point 1). There was no relationship between change in weighted degree and network size so we therefore stuck with weighted degree as the more commonly used metric rather than changing to average non-zero dyadic bond scores.

Reviewer #1:

This is an interesting paper about how the loss of a mother influences individuals future development, or rather the fact that it might not due to social connections with other group members. The paper examines a very comprehensive set of question, and on the whole I found it very well written, though I have some quibbles about how certain ideas are presented/supported. The dataset presented seems impressive. The methods used generally seem appropriate and robust, though I'd like to see some extra justifications for some of the approaches taken during network analysis. For more info, see detailed comments:

Introduction

I think an example of how the presence of the mother affects fitness in later life would be nice here, in order to lend weight to the statement about maternal loss having affects throughout an individual's lifetime. The third paragraph contains examples, and currently feels repetitive in terms of the points raised.

Thank for this suggestion – we fully agree and have reorganised the Introduction considerably in an attempt to make it more streamlined and less repetitive

Similarly, I think "these negative changes" could be expanded on in the preceding sentences with an example of negative changes to the social environment.

This has now been addressed in the reorganization of the Introduction.

I am not sure how convinced I am that the human example here adds to the authors' point. As mentioned above, I would prefer to see more general examples of these ideas/effects in social mammals (or animals in general). If the authors' really want to keep a human example, I'd like to see significantly more detail about the circumstances in which this result was obtained.

This has been removed – we agree that it does not tie in very well here.

It feels slightly odd to discuss sex specific consequences before the general consequences.

Agreed – Introduction reorganized as discussed above.

As above, I am slightly unconvinced by the need/relevance to relate these results back to humans. This might be a result of me not having an anthropology background of course. Related to this, I feel this is making a rather large assumption about the familiarity a reader might have with this literature. Some small details about what these populations are might help convince that they are relevant to build the argument in the same paragraph as killer whales. There are a few other instances of this in the Discussion too, but I'll avoid repeating myself.

Thank you for pointing this issue out. We have now added further information to the Introduction to explain why humans are an important example – they are potentially the only social mammal where social buffering does overcome the social adversity of maternal loss.

Any directional predictions going into the study? There are quite a few variables under investigation, and it would be nice to link them more strongly to the ideas articulated in the Introduction. Is social buffering more likely in certain group compositions of age/sex?

We have now more clearly stated our hypotheses going into the study to the Introduction. We hypothesize that as demonstrated in chimpanzees, gorillas may face greater fitness costs if they suffer maternal loss at an earlier age and that males may face greater costs from maternal loss than females due to their longer periods of mother-offspring co-residence. Alternatively, as observed in many human populations, the cohesive, stable social groups of mountain gorillas may enable social buffering from group members to compensate for the social costs of maternal loss with minimal fitness consequences to maternal loss. In particular, dominant males may take on crucial roles in buffering the social adversity faced by maternal orphans, as past research has demonstrated the strong bond between dominant males and young orphans who may regularly share a nest at night.

Results: The term age mates initially confused me a little, I thought it was a typo in the table. Perhaps "cohort" or similar accepted term?

This term was used as it is the same term used in the elephant paper investigating social responses to maternal loss. It’s also quite a common term in the human literature although is slightly confusing as it has nothing to do with mating. We’ve therefore added in “those within 2 years age of the orphan” where it is first used.

Discussion: I feel you could just say "non-breeder" or "virgin" here.

We have changed it to “pre-reproductive”. We thought that non-breeder could be confused with infertile older females and virgin had too many human social connotations.

Materials and methods

I am also not quite sure about the use of simple paired t-tests to address a rather complex question, with many potential confounding variables. While I am all for using simple stats where possible, it seems like there would be more going on (age difference of a dyad, group size, age of orphan, year of study) that should be controlled for/investigated. Given the detailed models that follow, perhaps some clarity about why this is not necessary for the question currently being addressed would be useful?

Additionally, is there a citation for this permutation approach over permuting the networks themselves?

Yes, we agree there are some issues with this previous t-test approach. We have now altered this analysis to compare the change in network metric between the pre- and post-maternal loss periods, testing whether orphans and non-orphans show different changes in network metrics (as described above) using node-based permutations between immature gorillas (both orphans and non-orphans) within the same group (Described in the Materials and methods and Results.). This tests the observed pattern against a null model in which the network metrics of orphans and non-orphans within the same group and the same set of networks do not change differently after the maternal loss incident. This therefore controls for differences in group size, group composition, age-related changes over time, year of study etc. This does not control for the age of the immature gorilla themselves, however ages were distributed fairly evenly across both orphans and non-orphans (the mean age ± SD of orphans was 5.12 ±1.49 years and for non-orphans was 4.71 ±1.71 years) and there is therefore no reason to believe that age could be driving a difference between orphans and non-orphans.

This suggests multiple models, but it seems only one model is described. I think I need to see these models/model formula laid out in a table for clarity. I'd also like to see a citation for the use of the random effects structure to control network independence.

Model formulas have now been added to the table legends in which results are reported (Table 5 and Supplementary file 4 and Supplementary file 5).

The random effects structures are used to control for the non-independence of pairwise relationships involving the same individual but each pairwise relationship is only included in each model once. Apologies, this section was poorly worded and was been altered (subsection “Relationship changes following maternal loss”). We don’t believe that random effect structures would adequately account for non-independence of network metrics, hence the permutation approach used for degree and centrality metrics.

Reviewer #2:

This is a novel, very well written and well-researched project, making use of a fantastic long-term dataset. It is interesting to a wide readership and of high scientific value. I do not really have any major concern but a few queries and suggestions in order improve the clarity in places.

List of slightly more substantial comments/queries:

1) Results – throughout – can you please include sample sizes when stats values are given?

Apologies, this has now been added more consistently throughout.

2) Results – can you please also give us an idea of how much time they spent in proximity and contact? It’s good to see the stats but it would be nice to get a feel for who much time we are talking about (and how big the actual change is).

We have added in this information for mother-offspring dyads to give a feel for exactly what losing that bond means for their time in proximity and their affiliative contact. Changes overall for the group are harder to interpret as it’s the sum of the proportion of time spent in contact or proximity of another individual so a value of 1 could mean they are always in physical contact with only 1 other individual or that they are in physical contact with 2 other individuals simultaneously but only half the time. We have therefore included this information as a percentage increase or decrease compared to the pre-maternal loss period (subsection “Changes in network position following maternal loss”).

3) Discussion – I don't follow this argument – can you explain the reduced siring opportunities?

We have reworded this section to clarify – other studies have demonstrated that for males, those that remain in their natal group have greater siring opportunities than those that disperse. If they remain, they can potentially mate while they are subordinate (although far less so than the dominant male) and they can also wait it out for the dominant male to die or become weak so they can eventually take over. In contrast, males that disperse will not be accepted into another group and are solitary until they are able to attract females and form a group of their own. Many solitary males never succeed in attracting females and therefore have almost no siring opportunities.

Also, can you explain why infant-orphans are doing BETTER than older age orphans? This seems counter-intuitive and I am wondering why this would be the case?

Further details have been added (Discussion). We believe it may be due to two things. Firstly, that individuals that suffer maternal loss at a younger age show greater strengthening of proximity-based relationships with group members. Secondly, that there is a greater period of time between maternal loss and the age at which they could disperse. Both of these things mean that these infant-orphaned gorillas may have stronger social relationships within their group by the time they reach the age at which they may choose to disperse.

4) Can you discuss a little who initiated the contact and proximity? Would that be on the initiative of the orphan or of the other group members?

We cannot say from this data who initiated the contact or proximity, which is an important caveat to the study. This has now been discussed further (Materials and methods).

5) Can you provide any thoughts on what happens after the 6 months? You are concluding that the strengthening of bonds for 6 months after the loss of maternal care is responsible for the lack of adverse effects – but do you have any evidence or indication that this strengthening is extending over a longer time period? I understand that you chose 6 months in order avoid other confounding variables in the analysis, such as changes to group composition – but in the discussion it reads like these bonds are strengthen “forever”. Some indicators that this indeed the case would be good to see if this generalization is warranted.

Thank you for highlighting this important caveat, we have now added some discussion of this into the manuscript (Materials and methods).

6) Materials and methods – a few more details about the data feeding into the networks would be good – which behaviours were included as affiliative behaviours? And how was the SRI calculated, ie did you use frequencies or durations of time in contact? More details on how focal sampling was carried out (length of focals and frequency of data recording) would be useful.

More details on the behaviours included as affiliative contact have now been added to the Results section where they are first mentioned. Further details on the focal sampling approach and the use of frequencies rather than durations of association have now been added to the Materials and methods.

7) I would also like to see more details on number of groups, group sizes and compositions used in this study (maybe as supplementary material); all of this can go in supplementary material – but would be nice to have. Also, information on observation times for each social group and how stable they were would be helpful.

This information has been added to the supplementary material (Supplementary file 4).

Reviewer #3:

The manuscript "Social groups buffer maternal loss in mountain gorillas" examines the potential consequences of maternal loss in mountain gorillas. The topic of early life adversity and consequences of maternal care that extends beyond weaning is of growing interest to researchers including behavioral ecologists and anthropologists. The results presented in the manuscript are an interesting and important contribution to that literature. Whereas most studies have found negative fitness consequences associated with maternal loss, the results reported here indicate that gorillas who experience early maternal loss do not face negative consequences in terms of survival, maturation (age at first birth), or an indicator of reproductive success (first offspring survival). Furthermore, rather than speculating the authors follow-up on social buffering as a potential explanation for this somewhat unexpected (given outcomes observed in other social species) result using social network analyses. My comments on the manuscript primarily concern organization and clarity to help the reader, particularly the reader of a broader journal such as eLife.

1) I realize that the structure of an eLife article has the methods at the end, but as currently presented most of the results are impossible to interpret without a lot of flipping back and forth searching for information. For example, the Results start out by briefly stating that survival was examined using a cox proportional hazards model, which was very helpful for interpreting the results. However, in the next section there is reference to model results without any information about what type of model it is. Later, there are t statistics and p values with no explanations and no indication they came from permutation-based tests.

Thank you for pointing this out – this additional information has now been added throughout the Results section.

2) Some of the abruptness of the transition from introduction to results might be helped by stating clearer predictions that help set up the outcomes you tested. I found the descriptions and expectations concerning the fitness outcomes clearer than the social network outcomes and suggests more information be given about the networks before the results of the analyses are presented.

Agreed – thanks for this suggestion, we have now added this information at the end of the Introduction and prior to the network results (Results).

3) I also have some questions concerning the social network analysis. One additional analysis that might be interesting is looking at binary degree along with weighted degree. High weighted degree can result from few strong connections or many weak connections and presenting both weighted and binary degree might indicate one strategy (find a strong buddy) versus another (cast a wide net). Furthermore, what was the variation in group size in these data? Did you take variation in group size into account when calculating network metrics. The maximum weighted degree of an individual in a group with 5 individuals is much lower than the maximum weighted degree of an individual in a group with 15 individuals.

Thank you very much for these constructive suggestions. We have reanalyzed this section entirely incorporating binary degree and eigenvector centrality (Described in the Materials and methods and Results). By looking at the change in these metrics rather than the raw values we minimize the issues relating to variation in group size, but also control for these by including random effects for each network and only permuting between individuals within the same network to generate p-values. Data on group sizes are provided in Supplementary file 4 and the Figure 2—source data 1.

4) Results: Regarding the results for dispersal – any reason to believe group size will influence likelihood of dispersal?

No evidence of this was found in previous studies so we have not included it in the models in the interests of simplicity. Robbins, (1995). Stoinski et al., (2009).

5) Materials and methods: Social buffering of maternal loss results: can you clarify whether the edge to mom included when weighted degree was calculated?

This has now been reanalyzed, as described above. The edge to the mother was not included when looking at the overall change in network metrics and clarification of this has been added. Only broad descriptive statistics are used for the mother-offspring bond of orphans to give an idea of the importance of that relationship pre-maternal loss.

6) Table 5:

- How was age included in the GAMM models? Age in years?

Yes – added.

- This table could be presented more clearly. Sometimes the bold is used to describe the two variables under it (Maternal orphan and Group member) and other times the bold itself is a variable with multiple levels under it. Maybe just write out "Orphan age" "Orphan Sex" and not include the extra bold rows?

This has been reformatted.

- Should the age/sex class – sibling interaction results be relative to the dominant male who is or isn't a sibling?

Yes, this has now been added.

- Where are the results of the smoothed term?

These have now been added below the table.

7) Figure 1: Curious about why each age category has its own line. Since maternal loss is a time varying covariate shouldn't the survival probability of individuals who lost their mother at 7 (for example) be the same as non-orphans until age 7? Apologies if what I described above is the actually the case. It is tricky to tell where the dashed lines start.

You are correct that we used time varying covariates for the Cox-proportional hazards model, however we still distinguished between the three categories to make our study comparable to other studies (e.g., Stanton et al., 2020). The analysis shows two things, that there are no proportional differences between the three orphan classes and the non-orphans, therefore a proportional hazards is not the correct assumption, and second, when running the Bayesian model, we confirmed that there are in fact no differences between the orphan classes.

To make the points at which orphans and non-orphans split clearer we have added Figure 1—figure supplement 1 which plots each age category by sex separately and indicates the point from which their survival is modelled separately.

8) Materials and methods: Can you explain why sampling variation warranted a smooth? Not necessarily questioning your decision – I find it interesting and looking for more information about it!

We think it is probably unnecessary but accounts for any potential non-linear relationship between sampling and relationship strength. The SRI index takes into account any linear relationship, but we wanted to be extra cautious because of some of the uneven sampling introduced by different focal sampling protocols over the years (as described in more detail above).

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

As outlined in the first decision letter, this manuscript provides an "in-depth assessment of the consequences of maternal loss for wild mountain gorillas" and was found to be a "rigorous and detailed examination" that was appreciated by the Editors and all the reviewers. During the first assessment by the reviewers and Editors, some potential problems with the wording of particular parts of the text were raised, and this revision appears to address all of these in full. However, the first assessment also raised some issues with the reported analyses, and unfortunately, there is still some lack-of-clarity and some issues that need to be resolved. To outline these issues in as clear a way as possible, the below text takes a 3-step approach for each issue: (A) it revisits the initial comment, (B) provides the authors' response (and associated manuscript text), and finally (C) describes why a problem still exists and what needs to be addressed.

1) Differences in social changes

1A) Revisiting the issue: "The use of paired t-tests for the examination of social changes is a bit confusing here. Specifically -on my part- I'm unsure how these properly control for repeated sampling of individuals when considering changes in dyadic association scores? Also, the changes reported here appear to just assess how orphan behaviour altered between the pre- and the post- maternal loss period, but how do we know that such changes aren't expected anyway over time? Ideally this analysis would directly compared orphaned to non-orphaned individuals. Note the comments from reviewer 1 about these paired t-tests need to be addressed too"

1B) Author response: "Yes, we entirely agree and have changed these analyses to instead compare whether network metrics (binary degree, weighted degree and eigenvector centrality) change differently in orphans relative to other immature gorillas within the same group during the same time period using node-based permutations constrained to swap only between individuals within the same incident of maternal loss (based on the same pair of networks). Described in Materials and methods and Results."

1C) Current Issue:

While we appreciate that the authors have taken the first comment onboard and reconsidered the approach, there are two remaining problems with the current approach. Namely:

1Ci) In the Results, the results of these tests are reported. But, when reporting these results, it is important to not entirely focus on the p-value generated from the null models, but to also report the full results of the standard statistical test as well (i.e., the estimate, the SE, the t value, and the standard p-value). The p-value from the null model is simply used to 'double-check' the results of the standard statistical test and shouldn't be fully relied upon for conclusions like this. For example, the first result reported here is "(t=1.721,p=0.029)" which isn't clear to the reader, as a t-value this low wouldn't generally result in a p-value this low. Instead, the text needs to report the actual observed statistical test results from the standard test (including the real p-value) and then also include a Pnull value (the null value generated from the node permutations). If the primary standard statistical test shows no significant difference (as may be the case here), then this primary result should be relied upon for drawing conclusions (not the null model p-value, as this is mainly just for double-checking tests, rather than drawing direct conclusions from them in this particular way).

Thank you for clarifying this point. We have now included the full model outputs as supplementary tables and reported the estimate, SE, t-value, standard p-value and Pnull, and adjusted the wording to focus less on the Pnull. (Results). Standard p-value and Pnull values correspond closely overall. With changes in the overall model (described below in our response to sections 2 and 3) results remain very similar. Metrics from networks based on affiliative contact do not increase in orphans relative to non-orphans. The eigenvector centrality of orphans in networks based on proximity still shows a strong increase relative to non-orphans (Est=0.169±0.037, t=4.594, P<0.001, Pnull<0.001), whilst weighted degree also increases but to a lesser extent (Est=0.075±0.032, t=2.337, P=0.021, Pnull=0.024). Binary degree of orphans no longer increases significantly relative to non-orphans in these proximity-based networks.

Also, as a small side-point, the method used here for actually calculating the node permutation p-value is quite a strange one. Specifically, it is stated in the revised text that "two-tailed p-values were calculated as two times the proportion of permutation t-values greater or smaller than the t-value from the real data set". But this is problematic and doesn't make clear sense? What is actually being done here?

We have reworded this section to make the approach we have used clearer. We used a two-tailed approach since we were interested in both whether the observed test statistic was larger than those in the null models (ie strengthening of relationships between orphans and other group members after maternal loss lead to increased network metrics) or smaller than those in the null models (ie orphans relationships with other group members weakened after losing their mothers relative to other immature gorillas in the group leading to decreased network metrics). If the observed t-value was below the median of the null model t-values, Pnull was calculated as:

2 x numberofnulltvalueslowerthanobservedtvaluetotal null t values

If the observed t-value was above the median of the null model t-values, Pnull was calculated as:

2 x numberofnulltvaluesgreaterthanobservedtvaluetotal null t values

Such that a p-value of <0.05 would indicate the observed t-value was within the lowest 2.5% of null values or highest 2.5% of null values.

1Cii) The text goes on to state that "within contact-based networks orphans did not show these same gains, with no significant increase in binary degree (t=0.099, p=0.730), weighted degree (t=1.110, p=0.397) or eigenvector centrality (t=0.129, p=0.741) relative to non-orphans". But it isn't clear where these results are actually derived from. We are assuming that they come from a subset of tests the authors refer to in their response, i.e., 'node-based permutations constrained to swap only between individuals within the same incident of maternal loss'. But, if the swaps are constrained between individuals who experience the same incident of maternal loss, and you subsequently try to look at the comparable difference between incident of maternal loss, then the tests will never find a difference because this is already controlled for earlier in the pipeline? Some clarity about what is actually going on here needs to be added to the manuscript (and also some thought about whether such tests are appropriate for testing the questions at hand).

Apologies –this was not worded very clearly by us. When we refer to “the same incident of maternal loss” we are meaning individuals within the same group when a group-member suffered maternal loss (same group, same time period, same set of networks) rather than specifically only swapping between individuals that suffered maternal loss themselves. Ie if there were 5 immature gorillas within a group and one suffered maternal loss, our node-based permutations would permute the orphan/non-orphan labels associated with each of those immature gorillas in that network. We have tried to make this clearer in the manuscript. These values come from the same models discussed above which are run twice: once on networks based on 2m proximity and once on networks based on affiliative contact. This has now been clarified further in the manuscript.

2) Social networks and observations

2A) Revisiting the issue: The first assessment of the paper flagged up that animal social networks are strongly influenced by observation, and requested some analytical consideration of this in the revisions.

2B) Author Response: The authors responded by providing more details about the data collection methods, which is very much appreciated. However, the revision doesn't include any attempts to actually control for/integrate consideration of differences in observation number. Instead, the response states that the Simple Ratio Index includes this (stating "variation in the number of focal scans per individual which is accounted for within the simple ratio index" and "we did not believe that these extra data would bias the estimate of the specific relationships based on SRI (as these account for the total number of observations)".

2C) Current Issue: Unfortunately, the SRI does not control for differences in observation number in the way that the authors lay out here. While the SRI takes some of the difference in observation into account during the calculation of the dyadic values, it in no way fully accounts for this; it just uses it for the calculation. For instance, even in simulations where all individuals have the same social phenotype, those individuals which are observed more will have higher centrality (e.g., higher degree) than those which are observed less. This is simply a product of how these metrics are calculated and how these networks emerge. Therefore, the previous comments from the Editors and the reviewers that the analyses should directly consider 'observation number' still stands.

Thank you for pointing this out. All network metric models now include a smoothing term for the number of focal scans per individual to account for possible increases in network metric with number of observations of an individual (described in the Materials and methods). A smoothing term was chosen as these relationships may not be linear, particularly at higher numbers of observations. The models of pairwise relationships include the total number of focals of both individuals involved in the relationship as a smoothing term (as was the case in the previous draft).

3) Outstanding analytical issues

3A/B) Revisiting the issue: In the reviewer comments, a few analytical (smaller) issues were pointed out, and in most cases the authors have addressed these fully and clearly. However, a couple of key points remain. Particularly, controlling for age and network differences within the models themselves, and also using 'change in metric' as the response.

3C) Current Issue: The revision didn't make any attempt to actually integrate age of individuals into the models, nor did it try to consider network groups, but the reviewers made solid points that these would be good to consider. We believe these should be integrated into the paper as supplementary analyses. Finally, the switch to considering 'change in metric' as the response is certainly useful in lots of ways, but have the authors' considered problems with 'regression to the mean' in this sense? Specifically, we can automatically expect this metric to be altered significantly by the initial value, whereby those which initially have a large SN metric value are likely to (just by chance) have large -ve change values, and whereby those which initially have a small SN metric value are likely to (just by chance) have large +ve change values. Have the authors assessed this as a driver/contributor to the results here? (Note this consideration relates to the above point about considering things like age, and network group, which might alter these initial values and thus affect the change values.)

Thank you for these suggestions. We have rerun all models of network metric change before and after maternal loss incorporating these suggestions. Age has now been directly included in the models and the metrics themselves have been adjusted relative to the specific network so that binary and weighted degree of each individual is proportional to the maximum value of binary or weighted degree within the network (as is already the case for eigenvector centrality), described in the Materials and methods. This should make network metrics more comparable across networks, but we also still included the network (or incident of maternal loss as referred to in earlier drafts) as a random factor in all models. The potential for initial values that are far from the mean to be driving the results is also now accounted for by including the deviance of the initial network metric value from the group mean (separate mean values calculated for each network as this was included as a random factor).

New Model run as a GAMM to allow inclusion of a smoothing term for number of focal scans:

Metric Change ~ Orphan + Age + Deviance of initial value from group mean + s(focal scans), random=~(1|Network))

As described in our response to section 1, these model changes did not alter the overall findings except that the increase in binary degree of orphans in proximity-based networks went from being weakly significant to no longer significant. This however still supports our previous conclusions that orphans primarily strengthen existing social bonds following maternal loss rather than forming new ones.

https://doi.org/10.7554/eLife.62939.sa2

Article and author information

Author details

  1. Robin E Morrison

    1. Dian Fossey Gorilla Fund, Musanze, Rwanda
    2. Centre for Research in Animal Behaviour, University of Exeter, Exeter, United Kingdom
    Contribution
    Data curation, Formal analysis, Visualization, Writing - original draft, Writing - review and editing
    For correspondence
    rmorrison@gorillafund.org
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9161-4734
  2. Winnie Eckardt

    Dian Fossey Gorilla Fund, Musanze, Rwanda
    Contribution
    Conceptualization, Data curation, Formal analysis, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
  3. Fernando Colchero

    1. Department of Mathematics and Computer Science, University of Southern Denmark, Odense, Denmark
    2. Interdisciplinary Center on Population Dynamics, University of Southern Denmark, Odense, Denmark
    Contribution
    Formal analysis, Visualization, Writing - review and editing
    Competing interests
    No competing interests declared
  4. Veronica Vecellio

    Dian Fossey Gorilla Fund, Musanze, Rwanda
    Contribution
    Data curation, Project administration
    Competing interests
    No competing interests declared
  5. Tara S Stoinski

    Dian Fossey Gorilla Fund, Musanze, Rwanda
    Contribution
    Conceptualization, Funding acquisition, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared

Funding

No external funding was received for this work.

Acknowledgements

We are grateful to the Rwandan Development Board (RDB) for their long-term support of the Karisoke Research Center and thank the Karisoke field staff for collection of the data. We are thankful to Edward Wright for his support with the Elo-rating analysis for male dominance hierarchies. We are grateful to Lauren Brent and her research group for providing insightful feedback and discussion on the project, and to members of the University of Exeter’s Social Network Club for their advice on statistical analyses. We thank Elizabeth Lonsdorf for her valuable feedback on the manuscript and the editors and reviewers at eLife for their constructive comments.

Ethics

Animal experimentation: The research presented here was non-invasive and did not involve any animal experimentation. It was approved by the Rwandan Development Board and conducted in accordance with the ethical standards of the Dian Fossey Gorilla Fund and the International Primatological Society's Code of Best Practices for Field Primatology.

Senior Editor

  1. Christian Rutz, University of St Andrews, United Kingdom

Reviewing Editor

  1. Josh Firth, Edward Gray Institute, United Kingdom

Publication history

  1. Received: September 10, 2020
  2. Accepted: February 16, 2021
  3. Version of Record published: March 23, 2021 (version 1)

Copyright

© 2021, Morrison et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 809
    Page views
  • 89
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Ecology
    Jakob Thyrring, Lloyd S Peck
    Research Article Updated

    Whether global latitudinal diversity gradients exist in rocky intertidal α-diversity and across functional groups remains unknown. Using literature data from 433 intertidal sites, we investigated α-diversity patterns across 155° of latitude, and whether local-scale or global-scale structuring processes control α-diversity. We, furthermore, investigated how the relative composition of functional groups changes with latitude. α-Diversity differed among hemispheres with a mid-latitudinal peak in the north, and a non-significant unimodal pattern in the south, but there was no support for a tropical-to-polar decrease in α-diversity. Although global-scale drivers had no discernible effect, the local-scale drivers significantly affected α-diversity, and our results reveal that latitudinal diversity gradients are outweighed by local processes. In contrast to α-diversity patterns, species richness of three functional groups (predators, grazers, and suspension feeders) declined with latitude, coinciding with an inverse gradient in algae. Polar and tropical intertidal data were sparse, and more sampling is required to improve knowledge of marine biodiversity.

    1. Ecology
    2. Evolutionary Biology
    Aspen T Reese et al.
    Research Article

    Domesticated animals experienced profound changes in diet, environment, and social interactions that likely shaped their gut microbiota and were potentially analogous to ecological changes experienced by humans during industrialization. Comparing the gut microbiota of wild and domesticated mammals plus chimpanzees and humans, we found a strong signal of domestication in overall gut microbial community composition and similar changes in composition with domestication and industrialization. Reciprocal diet switches within mouse and canid dyads demonstrated the critical role of diet in shaping the domesticated gut microbiota. Notably, we succeeded in recovering wild-like microbiota in domesticated mice through experimental colonization. Although fundamentally different processes, we conclude that domestication and industrialization have impacted the gut microbiota in related ways, likely through shared ecological change. Our findings highlight the utility, and limitations, of domesticated animal models for human research and the importance of studying wild animals and non-industrialized humans for interrogating signals of host–microbial coevolution.