Discussion
Studies have applied social network analysis to quantify and explore the
social structure of populations across numerous taxa (Brent, 2015;
Brent, Lehmann, & Ramos-Fernández, 2011; Sueur et al., 2011), but to
our knowledge this is the first to combine genetically-derived pedigree
data with social network analysis to infer social structure of wild
populations. Social network analyses are powerful and flexible methods
for investigating the complex networks of interconnections between
individuals within and between populations, providing numerous measures
of individual sociality (Brent, 2015; Wasserman & Faust, 1994). With a
large interconnected network of 1,562 nodes (individuals) and 1,866
edges (parent-offspring relationships) between individuals, it can be
difficult to identify significant differences within the network. By
bringing the familial network into a spatial framework and incorporating
aspatial node-based centrality metrics, we were able to identify
different levels of sociality within the network, with some local areas
composed of stable family groups and others that are less sedentary.
Comparing local area networks of management interest allowed us to
identify areas of higher and lower fitness and connectivity in the
overall boreal caribou familial network.
There are numerous network centrality metrics available, and many are
often correlated (Brent, 2015; Freeman, 1977; Newman, 2003; Sueur et
al., 2011; Wasserman & Faust, 1994; Wey & Blumstein, 2012). Indirect
measures of centrality are commonly used in social network analysis to
identify central individuals within a network (Brent, 2015; Wey,
Blumstein, Shen, & Jordán, 2008), but there has been little research
done to determine which of these metrics are suitable for measuring
centrality in pedigree-based familial networks. As familial networks are
not built on associations between individuals or constructed through
direct observation, measures of network centrality apply differently,
and not all metrics may be suitable for familial-based networks. By
identifying local areas within the network, we were able to gain a
better understanding of which areas contributed most to the familial
network. We found significant differences in centrality measures between
local areas in the full familial network, and these variations in
individual centrality would have remained hidden if only the full
familial network was examined. We used five centrality metrics in our
social network analysis of familial networks (Figure 2), and found that
alpha, betweenness, and eccentricity centrality were the most
informative measures of individual centrality (Figure S1.1). Degree
centrality in familial networks represents the parents of an individual
(in-degree) and the offspring of an individual (out-degree), giving a
direct measure of an individual’s reproductive output and fitness
levels. It is important to note, however, that inferred individuals in
the pedigree will always have an in-degree of 0, as it is not possible
to infer the parents of inferred individuals. Alpha centrality is an
important metric for familial networks, as it indicates those
individuals who are connected to individuals who themselves are highly
connected, giving an indication of individual status, even if that
individual does not have a lot of direct connections (offspring).
Reproductive output can be highly asymmetrical, with the number of
offspring varying between individuals (McFarlane et al., 2018), and
alpha centrality can indicate if that individual is part of a
high-fitness family if they are connected to highly connected
individuals. We found that local areas with high edge-to-node ratios had
a wider distribution of alpha and degree centrality, indicating that
more higher fitness individuals are found in these local areas than in
low edge-to-node local areas (Figure 2C), and are better connected to
other well-connected individuals. Three of the four high edge-to-node
ratio local areas we identified are located in the western part of
Saskatchewan’s Boreal Plains, which has the highest levels of both
anthropogenic and fire disturbance in the Boreal Plains (Figure S3.2),
and the tight family groups we observed in these areas may be a result
of decreased dispersal propensity due to high levels of fragmentation
between local areas.
Betweenness centrality is another important metric for network analysis,
as it captures the interconnectedness of subgroups; individuals with
high betweenness interact with individuals who do not interact with one
another, therefore making betweenness important for maintaining group
cohesion, and connecting disparate parts of the network (Brent, 2015).
Our familial network was not comprised of subgroups, as most individuals
(94.2%) had a betweenness centrality of 0, and 95.2% of all sampled
individuals formed one large familial network. Even after the removal of
edges with the highest edge betweenness, the overall network structure
did not change, with most individuals still connected in one main
network, with no clear subgroups (Figure S2.10). Our study species
displays a polygamous mating system, with individuals potentially having
multiple partners, producing a complex network of parent-offspring
relationships and full- and half-siblings, with high interconnectedness
among individuals across the network (Figure S2.1). Our highly
interconnected network with no evidence of subgroups and low average
betweenness centrality is the result of the polygamous mating system and
high dispersal ability.
The high eccentricity centrality and low closeness centrality informs on
the social structure and the presence of animals dispersing longer
distances in the Boreal Shield. The Boreal Shield is less fragmented
than the Boreal Plains, with significantly less anthropogenic
disturbance (Figure S3.2; Table S3.1). Very few parent-offspring
relationships occurred within or between the northern Boreal Shield
local areas (Figure 1). This suggests that individuals in the Boreal
Shield are not central to the familial network and have lower individual
fitness, not reproducing many offspring that survive until fall (low
degree centrality). Individuals in low edge-to-node local areas are not
from the same familial lines and are not highly related to any other
individuals in the network. The removal of high betweenness edges led to
some individuals becoming disconnected from the full network, but these
disconnected individuals were not from one local area, instead located
throughout both ecozones, again highlighting the interconnectedness of
the familial network.
In most animal network studies, nodes represent observed individuals,
with relationships between pairs of individuals (dyads) defined by an
association index (the time the pair of individuals spent together),
with edges representing observed relationships, forming an interaction
network (Morrison, 2016; Whitehead & Dufault, 1999). For many species,
it is not possible or feasible to directly observe rare and elusive
species, and therefore association information cannot be obtained.
Pedigree reconstruction can give direct information about dyads between
closely related individuals (parent-offspring and full-siblings), with
these relationships forming the basis of the familial network. In
comparison to association networks, in familial networks, only the
sampled individuals are known or observed, and the edges between
individuals and the unsampled individuals (parents) are inferred by the
data analysis (Morrison, 2016). Reconstructing a familial network from
genetically-derived pedigree data gives valuable information about the
number of mating partners, the number of offspring, and the structure of
the reproductive network of a population (McFarlane et al., 2018;
Pemberton, 2008). Pedigrees represent historical and evolutionary
connections between generations; these relationships have long been
recognized as reticulating but are instead commonly presented as
simplified trees instead of networks, where reticulations caused by
inbreeding are absent (Morrison, 2016). Pedigrees represent a network of
relationships, and therefore reconstructed pedigrees inherently contain
information that can be used to construct a network. With a wide
spectrum of mating systems present in species (Clutton-Brock, 1989),
almost all species have pedigree networks, with multiple partners and/or
offspring attributed to each individual, therefore creating a complex
network of familial relationships (Morrison, 2016). Although boreal
caribou have a skewed reproductive rate, with varying levels of
individual fitness (McFarlane et al., 2018), our network does not appear
to be vulnerable to sudden population crashes resulting from changes in
social structure or social isolation and inbreeding. Due to the
polygamous mating system and long-range dispersal ability, the boreal
caribou network is highly interconnected, and removal of edges with high
betweenness did not change the overall network structure or lead to
distinct clusters or social groups, although family groups can be
identified within the network, with varied levels of dispersal and
fitness levels among family groups.
Social networks are powerful methods to assist in wildlife conservation
(Snijders et al., 2017), but most wild populations cannot be directly
observed, and social association networks cannot be constructed. By
constructing a familial network based on genetically-derived
parent-offspring relationships, we calculated informative metrics to
draw a much finer picture of their individual fitness levels, pattern of
demographic structure, and relative contribution of local areas to the
larger population. The spatial application of the familial network
allowed us to identify areas of higher fitness levels and social
cohesion across the range in support of population monitoring and
recovery efforts.