The structure of stereotyped calls reflects kinship and social

DOI 10.1007/s00114-010-0657-z
The structure of stereotyped calls reflects kinship and social
affiliation in resident killer whales (Orcinus orca)
Volker B. Deecke & Lance G. Barrett-Lennard &
Paul Spong & John K. B. Ford
Received: 14 December 2009 / Revised: 2 February 2010 / Accepted: 13 February 2010
# Springer-Verlag 2010
Abstract A few species of mammals produce groupspecific vocalisations that are passed on by learning, but
the function of learned vocal variation remains poorly
understood. Resident killer whales live in stable matrilineal
groups with repertoires of seven to 17 stereotyped call
types. Some types are shared among matrilines, but their
structure typically shows matriline-specific differences. Our
objective was to analyse calls of nine killer whale matrilines in British Columbia to test whether call similarity
primarily reflects social or genetic relationships. Recordings
were made in 1985–1995 in the presence of focal matrilines
Electronic supplementary material The online version of this article
(doi:10.1007/s00114-010-0657-z) contains supplementary material,
which is available to authorized users.
V. B. Deecke (*)
Sea Mammal Research Unit, Scottish Oceans Institute,
University of St. Andrews,
St. Andrews,
Fife KY16 8LB Scotland, UK
e-mail: [email protected]
V. B. Deecke : L. G. Barrett-Lennard
Cetacean Research Laboratory, Vancouver Aquarium,
P.O. Box 3232, Vancouver, BC V6B 3X8, Canada
V. B. Deecke : L. G. Barrett-Lennard : J. K. B. Ford
Department of Zoology, University of British Columbia,
6270 University Blvd.,
Vancouver, BC V6T 1Z9, Canada
P. Spong
OrcaLab, Hanson Island,
P.O. Box 510, Alert Bay, BC V0N 1A0, Canada
J. K. B. Ford
Cetacean Research Program, Pacific Biological Station,
Fisheries and Oceans Canada,
Nanaimo, BC V9T 6N7, Canada
that were either alone or with groups with non-overlapping
repertoires. We used neural network discrimination performance to measure the similarity of call types produced by
different matrilines and determined matriline association
rates from 757 encounters with one or more focal matrilines. Relatedness was measured by comparing variation at
11 microsatellite loci for the oldest female in each group.
Call similarity was positively correlated with association
rates for two of the three call types analysed. Similarity of
the N4 call type was also correlated with matriarch
relatedness. No relationship between relatedness and
association frequency was detected. These results show
that call structure reflects relatedness and social affiliation,
but not because related groups spend more time together.
Instead, call structure appears to play a role in kin
recognition and shapes the association behaviour of killer
whale groups. Our results therefore support the hypothesis
that increasing social complexity plays a role in the
evolution of learned vocalisations in some mammalian
Keywords Vocal learning . Gene-culture coevolution .
Social network . Kin recognition
The reasons why some animals including humans have
evolved learned vocalisations are open to debate. One
possible explanation is that vocal learning evolved as a
means to increase the complexity of signals to meet an
increased need for the recognition of individuals, kin, or
social partners as social systems became more complex
(Janik and Slater 1997). Variation in the structure of
vocalisations often reflects social relationships between
individuals and thus potentially encodes information about
association patterns (e.g. Miller and Bain 2000; Rendell and
Whitehead 2003), membership of a social or reproductive
unit (e.g. Enggist-Dueblin and Pfister 2002; Watwood et al.
2004) or use of a common foraging or resting site (e.g.
Wright 1996; Boughman 1997).
Vocal dialects can be defined as differences in the
structure of vocalisations among animals that live in
sympatry and come into acoustic contact (Conner 1982).
Dialect variation has been documented in species from diverse
taxonomic background, e.g. songbirds (Trainer 1989;
Enggist-Dueblin and Pfister 2002), parrots (Wright 1996;
Bartlett and Slater 1999), bats (Boughman 1997), cetaceans
(Ford 1991; Rendell and Whitehead 2003; Watwood et al.
2004) and primates (Gouzoules and Gouzoules 1990;
Crockford et al. 2004). In the majority of cases where the
mechanisms of dialect development have been investigated,
vocal learning has been shown to play a role (Boughman
1998; Bartlett and Slater 1999; Deecke et al. 2000).
However, in many species, social groups consist of closely
related individuals making it difficult to dissociate the
relative effects of socially mediated vocal learning and
genetic relatedness on dialect variation.
Resident killer whales in the northeastern Pacific feed
almost exclusively on fish, primarily on Pacific salmon
Oncorhynchus spp. (Ford and Ellis 2006), and live in
stable social groups (matrilines) that consist of a female
and her offspring. The social structure of resident killer
whales is characterised by natal philopatry, and both
male and female offspring typically travel with their
mothers for their entire lives (Bigg et al. 1990). Because
killer whales are long-lived, matrilines often contain three
or even four generations related through maternal descent.
Groups may eventually split when the oldest female dies
and her daughters (who by then usually have offspring of
their own) spend more and more time apart from each
Killer whales emit three types of vocalisations: echolocation clicks, whistles and pulsed calls. Pulsed calls are
highly stereotyped, and many call types contain independently modulated high- and low-frequency components
(Ford 1989; Miller 2002). These calls exhibit groupspecific variation at the level of the vocal repertoire as
well as in the structure of individual call types (Fig. 1). All
members of a matriline share a common repertoire of seven
to 17 call types (Ford 1991). Some of these may be shared
with other, presumably related matrilines. However, shared
call types often show some degree of group-specific
structural variation (Ford 1991; Deecke et al. 1999; Miller
and Bain 2000; Nousek et al. 2006). While killer whale call
repertoires are stable over long periods of time (Ford 1991),
the structure of individual call types undergoes subtle
changes with time. These changes can be transmitted
between frequently associating matrilines through vocal
learning (Deecke et al. 2000).
This study focused on nine matrilines belonging to the
Northern Resident population of killer whales that inhabits
the inshore waters of British Columbia, Canada. Our
objective was to test the hypothesis that variation of
stereotyped calls encodes information about group relatedness and social affiliation by comparing patterns of
variation in call structure to association patterns of the nine
matrilines as well as to relatedness of the groups' matriarchs
determined from microsatellite variation. Including both
measures allows us to separate the effects of genetic and
social factors on call variation and to test whether differences in call structure primarily reflect the strength of social
bonds, kinship or a combination of the two.
Materials and methods
The study population
Northern Resident killer whales inhabit the coastal waters
of British Columbia from central Vancouver Island north to
the Alaskan Border. The population numbered 216 individuals in 1998 (Ford et al. 2000), and the animals are
grouped into three acoustic clans, with all members of a
clan sharing a portion of their call repertoire (Ford 1991).
One of these, A-clan, can be further subdivided into two
subclans. The nine matrilines of A-subclan that are the
focus of this study have vocal repertoires of 13 or 14 call
types, of which 11 are shared by all. They only have seven
call types in common with the members of B-subclan (Ford
Acoustic analysis
Recordings were made during 1985–1995 using a variety of
recording systems. All systems had a flat frequency
response from 0.1 to 7 kHz, although for some systems,
this extended up to 20 kHz. We only included recordings
made when a single focal matriline was within recording
range, or when the focal matriline was associated with
groups belonging to a different clan and thus known not to
make any of the call types analysed in this study (Ford
1991). Recordings were digitised with a sampling rate of
22.1 kHz. We selected calls with adequate signal-to-noise
ratio and assigned them to call types according to Ford
(1987). We generated spectrograms for each selected call
(FFT size and frame length: 1,024 samples; overlap
between frames: 87.5%; window function: Hamming) and
extracted a frequency contour of the low-frequency component using the method of Deecke et al. (1999). Frequency
contours were standardised to 100 equally spaced points,
Fig. 1 Spectrograms of examples of the N9 call type from six
matrilines belonging to A-subclan of the Northern Resident population
of killer whales (Orcinus orca). Versions of different matrilines show
consistent differences in call structure. Note for example the
pronounced terminal component (abrupt drop in pulse-repetition rate
towards the end of the call) in the versions of A08, A23 and A25
matrilines. This feature is absent in the N9 calls of A12 and A30
matrilines. The versions by A36 matriline tend to be intermediate with
only a short terminal component
and call duration was also included in the analysis creating
an input vector of 101 variables. Because the highfrequency component is highly directional (Miller 2002),
it is often indistinct, and including it in the analysis would
have severely restricted our sample size. However, previous
studies have shown that patterns of variation tend to be
broadly similar for both call components (Miller and Bain
2000; Nousek et al. 2006; Fig. 1).
We measured call similarity in pairwise comparisons by
training an artificial neural network to discriminate between
samples of input vectors of the same call type made by two
different matrilines. To capture within-group variation,
samples were selected to include calls from as many
independent recording sessions from each matriline as
possible. No sample from any matriline contained calls
from fewer than three independent sessions with that group.
To generate a similarity index, a single vector was removed
from the sample, a backpropagation neural network trained
to discriminate between the vectors of the two matrilines,
and the removed vector was used to test the neural network
performance. This procedure was repeated until all vectors
had served as test vectors (see Deecke et al. 1999, 2000 and
Nousek et al. 2006 for greater detail on the neural network
analysis). The network's classification error (the absolute
difference between expected and observed output for a
given trial) averaged over all trials served as a call
similarity index for the two groups. A large error (near
50% incorrect classification) therefore signifies very similar
calls, while a low error indicates consistent group-specific
differences in call structure and/or duration.
Analysis of association patterns
We calculated the half-weight index of association (Cairns
and Schwager 1987; Ginsberg and Young 1992) of the nine
matrilines from a database of sightings of Northern
Resident killer whales from the years 1990–1995. This
covers the period when the majority of the recordings
analysed in this study were made and thus best reflects
the social structure of A-subclan at the time call
Fig. 2 The interaction of dialect similarity with association behaviour
and relatedness of resident killer whales (Orcinus orca). The scatter
plots show the relationship between the similarity of three stereotyped
call types of different A-subclan matrilines and their half-weight index
of association (a–c) and the relatedness of the oldest females in each
group (d–f) determined by analysis of microsatellite variation. The
results of the Mantel's matrix permutation test are given in the top left
corner of each panel, and regression lines are shown for significant
matrix correlations. For the call similarity-relatedness correlations (d–
f), note the absence of points near the top and left of the graphs.
Although there are pairs of matrilines with above-average relatedness
but not very similar calls, the converse is not true: all pairs whose call
similarity falls above the group average also have above-average
relatedness values
variation was assessed. We only included encounters
during which one or more matrilines of A-subclan were
present and did not analyse associations with groups
outside of A-subclan.
Analysis of matriarch relatedness
DNA was obtained from biopsy samples taken from
identified individuals using a lightweight pneumatic dart
Fig. 3 The interaction between association behaviour and relatedness
in resident killer whales (Orcinus orca). The scatter plot shows the
relationship between the association patterns of the nine A-subclan
matrilines and the relatedness of the oldest females in each group
determined by analysis of microsatellite variation. The result of the
Mantel's matrix permutation test is given in the top left corner, and the
regression is shown in dashed grey. The plot suggests that there is no
clear relationship between the relatedness of matriarchs of groups and
their association frequency
(Barrett-Lennard et al. 1996) or, in one case (A09 matriline), from a sample taken during a necropsy. Genetic
relatedness of the oldest female in each group was assessed
by studying variation at 11 microsatellite loci. Details on
extraction protocols, primers and amplification conditions
are given in the Electronic supplementary material. We
used the method of Queller and Goodnight (1989) to
calculate pairwise coefficients of relatedness. Allele frequencies were estimated from a sample of 75 individuals
from A-clan of the Northern Resident Community.
Statistical analyses
The acoustic similarity matrices for the different call types
and matrices of association and relatedness for the nine
matrilines were compared by computing the matrix correlation coefficient (Zar 1996). Mantel's matrix permutation
test (Schnell et al. 1985) was used to test whether matrix
correlations were significant.
Results and discussion
We obtained sufficient sample size (>20 calls) from more
than three matrilines for three call types. Sample sizes for
the analysis of acoustic variation were N2 call type: four
matrilines, 21 calls each; N4 call type: nine matrilines, 24
calls each and N9 call type: six matrilines, 21 calls each.
Call similarity values ranged from 0.003 (N9 call, A08 and
A36 matrilines) to 0.482 (N9 call, A23 and A25 matrilines).
A total of 757 encounters in the sightings database involved
one or more matrilines of A-subclan. Half-weight indices of
association ranged from 0.239 (A09 and A11 matrilines) to
0.973 (A11 and A24 matrilines). Matriarch relatedness
ranged from −1.329 (A12 and A23 matrilines) to 0.561
(A09 and A36 matrilines).
Call similarity was significantly correlated with association frequency for two of the three call types analysed
(Fig. 2a–c). Call similarity was significantly correlated with
relatedness of the matriarchs for one of the three call types
(Fig. 2d–f). There was no significant relationship between
the groups' association patterns and relatedness of their
matriarchs (Fig. 3). Because significant correlations were
always found for the call type(s) for which samples from
the largest number of matrilines were analysed, it seems
likely that sample size is largely responsible for why some
call types gave significant correlations while others did not.
Structural variation in the two call types with the largest
sample size reflects the association patterns of the groups as
would be expected if modifications to call structure are
more likely to be transmitted between frequently associating groups (Deecke et al. 2000). However, variation in the
structure of the one call type for which samples from all
nine matrilines were available was also significantly
correlated with the relatedness of the groups' matriarchs,
even though there is no evidence that resident killer whale
matrilines associate preferentially with other closely related
matrilines. This rules out the possibility that kinship affects
call similarity only via association patterns. It appears that
killer whales do not copy call modifications indiscriminately but are selective as to who they copy from.
The strongest bonds in resident killer whale society are
those between mothers and their offspring, and individuals
spend their entire lives in close association with maternal
relatives. As calls are likely learned from maternal relatives,
we would expect maternal descent, rather than bi-parental
relatedness as given by the coefficient of relatedness using
microsatellite data, to be the best correlate of call similarity.
Unfortunately, analyses of mitochondrial DNA in resident
killer whales have found at most two very similar
haplotypes within a population and no variation within
clans (Hoelzel and Dover 1991; Yurk et al. 2002). We
therefore have no way to reconstruct maternal relatedness
prior to the onset of studies using photographic identification of individuals. The plot of N4 call similarity and
matriarch relatedness (Fig. 2e) shows that there are groups
with relatively closely related matriarchs, and relatively
dissimilar calls, but no groups with similar calls but only
distantly related matriarchs. This is consistent with a theory
of maternal transmission of call similarity—groups with
related matriarchs but dissimilar calls share paternal rather
than maternal ancestry, whereas all groups with similar calls
have recent common maternal ancestors.
An interesting finding of our study was the lack of a
clear link between association patterns and kinship at the
level of the matriline. Because we analysed bi-parental
relatedness, there is again the possibility that a clear link
between maternal relatedness and association patterns was
masked by patterns of paternal ancestry. However, since we
analysed association patterns for all nine matrilines and thus
had equal statistical power as in the analysis of the N4 call,
we can conclude that any link between relatedness and
association patterns is weaker than the link between
relatedness and call similarity. Our results suggest that
factors other than kinship take over in determining
association behaviour once matrilines split, while at least
some call types continue to encode information about
kinship, as shown by the correlation between similarity of
the N4 call type and matriarch relatedness.
The tight correlation between the similarity of most call
types and association patterns in the absence of any
relationship between matriarch relatedness and association
strength suggests that the similar dialects of closely related
groups are not simply an epiphenomenon mediated by
association behaviour. Instead, call similarity and association
strengths may both depend on some other as of yet
unidentified parameter. Alternatively, and more likely
perhaps, rather than being a result of association patterns,
dialect variation may actually drive the association behaviour of matrilines: call similarity provides a reliable
mechanism to identify maternal kin and may be a foundation
on which choices of social affiliates are based. Our results
therefore provide evidence in favour of the hypothesis that
vocal learning evolved to recognise kin and facilitate social
decisions in increasingly complex social networks.
Acknowledgements We thank David E. Bain, David A. Briggs,
Graeme M. Ellis, Alexandra Morton, Helena Symonds, Frank
Thomsen and Stephen Wischniowski for contributing additional
recordings for analysis. Helena Symonds and Nicola Rehn provided
valuable comments on the manuscript. VBD was supported by a
DAAD-Doktorandenstipendium aus Mitteln des 3. Hochschulsonderprogramms and a Marie-Curie Intra-European Fellowship.
Barrett-Lennard LG, Smith TG, Ellis GM (1996) A cetacean biopsy
system using lightweight pneumatic darts, and its effect on the
behaviour of killer whales. Mar Mamm Sci 12:14–27
Bartlett P, Slater PJB (1999) The effect of new recruits on the flock
specific call of budgerigars (Melopsittacus undulatus). Ethol Ecol
Evol 11:139–147
Bigg MA, Olesiuk PF, Ellis GM, Ford JKB, Balcomb KC (1990)
Social organization and genealogy of resident killer whales
(Orcinus orca) in the coastal waters of British Columbia and
Washington State. Rep Int Whaling Comm Spec Issue 12:383–
Boughman JW (1997) Greater spear-nosed bats give group-distinctive
calls. Behav Ecol Sociobiol 40:61–70
Boughman JW (1998) Vocal learning by greater spear-nosed bats.
Proc R Soc Lond, Ser B: Biol Sci 265:227–233
Cairns SJ, Schwager SJ (1987) A comparison of association indices.
Anim Behav 35:1454–1469
Conner DA (1982) Dialects versus geographic variation in mammalian vocalization. Anim Behav 30:297–298
Crockford C, Herbinger I, Vigilant L, Boesch C (2004) Wild
chimpanzees produce group-specific calls: a case for vocal
learning? Ethology 110:221–243
Deecke VB, Ford JKB, Spong P (1999) Quantifying complex patterns
of bioacoustic variation: use of a neural network to compare killer
whale (Orcinus orca) dialects. J Acoust Soc Am 105:2499–2507
Deecke VB, Ford JKB, Spong P (2000) Dialect change in resident
killer whales (Orcinus orca): implications for vocal learning and
cultural transmission. Anim Behav 60:629–638
Enggist-Dueblin P, Pfister U (2002) Cultural transmission of vocalizations in ravens, Corvus corax. Anim Behav 64:831–841
Ford JKB (1987) A catalogue of underwater calls produced by killer
whales (Orcinus orca) in British Columbia. Canadian Department of Fisheries and Oceans, Nanaimo, p 161
Ford JKB (1989) Acoustic behaviour of resident killer whales
(Orcinus orca) off Vancouver Island, British Columbia. Can J
Zool 67:727–745
Ford JKB (1991) Vocal traditions among resident killer whales
(Orcinus orca) in coastal waters of British Columbia, Canada.
Can J Zool 69:1454–1483
Ford JKB, Ellis GM (2006) Selective foraging by fish-eating killer
whales Orcinus orca in British Columbia. Mar Ecol Prog Ser
Ford JKB, Ellis GM, Balcomb KC (2000) Killer whales—the natural
history and genealogy of Orcinus orca in British Columbia and
Washington State. University of British Columbia Press, Vancouver
Ginsberg JR, Young TP (1992) Measuring associations between
individuals or groups in behavioural studies. Anim Behav
Gouzoules H, Gouzoules S (1990) Matrilineal signatures in the
recruitment screams of pigtail macaques, Macaca nemestrina.
Behaviour 115:327–347
Hoelzel AR, Dover GA (1991) Genetic differentiation between
sympatric killer whale populations. Heredity 66:191–196
Janik VM, Slater PJB (1997) Vocal learning in mammals. Adv Stud
Behav 26:59–99
Miller PJO (2002) Mixed-directionality of killer whale stereotyped calls:
a direction of movement cue? Behav Ecol Sociobiol 52:262–270
Miller PJO, Bain DE (2000) Within-pod variation in killer whale calls.
Anim Behav 60:617–628
Nousek AE, Slater PJB, Wang C, Miller PJO (2006) The influence of
social affiliation on individual vocal signatures of northern
resident killer whales (Orcinus orca). Biol Lett 2:481–484
Queller DC, Goodnight KF (1989) Estimation of genetic relatedness
using genetic markers. Evolution 43:258–275
Rendell L, Whitehead H (2003) Vocal clans in sperm whales (Physeter
macrocephalus). Proc R Soc Lond, Ser B: Biol Sci 270:225–231
Schnell GD, Watt DJ, Douglas ME (1985) Statistical comparison of
proximity matrices: applications to animal behaviour. Anim
Behav 33:239–253
Trainer JM (1989) Cultural evolution in song dialects of yellowrumped caciques in Panama. Ethology 80:190–204
Watwood SL, Tyack PL, Wells RS (2004) Whistle sharing in paired
male bottlenose dolphins, Tursiops truncatus. Behav Ecol Sociobiol 55:531–543
Wright TF (1996) Regional dialects in the contact call of a parrot. Proc
R Soc Lond, Ser B: Biol Sci 263:867–872
Yurk H, Barrett-Lennard LG, Ford JKB, Matkin CO (2002) Cultural
transmission within maternal lineages: vocal clans in resident
killer whales in Southern Alaska. Anim Behav 63:1103–1119
Zar JC (1996) Biostatistical analysis. Prentice Hall, Upper Saddle