2.4. Matching songs within and between populations
To investigate song sharing, we calculated two similarity indices: (1) the LSI and (2) DSI (see the following paragraph). Once theme assignments were verified, the LSI calculation was performed at the song level as the sequence of themes that made up each song type (table 1). For the LSI, phrase repetitions were omitted to extract the order of all themes per singer, regardless of the length of song. We made no attempt to divide the sequence of themes into individual songs for the analysis. The LSI was calculated for all theme sequences (n = 19) and the resulting similarity matrix was average-linkage clustered (and CCC calculated as above) and bootstrapped 1000 times (with multiscale bootstrap resampling (AU) and normal bootstrap probability (BP)) using the pvclust package [81] in R. Bootstrapping produced p-values for each split in the tree, which are regarded as significant if p > 95% for AU [81] and considered significant if p > 70% for BP [40] to ensure the tree was robust and stable.
As an alternative analysis to LSI, DSI was calculated based on the presence and sharing of themes (phrase types) among all singers. This analysis does not rely on any sequence information in the song, simply the presence and sharing of themes [82]. DSI is calculated as the number of shared themes divided by the sum of the total number of themes present in the song of singer 1 and singer 2 [82] (see electronic supplementary information, S1). DSI calculation was performed in R using custom-written code (available at https://github.com/ellengarland/dice_si). The similarity matrix was clustered (average-linkage) and bootstrapped 1000 times (as for the LSI), and the CCC calculated to ensure the resultant tree structure represented the connections in the data.
3. Results
Four song types (labelled 1–4) identified from 19 singers from French Polynesia (n = 9) and Ecuador (n = 10) were composed of 41 themes (table 1; electronic supplementary material, tables S1 and S3). Song type 1 (coloured blue; electronic supplementary material, figures S2–S4 and S7) was identified in French Polynesia in 2016 (3/3 singers) and was also present in French Polynesia in 2017 (1/3 singers) and 2018 (2/3 singers, both hybrid (singers 7 and 8)), as well as in Ecuador in 2018 (3/3 singers). Song type 2 (green; electronic supplementary material, figures S3 and S4) was described in French Polynesia in 2017 (2/3 singers) and 2018 (1/3 singers, part of hybrid singer 8). Song type 3 (orange; electronic supplementary material, figure S4) was sung by one singer (singer 9) in French Polynesia in 2018. Finally, song type 4 (grey; electronic supplementary material, figures S5 and S6) was identified in Ecuador in 2016 (3/3 singers) and 2017 (4/4 singers). Additionally, two singers in French Polynesia in 2018 (singers 7 and 8) sang hybrid songs combining themes from song types 1 (blue) and 3 (orange), and 1 (blue) and 2 (green), respectively (table 1; see electronic supplementary material, S1, for further details).
3.1. Song similarity in the central and eastern South Pacific
The LSI from hierarchically clustered and bootstrapped theme sequences for each singer revealed three major groupings (figure 2b, CCC = 0.962). Two initial branches were produced: one included song types 1 (blue), 2 (green) and 3 (orange) and the other song type 4 (grey). Song type 1 (blue) included all singers from French Polynesia 2016, two (of three) from 2017 and all Ecuador 2018 singers. Song types 2 (green; one singer from French Polynesia 2018) and 3 (orange; two singers from French Polynesia 2017) branched off the song type 1 (blue) cluster at a high level. Song type 4 (grey) included all Ecuador 2016 and 2017 singers and was entirely separated from the song type 1 (blue) cluster, which included the 2018 Ecuador singers. Both hybrid songs (singers 7 and 8) were grouped within song type 1 (blue; figure 2b). Singer 7 was placed on a separate branch to the rest of song type 1 (blue), while singer 8 was nested within the branch based on similarity in theme sequences (figure 2b).
DSI based on the presence and sharing of themes regardless of their theme sequence produced a similar grouping to LSI (hierarchically clustered and bootstrapped; figure 2c, CCC = 0.966). Song types 1 (blue) and 4 (grey) clustered in a similar way to LSI, but with additional confidence in placement of singers within the dendrogram (based on AU and BP p-values). However, hybrid singer 7 (which combined themes from song types 1 and 3; blue/orange) was grouped with song type 3 (orange) singer 9, due to sharing theme 22 regardless of the sequence order of themes. Finally, regardless of the method and fine-scale placement of hybrid singers, all analyses indicated that there were four song types present across the central and eastern South Pacific over the course of the study (figure 3).
4. Discussion
Humpback whale song continues to spread east from French Polynesia in the central South Pacific to the Ecuadorian breeding ground in the eastern South Pacific. Three themes from song type 1 (blue) were documented in both French Polynesia in 2016, 2017 and 2018 as well as in Ecuador in 2018, confirming the hypothesis of delayed eastward transmission of song. Thus, the unidirectional song transmission in the South Pacific not only extends from Australia to French Polynesia [41,56], but a further 8000 km distance to Ecuador. This finding extends the geographical bounds of the horizontal cultural transmission of Southern Hemisphere humpback whale song and demonstrates vocal connectivity among populations that are 14 000 km apart (i.e. the distance from eastern Australia to Ecuador). It reiterates that the patterns of migration are written into the whales' song [58], and the potential for a full circumpolar transmission of song is highly plausible [12,70].
Recent agent-based cultural evolution models of global song transmission suggest the unidirectional song transmission is driven by differences in population sizes as songs spread from large to small populations [70]. We speculate that this may result in song transmission in different directions in other ocean basins dependent on population sizes (e.g. this allows the potential for songs to spread west across the South Indian Ocean from the large western Australian population). However, the global model also suggested that once a song revolution took hold, it continued to spread in a single direction [70]. We speculate this may allow a song to be passed in a stepwise fashion through all eleven Southern Hemisphere populations before returning to the origin population. The song would evolve during this time period (as is shown in the South Pacific [41]) and would be substantially different and thus considered a ‘new’ song by the time it transited the globe.
Song in both French Polynesia and Ecuador changed through progressive evolution and song revolutions, as is common for South Pacific breeding populations [41,56]. Progressive evolution occurs at a slower pace (over years) and changes are small, while song revolutions are identified when all themes of one song type are replaced by a novel one, and the origin of the novel song themes can be traced to a neighbouring breeding population [41,56]. Out of four song types identified in French Polynesia and Ecuador between 2016 and 2018, two song types resemble previously known song in western and central South Pacific breeding populations (electronic supplementary material, table S4). Song type 1 (blue) qualitatively resembled song themes previously recorded in French Polynesia, the Cook Islands and Tonga in 2015 (from [58]: figure 1 ‘light blue song’; electronic supplementary material, table S4). Five themes from the current study were visually and aurally similar to song themes identified in Owen et al. [58] (electronic supplementary material, table S4), with three themes present only in French Polynesia 2016–2018 (themes 3, 4 and 8), and two themes (themes 1 and 5) present in both French Polynesia 2016–2018 and Ecuador 2018 (table 1; electronic supplementary material, table S4). This suggests that humpback whales wintering in Ecuador are vocally connected with the western and central South Pacific breeding populations to at least Tonga. However, it is highly likely that the connection reaches back to the Australian breeding grounds based on previous studies that documented consistent song revolution events spreading from western to eastern Australia, onto New Caledonia, Tonga and finally to French Polynesia [40,41,57,58,64,67,68,72,82]. By contrast, the origin of song type 4 (grey) recorded in Ecuador in 2016 and 2017 preceding the song revolution in 2018 is unknown. The song themes did not resemble any of the themes described in French Polynesia or the wider South Pacific in previous studies. We hypothesize sporadic song sharing is occurring between the eastern and central South Pacific followed by periods of acoustic isolation and rapid evolution of the song making themes unrecognizable, and/or song sharing is occurring between the eastern South Pacific populations and Brazil (located on the east coast of South America) on a regular or sporadic basis.