This study reveals that the ultrasonic vocalizations of the mouse have the characteristics of song. Qualitatively, this is apparent directly from playback of pitch-shifted audio recordings; we have also provided quantitative evidence for the usage of distinct syllable types arranged in nonrandom, repeated temporal sequences. These songs satisfy Broughton's sensu stricto
definition of song [13
], as well as many aspects of his sensu strictissimo
(see ). While courtship songs are common among birds, insects, and frogs, song has only rarely been documented in mammals, and to our knowledge only in humans, whales, and bats [3
]. However, some rodent species display a variety of calls [26
] and at least one other, the rat Dactylomys dactylilnus,
utters long sequences of vocalizations that contain some syllabic diversity [27
]. More generally, a number of Central and South American rodent species display complex vocalization (L. H. Emmons, personal communication), but none has been characterized in detail. However, it seems likely that song is more widely distributed than we currently appreciate. While the neural and motor mechanisms used to produce song and other communication sounds vary across species, recent work has indicated some commonality at the molecular level: the Foxp2 transcription factor, expressed in the brain of zebra finches during vocal learning [28
], seems to be required both for mouse ultrasonic vocalization [29
] and normal human speech [30
Subjectively, mouse song has a diversity and complexity that exceeds that of most insect and amphibian advertisement songs, which often contain only a single syllable type [1
], perhaps modulated in amplitude and cadence [31
]. At the syllable level, diversity in mouse song comes in two forms, discrete and continuous. Discrete categories of syllables exist, as evidenced by the appearance of distinct clusters, by two criteria: in terms of the sequence of stereotyped frequency jumps (see ), and by a comparison of the pitch waveforms of individual syllables (see ). Within syllable types, there also exists considerable continuous variability (see and A). Because of our adoption of a strict quantitative classification of types, we have not used this continuous variability to define subtypes. This does not, however, argue that additional types are not present, merely that our analysis does not yet support further subdivision of types. Our quantitative classification scheme may be stricter than that employed in some analyses. A comparison of both subjective and quantitative classification has been carried out for the song of swamp sparrows: subjective methods [32
] were used to classify notes into either 96 subtypes (which they termed the “splitter's classification”), or into six major categories (termed the “lumper's classification”). A later quantitative analysis carried out by the same laboratory, using techniques related to those employed here, found that notes clustered in general agreement with the major categories identified in the “lumper's classification,” with no evidence for further subdivision [21
The richness and complexity of mouse song appear to approach that of many songbirds. For example, in the zebra finch, a widely used model organism for studying song production, individuals have a number (3–7) of syllable types [25
] similar to the number of common types we find in mice (). There are other species, for example, canaries, whose vocal repertoire would appear to exceed that of mice [34
]. Both mice (see ) and birds [25
] exhibit regular temporal structure in their songs, including the production of repeated themes with sharp transitions between syllable types. However, mice also exhibit more gradual changes in syllable structure (see ). Overall, the tendency to repeat a syllable, with sharp transitions between types, appears to be stronger in some birds [34
] and whales [3
] than in mice. However, in birds these sharp transitions are a feature of the adult “crystallized” song; juvenile or isolation-reared birds are more experimental and less predictable in terms of the temporal structure of their song [33
]. Indeed, our pitch-shifted recordings of mouse song sound similar to the early “plastic” song of species such as swamp sparrows (Audio S5
). For this reason, any comparison between birds and mice should consider the development of mouse song over the lifetime of the animal. Such a study has been undertaken for properties like mean pitch and cadence over the first 2 wk of life [12
], but is lacking for the more complex features that compose song.
Because mouse songs are ultrasonic and therefore inaudible to human ears, it is worth noting that laboratory domestication has probably not acted to preserve the full richness of mouse song through generations inbreeding. One study documented considerable variability in the amount of vocalization by different laboratory strains [36
]. In contrast, domesticated bird populations have been subject to song selection, and indeed sub-strains such as the Waterschlager canary have been bred for particular vocal qualities. It therefore seems possible that wild mice might exhibit considerably greater diversity and/or more complex structure in their songs. Future comparisons between the songs of mice and birds may benefit from using wild mice.
A final question is whether mice, like birds, learn their songs through experience. The fact that different males have characteristic syllable usage and temporal structure to their songs (see ) is evidence for individual variability in song. Directly testing the role of experience will require that the auditory environment during development be explicitly controlled.
In sum, we have demonstrated that the ultrasonic vocalizations of mice are songs, containing different syllable types sequenced in regular temporal patterns. Different individuals sing recognizably different songs. These results open new possibilities for molecular and physiological studies of the production and perception of song in a well-studied laboratory organism.