Classifying sex and strain from mouse ultrasonic vocalizations using deep learning

Vocalizations are widely used for communication between animals. Mice use a large repertoire of ultrasonic vocalizations (USVs) in different social contexts. During social interaction recognizing the partner's sex is important, however, previous research remained inconclusive whether individual USVs contain this information. Using deep neural networks (DNNs) to classify the sex of the emitting mouse from the spectrogram we obtain unprecedented performance (77%, vs. SVM: 56%, Regression: 51%). Performance was even higher (85%) if the DNN could also use each mouse's individual properties during training, which may, however, be of limited practical value. Splitting estimation into two DNNs and using 24 extracted features per USV, spectrogram-to-features and features-to-sex (60%) failed to reach single-step performance. Extending the features by each USVs spectral line, frequency and time marginal in a semi-convolutional DNN resulted in a performance mid-way (64%). Analyzing the network structure suggests an increase in sparsity of activation and correlation with sex, specifically in the fully-connected layers. A detailed analysis of the USV structure, reveals a subset of male vocalizations characterized by a few acoustic features, while the majority of sex differences appear to rely on a complex combination of many features. The same network architecture was also able to achieve above-chance classification for cortexless mice, which were considered indistinguishable before. In summary, spectrotemporal differences between male and female USVs allow at least their partial classification, which enables sexual recognition between mice and automated attribution of USVs during analysis of social interactions.

Sexual dimorphism of AMBRA1-related autistic features in human and mouse

Ambra1 is linked to autophagy and neurodevelopment. Heterozygous Ambra1 deficiency induces autism-like behavior in a sexually dimorphic manner. Extraordinarily, autistic features are seen in female mice only, combined with stronger Ambra1 protein reduction in brain compared to males. However, significance of AMBRA1 for autistic phenotypes in humans and, apart from behavior, for other autism-typical features, namely early brain enlargement or increased seizure propensity, has remained unexplored. Here we show in two independent human samples that a single normal AMBRA1 genotype, the intronic SNP rs3802890-AA, is associated with autistic features in women, who also display lower AMBRA1 mRNA expression in peripheral blood mononuclear cells relative to female GG carriers. Located within a non-coding RNA, likely relevant for mRNA and protein interaction, rs3802890 (A versus G allele) may affect its stability through modification of folding, as predicted by in silico analysis. Searching for further autism-relevant characteristics in Ambra1^+/- mice, we observe reduced interest of female but not male mutants regarding pheromone signals of the respective other gender in the social intellicage set-up. Moreover, altered pentylentetrazol-induced seizure propensity, an in vivo readout of neuronal excitation-inhibition dysbalance, becomes obvious exclusively in female mutants. Magnetic resonance imaging reveals mild prepubertal brain enlargement in both genders, uncoupling enhanced brain dimensions from the primarily female expression of all other autistic phenotypes investigated here. These data support a role of AMBRA1/Ambra1 partial loss-of-function genotypes for female autistic traits. Moreover, they suggest Ambra1 heterozygous mice as a novel multifaceted and construct-valid genetic mouse model for female autism.

A humanized version of Foxp2 does not affect ultrasonic vocalization in adult mice

The transcription factor FOXP2 has been linked to severe speech and language impairments in humans. An analysis of the evolution of the FOXP2 gene has identified two amino acid substitutions that became fixed after the split of the human and chimpanzee lineages. Studying the functional consequences of these two substitutions in the endogenous Foxp2 gene of mice showed alterations in dopamine levels, striatal synaptic plasticity, neuronal morphology and cortico-striatal-dependent learning. In addition, ultrasonic vocalizations (USVs) of pups had a significantly lower average pitch than control littermates. To which degree adult USVs would be affected in mice carrying the 'humanized' Foxp2 variant remained unclear. In this study, we analyzed USVs of 68 adult male mice uttered during repeated courtship encounters with different females. Mice carrying the Foxp2(hum/hum) allele did not differ significantly in the number of call elements, their element structure or in their element composition from control littermates. We conclude that neither the structure nor the usage of USVs in adult mice is affected by the two amino acid substitutions that occurred in FOXP2 during human evolution. The reported effect for pup vocalization thus appears to be transient. These results are in line with accumulating evidence that mouse USVs are hardly influenced by vocal learning. Hence, the function and evolution of genes that are necessary, but not sufficient for vocal learning in humans, must be either studied at a different phenotypic level in mice or in other organisms.

A single gene defect causing claustrophobia

Claustrophobia, the well-known fear of being trapped in narrow/closed spaces, is often considered a conditioned response to traumatic experience. Surprisingly, we found that mutations affecting a single gene, encoding a stress-regulated neuronal protein, can cause claustrophobia. Gpm6a-deficient mice develop normally and lack obvious behavioral abnormalities. However, when mildly stressed by single-housing, these mice develop a striking claustrophobia-like phenotype, which is not inducible in wild-type controls, even by severe stress. The human GPM6A gene is located on chromosome 4q32-q34, a region linked to panic disorder. Sequence analysis of 115 claustrophobic and non-claustrophobic subjects identified nine variants in the noncoding region of the gene that are more frequent in affected individuals (P=0.028). One variant in the 3'untranslated region was linked to claustrophobia in two small pedigrees. This mutant mRNA is functional but cannot be silenced by neuronal miR124 derived itself from a stress-regulated transcript. We suggest that loosing dynamic regulation of neuronal GPM6A expression poses a genetic risk for claustrophobia.

Ultrasonic vocalizations in mouse models for speech and socio-cognitive disorders: insights into the evolution of vocal communication

Comparative analyses used to reconstruct the evolution of traits associated with the human language faculty, including its socio-cognitive underpinnings, highlight the importance of evolutionary constraints limiting vocal learning in non-human primates. After a brief overview of this field of research and the neural basis of primate vocalizations, we review studies that have addressed the genetic basis of usage and structure of ultrasonic communication in mice, with a focus on the gene FOXP2 involved in specific language impairments and neuroligin genes (NL-3 and NL-4) involved in autism spectrum disorders. Knockout of FoxP2 leads to reduced vocal behavior and eventually premature death. Introducing the human variant of FoxP2 protein into mice, in contrast, results in shifts in frequency and modulation of pup ultrasonic vocalizations. Knockout of NL-3 and NL-4 in mice diminishes social behavior and vocalizations. Although such studies may provide insights into the molecular and neural basis of social and communicative behavior, the structure of mouse vocalizations is largely innate, limiting the suitability of the mouse model to study human speech, a learned mode of production. Although knockout or replacement of single genes has perceptible effects on behavior, these genes are part of larger networks whose functions remain poorly understood. In humans, for instance, deficiencies in NL-4 can lead to a broad spectrum of disorders, suggesting that further factors (experiential and/or genetic) contribute to the variation in clinical symptoms. The precise nature as well as the interaction of these factors is yet to be determined.

Female mice respond to male ultrasonic 'songs' with approach behaviour

The ultrasonic vocalizations of mice are attracting increasing attention, because they have been recognized as an informative readout in genetically modified strains. In addition, the observation that male mice produce elaborate sequences of ultrasonic vocalizations (‘song’) when exposed to female mice or their scents has sparked a debate as to whether these sounds are—in terms of their structure and function—analogous to bird song. We conducted playback experiments with cycling female mice to explore the function of male mouse songs. Using a place preference design, we show that these vocalizations elicited approach behaviour in females. In contrast, the playback of pup isolation calls or whistle-like artificial control sounds did not evoke approach responses. Surprisingly, the females also did not respond to pup isolation calls. In addition, female responses did not vary in relation to reproductive cycle, i.e. whether they were in oestrus or not. Furthermore, our data revealed a rapid habituation of subjects to the experimental situation, which stands in stark contrast to other species' responses to courtship vocalizations. Nevertheless, our results clearly demonstrate that male mouse songs elicit females' interest.

Neuroligin-3-deficient mice: model of a monogenic heritable form of autism with an olfactory deficit

Autism spectrum disorder (ASD) is a frequent neurodevelopmental disorder characterized by variable clinical severity. Core symptoms are qualitatively impaired communication and social behavior, highly restricted interests and repetitive behaviors. Although recent work on genetic mutations in ASD has shed light on the pathophysiology of the disease, classifying it essentially as a synaptopathy, no treatments are available to date. To develop and test novel ASD treatment approaches, validated and informative animal models are required. Of particular interest, in this context are loss-of-function mutations in the postsynaptic cell adhesion protein neuroligin-4 and point mutations in its homologue neuroligin-3 (NL-3) that were found to cause certain forms of monogenic heritable ASD in humans. Here, we show that NL-3-deficient mice display a behavioral phenotype reminiscent of the lead symptoms of ASD: reduced ultrasound vocalization and a lack of social novelty preference. The latter may be related to an olfactory deficiency observed in the NL-3 mutants. Interestingly, such olfactory phenotype is also present in a subgroup of human ASD patients. Tests for learning and memory showed no gross abnormalities in NL-3 mutants. Also, no alterations were found in time spent in social interaction, prepulse inhibition, seizure propensity and sucrose preference. As often seen in adult ASD patients, total brain volume of NL-3 mutant mice was slightly reduced as assessed by magnetic resonance imaging (MRI). Our findings show that the NL-3 knockout mouse represents a useful animal model for understanding pathophysiological events in monogenic heritable ASD and for developing novel treatment strategies in this devastating human disorder.

Resistance against heterogeneous sequential infections: experimental studies with a tapeworm and its copepod host

Parasite heterogeneity is thought to be an important factor influencing the likelihood and the dynamics of infection. Previous studies have demonstrated that simultaneous exposure of hosts to a heterogeneous mixture of parasites might increase infection success. Here this view is extended towards the effect of parasite heterogeneity on subsequent infections. Using a system of the tapeworm Schistocephalus solidus and its copepod intermediate host, heterogeneity of the tapeworm surface carbohydrates is investigated, i.e. structures that are potentially recognized by the invertebrate host's immune system. With lectin labelling, a significant proportion of variation in surface carbohydrates is related to differences in worm sibships (i.e. families). Tapeworm sibships were used for experimental exposure of copepods to either homogeneous combinations of tapeworm larvae, i.e. worms derived from the same sibship or heterogeneous mixtures of larvae, and copepods were subsequently challenged with an unrelated larva to study re-infection. Contrary to expectation, neither an effect of parasite heterogeneity on the current infection, nor on re-infection were found. The effect of parasitic heterogeneity on host immunity is therefore complex, potentially involving increased cross-protection on the one hand, with higher costs of raising a more heterogeneous immune response on the other.

The harmonic-to-noise ratio applied to dog barks

Dog barks are typically a mixture of regular components and irregular (noisy) components. The regular part of the signal is given by a series of harmonics and is most probably due to regular vibrations of the vocal folds, whereas noise refers to any nonharmonic (irregular) energy in the spectrum of the bark signal. The noise components might be due to chaotic vibrations of the vocal-fold tissue or due to turbulence of the air. The ratio of harmonic to nonharmonic energy in dog barks is quantified by applying the harmonics-to-noise ratio (HNR). Barks of a single dog breed were recorded in the same behavioral context. Two groups of dogs were considered: a group of ten healthy dogs (the normal sample), and a group of ten unhealthy dogs, i.e., dogs treated in a veterinary clinic (the clinic sample). Although the unhealthy dogs had no voice disease, differences in emotion or pain or impacts of surgery might have influenced their barks. The barks of the dogs were recorded for a period of 6 months. The HNR computation is based on the Fourier spectrum of a 50-ms section from the middle of the bark. A 10-point moving average curve of the spectrum on a logarithmic scale is considered as estimator of the noise level in the bark, and the maximum difference of the original spectrum and the moving average is defined as the HNR measure. It is shown that a reasonable ranking of the voices is achievable based on the measurement of the HNR. The HNR-based classification is found to be consistent with perceptual evaluation of the barks. In addition, a multiparametric approach confirms the classification based on the HNR. Hence, it may be concluded that the HNR might be useful as a novel parameter in bioacoustics for quantifying the noise within a signal.

Local variation in Barbary macaque shrill barks

We examined vocalizations of Barbary macaques, Macaca sylvanus, given in response to a dog, in two populations, at Rocamadour, France, and Salem, Germany. Calls were recorded from 16 individuals in Rocamadour and 23 individuals in Salem. Despite an overall similarity, an acoustic analysis revealed significant differences in the call structure between populations. To test the perceptual salience of these acoustic differences, we conducted playback experiments in both populations in which calls from the own or the other population were broadcast. The overall response pattern did not differ significantly between the populations with regard to the origin of the call. However, subjects responded slightly, but significantly longer after playback of calls from the other group. Although call function apparently determined the general response of subjects, they none the less discriminated between calls from different origins. These results suggest a small but possibly meaningful plasticity in call production. Copyright 1998 The Association for the Study of Animal Behaviour.