We recorded pup isolation calls from Gnptab^−/− on P5 and P8 and compared them to wild type littermates Gnptab^+/+ (Figures 1A,B). Gnptab^−/− vocalized were found to emit pup isolation calls. Their calls included bouts of vocalization as well as high, low and pitch jump calls. No differences were found in either the number of calls (P5; t-test; t = 1.154, p = 0.254, WT, n = 40; KO, n = 13, P8; t-test; t = 0.0004, p = 0.997, WT, n = 41; KO n = 15 Figure 1C), or the number of clicks (short broadband noise pulses) (P5; t-test; t = 0.820, p = 0.416, P8; t-test; t = 1.065, p = 0.291) emitted by mice with the two different genotypes. Further analysis revealed a significant difference in the mean duration of the calls emitted by the Gnptab^−/− pups (0.046 ± 0.002 s) compared to the Gnptab^+/+ pups (0.052 ± 0.002 s) at P8 (t-test; t = 2.332, p = 0.024; ~88% of wild type duration, Figure 2A). This is analysis was performed by taking the mean duration of all calls for each animal and comparing them across groups. A more sensitive measurement, wherein each call duration is averaged over the genotype instead of over the animal (which increases n considerably and can therefore pick up smaller differences) shows that this difference is present in P5 recordings as well (Figure 2B).

Next the pauses in the pup vocalizations were analyzed. Not unlike human vocalizations where there are time periods of quietness between sentences or thoughts and shorter brief pauses between words or phrases, pup song contains long pauses, called inter-bout pauses interspersed with bouts of calls. This can be seen in the example sonogram in Figure 1B and in the long tail of the histogram showing pause durations (Supplemental Figures 1A–C). There are also brief pauses, termed intra-bout pauses, between calls during bouts.

We first looked at all pauses (both inter-bout and intra-bout pauses). We found no significant difference in the mean pause duration at P5 (WT = 0.806 ± 0.135 s, KO = 1.526 ± 0.566 s, t-test; t = 1.765, p = 0.084) and P8 (WT = 0.615 ± 0.061 s, KO = 0.739 ± 0.166 s, t-test; t = 0.854, p = 0.397, Figure 3A). We next separately analyzed inter-bout pause durations and intra-bout pause durations. Inter-bout pause durations were not significantly different on P5 (WT = 2.61 ± 0.228 s, KO = 3.78 ± 1.37 s, t-test; t = 1.308, p = 0.197) or P8 (WT = 2.31 ± 0.169 s, KO = 1.96 ± 0.360 s, t-test; t = 0.974, p = 0.334, Figure 3B). Analyzing the intra-bout pause durations (inter- call interval in a series of calls), Gnptab^−/− mice showed a significantly increased duration of intra-bout pause compared to Gnptab^+/+ on P5 (WT = 0.157 ± 0.003 s, KO = 0.170 ± 0.005 s, t-test; t = 2.442, p = 0.018) and P8 (WT = 0.133 ± 0.002 s, KO = 0.144 ± 0.003 s, t-test; t = 2.841, p = 0.006, Figure 3C). The proportion of large pauses in the Gnptab^−/− was increased for a subset of pause lengths compared to Gnptab^+/+. This can be seen in Figure 3D where the criterion for large pauses steadily increases along the x-axis.

We next examined the power spectrum of the vocalizations to see if there was any systematic difference in the pitch or loudness of the calls made by each genotype that could explain some of the findings above. The total power summed over all frequency bands was not significantly different between groups on either day of recording (t-test; P5, t = 1.809, p = 0.0764, P8, t = 1.063, p = 0.2923; Figure 4A). The power spectrum of the pup isolation calls is shown in Figure 4B. When broken down by frequencies, we found a significant decrease in the Gnptab^−/− compared to wildtype for the power in all frequency bands except 65–85 KHz range on P5. On P8, we found significant decreases in Gnptab^−/− compared to wildtype in the 35–65 KHz range and in the 80–100 KHz range.

We also tested whether the alterations in Gnptab^−/− vocalizations might be might be attributed to changes in overall physical health. Gnptab^−/− pups gross appearance was normal (Figure 5A). However, P5 Gnptab^−/− animals showed a tendency toward slightly reduced body weight (WT = 3.25 ± 0.09 grams, KO = 2.97 ± 0.10 grams; t-test; t = 1.734, p = 0.09). By P8, this difference was significant (WT = 4.34 ± 0.11 grams, KO, 3.85 ± 0.14 grams; t-test; t = 2.554, p = 0.015, Figure 5B). This is consistent with previous reports (Gelfman et al., 2007; Idol et al., 2014). Due to this significant difference in weight between the Gnptab^+/+ and Gnptab^−/− pups on P8, we performed a regression analysis of Gnptab^+/+ pups for each of our parameters to determine if they directly correlate with the weight of the animal (Table 1).

We found that for intra-bout pause duration (P5), number of bouts (P5), percent of isolated calls (P8), and mean calls per bout (P8), there was a significant correlation. Note that there was also a trend in the mean calls per bout on P5 making this variable the most likely to be explained by weight differences. The explanatory power if each variable can be seen in Table 1.

Abrupt pitch jumps are a naturally occurring feature of mouse vocalizations as can be seen in Figure 1B (Liu et al., 2003; Holy and Guo, 2005). The incidence of pitch jumps were significantly reduced in the Gnptab^−/− pup calls compared to the Gnptab^+/+ pup calls on both P5 and P8 over a wide range of criteria used to define the minimum size pitch jumps for both positive and negative jumps (Figures 6A,B; Supplemental Table 2). Likewise the magnitude of the largest pitch jump in each call was also decreased in the Gnptab^−/− pup calls compared to Gnptab^+/+ pups on P5 (WT = 11485 ± 550 Hz, KO = 8947 ± 1141 Hz; t-test; t = 2.139, p = 0.0373), and P8 (WT = 12689 ± 746 Hz, KO = 8687 ± 953 Hz, t-test; t = 2.886, p = 0.0056; Figure 6C).

The mean number of bouts per recording (P8; WT = 69.63 ± 3.45, KO = 98.93 ± 10.76, t-test; t = 3.326, p = 0.0016) increased significantly in the Gnptab^−/− pups compared to the Gnptab^+/+ pups by P8 (Figure 7A). This was at least in part due to a significant increase in the percent of isolated calls (i.e., calls surrounded by inter-bout pauses) per recording (P8; WT = 7.8% ± 0.7%, KO = 12.9% ± 1.5%, t-test t = 3.475, p = 0.001; Figure 7B). The tendency of the vocalization data recorded on P5 was similar to that of P8 for the mean number of bouts per recording (P5; WT = 59.83 ± 3.50, KO = 75.31 ± 11.20, t-test; t = 1.714, p = 0.092), though not significant. On P5, the percentage of isolated calls differed (P5; WT = 10.6% ± 1.7%, KO = 17.9% ± 3.1%, t-test; t = 2.051, p 0.045). Relatedly, on both P5 and P8 of recording, the Gnptab^−/− pups had significantly fewer calls per bout when compared to the Gnptab^+/+ pups (P5, WT = 6.04 ± 0.449, KO 3.69 ± 0.45; t-test; t = 2.776 p = 0.007; P8, WT = 5.28 ± 0.243, KO = 3.66 ± 0.26; t-test; t = 3.678, p = 0.0005); Figure 7C).

Next, calls were classified based on a previously published classification (Liu et al., 2003; Barnes et al., 2016) (Figure 8A). The breakdown of call type is shown in Figures 8B,C. Entropy of call type usage was calculated from the proportion of different call types (see Methods). Gnptab^−/− call type entropy was significantly decreased in both P5 (P5; WT = 1.915 ± 0.044, KO = 1.692 ± 0.062 t-test; t = 2.593, p = 0.012); and P8 (P8; WT = 2.004 ± 0.053, KO = 1.742 ± 0.098 t-test; t = 2.443 p = 0.018 Figure 8D). We also looked at the diversity of call type sequences by analyzing the entropy in the first-order Markov process. We found this too to be significantly decreased in the Gnptab^−/− mice on both days (P5; WT = 1.480 ± 0.057, KO = 1.202 ± 0.084 t-test; t = 2.459, p = 0.017; P8, WT = 1.582 ± 0.052, KO = 1.321 ± 0.085; t-test; t = 2.573, p = 0.013); Figure 8E).

So far, group differences have been identified in the different genotypes. Next we asked whether using these features we can identify a particular recording as coming from either a Gnptab^+/+ or Gnptab^−/− mouse. We developed an automated classification scheme using the following variables: normalized intra-bout pause duration, largest pitch jump per call, bouts per recording, calls per bout, and percent isolated calls. The number of animals in each data set was first matched. Next, the data was projected into 5 dimensional space. Individual recordings were then classified as Gnptab^−/− or Gnptab^+/+ based on the majority identity of its 3 nearest neighbors, leaving the recording itself out of the comparison set. We found that a larger number of recordings could be correctly classified compared to chance on P8 (80%; c.i., 67–90%). Results were similar for classifications based on 1, 3, and 5 nearest neighbors.

We examined the features of the heterozygous pup calls (Gnptab^+/−). In all of the features we analyzed Gnptab^+/− were not significantly different from that of wild type Gnptab^+/+ mice. Likewise, features that were significantly different between Gnptab^+/+ mice and Gnptab^−/− homozygous knockout mice, were also significantly different between Gnptab^+/− heterozygous mice and Gnptab^−/− homozygous mice in all but two cases (call duration on P8 and percent isolated syllables on P5, Supplemental Table 1, Figure 9).

We examined the isolation calls of Gnptab^−/− pups and compared them with the isolation calls of their wild type littermates, Gnptab^+/+. Gnptab^−/− mice cannot make the first enzyme in the pathway that targets lysosomal enzymes to their lysosomes. Instead, in these mice, lysosomal enzymes end up in the extracellular space (Gelfman et al., 2007; Idol et al., 2014; Paton et al., 2014).

In adult Gnptab^−/− mice, many deficits— physical, histological, and behavioral—have been documented (Gelfman et al., 2007; Idol et al., 2014; Paton et al., 2014). Fewer abnormalities have been identified for younger mice. Physically, studies have found Gnptab^−/− mice to have a normal weight at 2 weeks and 4 weeks. By 6 weeks (Gelfman et al., 2007)—8 weeks (Paton et al., 2014) Gnptab^−/− showed a significant decrease in their weight compared to Gnptab^+/+ mice. Histological abnormalities have been found at 1 month of age including Gnptab^−/− mice having an increase in the number of cells undergoing apoptosis (Gelfman et al., 2007). Behaviorally, 1 month Gnptab^−/− mice preform normally on a 1 h locomotor test but show deficiencies on 3/7 locomotor assays (Idol et al., 2014). Here we find abnormalities both in weight and vocalizations as early as P5. As early as P5, we found a trend (p = 0.09) for Gnptab^−/− pups to weigh less. By P8 this difference was significant. The difference between the weights found in previous studies and our study, could be due to variety of factors including statistical power (Gelfman et al., 2007) or background strain differences (Paton et al., 2014).

Behaviorally, we found that the vocalizations of Gnptab^−/− mouse pups were grossly normal. Gnptab^−/− pups had similar call rates (both in the number of calls and number of clicks per recording) and call volumes. Gnptab^−/− pups were able to make both high and low frequency calls. It appears that the vocal tract is intact capable of making the same repertoire of vocalizations.

Many features of the Gnptab^−/− isolation calls were distinct from Gnptab^+/+. Duration of the mouse pup isolation calls in the Gnptab^+/+, and Gnptab^−/− animals differed significantly. The intra-bout pause duration was increased in the Gnptab^−/− animals though this could be due to the weight differences between the two groups as this measure was always significantly correlated with the weight of the pup. The difference in weight of Gnptab^−/− may be causing the difference in USVs, or the mutation may independently cause the difference in weight and difference in USVs.

Pitch jumps sizes were markedly different between genotypes, with Gnptab^−/− pups showing significantly smaller and fewer abrupt pitch jumps. Overall, the Gnptab^−/− pups were not quieter than their wild-type litter mates, but they did have less power in the higher frequencies. This fits well with having smaller and fewer pitch jumps. Though there we no difference in the number of calls emitted during the recordings, the number of bouts were greater in the Gnptab^−/− pups compared to Gnptab^+/+ pups. This was in part due to the increase in isolated calls observed in the Gnptab^−/− recordings. Gnptab^−/− pups were physically able and often did string many calls together, but their vocalizations more often included single calls. Entropy both for the proportion of different call types and the entropy of the temporal sequencing were decreased in the Gnptab^−/− mouse pup calls. This may be in part due to the decrease in pitch jumps, a defining feature of many call types.

Heterozygous pups Gnptab^+/− were not significantly different, in any of the features examined, from wild-type littermates Gnptab^+/+. This fits well with human findings, as humans with one copy of the GNPTAB gene missing are generally phenotypically normal.

By P8, the Gnptab^−/− pups showed many abnormalities compared to the Gnptab^+/+ pups. In the cases where recording from P5 pups did not show a statistically significant difference between Gnptab^+/+ and Gnptab^−/− animals, the recordings often exhibited the same general trend, suggesting the differences were growing as time passed. It is possible that some of the difference found in the Gnptab^−/− pups was due to a maturation delay, rather than a long term deficiency. Indeed the mean call length on the Gnptab^−/− pups calls on P8 (0.046 s) was similar to that of the Gnptab^+/+ pup calls on P5 (0.044 s). A more longitudinal study might be able to determine if juvenile and adult vocalizations are similarly affected.

The speech abnormalities in patients with Mucolipidosis II/III have not been well-quantified. Otomo et al found that, out of 13 patients diagnosed with Mucolipidosis II, 12 were unable to utter single words (Otomo et al., 2009).

These data can be compared to data recently published on pup isolation calls in the Gnptab^mut/mut mice (Barnes et al., 2016). These mice were engineered to carry a homozygous Glu1179Lys mutation in Gnptab homologous to the Glu1200Lys mutation in human GNPTAB well characterized for its role in stuttering (Kang et al., 2010; Fedyna et al., 2011). Gnptab^−/− pups showed more varied abnormalities and in features of their calls than the Gnptab^mut/mut mice. Gnptab^mut/mut showed a difference in the number of vocalizations per unit time, while the Gnptab^−/− were normal for this measure.

Both modifications had effects on the temporal structure of the calls as well. They caused a decrease in first order entropy of the call sequences. Pauses between calls were also affected in both cases. In the knockout, the intra-bout pause duration was affected while the missense mutation increased the occurrences of longer pauses. Both modifications resulted in an increase in the number of isolated syllables. The pathways by which these modifications can affect such specific features of mouse vocalizations are unknown, as is the reason why the knockout fails to affect the number of calls and other features perturbed by the missense mutation.

These data show that mutations in the lysosomal enzyme targeting pathway affect mouse pup isolation calls. Persons who suffer from Mucolipidosis II also have speech abnormalities, suggesting that in regards to vocalizations, this mutation affects the vocalizations of both species from a very young age.

All analyses were done using in-house MATLAB programs, some of which are available online at http://holylab.wustl.edu/. Waveforms were pre-processed, band-pass filtered (25–110 kHz), and calls identified using mean frequency, “spectral purity” (fraction of total power concentrated into a single frequency bin), and the “spectral discontinuity” (the change in the allocation of power across frequencies between two adjacent time bins) (Barnes et al., 2016). Stored acoustical waveforms were processed using MATLAB to compute the sonogram (512 samples/block, half-overlap, resulting in a time resolution of 1.02 ms and a frequency resolution of 0.98 kHz). Clicks were defined as milliseconds where fewer than 200 of the 512 samples were empty.

Analysis code implemented a fully-automated algorithm and was therefore blind to genotype. To calculate the number of calls and the duration of calls and pauses, each vocalization or pause contributed to the mean for each animal or subject; each individual's mean was then averaged to obtain the group average. A t-test was then performed to compare groups with each individual's mean. All t-tests were two-tailed. An alternative analysis, where each call contributes to the overall mean was also performed and produced comparable results, as shown.

For the call analysis, the definitions set forth in Arriaga et al. were used, with the exception of classes “i, j, k” these call types were grouped into the category “other” (Liu et al., 2003; Barnes et al., 2016). This classification was done using a fully-automated algorithm.

Bout-level analyses defined bouts based on histograms of pause lengths for all groups of mice in each day we recorded. Histograms were constructed with a range of bin sizes (50 to 300 ms). The middle of minimum bin in the range of 0.15–0.32 s was averaged across all bin sizes to determine the criteria for an inter-bout pause. The resulting minimum intra-bout/inter-bout cutoff was determined to be 0.273 s in P5 and 0.239 s in P8 recordings.

Entropy of call usage was calculated from the proportion of different call types; entropy for the temporal sequence (modeled as a first-order Markov process) was given by H2 = −Σp(X) Σp (X|Y) log2p (X|Y) with X and Y being each call type (Ey et al., 2013).

For the classification of genotype analysis, data were first normalized. A random sample of Gnptab^+/+ recordings were chosen to match the n of the Gnptab^−/− recordings. We then categorized each recording as being either from either a Gnptab^+/+, mouse or Gnptab^−/− mouse based on nearest neighbors. Choices of 1, 3, and 5 closest neighbors all yielded very similar results. This analysis was performed 10,000 times each time selecting a random sample of Gnptab^+/+ recordings.

To calculate the power spectrum, all time periods during a call were grouped and then the power spectrum was calculated per subject. The subjects' averages were then bootstrapped 1000 times to determine the 95% confidence intervals. We reported as significant, significant bins adjacent to at least one other significant bin.

For each call, the number of pitch jumps was determined. Jumps were defined as abrupt changes in pitch exceeding a minimum threshold. We explored thresholds of 500, 1000, 3000, 5000, 10,000, 15,000, and 20,000 Hz, and found that the significance of the results was not dependent upon the particular choice of threshold. Pitch was defined as the dominant frequency as a function of time, discarding periods of dropouts (Barnes et al., 2016). P values for Figure 6 can be found in Supplemental Table 1. In a separate measurement, the size of the largest pitch jump in each call was determined and then averaged across calls for each animal. We also tested for effects of gender and found no significant effect of gender on phenotype for any of the features studied.

A regression analysis was done of wild-type pups for each of our parameters to determine if the parameters directly correlated with weight. The Gnptab^−/− and Gnptab^+/− pups were not included because the degree that each pup was affected could be correlated with both the weight of the animal and the severity of the deficiency. In fact several of the parameters for the knockout did correlate with body weight (Supplemental Figure 2).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.