High-throughput analysis of vocalizations reveals sex-specific changes in Fmr1 mutant pups

Abstract

There have been several reports that individuals with Fragile X syndrome (FXS) and animal models of FXS have communication deficits. The present study utilized two different call classification taxonomies to examine the sex-specificity of ultrasonic vocalization (USV) production on postnatal day (PD8) in the FVB strain of Fmr1 knockout (KO). One classification protocol requires the investigator to score each call by hand, while the other protocol uses an automated algorithm. Results using hand-scoring protocol indicated that male Fmr1 KO mice exhibited longer calls (p = 0.03) than WTs on PD8. Male KOs also produced fewer complex, composite, downward, short and two-syllable call-types, as well as more frequency steps and chevron call-types. Female heterozygotes exhibited no significant changes in acoustic or temporal aspects of calls, yet showed significant changes in call-type production proportions across two different classification taxonomies (p < 0.001). They exhibited increased production of harmonic and frequency steps calls, as well as fewer chevron, downward and short calls. According to the second high-throughput analysis, female heterozygotes produced significantly fewer single-type and more multiple-type syllables, unlike male KOs that showed no changes in these aspects of syllable production. Finally, we correlated both scoring methods and found a high level of correlation between the two methods. These results contribute further knowledge of sex differences in USV calling behavior for Fmr1 heterozygote and KO mice, and provide a foundation for the use of high-throughput analysis of neonatal USVs.

INTRODUCTION

Fragile X syndrome (FXS) is a neurodevelopmental disorder, caused by a CGG expansion mutation in the FMR1 gene on the X chromosome. The prevalence of FXS shows a quantitative difference in rates across the two sexes (1 in 4000 males vs. 1 in 8000 females) ¹. Females with Fmr1 mutations are less likely to exhibit stereotypy and exhibit an increased likelihood of a comorbid schizophrenia diagnosis, while males are more likely to demonstrate severe intellectual disability ^1–3. These sex differences in the clinical phenotype are mirrored in the Fmr1 knockout (KO) mouse model [reviewed in ⁴] ^5–10. Given that the Fmr1 mutation is an X-linked mutation, females with the mutation are typically heterozygous and the functional copy of the gene is still expressed throughout the body. Thus, sex differences in the clinical phenotype are often attributed to gene dosage effects. However, when studies compare male and female mice with full loss of Fmr1, there are still sex differences in their behavioral phenotype, suggesting that gene dosage effects are not the only factor involved ^5,6,8,9. Lack of studies that comprehensively understand sex differences in the Fmr1 KO model can significantly limit reproducibility of the research and the generalizability for development of potential therapeutics¹¹.

One area that has received limited research in sex differences of Fmr1 mice are isolation-induced ultrasonic vocalizations (USVs). In pups, the peak number of USVs emitted is typically between postnatal day (PD) 7 and PD9. The calls serve as a signal to the mother that the pup is in distress, increasing the probability of maternal retrieval to the nest ^12–14. Given the important role of these calls in pup survival and development, it is critical to understand how vocalization production might be dysregulated in pups lacking Fmr1, and how sex may interact with these genotype effects to influence the presentation of this phenotype. A few studies have characterized this behavior during the early neonatal time period in the Fmr1 KO model, demonstrating significant sex-specific effects ^15,16. These prior studies demonstrate communication alterations in male and female Fmr1 mutants, therefore helping to elucidate the impact of sex.

The most common clinical form of FXS in females is a heterozygous genotype, since homozygosity for the Fmr1 mutation in females is extremely rare. However, only one other study has examined the potential sex-specificity of the USV phenotype in Fmr1 heterozygotes (HETs), reporting no difference in neonatal vocalization behavior between Fmr1 wildtype (WT) and HET pups on PD8 on a C57BL/6J background strain ¹⁰. It should be noted that significant differences exist between the Fmr1 KO phenotypes on the different background strains commercially available, like many transgenic models ^17,18. Even in non-transgenic animals, substrains of the same background strain (i.e. C57BL/6J and C57BL/6N) demonstrate significant differences on a battery of tests ¹⁹. As such, it is conceivable that the lack of findings in the previous paper may not generalize to the strain used in the present study (i.e. FVB). Thus, further investigation of USV behavior in the HET model among different background strains is warranted.

Similar to expanding studies to multiple strains, the use of automated methodology can improve precision of measurements and enhance reproducibility. Conducting vocalization analyses is time-consuming and often subject to software malfunction, both of which may contribute to the paucity of studies. Recent efforts have focused on automation of vocalization analysis, though these methods have yet to be applied in neonatal USV studies ^20,21. For other forms of behavioral analysis, large efforts have been made to automate data collection, such as video tracking in C. elegans models ²² or gesture tracking in mice ²³. This effort is not limited to preclinical models, as recent studies from the field of neurology have focused on automated algorithms for diagnosing dementia and detecting biomarkers of Alzheimer’s in clinical samples ^24,25. Thus, it is important to discern the potential utility of automated methods of data analysis in studies of pup vocalization production. In summary, the purpose of the present study was trifold: 1) to extend previous findings to heterozygous females, 2) to extend previous findings to another strain by characterizing vocalization behavior in female wildtype (WT) and Fmr1 HETs in the FVB.129 strain and 3) compare the detection rate of a high-throughput automated analysis method to previously described methods of hand-cleaning in the present model. To assess this, vocalizations were collected on postnatal day 8 in male Fmr1 KO and female Fmr1 HET pups and a variety of analyses were used to highlight the impact of loss of Fmr1 on communication behavior. Moreover, automated analyses were piloted and compared directly to determine the viability of this method in the present model.

MATERIALS AND METHODS

Animals

Male Fmr1 WT and female Fmr1 HET FVB.129P2-Pde6b+Tyrc-ch Fmr1tm1Cgr/J (Jackson Labs Stock No: 004624) mice were used as breeders to produce the following groups: male WT and male KO pups, female WT, and female HET pups. The final sample sizes were as follows: n_{male WT} = 17, n_{male KO} = 13, n_{female WT} = 12, n_{female HET} = 12. All pups were housed in individual cages with parents and littermates. The light cycle was kept on a 12 hr. light/dark cycle, and the colony room was kept at an ambient temperature of 22° C. Animals had ad libitum access to food and water. All procedures performed were in accordance with Baylor University Institutional Animal Care and Use Committee, as well as the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals. All testing was conducted during the light cycle, specifically between 8 am and 5 pm. On postnatal day 7 (PD7), two toes were clipped for identification and genotyping analysis. Pups were then returned to the home cage until testing on PD8.

Isolation-Induced Ultrasonic Vocalizations

To expand on our previous study and maximize the potential to detect subtle differences following loss of Fmr1, we recorded vocalizations on PD8. On PD8, pups were brought down from the colony room in their home cage and allowed to habituate to the testing room for 30 minutes prior to the testing period. At the time of testing, pups were separated from their parents into a clean housing pan with fresh bedding, warmed to ambient nesting temperature (approximately 35° C) using a heating pad. Vocalizations were recorded using a condenser microphone (CM16/CMPA, Avisoft Bioacoustics, Germany). This microphone was connected to a recording interface (UltraSoundGate 116Hb, Avisoft Bioacoustics) that records USVs on a continuous spectrum from 0 – 125 kHz. Vocalizations were recorded for 2 minutes for each pup. Each pup was chosen randomly, recorded for 2 minutes, weighed, marked and then returned to the warmed cage with their littermates. An experimenter remained in the room to monitor the gain and acquisition of the microphone. Pups were not away from the dam for longer than 30 minutes. After the last pup was tested, all littermates were returned to the home cage with the parents.

Avisoft Analysis

Following the conclusion of all testing, files were collected and Avisoft SASLab Pro software (Avisoft Bioacoustics, Germany) was used to convert all USV files (.wav) into spectrograms using a fast Fourier transformation procedure. The parameters were set as follows for all lab studies to maintain consistency between experiments: FFT length = 1024, time window overlap = 75% (100% Frame, Hamming window), time resolution = 1 ms. Sampling frequency was set at 22050. Files were cleaned by an experimenter blinded to treatment identity, and any extraneous noise was removed from the spectrogram. Additionally, call-types were manually identified using a previously described call-type taxonomy ²⁶. In addition to collecting information on call-type production, the following acoustic and temporal parameters were also measured: duration, peak frequency, peak amplitude and fundamental frequency across all calls using this analysis.

MATLAB Analysis

A recently published article in the Journal of Visualized Experiments (JOVE) described an automated technique using MathWork’s MATLAB software to score vocalizations from adult animals ²⁰. In order to determine whether the automated method would be suitable for neonatal vocalizations, we measured syllable production, as well as acoustic and temporal properties, then compared the results to the results obtained using previously described methodology. All WAV files were converted to sonograms using MATLAB syntax, freely available from: (http://jarvislab.net/research/mouse-vocal-communication/). Sonograms were then processed using the graphical user interface as described previously ²⁰. In the Sonogram Parameters section, we set Min Frequency to 15,000 Hz, Max Frequency to 125,000 Hz, the sampling frequency to 256 kHz, and the Threshold to 0.3. In the Whistle Options section, we set the Purity Threshold to 0.075, the Min Duration of the syllable to 3 ms, the Min Frequency sweep to 20,000 Hz, and the Filter Duration to 3 ms. For consistency, we also set the Min Note Duration to 3 ms, and the Min Note Count to 1. Following sonogram processing, density inter-syllable interval (ISI) was determined using the song-analysis guided analysis Excel file available at the same website. The previous work set it at 0.25, and for the purposes of this study, we set it at 0.25 as well ²⁰. The following quantitative parameters were also included for analyses: spectral purity and sequence length. The system also identified distinct syllable types according to a previously described taxonomy and quantified the relative proportion of each type ²⁷.

Statistical Analysis

All data were analyzed using GraphPad Software 7.0 (San Diego, CA) or IBM SPSS Statistics 23 (Aramonk, NY). Given that the males included only wildtype and knockout genotypes while the females included only wildtype and heterozygote genotypes, the effect of genotype was analyzed separately for each sex. Results were evaluated using a Student’s t-test (Genotype [WT, (HET or KO)]) on each of the dependent variables. For the Avisoft analysis, the dependent variables were as follows: number of vocalizations, total duration of calls, average duration of calls, average fundamental frequency, average peak amplitude and average peak frequency. Individual experimenters blinded to the condition of the animal recorded also labeled the type of calls present within the file using the Scattoni call-type taxonomy ²⁶. The relative proportion of different call-types were tested using a χ² analysis with Z-testing for column proportions. For the MATLAB analysis, the dependent variables were sequence length and spectral purity. For the quantitative analysis, the relative frequency of syllables were also assessed according to the Holy and Guo et al (2005) call-type taxonomy ²⁷. Similar to above, the relative proportion of these different syllable types were tested using a χ² analysis with Z-testing for column proportions. To determine the sensitivity of hand-scoring in Avisoft compared to automated processing in MATLAB, a Spearman’s correlation was performed. Results were described as small (.00 – 0.39), moderate (.40 - .59), or strong (.60 +) ²⁸. We selected the Spearman’s correlation because some of the variables included had non-normal distributions as assessed by the Shapiro Wilk test.

RESULTS

Fmr1 KO males, but not Fmr1 HET females, exhibit differences in quantitative aspects of vocalization behavior on PD8

Statistical analyses in males revealed one major difference in the quantitative aspects of calling behavior. Specifically, Fmr1 male KOs had significantly longer calls than their WT counterparts, t(26) = 2.25, p = 0.03 (Figure 1a. Fmr1 male KOs also exhibited marginally fewer calls than WTs, t(26) = 1.77, p = 0.09 (Figure 1b). Fmr1 male KOs did not differ from WTs on the following dependent variables: cumulative duration of calling behavior, t(26) = 0.96, p = 0.35; average fundamental frequency, t(26) = 0.81, p = 0.42; average peak amplitude, t(26) = 1.16, p = 0.26; average peak frequency, t(26) = 0.12, p = 0.91; spectral purity, t(26) = 1.64, p = 0.11; and sequence length, t(26) = 0.86, p = 0.40 (Figure 1c–h).

Figure 1. Fmr1 KO males, but not Fmr1 HET females, exhibit differences in quantitative aspects of vocalization behavior on PD8.

a. Male KOs produced marginally fewer calls than male WTs, while female HETs were not different from their WT counterparts. b. Male KOs produced significantly longer calls on average than their WT male counterparts, while female HETs were not different from WT females. c. No differences were detected in the cumulative duration of calls. d. No difference in fundamental frequency, or pitch, was detected between any groups. e. No difference in peak amplitude was detected between any groups. f. No difference between peak frequency was detected between any groups. g. No difference in sequence length was detected between any groups. h. No differences were detected in spectral purity between any groups. Data are expressed as mean ± SEM. * = P < 0.05. # = P < 0.10.

No differences between female Fmr1 HETs and female WTs were noted on any quantitative variables: number of calls, t(22) = 0.12, p = 0.91; cumulative duration of calls, t(22) = 0.46, p = 0.65; average duration of calls, t(22) = 1.69, p = 0.11; average fundamental frequency, t(22) = 10.02, p = 0.99; average peak amplitude, t(22) = 0.21, p = 0.84; average peak frequency, t(22) = 0.02, p = 0.98; spectral purity, t(22) = 0.49, p = 0.63; and sequence length, t(22) = 0.61, p = 0.55 (Figure 1c–h).

Both Fmr1 KO males and Fmr1 HET females exhibit differences in qualitative aspects of vocalization behavior on PD8 according to two call-type taxonomies

The first call-type taxonomy came from the Avisoft analysis according to previous methods ²⁶. Chi-square analysis in males revealed that the distribution of call-types utilized between Fmr1 KOs and their WT counterparts was significantly different on PD8, χ2 (9) = 151.45, p < 0.001. Z-tests for column proportions indicated that several call-type proportions significantly differed from each other at the level of p < 0.05. Fmr1 male KOs produced proportionally more chevron and frequency steps call-types, as well as proportionally fewer complex, composite, downward, harmonics, two-syllable, and short call-types (Figure 2a).

Figure 2. Both Fmr1 KO males and Fmr1 HET females showed differences in their call-type utilization profiles compared to their WT counterparts.

a. The distribution of call-types utilized between male KOs and their WT counterparts was significantly different on PD8. Male KOs produced proportionally more chevron and frequency steps call-types, as well as proportionally fewer complex, composite, downward, harmonics, two-syllable, and short call-types. b. The distribution of call-types utilized between female HETs and their WT counterparts were significantly different on PD8. Specifically, female HETs produced proportionally more frequency steps call-types, as well as proportionally fewer chevron, downward, harmonics, and short call-types. * = P < 0.05.

Chi-square analysis in females revealed that the distribution of call-types utilized between Fmr1 female HETs and WTs was significantly different on PD8, χ2 (9) = 117.96, p < 0.001. Z-tests for column proportions indicated that several call-type proportions significantly differed from each other at the level of p < 0.05. Fmr1 HETs produced proportionally more frequency steps call-types, yet proportionally fewer chevron, downward, harmonics, and short call-types (Figure 2b).

Using the other call-type taxonomy with the MATLAB analysis, chi-square analysis in males revealed that the distribution of syllables utilized between Fmr1 KOs and WTs was not significantly different on PD8, χ2 (3) = 4.46, p = 0.22. Z-tests for column proportions indicated that Fmr1 KOs produced proportionally fewer “d” syllable types than WTs at the level of p < 0.05 (Figure 3a).

Figure 3. Fmr1 HET females, not Fmr1 KO males, showed differences in their syllable utilization profiles compared to their WT counterparts.

a. The distribution of syllables utilized between male KOs and their WT counterparts was not significantly different. b. Chi-square analysis revealed that the distribution of syllables utilized between female HETs and their WT counterparts was significantly different on PD8. Female HETs produced proportionally fewer “s” syllable types, as well as proportionally more “m” syllable types. * = P < 0.05.

Chi-square analysis in females revealed that the distribution of syllables utilized between Fmr1 HETs and their WTs counterparts was significantly different on PD8, χ2 (3) = 13.78, p < 0.005. Z-tests for column proportions indicated that several syllable proportions significantly differed from each other at the level of p < 0.05. Specifically, Fmr1 HETs produced proportionally fewer “s” syllable types, as well as proportionally more “m” syllable types (Figure 3b).

Comparison of the two methods of USV data extraction suggests similar agreement.

Spearman correlations were performed on data across three different parameters measured by both systems: number of USVs, fundamental frequency (kHz), and average duration of calls. Results indicated a significant association between the number of USVs detected in a file using hand-scoring methods and the number detected using MATLAB processing, r_s(52) = 0.85, p < 0.0001 (Figure 4a). With regard to fundamental frequency, results indicated another significant association across the two systems, r_s(52) = 0.91, p < 0.0001 (Figure 4b). The average duration of calls detected in the two systems was also significantly associated, r_s(52) = 0.42, p = 0.002 (Figure 4c). Overall, these results indicate that both systems are sensitive for analysis of USVs.

Figure 4. Comparison of the two methods of USV data extraction suggests similar agreement.

a. The number of calls detected by the two methods of analysis were nearly identical, r_s(52) = 0.85, p < 0.0001. b. The average frequency of the calls detected was similar between the two systems, r_s(52) = 0.91, p < 0.0001. c. Average duration was also similar between the two systems, r_s(52) = 0.42, p = 0.002. Data points represents individual subject scores.

DISCUSSION

The present results corroborate previous findings in Fmr1 KO male pups and extend these findings to Fmr1 HET female pups on PD8. Previously, behavioral characterizations following the loss of Fmr1 have typically only examined males. Increasing evidence supports that female Fmr1 KOs exhibit behavioral changes that are qualitatively different than their male counterparts, similar to what is typically seen in clinical datasets ^5,6,8,9. Only one study to date has investigated differences in USV production between female Fmr1 HETs and WTs early in development ¹⁰. Our data finds that Fmr1 KO males and Fmr1 HETs females utilized significantly different call-type and syllable repertoires compared to their respective WT counterparts, and these changes were not the same across the sexes. Specifically, male KOs also produced fewer complex, composite, downward, short and two-syllable call-types, as well as more frequency steps and chevron call-types. Female HETs exhibited increased production of harmonic and frequency steps calls, as well as fewer chevron, downward and short calls. Moreover, when vocalizations were assessed using the Holy and Guo (2005) call-type taxonomy we found that only Fmr1 HET females exhibit differences (fewer single-type and more multiple-type), while males did not show any changes according to this taxonomy. These call-type distributions are important, as a previous study found that atypical call sequences can significantly impact maternal approach ²⁹. In summary, findings of the current study corroborate the subtle female phenotype previously demonstrated in adulthood following heterozygous loss of Fmr1 in adulthood.

The present study also sought to extend previous findings by further characterizing vocalization behavior in the FVB background strain. A previous study conducted in the C57/BL6J strain using the same recording setup on PD8 did not find any changes between female Fmr1 HETs and WTs in number and mean duration of USVs, peak amplitude, peak frequency, and duration of calls between subsequent calls ¹⁰. In the present study, comprehensive analysis detected a significant increase in the average duration of these calls in male KO pups, as well as a marginal decrease in the number of calls emitted in male mutants only. However, in females, changes in vocalization behavior were limited to changes in call-type and syllable repertoires. The discrepancies in findings across the two studies could be due to differences in variables measured between the two studies, as the previous one in the C57BL/6J strain did not assess call-type production. However, the differences in findings may be more likely due to the different background strains used ^17,18. In non-transgenic mouse lines, previous studies indicate several behavioral differences between strains, such as increased spontaneous hyperactive in the FVB strain and enhanced motor coordination in the C57BL/6J strain ³⁰. The use of inbred mouse strains for genetic knockouts is quite prevalent throughout the scientific community, and the results of the present study corroborate the notion that background strain and origin of mouse colonies should be considered when designing an experiment and comparing to existing literature to enhance scientific rigor and reproducibility.

The present study also sought to examine the viability of high-throughput analyses in a neonatal USV model. According to our results, there was significant agreement between the two methods. The number of vocalizations detected and their average frequency were similar across both analyses, supporting the use of the automated analyses in future studies. It should be noted that while statistically significant, the correlation between average duration of calls measured by the two methods was only moderately sized. This may be due to the use of the two different call-type taxonomies across the two methods. Specifically, the way the calls are categorized across the two taxonomies may lead to specific aspects of the spectrogram being differentially grouped as a call or syllable or not, leading to variability in the duration. Though this may have impacted the reliability of the duration parameters, the inclusion of the Holy and Guo (2005) call-type taxonomy is also a strength of the present work, as it has never been applied to neonatal USVs in the Fmr1 model ²⁷. Using this analysis, female HET specific differences were detected in syllable type production, and these were not present in the male KO model. Future work could examine whether this type of analysis is suitable for multiple strains of genetic knockout mice, as many spectral and temporal aspects of vocalizations can be affected by strain ³¹.

The present study highlights two conclusions. First, the automated analysis delineated by the previous study is sufficient to detect changes in the isolation-induced USV paradigm. The use of this method allows processing of larger volumes of data, thereby allowing research to move at a faster pace. The previous study outlined appropriate methodology for automated syllable detection using MathWorks MATLAB software, allowing investigators to analyze and extract large amounts of information from datasets in less time than previous methods, which can take up to 2 hours on average to process a 2-minute recording ²⁰. Secondly, the use of both systems may be advantageous in identifying the changes following loss of Fmr1. This finding extends generally to a number of fields that utilize vocalization recordings and analysis as a behavioral testing paradigm, such as drugs of abuse ³². For FXS research, it is paramount to characterize the full spectrum of potential changes in the present model, particularly early in development. Understanding the changes characteristic of this period gives us more information about the baseline for developing and screening potential treatments. The use of high-throughput methodology, the inclusion of both sexes and the consideration of the background strain are all factors necessary to advance the field in a rigorous and reproducible manner that allows for the development of suitable treatments to improve the lives of those with neurodevelopmental disorders.