Selection on vocal output affects laryngeal morphology in rats

Abstract Although laryngeal morphology often reflects adaptations for vocalization, the structural consequences of selection for particular aspects of vocal behavior remain poorly understood. In this study, we investigated the effects of increased ultrasonic calling in pups on the adult larynx morphology in selectively bred rat lines. Laryngeal morphology was assessed using multiple techniques: mineralized cartilage volumes were compared in 3D‐models derived from microCT scans, internal structure was compared using clearing and staining procedures combined with microscopy, cellular structure was compared using histology and microscopy, and element composition was assessed with scanning energy dispersive X‐ray spectroscopy. Our results show that adult rats from lines bred to produce ultrasonic calls at higher rates as pups have shorter vocal folds and a more mineralized thyroid cartilage compared to rats bred to produce ultrasonic calls at lower rates. The change in vocal fold length appears to account for differences in low‐frequency calls in these two rat lines. We suggest that the observed increases in mineralization of the thyroid cartilage in the high‐ultrasound lineage provide increased reinforcement of the laryngeal structure during ultrasonic call production. Our findings therefore demonstrate an effect of selection for vocal behavior on laryngeal morphology, with acoustic consequences.

syrinx, which functions more purely in sound production due to its location deep in the chest, exhibits a much higher degree of diversity across species (Fitch & Hauser, 2006). Thus, the fact that the larynx must satisfy requirements for at least two critical functions can be estimated to have constrained its evolution and development, particularly with respect to modifications made solely for the purpose of vocal communication.
The mammalian larynx consists of a cartilaginous thyroid, arytenoids, cricoid, epiglottis, and a supporting set of hyoid bones (Harrison, 1995;Schneider, 1964). Despite a high degree of interspecific conservation, a number of impressive modifications of the mammalian larynx have been documented (Fitch, 2016). Among such adaptations we find enlargements of the larynx itself in howler monkeys, pads on the vocal folds in lions, or the ability to produce high frequency whistles in rodents (Dent & Fay, 2018;Dunn et al., 2015;Klemuk et al., 2011). In some rodents, such as rats and mice, another structure, the alar cartilage is added to the "standard" structures of the mammalian larynx (Inagi, Schultz & Ford, 1998).
Murine rodents have two distinct vocalization ranges, defined by low-frequency calls and ultrasonic vocalizations (USVs) (Brudzynski, 2005(Brudzynski, , 2018. This provides a very interesting and unusual opportunity to analyze functional changes to larynx morphology. USVs in rats can further be categorized into two distinct ultrasonic vocalization ranges used to communicate different affective states of the caller: 22 kHz vocalizations communicate negative affect whereas 50 kHz calls communicate positive affect (Portfors, 2007). These two ranges are specific to adult rats; when isolated, rat pups produce USVs in a single range centered around 40 kHz (Portfors, 2007). Apart from these well-studied ultrasonic vocalizations, rats also produce low-frequency calls within the human hearing range (Brudzynski, 2010(Brudzynski, , 2018. Extrapolating from what is known about vocalization in other mammals, the production mechanism of low-frequency calls in rodents is predictable on the basis of "ordinary" laryngeal vocal fold vibration (Herbst et al., 2012;Riede et al., 2011;Roberts, 1975b).
Previous studies agree that USVs are also produced in the larynx, but by a different "whistle" mechanism (Johnson et al., 2010;Mahrt et al., 2016;Riede, Borgard & Pasch, 2017;Sanders et al., 2001). However, debate remains over the nature of this mechanism: Mahrt and colleagues suggest a planar impinging jet. In contrast, Riede and colleagues argue for an edge-tone whistle mechanism (Mahrt et al., 2016;Riede, Borgard & Pasch, 2017). These two different mechanisms are associated with different types of morphological changes in the larynx, for example in the alar cartilage (thought to be essential in the production of USVs in rat communication (Inagi, Schultz & Ford, 1998;Riede, Borgard & Pasch, 2017). This leads to the question addressed by our study: how does selection for USV production rate influence laryngeal anatomy, including both USVrelated anatomy, and morphological characteristics associated with low-frequency vocal production?
We address this question here by analyzing vocal anatomy in two rat lines that have been selected for over 50 generations to produce low or high rates of USVs as pups, during a maternal separation paradigm (Brunelli et al., 1997;Brunelli & Hofer, 2007). This sustained artificial selection upon USV call rate in pups yielded two distinct rat lines with both different vocal behavior and distinctive stress responses (Brunelli & Hofer, 2007;Zimmerberg, Brunelli & Hofer, 1994). An earlier study of rats from these two selected lines revealed clear effects of selection on vocal acoustics (Lesch et al., 2020), but the morphological basis of the line-specific acoustic differences documented there remain unknown.
In particular, our previous study showed that selection on rat pup USV rate did not affect acoustic parameters of adult USVs, but rather affected acoustic aspects of adult low-frequency calls. Specifically, we found that "high line" individuals (i.e., pups selected for increased USV production) were heavier in body weight than "low line" rats (i.e., pups selected for decreased USV production), but did not produce lower frequencies in their calls. This apparent exception to a general correlation (allometry) between body mass and fundamental frequency of low-frequency calls led us to formulate the hypothesis that selection for acoustic output in pups will lead to changes in adult laryngeal morphology. In the present study, we used multiple visualization and measurement techniques to quantitatively demonstrate an effect of artificial selection for a specific aspect of vocal behavior on the laryngeal morphology of rats. We found that larynges of high line adults have shorter vocal folds and a more mineralized thyroid cartilage compared to low line adults. These results support our hypothesis that selection on pup calling will lead to differential changes in adult laryngeal morphology, depending on rat line.

| Rat lines
The rat lines studied here were initially described in Brunelli et al., (1997) and have been bred since the 1990s (57-58 generations) on the basis of differences in the rate of USV calls produced by pups in response to a maternal separation paradigm. Both lines were developed from the same founding population of N:NIH Rattus norvegicus domestica. The N:NIH strain was developed in the 1980s to provide an outbred and heterogeneous strain from eight inbred strains (including the Wistar lineage; Hansen & Spuhler, 1984). From a founding population of these N:NIH rats one line was bred to produce high rates of pup USV calls during a maternal separation paradigm ("high line"), while the "low line" was bred to produce few calls during the same paradigm. In the maternal separation paradigm, the dam was separated from the litter 20 minutes prior to testing, to elicit separation induced calls from the pups. During a two minute test phase each pup's USV call rate was measured. This rate was used as the basis from which the high-and low-lines were bred. Within five generations the high line pups produced up to 300 USVs within two minutes whereas the low line pups produced less than 50 calls. In the 58th generation high line individuals produced on average over 300 calls with some individuals reaching over 400 USVs; low line individuals produced less than 50 USVs with the exception of one individual that produced 86 USVs (Lesch et al., 2020). The breeding was conducted at Williams College Animal Facility and all procedures were approved by the Williams College Animal Care and Use Committee.

| Larynx extraction
We dissected larynges from 51 adult rats (between 133 and 135 days old; high line: 14♂, 13♀; low line: 12♂, 12♀) that were culled from the breeding population of the two rat lines main-

| Overview
Out of the 51 larynges we randomly chose 8 larynges of each group (high male, high female, low male, low female), adding up to a total of 32 larynges. All 32 larynges were used in the microCT analysis and after the scan four of these larynges were stained, three were used for the energy dispersive X-ray (EDX) and one underwent sectioning.

| microCT scans
Our primary method for comparing larynges between lines was based on the volumes of key cartilages. Each laryngeal specimen was scanned using microCT (µCT) to obtain a three-dimensional computer model that allowed us to quantitatively determine the volume of mineralized tissue in the hyoid bone, and each of the main laryngeal cartilages (cricoid, arytenoids and thyroid).

| Specimen preparation
Thirty-two larynges (8 per sex and line combination) were prepared for microCT scanning at the University of Vienna. The individual larynges were thawed and further dissected for the scanning procedure. The majority of the tongue was cut off and other surplus tissues were removed. We flushed the larynx with 0.9% saline solution to remove any tissue or fluid from within the larynx and trachea.
Afterward, we placed the larynx in a plastic tube, stabilized its position with plastic straws and added a drop of saline at the bottom to prevent dessication. The tubes were sealed with parafilm and mounted on the scanner platform for scanning.

| Scanning process and visualization
We scanned all larynges with an XRadia MicroXCT-400 (Carl Zeiss X-ray Microscopy, Pleasanton, CA, USA) using the 0.4x detector as- DICOM image stacks were loaded in the Amira software package (version 6.4.0) and the following structures were segmented manually: thyroid, arytenoids, cricoid, tracheal rings, and the hyoid apparatus. Since the microCT scans only reliably capture mineralized structures, these reconstructions only show the mineralized parts of the cartilages and hyoid apparatus. We used the threshold tool in the Amira segmentation editor to aid manual segmentation of the cartilages and hyoid apparatus (basihyal, ceratohyal, chondrohyal, hypohyal and thyrohyal; Sharma & Sivaram, 1967). The threshold tool was adjusted for each individual to perfectly capture the hyoid apparatus and these same settings were then used for the rest of the (mineralized) larynx. We measured the mineralized volume of the hyoid, cricoid, arytenoids and thyroid and vocal fold length on the reconstructed 3D models of all larynges. Vocal fold length was approximated based on the mineralization patterns of the 3D reconstructions. We placed landmarks on the dorsal border of each arytenoid cartilage, and in the area of vocal fold attachment on the thyroid cartilage; we then measured the distance between the arytenoid and thyroid landmarks, resulting in two vocal fold measurements per individual (right and left; Bowling et al., 2020). We recorded the average value as "mean vocal fold length" for each individual to be used for all further statistical analysis.

| Clearing and staining
To visualize the entire (and not just mineralized) laryngeal anatomy in our specimens, we selected four larynges (high male: #18, high female: #42, low male: #34, low female: #2; microCT scans and 3D reconstructions were done for all specimens) to create cleared and stained specimens, according to an adapted staining protocol based on Rigueur and Lyons (Rigueur & Lyons, 2014). Excess tissue was mechanically removed for faster preparation. The main steps in the preparation were fixation of frozen larynges in 4% formaldehyde, followed by the first staining step in an ethanolic Alcian blue solution for staining cartilage. Afterwards, soft tissue was preincubated and macerated in potassium hydroxide (KOH) solution, followed by staining in aqueous Alizarin Red solution in KOH. Additional soft tissue removal and clearing was then accomplished using KOH. Whole specimens were finally transferred into glycerol for documentation and analysis with a Nikon SMZ25 stereomicroscope equipped with a Nikon DsRi-2 camera (Nikon Instruments, Tokyo, Japan) or a Hirox RH-2000 digital microscope system (Hirox, Limonest, France).

| Electron microscopy and energy dispersive X-ray spectroscopy (EDX)
To assess the elemental composition of the mineralization we used three larynges (high line male #19, low line male #26, high line female #15) for elemental analysis. The distribution of elements was identified using EDX throughout cartilages of the larynx and the hyoid bone. Both the cricoid and thyroid cartilage were cut in half and attached to carbon adhesive discs on aluminum stubs with their cut surfaces facing upward. The hyoid bones were attached to the carbon disks as a whole. Additional conductive carbon cement (Leit-C) was applied to enable better contact to the stubs and prevent excessive charging. Prior to analysis, specimens were carbon coated with a Leica EM MED20 (Leica Microsystems, Wetzlar, Germany). Analysis was conducted with a Jeol-IT3000 scanning electron microscope with the following parameters: BED-C, 20 kV, WD 11 mm, STd.P.C 63.7-67.7, 100 Pa and 350-600 times magnification.

| Histological analysis
To distinguish between calcification and ossification of the mineralized parts, one larynx (high line male #25) was embedded in epoxy resin for sectioning and histological analysis. The specimen was first fixed as described in the clearing and staining section. It was then decalcified with 20% EDTA and dehydrated with acidified dimethoxypropane prior to embedding into Agar LVR resin (Agar Scientific, Stansted, UK) using acetone as an intermediate. Sections of cured resin blocks were sliced at 1 µm section thickness with a Leica UC6 ultramicrotome equipped with a diamond knife (Diatome, Nidau, Switzerland) and analyzed with a Nikon NiU compound microscope.

| Statistical analysis
To determine the effect of line breeding on vocal fold length and mineralized cartilage volumes we constructed separate generalized linear models. The models compare each of the following measurements between lines: cricoid mineralized volume, thyroid mineralized volume, arytenoid mineralized volume and vocal fold length. A Gamma log distribution was used for all models. The significance of line specific differences was assessed by comparing the predictive value of full models against null models lacking the relevant predictor. In the full models we included the covariate bodyweight and a dummy coded variable representing line and the respective sex within line. For all model comparisons, the null model only included the covariate 'body weight' due to our default allometric expectation that larger individuals should have larger vocal structures, and accounting for the fact that the different selected rat lines showed significant differences in body weight. The covariate bodyweight was scaled and centered before running any analysis. All models were inspected and plotted to determine if model assumptions were satisfied. All variance inflation factors were <4 and overdispersion was <0.4.
Statistical analyses were performed in R (Version 3.6.1; www.r-proje ct.org/) and R-

| Cartilage volumes, vocal fold lengths, and bodyweight allometry
The reconstructed larynx models allowed us to measure a proxy of vocal fold length and quantify the mineralized volumes of each cartilage and the hyoid apparatus ( Figure 1). The full models for vocal fold length as well as arytenoid, thyroid, and cricoid volume were all significantly superior to the null models (Tables 1-4); only the full model for the hyoid was not significantly better than the null model (Table 5). Therefore differences between lines were identified for    (Lesch et al., 2020).

| DISCUSS ION
USVs are at the center of research associated with emotional states, and are often used as an easily measured proxy for stress response styles in rats (Branchi, Santucci & Alleva, 2001;Brunelli, 2005;Simola & Brudzynski, 2018). Therefore it is important to document and understand possible changes to the vocal production mechanisms that may arise as a result of selective breeding (and, by extension, natural selection). In Lesch et al. (2020), analyses of vocal outputs from rat lines selectively-bred for USVs as pups showed that line breeding affected the fundamental frequency of low-frequency calls but not USVs in adults. The likely mechanism by which these low-frequency calls are produced is standard vocal fold vibration in the larynx, as detailed in the myo-elastic aerodynamic theory (Elemans et al., 2015;Titze, 2006), implicating potential differences in vocal fold length as the main cause of these differing fundamental frequencies (Herbst et al., 2012;Riede et al., 2011;Roberts, 1975a). Our approximate measurements of vocal  (Boskey, 1988). The histological structure, however, shows patchy distribution of the mineralized areas between chondrocytes, indicating calcified cartilage rather than bone. Definitive distinction of calcified cartilage vs bone would require specific staining against collagen I for bone and collagen II for cartilage (Fratzl, 2008;Hall, 2015).
Based on a comprehensive review of mammalian laryngeal anatomy, Schneider (1964) concluded that there is no generalizable rule for the process of calcification in the mammalian larynx, and suggested that the thyroid cartilage in eutherians tends to first calcify in areas that are under less deformation pressure, for example, inferior and superior thyroid horn (Schneider, 1964). This prediction does not match the pattern of calcification we found in our rat larynges, which were mainly calcified in areas that can be under significant deformation pressure, specifically the specific areas of vocal fold attachment. Carter (2020) compared calcifications in the trachea and larynx across laryngeally echolocating bats and found that higher-intensity vocalizing species tend to have a more mineralized thyroid cartilages than lower-intensity species.
Thyroid mineralization in the high line rats might help to facilitate and sustain USV production, similar to that observed in laryngeally echolocating bats. However, it remains unclear whether this mineralization is genetically determined irrespective of vocalization, or whether increased rates of vocalization in young animals might have led to the increased mineralization of thyroid cartilages that we observe in adults. Detailed anatomical investigations through development, for example, using microCT, would be necessary to test this prediction.
A limitation of this study is the fact that we cannot determine whether the line specific differences that we have documented  (Brunelli et al., 1997). But this does not necessarily indicate that the anatomical changes occurred equally rapidly. Unfortunately, addressing this question would entail starting the selection experiment afresh from unselected lines, and acquiring anatomical specimens at each generation, a research program far beyond the scope of our study.
It is well known that, despite its relatively conservative structure, the form of the mammalian larynx can be modified to suit its particular functions (Charlton & Reby, 2016;Fitch, 2016).  (Dunn et al., 2015). The large "roaring cats" (lions, tigers, jaguars, and leopards) have several modifications of vocal fold structure and hyoid morphology that again support the production of loud, low-frequency vocalizations (Klemuk et al., 2011;Titze et al., 2010). Many similar modifications for loud low-frequency vocalizations have been documented in a range of ungulates (Frey et al., 2007;Frey et al., 2011;Frey & Hofmann, 2000;Frey & Riede, 2003). Nonhuman primates show a suite of specific vocal modifications, including many types and sizes of laryngeal air sac (Fitch, 2016;Schön Ybarra, 1995), and large-scale comparisons of laryngeal morphology in primates and carnivores suggest that the primate larynx has evolved more rapidly than that of carnivores (Bowling et al., 2020).
Given these well-known modifications on an evolutionary timescale, it is clearly important to better understand how selection for specific vocal traits (whether lower-or higher-frequency calls, louder calls, or more frequent calling) affects the morphology and development of the main sound producing organ, the larynx (that is, to examine both the ontogeny and phylogeny of changes in vocal anatomy). Our study takes a first step in this direction and reveals that two specific aspects of laryngeal structure -vocal fold length and thyroid mineralization -can be modified by artifical selection on rate of vocalization alone, and in a relatively short time span.
Furthermore, it should be noted that this targeted selection produced a host of other "side-effects," including, for example, on voice fundamental frequency, body size, and temperament. This implies that selection on the voice can have a considerable diversity of effects that are of clear relevance for an animal's behavior and fitness.
Finally, our results also show that the expected allometric correlation between overall body size and the size of specific organs can easily be overridden by selection, at least with respsect to laryngeal size.
Although our high line rats were larger in overall body size, they had shorter vocal folds and produced vocalizations with higher fundamental frequencies. Thus, our results are consistent with the hypothesis that laryngeal morphology, and the vocal acoustic parameters that depend on it such as fundamental frequency, is relatively unconstrained by overall body size in mammals (Fitch & Hauser, 2006;Garcia et al., 2017). Put differently, under strong selection for vocal output, laryngeal size can change independent of body size. This offers comparative insight into the peculiar fact that human males, due to laryngeal hypertrophy, have a much lower fundamental frequency than women (about 50%) despite being on average only 20% larger (Pisanski et al., 2014), and belies the common assumption that fundamental frequency automatically reflects body size as "a law of physics" (Morton, 1977). The depends strongly on biological factors, specifically whether laryngeal morphology correlates with overall body size. This is a matter of anatomy and physiology, not of physics.
To sum up, our data support the hypothesis that rats selectively bred for USV production as pups show changes in larynx morphology in adulthood, which in turn have clear and predictable results on vocal production. This finding is clearly relevant to researchers using USVs to gain insight into the emotional state and stress response, as well as to broader questions in the evolution of communication (cf. Bowling et al., 2020;Dunn et al., 2015;Fitch, 2016).

CO N FLI C T O F I NTE R E S T
The authors state no conflict of interest.

O PE N R E S E A RCH BA D G E S
This article has earned an Open Data badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at http://doi.org/10.5281/ zenodo.4415263.