Abstract
The identity and location of vocalization pattern generating (VPG) circuits in mammals is debated. Based on physiological experiments, investigators suggested anterior brainstem circuits in the reticular formation, and anatomic evidence suggested the nucleus retroambiguus (NRA) in the posterior brainstem, or combinations of these sites as the putative mammalian VPG. Additionally, vocalization loudness is a critical factor in acoustic communication. However, many of the underlying neuronal mechanisms are still unknown. Here, we evoked calls by stimulation of the periaqueductal gray in anesthetized male rats, performed a large-scale mapping of vocalization-related activity using the activity marker c-fos, and high-density recordings of brainstem circuits using Neuropixels probes. Both c-fos expression and recording of vocalization-related activity point to a participation of the NRA in vocalization. More important, among our recorded structures, we found that the NRA is the only brainstem area showing a strong correlation between unit activity and call intensity. In addition, we observed functionally diverse patterns of vocalization-related activity in a set of regions around NRA. Dorsal to NRA, we observed activity specific to the beginning and end of vocalizations in the posterior level of the medullary reticular nucleus, dorsal part, whereas medial and lateral to the NRA, we observed activity related to call initiation. No clear vocalization-related activity was observed at anterior brainstem sites. Our findings suggest a set of functionally heterogeneous regions around the NRA contribute to vocal pattern generation in rats.
SIGNIFICANCE STATEMENT Vocalization patterns are shaped in the mammalian brainstem, but the identity and location of the circuits involved is debated. Additionally, the neuronal mechanisms of vocal intensity control are still unknown. This study consisted of a large-scale mapping of brainstem vocalization circuits based on the activity marker c-fos and high-density recordings with Neuropixels probes. The results confirm the role of nucleus retroambiguus in call production and point to a key role of neurons in this nucleus in loudness control. Dorsal to the nucleus retroambiguus and in the posterior medulla, the authors identify neurons with activity specific to the beginning and end of vocalizations. The results point to specific neural dials for various aspects of rat vocalization control in the posterior brainstem.
Introduction
Vocalizations play a crucial role in social interactions among mammals (Brudzynski, 2010, 2021) and may signal fear, excitement (Burgdorf et al., 2007), or courtship (Egnor and Seagraves, 2016). Given the immense importance of vocalizations, it is not surprising that mammals developed elaborate mechanisms for generating vocalizations. In fact, vocalizations are among the most complicated motor patterns generated by mammals as they require highly precise control of the larynx, facial muscles, and respiration. The study of vocalization control goes back a long time and had a dramatic onset in Broca's (1861) discovery of the left hemispheric cortical localization of speech, a finding that marks the beginnings of modern neuroscience. The investigation of vocalization control is very much a success story with respect to high-level vocalization control, and there is good consensus about neural structures critically involved in mammals. Cortical areas involved include orofacial motor and somatosensory cortices and the anterior cingulate cortices (Jürgens, 2009; Dichter et al., 2018; Bennett et al., 2019). At the midbrain level there is broad agreement that the periaqueductal gray (PAG) is a decisive control structure for vocalizations (Jürgens, 1994). Activation of these high-level control structures can evoke vocalizations and typically such evoked vocalizations consist of complete species-specific vocalizations.
There is also a consensus that the actual vocal pattern generation for mammalian vocalizations occurs in brainstem circuits. The evidence for this conclusion comes from both stimulation experiments, which suggested that brainstem-evoked vocalizations are often more elementary non-species-typical vocalization patterns (Zhang et al., 1992, 1995; Hartmann and Brecht, 2020) and from interference approaches, which point to the necessity of brainstem structures for vocalizations (Zhang et al., 1995; Shiba et al., 1997; Hartmann and Brecht, 2020). It has been very difficult, however, to delineate the precise identity of brainstem circuits involved in vocal pattern generation. Specifically, different authors have made different suggestions for the identity of the mammalian vocal pattern generator (VPG). Anatomical analysis (i.e., tracing of descending projections from the PAG) pointed early to a crucial role of nucleus retroambiguus in the posterior brainstem (Holstege, 1989), and this suggestion received support from interference experiments (Zhang et al., 1995; Shiba et al., 1997; Hartmann and Brecht, 2020). Other authors suggested, based on unit recording and interference experiments, the reticular formation in the anterior brainstem as the VPG (Jürgens, 2000; Hage and Jürgens, 2006a; Hartmann and Brecht, 2020). A recent paper provided evidence for yet another region, the intermediate reticular oscillator, as the pattern generator for pup calls in mice (Wei et al., 2022). Various suggestions have been made on how to reconcile the conflicting evidence about an anterior brainstem or posterior brainstem VPG; thus, a distributed model of vocal pattern generation spanning from the anterior to posterior brainstem has been suggested (Hage, 2010), and Hartmann and Brecht (2020) suggested a bipartite VPG with the anterior vocalization region of the parvicellular reticular formation (VoPaRt) region being specialized for high-frequency calls and a posterior nucleus retroambiguus (NRA) region being specialized for low-frequency calls. Finally, it is well known that voice frequency is primarily determined by the tension of the vocal folds, whereas the loudness of vocalizations is dependent on the degree of expiratory effort. Both vocalization frequency and intensity are widely accepted as important components of speech and vocalization; however, much previous work has focused on how neurons of VPG are related to vocal frequency, whereas the precise neuronal mechanism underlying control of vocal loudness remains largely unexplored.
In this article, we direct powerful mapping techniques to further elucidate the identity of the brainstem vocal pattern generator in rats. Specifically, we use c-fos expression and high-density Neuropixels recordings to map vocalization circuits. We ask the following: (1) Which brainstem neurons show c-fos expression after inducing intense vocalizations? (2) Which brainstem neurons show neural responses before and during vocalizations? (3) How is the vocalization-related activity of brainstem neurons patterned? (4) How do brainstem neurons respond to auditory stimuli? (5) How is the activity of brainstem neurons related to breathing? (6) Where and how is vocalization loudness controlled in the rat brainstem?
Materials and Methods
Animals
All animal experiments and procedures were conducted according to German law for animal welfare and approved by the State Office for Health and Social Affairs Committee in Berlin (Animal license number G0279/18). Male Long-Evans rats were ordered from Janvier Labs and kept in a temperature- and humidity-controlled room on a 12 h light/dark cycle. Animals were allowed clean food and water ad libitum in standard rat cages.
Surgery
For surgery, animals were deeply anesthetized by applying intraperitoneal injections of urethane (1.5 × g/kg body weight. The fur overlying the dorsal aspect of the animal skulls was shaved. Then the rat was placed in a standard stereotaxic surgical apparatus (Narishige). The body temperature of the animal was measured with a rectal probe and kept at 36°C ± 0.5°C by a homeothermic blanket (FHC). Before the surgical incision, the scalp of the animal was locally anesthetized by injecting 2% lidocaine solution. To access the brainstem and PAG, the skin was cut anteroposteriorly along the midline, and the remaining connective tissue on the skull was removed. The anchoring screws were inserted to the skull bone and a head-fixation post was then secured to these screws using UV-curable adhesive glue (OptiBond, Altschul Dental) and dental cement (Heraeus Kulzer). For the c-fos experiments, a craniotomy of ∼1.5 mm diameter was made in the skull over PAG using a dental drill. A single tungsten electrode (impedance 1 MΩ) microprobe was then targeted vertically into the left PAG (anteroposterior, −6.5 to −7.0 mm from bregma; mediolateral, 0.75–0.8 mm; dorsoventral, 4–4.5 mm). To produce vocalizations and induce c-fos expression in the rat, 1-s-long stimulation trains (100 Hz, 100 pulses, 100 µA) with 10 s intervals were applied for 30 min. In the nonvocalized animal group, the same stimulation protocol was used, but no calls were evoked. For the neuronal recording experiments, an additional craniotomy of ∼1.5 mm diameter was performed in the skull above the desired anterior brainstem areas or posterior brainstem areas. Then, a 960-site, 384-channel Neuropixels probe (Phase 3B) was coated with lipophilic carbocyanine fluorescent dyes DiO, DiI, or DiD and lowered slowly into the anterior brainstem regions vertically, or into posterior brainstem regions at a 45° angle. To coat the Neuropixels probe before each penetration, the tip of the probe was dipped 15–20 times into a 5% (w/v) DiO, DiI or DiD solution and air-dried for ∼8 s between dips. To avoid drifting of the probe, we applied a layer of 4% agarose between the probe and the exposed brainstem.
Histology
At the end of c-fos or extracellular recording experiments, small electrolytic lesions were induced by applying a direct current (8–10 µA for 10 s) in the PAG region to mark stimulation sites. For c-fos immunohistochemistry, the rats remained in the apparatus for an additional 90 min, after which animals received a fatal dose of urethane and were perfused transcardially with a prefixative solution (0.9% NaCl, 0.02 m phosphate buffer) and 4% paraformaldehyde in 0.1 m phosphate buffer solution (PFA). For extracellular recording experiments, the same procedures of overdose anesthesia and perfusion were performed immediately following the end of the experiment. The brains of the animals were then removed from the skull and postfixed in a 4% PFA solution for at least 24 h.
The brains were placed in 2–4% agarose/Ringer's solution and cut into 60- to 80-µm-thick coronal sections with a Vibratome (Microm HM 650 V, Thermo Fisher Scientific). For c-fos and NeuN double staining, slices were washed once with PBS and two times with PBS containing 0.5% Triton X-100 (5–10 min each wash) at room temperature. Subsequently, the sections containing interested brainstem areas were immersed in blocking solution (2.5% BSA and 0.75% Triton in PBS) for 1 h at room temperature, followed by incubation in the blocking buffer including rabbit anti-c-fos (1:1000; catalog #226003, Synaptic Systems) and mouse anti-NeuN (1:1000; catalog #MAB377, Chemicon) primary antibodies for 2 d at 4°C. After washing three times with 0.1 m PBS, the brain slices were then incubated in the secondary antibodies Alexa Fluor 488 donkey anti-rabbit (1:500; catalog #A21206, Invitrogen) and Alexa Fluor 633 goat anti-mouse (1:500; catalog #A21050, Invitrogen) at 4°C for 24 h. Slices were then washed three times with PBS and mounted on glass slides with Fluoromount mounting medium (Biozol). Images were taken on Leica SP8 confocal microscope (Leica Microsystems) with a 10× objective and Leica DM5500 fluorescence microscope (Leica Microsystems) with a 5× objective.
For identifying the electrode tracks in Neuropixels recordings, the coronal slices were directly mounted on slides with Fluoromount mounting medium (Biozol). Images of DiI-, DiD- or DiO-marked Neuropixels probe tracks were captured at 5× magnification using Leica DM5500 fluorescence microscope (Leica Microsystems). A typical fluorescent dye-labeled track can be observed across several sections, and recording sites along the Neuropixels probe were carefully aligned with an adult rat brain atlas (Paxinos and Watson, 2006). In all cases, 20% shrinkage factor of the fixed brain and penetration angle of Neuropixels probes were taken into account during the alignment.
c-fos quantification
To detect and quantify c-fos- and NeuN-expressing neurons, images were analyzed by using an automated and customized ImageJ plug-in. In our current study, we concentrated on the brainstem structures that may serve as a vocal pattern generator. Boundaries of each brain areas were carefully defined according to the rat brain atlas (Paxinos and Watson, 2006). c-fos- and NeuN-expressing neurons within the defined brain areas were then automatically detected and counted with our customized ImageJ macro. Two to three slices per brain region of each rat were applied for analysis, and the ratio of c-fos-expressing cells was estimated as the number of double-labeled positive neurons to the number of NeuN-positive neurons. The ratio of c-fos-expressing neurons was compared statistically using a one-way ANOVA test.
Spike sorting and assignment of units to histologically identified subdivisions
Spikes were detected from the high-pass filtered data using Kilosort 2.0 (Pachitariu et al., 2016), and then the output clusters were manually adjusted using the Phy GUI function (https://github.com/cortex-lab/phy). Spikes occurring during stimulation were not included to preserve the waveform shape. Clusters of neurons were assessed qualitatively in terms of their autocorrelogam (little presence of short-latency Interspike intervals), spike amplitude, and presence of a clear waveform modulation across channels. Neighboring clusters (up to 10 channels apart) were directly compared with each other in terms of cross-correlogram, waveform similarity per channel, and firing rate patterns (the latter, to avoid classifying as separate unit clusters that do not overlap in time). Clusters with a high similarity index were also compared in the same manner. Only clusters satisfying all these criteria were considered. Once units were sorted, they were assigned to the channel that showed the largest waveform amplitude. Probes were visually superimposed on the slices after shrinkage and angle correction, and then channels were located in the slice matching the fluorescent track. This procedure allowed us to assign all the units to the area where its channel belongs.
Acoustic stimuli
All auditory stimuli in the experiment were administered through an ultrasonic dynamic Vifa speaker (Avisoft Bioacoustics), positioned 10 cm away from the head of the rat. The following eight set of auditory stimuli, each with a duration of 500 ms, were presented to anesthetized rats: (1) 5 kHz pure tone, (2) 10 kHz pure tone, (3) 20 kHz pure tone, (4) 40 kHz pure tone, (5) 80 kHz pure tone, (6) white noise, (7) 22 kHz ultrasonic vocalizations (USVs), and (8) frequency-modulated 50 kHz USVs. For each Neuropixels recording, each type of auditory stimulus was repeated at least 20 times. During the experiment, auditory stimuli were recorded and monitored by an Avisoft condenser microphone (CM16/CMPA-5 V, frequency range 10–200 kHz, Avisoft Bioacoustics). Avisoft-RECORDER USGH software (Avisoft Bioacoustics, Berlin, Germany) was used for both auditory stimuli and recording.
Ultrasonic vocalization recording
Ultrasonic vocalizations emitted by the rats were monitored using a condenser ultrasound microphone (CM16/CMPA-5V, frequency range 10–200 kHz, Avisoft Bioacoustics) situated 15 cm in front of the rat. Data were collected at a sampling frequency of 250 kHz and 16-bit resolution using Avisoft-RECORDER USGH software (Avisoft Bioacoustics). Calls were detected using DeepSqueak version 3 (https://github.com/DrCoffey/DeepSqueak), and the spectral data were manually checked to refine frequency and time limits of each call to ensure that no call was missing and that the whole length of the call was detected (this was especially important as calls could last >700 ms in some of the recordings). There is a large literature to suggest a functional division of rat USVs into high-frequency calls with positive valence (Panksepp and Burgdorf, 2003) and low-frequency (i.e., 22 kHz) calls (Brudzynski, 2009). In our study, calls were classified manually as high or low frequency depending on whether they were below or above 35 kHz. In addition, the length and shape of calls were also considered for classifying call types. A recent study has shown that unanesthetized cats could produce different types of calls by stimulating different columns of PAG (Subramanian et al., 2021). In our experiments, most calls are evoked by stimulation of the lateral and ventrolateral PAG columns; both high-frequency and low-frequency calls can be emitted by stimulating the same area of the PAG.
Breathing assessment
Breathing was measured using a piezoelectric sensor (LDT0-028K, Measurement Specialties) located in the dorsolateral part of the trunk and sampled at 1000 Hz. Signal was z-scored using Hanning windows of 10 s to normalize for possible displacements of the instrument along the experiment. Amplitude and phase were obtained using a Hilbert transform. A 3 min baseline without stimulation was recorded at the beginning of each experiment.
Classification of response types
To find call responsive units, their spikes were aligned to the beginning and end of each call that occurred at least 50 ms away from any stimulation period. The alignment was done for low-frequency and high-frequency calls. Neurons that showed spike modulation either at the beginning or at the end of the calls were classified as call responsive. To classify breathing responsive units, breathing peaks (maximum inhalation amplitude) were detected during baseline, and spikes were aligned according to them. Also, the distribution of breathing phases at the moment of each spike was compared with the distribution of phases along the whole baseline period. Neurons that showed a response aligned to the inhalation peak or a clear phase preference were classified as breathing responsive units. Finally, neurons were classified as auditory responsive if they showed a clear modulation at the beginning or end of the audio stimulus.
According to the neuronal activities during vocalization and breathing, and their responses to auditory stimuli, these recorded brainstem cells were separated into the following different types: (1) neurons that were only activated during vocalization (CALL), (2) neurons that only responded to auditory stimuli (AUDITORY), (3) neurons that were only time locked to breathing (BREATHING), (4) neurons that were activated during vocalization and time locked to breathing (CALL + BREATHING), (5) neurons that were activated during vocalization and responded to auditory stimuli (CALL + AUDITORY), (6) neurons that were time locked to breathing and responded to auditory stimuli (BREATHING + AUDITORY), and (7) neurons that were activated during vocalization and time locked to breathing and responded to auditory stimuli (CALL + BREATHING + AUDITORY).
To determine the range of latencies of call responsive neurons at the onset and offset of the calls, we computed the time where neurons showed the first increase of activity over 2.5 SDs before the beginning and end of the calls, respectively. The intervals reported correspond to the 25th and 75th percentiles of the distributions obtained.
Finally, for population and call intensity analysis, all call responsive units were analyzed together; that is, we included call-related units that could also show auditory or breathing responses.
Experimental design and statistical analysis
A total of nine male rats were used for the c-fos study, in which the rats were randomly classified into three experimental groups of three each—vocalized group, nonvocalized group, and naive group. c-fos data analysis was performed by using GraphPad Prism 7 software. For comparisons among these three groups, one-way ANOVA with Tukey's multiple-comparisons test was used.
A total of 10 male rats were used of the Neuropixels mapping experiments. All spike analyses were performed using MATLAB (MathWorks) custom scripts. To visually evaluate call-intensity-related activity, call intensity within each recording session was z-scored, and the firing rate of call-related units was estimated locally using a 1 s sliding window (where only spikes happening in the past were considered). Neural activity aligned to the beginning of each call was plotted and sorted by call intensity (see Fig. 6C). To assess the relation between neural activity and call intensity, we estimated the Pearson correlation between call intensity and the firing rate in the 25 ms window before the call (see Fig. 6D). This procedure was repeated for all call-related neurons in facial nucleus (FN); VoPaRT; intermediate reticular nucleus (IRt); medullary reticular nucleus, dorsal part (MdDP); and NRA, where we obtained distributions of r2 for each brain region. A Kruskal–Wallis test was performed using brain region as factor, and Wilcoxon signed-rank test was used to compare the median of the groups (see Fig. 6E). Highly responsive cells were defined as neurons showing a significant (p < 0.05) r2 higher than 0.1 (see Fig. 6F). We analyzed the relation between the firing rate before call onset and during the call also using Pearson's correlation (see Fig. 6G).
Results
Approach
The pattern generator for mammalian vocalizations is known to be situated in the brainstem, but we do not know the exact identity of the brainstem circuits involved (Hage, 2010). We sought to address this issue using complementary techniques that can resolve circuits with cellular resolution but are still able to sample large parts of the brainstem. To this end, we first evoked calls in the rats by electrically stimulating the midbrain PAG and performed large-scale recordings in different brainstem areas with Neuropixels probes (Fig. 1A). In each experiment, we also examined whether the recorded neurons of these brainstem nuclei respond to external acoustic stimuli such as recorded natural USVs, sine wave pure tones, and white noise (Fig. 1A; see above, Materials and Methods). The calls and respiration of the animals were monitored during the whole experiment. Using unilateral stimulation of PAG, we found two major classes of calls, modulated high-frequency USV (>35 kHz) and low-frequency USV (<35 kHz; Fig. 1B). Figure 1C shows an example cell of the NRA that exhibited increased neuronal activity just before and during vocalization.
Large-scale mapping vocalization-evoked brainstem activity using Neuropixels probes and c-fos expression. A, Schematic of vocalization mapping by electrical stimulation in the PAG and Neuropixels recordings from brainstem neurons of an anesthetized rat. PAG stimulation-elicited vocalizations and respiration were simultaneously recorded throughout the experimental session. Various acoustic stimuli were presented and repeated 20 times by an ultrasonic dynamic speaker. µA, microampere; USV, ultrasonic vocalization. B, Representative spectrogram of ultrasonic vocalizations. Left, Modulated high-frequency vocalization; sounds with a fundamental frequency >35 kHz were classified as high-frequency vocalizations. Right, A low-frequency vocalization, for example, fundamental frequency <35 kHz. C, Vocalization-related respiration and neuronal activity in the NRA in the caudal medulla. Top, Sonogram of a low-frequency call, evoked by microstimulation in PAG. Bottom, A well-isolated unit in the NRA shows vocalization-related excitation during vocalization. D, Schematic showing stimulating electrodes placed in the PAG in the c-fos experiments. E, Schematic of the c-fos experimental procedure. F, Representative immunofluorescence section of the rat NRA stained with the neuronal nuclear marker NeuN (green) and the neuronal activation marker c-fos (red), and the merged confocal picture in the vocalized animal group.
Overview of vocalization-related activity and vocalization-related c-fos expression
In a second approach, we made use of an unbiased c-fos mapping method for further identifying brain areas that may elicit calls in rats (Fig. 1D–F). In the experimental group of rats, USVs were induced by brief high-frequency (100 Hz) train stimulation at a current intensity of 100 µA for half an hour (Fig. 1D; see above, Materials and Methods). The rats were perfused 90 min after the end of stimulation, and c-fos and NeuN expression was examined in the brainstem with immunohistochemistry (Fig. 1E,F). In our experimental design, two control groups of rats were used, the nonvocalized animal group, in which animals were treated with same stimulation parameters but no calls were produced during stimulation, and the naive group, in which animals were allowed to stay in their own cage (Fig. 1E).
An overview of vocalization-related activity as detected by Neuropixels recordings and c-fos mapping is given in Figure 2. To study the localization of the vocal pattern generator in the mammalian brain, we recorded extracellular spikes of the neurons in different brainstem areas during vocalization and the presentation of external auditory stimuli, using the advanced high-density extracellular probe, Neuropixels (Fig. 2A1; Jun et al., 2017). We assessed vocalization and auditory and respiratory responses in either the anterior (Fig. 2A1,A2) or posterior (Fig. 2A2,A3) brainstem regions in head-fixed anesthetized rats. In our anterior brainstem-recording experiments, the spiking activity in FN; gigantocellular reticular nucleus (Gi); IRt, α part (IRtA); superior olivary nuclei; intermediate reticular nucleus, α part (PCRtA); vocalization region of the parvicellular reticular formation (VoPart), and other anterior brainstem areas with few units (other) were recorded with Neuropixels probes in the left hemisphere (Fig. 2A1,A2). In the posterior brainstem-recording experiments, the neuronal activity in IRt; MdD; medullary reticular nucleus, ventral part (MdV); NRA; spinocerebellar tract (SCT); and spinal trigeminal nucleus, caudal part (Sp5C) were recorded (Fig. 2A2,A3). A total of 3345 units in the anterior and posterior brainstem were analyzed for their neuronal activity in 10 rats (Fig. 2B). Analyzed neurons were located in the facial nucleus (326 units), Gi (104 units), IRtA (197 units), superior olivary nuclei (117 units), PCRtA (116 units), VoPart (206 units), IRt (81 units), MdD (289 units), MdV (54 units), NRA (163 units), SCT (245 units), and Sp5C (77 units; Fig. 2A2,B). We classified neurons as call responsive, breathing responsive, or auditory responsive, and assigned them to one of the regions according to the location of the channel where they were detected. We decided to label neurons as call responsive only if they showed call-related activity but neither breathing or auditory activity. We use the same criterion for classifying neurons as breathing or auditory responsive (see above, Materials and Methods). It is worth noting that a high percentage of call-correlated neurons was found in facial nucleus (11.7%) and most posterior brainstem nuclei such as Sp5C (15.6%), MdD (15.6%), IRt (16.0%), and NRA (13.5%; Fig. 2B).
Overview of vocalization-related activity detected by large-scale recording and c-fos expression. A1, Schematic representation of a penetration in a coronal section of the rat anterior brainstem. A2, A drawing (horizontal view) of the rat brainstem modified from a rat brain atlas (Paxinos and Watson, 2006). Three different areas of the posterior brainstem are colored differentially. A3, Schematic representation of a penetration in a coronal section of the rat posterior brainstem. B, Bar graphs illustrating the proportion of seven different types of responses of units in each brainstem region. A total of 3345 well-isolated single units were included in the data shown here. C, Quantitative analyses of the ratio between total number of double-labeled neurons (c-fos/NeuN-positive cells) and total number of neurons (NeuN-positive cells) in each experimental group. The density of activated neurons was evaluated as none if no double-labeled neurons were found in a brain region, low (+) if 0–20% double-labeled neurons were detected in a brain region, modest (++) if 20–40% double-labeled neurons were detected in a brain region, and high (+++) if 40–60% double-labeled neurons were detected in a brain region. 7n, Facial nerve.
We also used c-fos and NeuN staining to map the brainstem areas that were active during vocalization (Fig. 2C). The percentage of c-fos-immunopositive cells in a specific area was calculated as the number of total double-stained cells to the number of total NeuN-expressing cells in this region. Figure 2C summarizes the relative ratio of c-fos-positive cells in different brainstem regions in each group. Electrical stimulation in the PAG increased the level of c-fos expression in most listed brainstem areas relative to home-cage control (Fig. 2C). The vocalized group has a significantly higher expression level of c-fos in the NRA than the animals of the home-cage group and nonvocalized group (p = 0.00000068, F(2,24) = 27.21). There was no significant difference in c-fos expression for other brainstem areas listed between vocalized group and two control groups.
A synopsis of the data presented in Figure 2A–C, suggests the following conclusions. (1) Vocalization-related brainstem activity is sparse. (2) The clearest activity patterns suggestive of vocal pattern generation (strong unit activity and c-fos expression) was observed in the NRA; as the NRA is a long-standing candidate structure for vocal pattern generation (Holstege, 1989), this finding instills confidence in our overall approach. (3) A range of structures may have shown incidental activations along with vocalizations, such as the superior olivary complex (putative auditory activation) or the trigeminal principalis nucleus (putative somatosensory activation by call-associated whisker movements). (4) Anterior brainstem structure showed little activation, although there is good evidence for a role of the structures in vocal pattern generation (Hage and Jürgens, 2006a; Hage, 2010; Hartmann and Brecht, 2020); the reasons for this negative finding are unclear. (5) A number of structures close to the nucleus retroambiguus also showed vocalization-related activation, although not as clearly the NRA itself; the activity patterns of the structure are discussed in detail in the figure legends.
Vocalization-related activity in the nucleus retroambiguus
We found vocalization-related excitation and inhibition of neurons from the posterior brainstem. During each recording session, rat produced between 44 and 1097 vocalizations (median, 290 calls), from 20 to 60 kHz, although most calls were between 20 and 35 kHz. Rat produced sequences of one to seven calls once PAG was stimulated. To avoid any confounding factor introduced by the stimulation, only calls occurring after the stimulation period were considered for the analysis (see above, Materials and Methods). Call-related activity of an NRA neuron can be found in Figure 3A–D. DiI labeling of the Neuropixels probe allowed to assign the recording site unambiguously to the NRA (Fig. 3A). This neuron showed a clear excitation that could be observed even on the raw signal (Fig. 3B) and also across several vocalizations (Fig. 3C,D; 464 outside stimulation calls, median duration 242.9 ms). This neuron started firing around 16 ms before the beginning of the call (median latency, 16 ms), and stopped firing around 75 ms before the end (median latency, 74.5 ms) of the call (Fig. 3D). Similarly, we also found NRA neurons that were inhibited during calls; an example is shown in Figure 3E–H. The respective recording sites were located in the NRA again by dye tracing of the probe (Fig. 3E). The neural response could be observed both at the raw signal (Fig. 3F) or during multiple vocalizations (Fig. 3G,H; 254 outside stimulation calls, median duration 269 ms). In this case the neuron stopped firing ∼72 ms before the beginning of the call (median latency, 72.8 ms; Fig. 3G) and started firing again either before or after the end of the call (Fig. 3H). Note that both excitatory and inhibitory activity was sustained during the calls, a response pattern characteristic of the NRA.
A vocalization-activated and a vocalization-inhibited nucleus retroambiguus neuron. A, Left, Neuropixels fluorescent track and superimposed probe (black-white dashed line). Red circle represents assigned channel for the units depicted in B–D. Right, Reference brainstem coronal view as in Figure 2A3. B, Vocalization-excited NRA neuron. Top, Ultrasonic vocalization. Bottom, Raw recording trace; blue sorted waveforms from the unit. C, Neural responses of the excited neuron triggered to vocalization onset. Trials are sorted by call length. D, Neural responses of the excited neuron triggered to vocalization offset. Trials are sorted by call length. E, Left, Neuropixels fluorescent track and superimposed probe (black-white dashed line). Red circle represents assigned channel for the units depicted in F–H. Right, Reference brainstem coronal view as in Figure 2A3. F, Vocalization-inhibited NRA neuron. Conventions as in B. G, Neural responses of the inhibited neuron triggered to vocalization onset. Trials are sorted by call length. H, Neural responses of the inhibited neuron triggered to vocalization offset. Trials are sorted by call length.
Vocalization-related activity in the posterior brainstem structures close to nucleus retroambiguus
We also studied IRt and MdD activity, both regions surrounding the NRA (Fig. 2A3). A representative IRt neuron is depicted in Figure 4A–D. The neuron could be assigned to the diagonal reticular band (around the NRA) by fluorescence tracing (Fig. 4A). Similar to some NRA neurons, IRt showed a sustained excitatory response, observable both at the raw (Fig. 4B) and across several vocalizations (Fig. 4C,D; 464 outside stimulation calls, median duration 242.9 ms). This neuron started firing ∼39 ms before the beginning of the call (median latency, 39 ms; Fig. 4C) and stopped firing ∼50 ms before the end (median latency, 50; Fig. 4D). The more dorsally and posteriorly located MdD cells (Fig. 4E) showed completely different pattern of responses (Fig. 4E,F); hence, we name this region posterior level of MdD (MdDP). As can be observed in the example, these neurons showed a remarkable sparse response at the beginning (Fig. 4G) and end (Fig. 4H) of the calls that was consistent across the different calls (393 outside stimulation calls, median duration, 268 ms).
Response patterns of IRt and MdDP neurons. A, Left, Neuropixels fluorescent track and superimposed probe (black-white dashed line). Green circle represents assigned channel for the units depicted in B–D. Right, Reference brainstem coronal view as in Figure 2A3. B, Mainly activated IRt neurons were observed. Top, Ultrasonic vocalization. Bottom, Raw recording trace; blue sorted waveforms from the unit. C, Neural responses of the IRt neuron triggered to vocalization onset. Trials are sorted by call length. D, Neural responses of the IRt neuron triggered to vocalization offset. Trials are sorted by call length. E, Left, Neuropixels fluorescent track and superimposed probe (black-white dashed line). Purple circle represents assigned channel for the units depicted in F–H. Right, Reference brainstem coronal view. F, Representative activated MdDP neuron. Conventions as in B. G, Neural responses of the MdDP neuron triggered to vocalization onset. Trials are sorted by call length. H, Neural responses of the MdDP neuron triggered to vocalization offset. Trials are sorted by call length.
Population responses of posterior brainstem structures
The pattern of responses described previously were also observable at the population level and even after grouping tens of neurons (Fig. 5A–F). For the population analysis, we included neurons that could also be breathing or auditory responsive to capture the whole dynamic of this region during vocalization. The neurons on IRt were activated between 20 and 150 ms before the onset of the call (Fig. 5A) and were mostly excitatory (25% inhibitory and 75% excitatory responses). The responses were mostly sustained, as most of the neurons fired almost until the end of the call (Fig. 5B). NRA neurons showed a consistent pattern across the almost 50 units recorded (Fig. 5C,D), whereas most units showed an excitatory or inhibitory sustained response. The latencies of the excitatory neurons were between 0 and 90 ms, whereas the inhibitory ones were beginning to decrease their activities between 30 and 300 ms before the beginning of the call. As on IRt, the majority of neurons were excitatory (25% inhibitory and 75% excitatory responses). MdDP cells, on the other hand, showed a transient activation that could occur only at the beginning of the call (onset response) or both at the beginning and at the end (onset-offset response). Neurons start firing between 5.0 and 23.8 ms before the beginning of the calls and got activated again (in case of onset-offset response) between 43.3 ms before up to 22.8 ms after the end of the call. These neurons were exclusively excitatory.
Population responses from NRA, IRt, and MdDP. A, IRt population onset responses. Top, Single neuron responses with z-scored firing rate values. Bottom, Mean response using raw firing rate values. Red, activated neurons; blue, inhibited neurons. B, IRt population offset responses. Top, Single neuron responses with z-scored firing rate values. Bottom, Mean response using raw firing rate values. Red, activated neurons; blue, inhibited neurons. C, NRA population onset responses. Same as A. D, NRA population offset responses. Same as B. E, MdDP population onset responses. Same as A. F, MdDP population offset responses. Same as B.
Activity-related vocalization loudness/amplitude
Despite the recognized role of NRA on vocal control, very few experiments have assessed the influence of NRA on call features. It has been hypothesized, for example, that NRA may participate in the regulation of call intensity, given it can control respiratory behavior during vocalizations (Subramanian and Holstege, 2009; Hage, 2010). We sought to test this hypothesis by analyzing NRA firing-rate modulation before and during calls. Some NRA neurons increased their firing rate during high-intensity calls (Fig. 6A,B). The effect started around 30 ms before the onset of the calls, and it was proportional to call intensity (Fig. 6C). In fact, in some units the firing rate in the 25 ms window presiding the initiation of the call was significantly correlated with the subsequent call intensity (Fig. 6D), suggesting that the call intensity was set before the beginning of the call by some NRA neurons. We further tested whether the intensity control was a unique feature of NRA or involved other call-related brainstem regions. We computed r2 values for all call-modulated units in several call-modulated brainstem regions, that is, facial nucleus, IRt, MdDP, VoPaRt. and NRA (Fig. 6E). NRA showed a higher r2 (Kruskal–Wallis test, χ2 = 26.5, df = 4, p = 2.5 10−5), with an r2 significantly higher than MdDP (Z = 2.80, rank sum = 3726, p = 0.005), facial nucleus (Z = 4.3657, rank sum = 542, p = 2.27 10−5) and VoPaRt (Z = 3.18, rank sum = 2284, p = 0.0015). We didn't find significant differences between NRA and IRt (Z = 0.5868, p = 0.5573). These results were still significant after correcting for multiple comparison (four comparisons, Bonferroni corrected α = 0.0125). Furthermore, IRt showed only slightly higher values than VoPaRt (Z = 1.97, p = 0.05) and facial nucleus (z = 2.2155, p = 0.03) but these results are not significant after controlling for multiple comparisons. Together, these observations suggest that NRA is the main responsible structure controlling call intensity. In fact, high correlated units with r2 > 0.1 were mainly observed in NRA (Fig. 6F), comprising 18.18% of all NRA call responsive units. Finally, because the firing rate before call onset correlates with the firing rate during the calls (Fig. 6G; r2 = 0.9435, p = 0.001), we also tested whether NRA also participated in sustaining call intensity during calls. We repeated the same analysis now using the mean rate during calls instead of the firing rate during the 25 ms window before call onset (data not shown). Through this procedure we found the same statistical differences between brainstem regions, that is, a higher r2 in NRA call responsive units when compared with other brainstem regions.
Nucleus retroambiguus discharges related to call intensity. A, Top, Waveform representation of a loud call from a recording dataset. Bottom, Discharges of a well-isolated unit in the NRA showing vocalization-related excitation; sorted spike waveforms are depicted in blue. B, Top, Waveform representation of a low-amplitude call from the same recording dataset shown in A. Bottom, Discharges of the same unit in the NRA as shown in A. C, Raster plot of NRA neurons, where calls are sorted by intensity (red window reflects the time window used to compute the correlation on B). Upper blue line represents the example in A, and lower blue line represents the example in B. Red line represents the beginning of the call. D, Scatter plot of firing rate and call intensity. Red line represents linear regression between the two variables. r2 = 0.2745. E, Population r2. Box plots of the cumulated r2 by region. Crosses represent outliers. *p < 0.05, ***p < 0.001. F, Percentage of neurons per region with significant r2 > 0.1; NRA, 18% (10 of 55 units; MdDP, 1.6% (1 of 61 units). G, Scatter plot of firing rate before and during the call. r2 = 0.9435, p = 0.001. Fr, firing rate
Discussion
Summary
We performed a large-scale mapping of vocalization-related immediate early gene expression and neural activity in the brainstem. The results of the two approaches support the notion that nucleus retroambiguus is involved in vocal production in rats. Nucleus retroambiguus cells are tonically activated or inhibited before and during calls, and a subset of such cells discharges in a tight relation to call intensity. More important, NRA appears to play a key role in vocal intensity control. A set of sites around the nucleus retroambiguus shows functionally diverse vocalization-related neural activity. The most striking of such discharges were call onset/call offset discharges in MdDP. For reasons not entirely clear, we found little evidence for vocalization-related activity in the anterior brainstem.
The posterior brainstem is involved in vocal pattern generation in rats
Several anatomic studies have shown that the NRA and its surrounding brain regions receive direct projections from midbrain PAG (Holstege, 1989; Oka et al., 2008; Tschida et al., 2019). Bilateral neurochemical lesions in the medullary reticular formation including NRA abolish PAG stimulation-produced calls (Shiba et al., 1997). In agreement with this, other data demonstrate that electrical and chemical stimulation of the NRA area was sufficient to cause abnormal vocalization in cat (Zhang et al., 1992, 1995) and rat (Hartmann and Brecht, 2020). Further supporting evidence from a recent study found that unilateral optogenetic stimulation of the pathway from the PAG to the NRA was able to produce vocalizations in mice (Tschida et al., 2019), whereas another piece of work suggests that chemical stimulation of NRA region barely yielded vocalization in guinea pig (Sugiyama et al., 2010).
Surprisingly, in a systematic literature search, we found only one study that performed single-unit recordings in the posterior brainstem of squirrel monkeys and described call-correlated neurons in the NRA and central reticular formation (Lüthe et al., 2000). Our data greatly extend the findings of Lüthe et al. (2000), and we also find 25% vocalization correlated NRA cells, which decreased their discharge rate in relation to call production. Overall, we find it odd that (our work included) that only two studies recorded call-correlated neurons in the NRA, presumably one of the important brain structures in mammalian vocal pattern generation, when at the same time thousands of studies investigate high-level vocal motor control structures such as Broca's area. We conclude that we need broader efforts to understand the posterior brainstem and cellular mechanisms of mammalian vocal pattern generation.
A diversity of vocalization-related posterior brainstem structures
On the basis of the results mentioned above, we have reasons to believe that different brainstem areas in the caudal medulla may play distinct roles in the vocal pattern-generating process (Fig. 7). We have further specified the vocal functions of these brainstem regions (Fig. 7A). We found the firing patterns of some NRA neurons were tonically increased or decreased just before and during vocalization; these data in electrophysiology, together with our c-fos results, suggest that NRA cells may participate in driving animal vocalizations (Zhang et al., 1992, 1995; Shiba et al., 1997). Similarly, as part of our recorded IRt cells showed sustained increased neuronal activity beginning very early before vocal onset and lasting through part of vocalization, it is highly likely that IRt would be involved in call preparation of animal. One notable aspect of our findings is that some MdDP neurons in our study showed increased neuronal activity immediately before and/or after vocalization end; it can be deduced that these neurons may participate in the call beginning and/or termination. Intriguingly, it seems that in Lüthe et al.'s (2000) squirrel monkey study, all the brain regions mentioned above exhibited more diverse reaction types in vocalization-related neurons, whereas our results suggest that each brain area in rats tends to be involved in call production in its own way with a unique firing pattern. It is thus reasonable to assume that the more sophisticated vocalizations produced by squirrel monkeys may rely on more differentiation of the neuronal activity in the brainstem vocal pattern generators. Additionally, at the population level, NRA an IRT are activated earlier than the MdDP (Fig. 7B), which further supports our proposal that these different nuclei in the posterior brainstem appear to be accountable for distinct aspects of rodent vocalizations. Finally, it needs be noted that compared with other mammalian species (e.g., bats, cetaceans, and primates), call production in rats appears to be predominantly laryngeal, with only minor involvement of the articulators. In light of this and previous evidence for a role of cranial motorneuron pools in controlling more featured calls in other mammalian species (Hage and Jürgens, 2006b), our findings might be a specific phenomenon in rats that cannot be easily generalized to other species.
Summary figure. A, Overview of the possible role of different caudal medulla areas involved in vocalization, based on present data. B, Population average onset responses of NRA, IRt, and MdDP.
Absence of strong vocalization-related activity in anterior brainstem structures
It has been proposed that the brain structures in the anterior brainstem are also dedicated to pattern generation in vocal production (Hage and Jürgens, 2006a; Hage, 2010; Hartmann and Brecht, 2020). Jürgens (2000) showed that frequency-modulated calls were affected by pharmacological inactivation of the anterior pontine brainstem. Similarly, local cooling pointed to an involvement of the anterior pontine brainstem in frequency modulation (Hartmann and Brecht, 2020). Furthermore, abnormal calls could also be evoked by stimulation in the anterior brainstem (Siebert and Jürgens, 2003). These identified anterior areas include the parvocellular pontine reticular formation above the superior olivary complex, the reticular formation around the facial nucleus, and a subsection of the parvicellular region of the reticular formation. However, our current electrophysiology and c-fos data of these areas do not show strong correlations with rat vocalizations. The meaning of this negative result is not entirely clear because there are multiple lines of evidence for a participation of the anterior brainstem in vocal pattern generation. It may be that these structures are very small or were not robustly activated under our experimental conditions.
Vocal intensity control
The vocal intensity is an important component of acoustic communication that is carefully controlled by the brain. Controlling voice intensity is a natural ability mainly reliant on laryngeal adjustment and respiratory effort (Jürgens, 2002; Makiyama et al., 2005; Subramanian and Holstege, 2009; Hage, 2010). Surprisingly, despite its functional importance, where and how the voice intensity is controlled in the mammalian brain and the neuronal mechanisms underlying it remain poorly understood. According to our knowledge, there are only two pieces of monkey literature reporting that the activity of some units in the midbrain PAG was positively correlated with vocal loudness (Larson and Kistler, 1986; Larson, 1991). In other words, some PAG neurons increased their firing rate for increased vocalization intensity. However, these neurons found in the PAG only showed burst firing just before vocalization onset without reflecting further detailed information about call patterns. Several lines of evidence suggest that PAG does not directly participate in vocal motor coordination in a strict sense but rather plays a crucial role in gating downstream vocal pattern-generating networks (Gruber-Dujardin, 2010; Tschida et al., 2019). In our present study, with the high-density Neuropixels recordings in the brainstem, we first reported that 18% of NRA call responsive neurons showed a tonic firing rate that is highly correlated with the intensity of vocalization. It is noteworthy that this significant correlation was discovered before and during vocalizations. However, the vocal intensity was not found to be strongly correlated with the firing rate of units in other recorded brainstem structures. Furthermore, the NRA is known to be densely and broadly connected with the brain areas recruited in vocal pattern generation. Together, our results suggest that the PAG–NRA motoneuronal pathway plays a pivotal role in voice intensity control and may serve as a classical pathway for call production in rats.
Conclusions
Our work points to an involvement of the posterior brainstem in vocal pattern generation in rats. We confirm that tonic excitatory and inhibitory discharges in the nucleus retroambiguus might be at the heart of this process, and this nucleus appears to also play a key role in vocal intensity control. More than that, we show elaborate vocalization-related activity in a set of structures around the nucleus retroambiguus. The most notable such activity pattern is the call-onset/call-offset discharges in the MdDP. Because our experiments were performed in the anesthetized rats, they cannot offer insights into mood-dependence of rat vocalizations. Accordingly, future research is needed to characterize the neuronal activity of the NRA, IRt, and MdDP during naturally induced calls in freely moving animals. It also needs be noted that as more low-frequency than high-frequency calls were evoked under our experimental conditions, our conclusions might have been biased by the call samples elicited.
Footnotes
This study was supported by the Bernstein Center for Computational Neuroscience Berlin, Humboldt-Universität zu Berlin, Deutsche Forschungsgemeinschaft (German Research Foundation) Grant 501100001659 under Germany's Excellence Strategy (EXC-2049 390688087), Bundesministerium Bildung und Forschung Grant 501100002347, a BrainPlay-ERC-Synergy grant, and Deutsche Forschungsgemeinschaft Grant SFB 1315-327654276. We thank Undine Schneeweiss and Tanja Wölk for technical assistance.
The authors declare no competing financial interests.
- Correspondence should be addressed to Michael Brecht at michael.brecht{at}bccn-berlin.de
