Intensity defines one fundamental aspect of sensory information and is specifically represented in each sensory modality. Interestingly, only in the central auditory system are intensity-selective neurons evolved. These neurons are characterized by nonmonotonic response-level functions. The synaptic circuitry mechanisms underlying the generation of intensity selectivity from nonselective auditory nerve inputs remain largely unclear. Here, we performed in vivo whole-cell recordings from pyramidal neurons in the rat dorsal cochlear nucleus (DCN), where intensity selectivity first emerges along the auditory neuraxis. Our results revealed that intensity-selective cells received fast-saturating excitation but slow-saturating inhibition with intensity increments, whereas in intensity-nonselective cells excitation and inhibition were similarly slow-saturating. The differential intensity tuning profiles of the monotonic excitation and inhibition qualitatively determined the intensity selectivity of output responses. In addition, the selectivity was further strengthened by significantly lower excitation/inhibition ratios at high-intensity levels compared with intensity-nonselective neurons. Our results demonstrate that intensity selectivity in the DCN is generated by extracting the difference between tuning profiles of nonselective excitatory and inhibitory inputs, which we propose can be achieved through a differential circuit mediated by feedforward inhibition.
In the sensory system of each modality, stimulus intensity must be represented as a fundamental aspect of sensory input. Most sensory neurons have monotonic spike rate versus intensity level functions (i.e., they encode intensity by increasing spike rate as intensity is increased). In the central auditory system, however, another strategy is used by intensity-selective neurons. The spike rate of these neurons initially increases and peaks as sound intensity is increased, and then decreases as sound intensity is further increased, resulting in a nonmonotonic response-level function (Phillips and Kelly, 1989; Phillips et al., 1995). Intensity-selective neurons are observed at every stage of the ascending central auditory pathway (Greenwood and Maruyama, 1965; Brugge et al., 1969; Aitkin and Webster, 1972; Young and Brownell, 1976; Rouiller et al., 1983; Aitkin, 1991; Schreiner et al., 1992; Kuwabara and Suga, 1993; Phillips et al., 1995). In the rat, the number of intensity-selective cortical neurons was found to increase in animals trained to perform an intensity discrimination task, suggesting that intensity-selective neurons may be required for the precise coding of sound loudness (Polley et al., 2004, 2006).
The neural basis for intensity selectivity remains not well understood. Previous studies have mostly focused on late stages of the auditory neuraxis, in particular the cortex. There, several synaptic mechanisms have been proposed for sharpening intensity selectivity or even generating intensity selectivity de novo (Shamma, 1985; Ojima and Murakami, 2002; Sutter and Loftus, 2003; Wu et al., 2006; Tan et al., 2007; de la Rocha et al., 2008). Intensity selectivity in the cortex may be partially inherited from intensity-selective outputs of previous processing stages, as evidenced by the intensity-tuned excitatory inputs to cortical neurons (Wu et al., 2006; Tan et al., 2007). Along the auditory neuraxis, intensity-selective neurons are first observed in the dorsal cochlear nucleus (DCN) (Young and Brownell, 1976). Because the ascending input to cochlear nuclei is provided by the auditory nerve (AN), which provides nonintensity-tuned responses (Kiang et al., 1965; Sachs and Abbas, 1974), it is interesting to examine DCN neurons to understand how intensity selectivity is initially generated (Fig. 1A). Based on spike response properties of DCN neurons, it has been proposed that the generation of intensity selectivity in the DCN may be attributed to a lower intensity threshold of excitation than inhibition (Voigt and Young, 1980; Nelken and Young, 1994; Davis et al., 1996; Spirou et al., 1999; Young and Davis, 2001). However, because sound-evoked excitatory and inhibitory synaptic inputs to individual DCN neurons have not been experimentally elucidated, this theory remains to be tested.
In the current study, we performed in vivo whole-cell recordings from DCN neurons to directly examine the synaptic mechanisms underlying intensity tuning. By isolating the excitatory and inhibitory inputs evoked by the same sound stimuli, we found that the difference in intensity threshold between excitation and inhibition alone is not sufficient for explaining intensity selectivity in the DCN. On the other hand, the differential intensity tuning profiles of excitation and inhibition is key to generating intensity selectivity.
Materials and Methods
Animal preparation and mapping of the DCN
All experimental procedures used in this study were approved under the Animal Care and Use Committee at the University of Southern California. Experiments were performed in a sound-attenuation booth (Acoustic Systems). Female Sprague Dawley rats (∼3 months old and weighing 250–300 g) were anesthetized with ketamine and xylazine (ketamine, 60 mg/kg; xylazine, 8 mg/kg; i.p.). The body temperature was maintained at 37.5°C by a feedback heating system (Harvard Apparatus). The animal was positioned with the left ear facing a calibrated free-field speaker (Vifa), and with a sound-attenuating plug placed in the right ear. After opening the left part of the occipital bone, part of the cerebellum was aspirated to expose the left DCN. Artificial CSF (ACSF; in mm: 124 NaCl, 1.2 NaH2PO4, 2.5 KCl, 25 NaHCO3, 20 Glucose, 2 CaCl2, 1 MgCl2) was applied when necessary during the surgery and the experiment to clean the DCN surface. To map the tonotopy, multiunit spike responses were recorded with a parylene-coated tungsten microelectrode (2 MΩ, FHC) at 100–300 μm below the DCN surface. Signals were amplified, bandpass filtered between 0.3 and 10 kHz (Plexon), and then thresholded to extract the timing of each spike using custom-made software (LabView, National Instrument). To obtain tonal receptive fields, pure tones (0.5–64 kHz at 0.1 octave intervals, 50 ms duration, 3 ms ramp) at eight 10 dB spaced sound intensities (sound pressure level, 0–70 dB) were delivered in a pseudorandom sequence. The time interval and intensity difference between two sequential tones were set at 0.5–1 s and ≤30 dB, respectively, as to minimize cross-interactions between the two stimuli. Characteristic frequency (CF), the frequency at which neurons reliably responded to tones with the minimum intensity, was determined for each recording site. As illustrated in Figure 1B, a clear low-frequency-to-high-frequency gradient along the lateral–medial axis can be observed in the rat DCN, similar as described previously (Yajima and Hayashi, 1989; Kaltenbach and Lazor, 1991).
In vivo whole-cell and loose-patch recordings
Whole-cell recordings (Wu et al., 2008; Sun et al., 2010; Zhou et al., 2010, 2012) were made with an Axopatch 200B amplifier (Molecular Devices). Patch pipettes made from borosilicate glass capillaries (Kimax) had an impedance of 3–4 MΩ. Pipettes contained a potassium-based solution composed of (in mm) the following: 130 K-gluconate, 4 MgATP, 0.3 GTP, 8 phosphocreatine, 10 HEPES, 11 EGTA, 5 KCl, 1 CaCl2, 0.2 fluorescein dextran, pH 7.3. The patch pipette was lowered into the DCN at an angle of ∼85°. The brainstem was then covered with 3.5% agar prepared in warm ACSF. Whole-cell capacitance was fully compensated and the initial series resistance (<30 MΩ) was compensated for 40–60%. Signals were filtered at 5 kHz and sampled at 10 kHz. Only cells with resting membrane potential lower than −50 mV were studied. A 10 mV junction potential was corrected. Excitatory and inhibitory currents were recorded by clamping the cell at −70 and 0 mV respectively. The morphology of some recorded cells was reconstructed by confocal fluorescence imaging of the cells filled with fluorescein dextran (0.2 mm, Invitrogen) in fixed brainstem sections.
In this study, we specifically focused on fusiform cells in the pyramidal cell layer. We controlled our recording depths according to the travel distance of the pipette tip. Because there is no pial tissue covering the DCN and pipette penetrations caused little dimpling of the DCN surface, there was relatively precise correspondence between travel distance and recording depth. Also, for consistent control of recording depth, we chose to record from the middle frequency representation region because the DCN surface in this region is parallel to the horizontal plane. Our reported pyramidal neurons were recorded at 100–250 μm depths below the surface, corresponding to the pyramidal cell layer (Mugnaini et al., 1980; Wouterlood and Mugnaini, 1984). Due to the depth control, giant cells, another type of DCN projection neurons, were presumably rarely encountered. The pyramidal cell type was confirmed by completely constructed dendritic morphologies in four cells (Fig. 2A; Hancock and Voigt, 2002), and the somata location in the pyramidal cell layer was also confirmed by six partially recovered cells (data not shown). In addition, the response properties of the recorded pyramidal cells were consistent with those previously reported (Evans and Nelson, 1973; Young and Brownell, 1976).
For loose-patch recordings, pipettes were filled with ACSF. Loose seal (100–200 MΩ) was made from neurons, allowing spikes only from the patched cell to be recorded. Recording was under the voltage-clamp mode. A command voltage was applied to achieve a zero baseline current. Signals were filtered at 0.3–10 kHz and sampled at 20 kHz. For recordings from inhibitory neurons, pipettes of smaller tip openings (impedance, ∼10 MΩ) were used to overcome a bias toward larger cells (Wu et al., 2008). We specifically searched for a type of inhibitory neurons, type II cells, at 250–700 μm depths (Mugnaini et al., 1980; Rhode, 1999) using tone and noise stimuli. We identified type II cells by their lack of spontaneous activity and absence of responses to broadband noise stimuli at 60 dB sound pressure level. Cells with complex spikes, potentially cartwheel cells (Portfors and Roberts, 2007), were excluded from the analysis in this study.
Spikes recorded in the loose-patch configuration could be detected without ambiguity because their amplitudes were normally >100 pA, while the baseline fluctuation was <5 pA. Tone-driven spikes were counted within a 4–60 ms time window after the tone onset. The average spontaneous spike number within the same time duration was subtracted from the number of evoked spikes. Intensity threshold was defined at the tip of the tonal receptive field (TRF) by the minimal intensity level at which CF tones reliably evoked responses. The evoked response should be higher than the average baseline activity by 3 SDs of the baseline fluctuation. All the synaptic response traces evoked by the same test stimulus were averaged. The peak amplitude of the average trace was determined by averaging pixels within a ±5 ms time window centered on the maximum value. The integrated conductance was calculated by deriving the synaptic conductance and then integrating conductance within 4–60 ms after the tone onset.
Excitatory synaptic conductance and inhibitory synaptic conductance were derived (Borg-Graham et al., 1998; Anderson et al., 2000; Zhang et al., 2003; Wu et al., 2008; Sun et al., 2010) according to ΔI = Ge(V − Ee) + Gi(V − Ei). ΔI is the amplitude of the synaptic current response at any time point after subtraction of the baseline current; Ge is the excitatory synaptic conductance; Gi is the inhibitory synaptic conductance; V is the holding voltage; and Ee (0 mV) and Ei (−70 mV) are the reversal potentials. The clamping voltage V was corrected from the applied holding voltage (Vh): V = Vh − Rs*I, where Rs is the effective series resistance. By holding the recorded cell at two different voltages (the reversal potentials for excitatory and inhibitory currents respectively), Ge and Gi were resolved from the equation.
Derived membrane potential responses.
The membrane potential change caused by synaptic conductances could be derived according to an integrate-and-fire neuron model (Somers et al., 1995; Liu et al., 2007; Zhang et al., 2011a): Vm(t + dt) = −(dt/C)[Ge(t) * (Vm(t) − Ee) + Gi(t) * (Vm(t) − Ei) + Gr(Vm(t) − Er)] + Vm(t), where Vm(t) is the membrane potential at time t, C is the whole-cell capacitance, Gr is the resting leaky conductance, and Er is the resting membrane potential (−60 mV). To simulate the spike response, 20 mV above the resting membrane potential was set as the spike threshold and a 5 ms refractory period was used. C (20–50 pF) was measured during the experiment and Gr was calculated based on the equation Gr = C*Gm/Cm, where Gm, the specific membrane conductance, is 2e − 5 S/cm2, and Cm, the specific membrane capacitance, is 1e − 6 F/cm2 (Hines, 1993; Stuart and Spruston, 1998).
Intensity tuning functions of rat DCN neurons
It was previously proposed in the cat and gerbil DCN that intensity-selective neurons have lower intensity thresholds of spike response than inhibitory neurons in DCN, and that this difference in intensity threshold can explain the generation of the selectivity (Voigt and Young, 1980; Nelken and Young, 1994; Spirou et al., 1999; Young and Davis, 2001; Oertel and Young, 2004). We examined this proposed mechanism in the rat by first characterizing intensity tuning functions of its DCN neurons. After exposing the dorsal surface of the cochlear nucleus in the brainstem, in vivo cell-attached loose-patch recordings were made from individual neurons in the DCN to obtain their spike responses evoked by tones of different frequencies and intensities (see Materials and Methods). In this study, we specifically focused on fusiform/pyramidal cells, the DCN projection neurons located in layer 2 (Mugnaini et al., 1980; Wouterlood and Mugnaini, 1984; see Materials and Methods). The frequency-intensity TRF was reconstructed from recorded spikes, and the CF was determined at the intensity threshold of the spike TRF (see Materials and Methods). The response-level function (i.e., intensity tuning) of the cell was then tested by applying CF tones at different intensities for a number of repetitions. Two types of pyramidal neurons were observed. Examples are given in Figure 1C,D. For the first type, neurons exhibited typical V-shaped spike TRFs, and inhibitory side bands were apparent when the spontaneous firing rate was relatively high (Fig. 1C, top and color map). They exhibited monotonic response-level functions, as indicated by a continuous increase in response level as the intensity of CF tones increased (Fig. 1C, lower right). These neurons can thus be viewed as intensity-nonselective, and are functionally similar to previously reported type III or type I/III cells (Evans and Nelson, 1973; Young and Voigt, 1982; Davis et al., 1996). In this work, we did not further separate them into type III (with inhibitory sidebands) and type I/III (without apparent inhibitory sidebands) neurons. For the second type, the response-level function for CF tones was nonmonotonic: the spike rate first reached a peak and then declined at higher intensity levels (Fig. 1D). These neurons can be viewed as intensity-selective, and are functionally consistent with previously reported type IV neurons (Evans and Nelson, 1973; Young and Voigt, 1982; Davis et al., 1996). Their TRFs were not typical V-shaped, since the responses at or near CF at high intensity levels were reduced.
We used an intensity selectivity index (ISI) to quantify the selectivity level. It was defined as the ratio of the response at 30 dB above the preferred intensity or at the highest testing intensity over the maximum response. A perfect monotonic response-level function would generate an ISI of 1. As shown by the distribution of ISIs, pyramidal cells appeared to cluster into three groups (Fig. 2A). The first group (“NS”) had ISIs of 1 or close to 1, and the average response-level function for this group was clearly monotonic (Fig. 2B). According to the ISI distribution and the previous criterion for defining intensity-selective neurons (Ding and Voigt, 1997; Wu et al., 2006), we defined intensity-selective neurons in this study as having an ISI of <0.6. The second and third group (Fig. 2A, “Si” and “Sii”) were therefore collectively intensity-selective neurons, with their average response-level function exhibiting a stronger nonmonotonicity (Fig. 2C). The “Si” group, however, exhibited much strong intensity selectivity than the “Sii” group. In the “Si” group, CF tones at high intensities could completely suppress spiking activity, resulting in a negative response relative to the baseline activity (Fig. 2E). In the “Sii” group, on the other hand, spike rate at the highest testing intensity was only reduced to ∼40% of the maximum (Fig. 2F). These two groups, one strongly selective and the other moderately selective, may correspond to previously reported “type IV” and “type IV-t” neurons respectively (Evans and Nelson, 1973; Davis et al., 1996). Within the pyramidal cell population, only 13% (17 of 129) of cells were intensity selective. This relatively small percentage of intensity-selective neurons was also reported in other rodents (Davis et al., 1996; Navawongse and Voigt, 2009; Ma and Brenowitz, 2012).
Previously, type II neurons have been thought to play an important role in creating intensity selectivity. They have been demonstrated to be vertical cells in morphology and be inhibitory in function (Davis and Voigt, 1997; Rhode, 1999). We specifically searched type II cells in the deep layer (see Materials and Methods). They could be identified by little or no spontaneous firing activity, absence of response to broadband noise stimuli, and strong response to CF tones (Evans and Nelson, 1973; Young and Voigt, 1982; Young and Davis, 2001). An example type II neuron is shown in Figure 1E. The cell exhibited almost zero spontaneous activity, a narrow V-shaped TRF, and monotonic intensity-tuning function. A summary of response-level function for 20 type II neurons is shown in Figure 2D. All type II neurons were intensity-nonselective with ISIs >0.6 (Fig. 2D, inset). Compared with pyramidal cells (both intensity nonselective and selective), which have 20–30 Hz spontaneous spike rates, the extremely low spontaneous rate of type II neurons made them easily distinguished (Fig. 2G). The type II neurons had significantly higher intensity thresholds than both the intensity-nonselective and intensity-selective pyramidal cells (Fig. 2H), consistent with previous reports in cats and gerbils. The two types of pyramidal cells did not differ in intensity threshold (Fig. 2H). This result suggests that in both types of pyramidal neurons the intensity threshold of inhibition may be higher than that of excitation. Therefore, the generation of intensity selectivity may not be simply attributed to a difference in intensity threshold between excitation and inhibition.
Excitatory and inhibitory synaptic inputs to DCN neurons
To reveal the synaptic basis for intensity selectivity, we performed whole-cell current-clamp and voltage-clamp recordings from the same DCN neuron (see Materials and Methods). First, in the current-clamp recording mode, the spike TRF was obtained to determine the cell's CF and whether the cell was intensity-selective. CF tone-evoked excitatory and inhibitory synaptic responses at different intensities were then recorded under the voltage-clamp mode. An example intensity-nonselective cell is shown in Figure 3A–C. The mapping of spike TRF revealed that the cell's CF was at 15.3 kHz (Fig. 3A). At this frequency, changes of spike response with increasing intensity indicated that the cell was monotonic (Fig. 3B, Sp). The subsequent voltage-clamp recording revealed that at the intensity threshold of spike response (marked as “0 dB”), there was significant tone-evoked excitatory input, but not inhibitory input (Fig. 3B, Ex and In, respectively). From the plotting of peak amplitude of evoked synaptic currents (Fig. 3C), it is clear that both the excitatory and inhibitory inputs were gradually strengthened as intensity increased, and that the intensity threshold of the inhibitory input was higher than the excitatory input. An example intensity-selective cell is shown in Figure 3D–F. In this cell, the amplitudes of excitatory and inhibitory inputs also increased with intensity increments, indicating that synaptic inputs per se are not intensity-selective. However, different from the intensity-nonselective neuron, its excitatory input appeared to quickly reach a plateau level as soon as the intensity threshold for spike response was reached (Fig. 3F).
More example intensity-nonselective and intensity-selective neurons are shown in Figure 3G,H and Figure 3I,J, respectively. For the intensity-nonselective neurons, excitatory and inhibitory inputs increased with similar paces. In contrast, for the intensity-selective neurons whose spike responses decreased at high intensities, the excitatory response increased faster than the inhibitory response with increasing intensity.
The whole-cell recording allowed labeling of the recorded cells with dyes. Complete dendritic morphologies were reconstructed for two intensity-nonselective and two intensity-selective neurons (Fig. 3K). The cells were all pyramidal cells located in layer 2 (i.e., pyramidal cell layer), as evidenced by the presence of apical and basal dendrites as well as the spindle-like cell body shape (Young and Davis, 2001). Six partially recovered cells were also located in the pyramidal cell layer and were identified as fusiform cells based on the soma shape (data not shown). There was no apparent difference in morphology between intensity-nonselective and intensity-selective pyramidal neurons (Fig. 3K). We also estimated the quality of our voltage-clamp recordings by plotting the current–voltage (I–V) relationship for the recorded synaptic currents. As shown by an example cell (Fig. 3L), synaptic currents evoked by CF tones were recorded under four different membrane potentials. The current amplitude at 1 ms after the response onset (Fig. 3L, red) changed linearly with voltage, with the reversal potential matching that expected for excitatory currents (0 mV). The current amplitude at 11 ms after the onset (Fig. 3L, black) also changed linearly with voltage, but the reversal potential was −46 mV, indicating a combination of excitatory and inhibitory currents at this time point. We next examined CF tone-evoked synaptic currents at 1 ms after the response onset in all the recorded pyramidal cells (Fig. 3M). The currents were all close to zero when the membrane was clamped at the excitatory reversal potential. The linear I–V relationship as well as the relatively precise excitatory reversal potential as revealed suggest that voltage-clamp quality was reasonably good in our experiments.
Synaptic tuning profiles and inhibition/excitation ratio
We summarized synaptic tuning profiles from nine intensity-nonselective and five intensity-selective pyramidal neurons. As shown in Figure 4A,B by the normalized peak response amplitude, intensity-nonselective and intensity-selective neurons differed dramatically in the relationship between excitatory and inhibitory intensity tuning profiles. Excitation and inhibition were similarly increasing with intensity increments in nonselective neurons, while in selective neurons excitation saturated much faster than inhibition. The same conclusion can be made when the integrated conductance was measured (Fig. 4C,D, see Materials and Methods). Between intensity-selective and intensity-nonselective cells, there was no difference in inhibitory tuning profile, while their excitatory tuning profiles significantly differed in terms of slope (Fig. 4E,F). Consistent with the loose-patch recording results, excitation had a lower intensity threshold than inhibition in both intensity-selective and intensity-nonselective neurons (Fig. 4G). This intensity threshold difference between excitation and inhibition was not different between the two types of pyramidal cells (p > 0.1, t test).
In addition to intensity tuning profile, the relative and absolute strengths of excitatory or inhibitory inputs may also contribute to intensity selectivity. We further compared the amplitudes of excitatory and inhibitory inputs between the two types of pyramidal neurons. As shown in Figure 5A, at the intensity threshold, excitation was stronger in selective than in nonselective neurons. However, at 50 dB above the intensity threshold, excitation in selective neurons became weaker than in nonselective neurons, due to its early saturation with intensity increments. On the other hand, the absolute level of inhibition was not significantly different between the two types of pyramidal neurons at either intensity (Fig. 5A). Figure 5B plots the change of inhibition/excitation (I/E) ratio with intensity increments. The I/E ratio was more or less constant above the intensity threshold in nonselective neurons, whereas it kept increasing in selective neurons. At the highest testing intensity, the selective neurons received relatively stronger inhibition compared with the nonselective neurons (Fig. 5B). As for the temporal properties of synaptic responses, we found that the onset delay of inhibition relative to excitation reduced with intensity increments for both intensity-selective and intensity-nonselective neurons (Fig. 5C). This shortening of relative delay of inhibition can be accounted for at least partially by the intensity-dependent changes of the excitatory response onset and the spike response onset of inhibitory type II neurons (Fig. 5D).
Neuronal modeling of the generation of intensity selectivity
We have shown that intensity-selective and intensity-nonselective neurons differ in excitatory tuning profile and absolute excitatory strength. Intuitively, the differential intensity tuning profile of excitation may fully account for the functional difference between intensity-selective and intensity-nonselective neurons. In the intensity-selective neuron, excitation is close to saturation at the intensity threshold (it reaches >80% of the peak level), whereas at this intensity level inhibition is nearly the lowest (Fig. 4A,B). This may lead to a strong output response at the intensity threshold. With intensity further increasing, inhibition catches up, and the level of spike response may gradually fall.
To demonstrate how much the observed excitatory and inhibitory tuning profiles can account for the intensity tuning profile of output responses in a quantitative manner, we used a conductance-based single-compartment integrate-and-fire neuron model (Liu et al., 2007; Zhou et al., 2012). Feeding it with experimentally obtained excitatory and inhibitory synaptic conductances, we derived the expected membrane potential and spike responses at different tone intensities on a cell-by-cell basis (see Materials and Methods). As shown in Figure 5E,F, the intensity tuning profile of derived spike responses could largely match that of recorded responses for both intensity-nonselective and intensity-selective neurons. This suggests that the interplay of somatically recorded excitatory and inhibitory responses can qualitatively determine the tuning property of the cell. Next, to test whether the differential intensity tuning profiles of excitatory input play a critical role in determining intensity-selective or intensity-nonselective properties, we exchanged the excitatory tuning profiles between the intensity-selective and intensity-nonselective groups. This was achieved in each individual cell by keeping the excitatory response at the highest intensity fixed while scaling the response amplitudes at other intensities according to the average excitatory tuning profile for the other neuronal type (Fig. 5G,H, top). The scaled excitatory response was then integrated with the original inhibitory response to derive spike response. Through this one-step swapping procedure, the tuning of intensity-nonselective neurons became slightly nonmonotonic (Fig. 5G, magenta), while that of intensity-selective neurons was changed to monotonic (Fig. 5H, magenta).
We noticed that the swapping of excitatory tuning profiles only produced modestly nonmonotonic spike responses. This apparent deviation from recorded intensity-selective responses may be due to a difference in absolute strength of synaptic inputs between the two types of pyramidal neurons (Fig. 5A). For example, in nonselective neurons, the excitatory response at the highest testing intensity is strong and may result in nearly saturated spike response, leaving little room for the input at lower intensities to generate even stronger spike response. In the following procedure, in addition to swapping the excitatory tuning profile, we also scaled the excitatory response amplitudes in general so that the I/E ratio at the highest intensity was the same as that observed for the other neuronal type. In other words, we swapped the I/E ratios of the two types of cells. As shown in Figure 5G,H (cyan), the two-step swapping procedure converted the intensity-nonselective responses to more in vivo-like intensity-selective responses, while the intensity-selective responses were completely switched to monotonic responses. These modeling results demonstrate that the differential tuning relationship between excitation and inhibition plus the differential excitatory input strength can explain the functional difference between intensity-selective and intensity-nonselective neurons. Thus, these two factors may play key roles in generating intensity selectivity in the DCN.
In this study, we applied in vivo whole-cell recordings to examine synaptic mechanisms underlying the creation of intensity selectivity in the DCN. To our knowledge, this is the first time excitatory and inhibitory synaptic inputs underlying functional selectivity at this earliest stage of central auditory processing are elucidated. Our results confirmed that synaptic inputs to DCN neurons all exhibit monotonic intensity tuning functions. Two synaptic mechanisms contribute to the converting of monotonic inputs to nonmonotonic outputs. First, the differential intensity tuning profiles of excitation and inhibition (i.e., fast-saturating excitation but slow-saturating inhibition) are a fundamental basis for the generation of intensity selectivity. Second, the lower level of excitation or relatively stronger inhibition at high intensities further strengthens the intensity selectivity of output responses.
Intensity selectivity in central auditory system
Nonmonotonic intensity-tuned neurons have been widely observed along the ascending pathway of the central auditory system [e.g., in the cochlear nucleus (Greenwood and Maruyama, 1965; Young and Brownell, 1976, Ramachandran et al., 1999), inferior colliculus (Aitkin, 1991; Kuwabara and Suga, 1993), medial geniculate body (Aitkin and Webster, 1972; Rouiller et al., 1983), and auditory cortex (Davies et al., 1956; Evans and Whitfield, 1964; Brugge et al., 1969; Schreiner et al., 1992; Phillips et al., 1995; Wu et al., 2006; Tan et al., 2007; Sadagopan and Wang, 2008)]. It has been thought that the selectivity can be generated de novo at later stages of the auditory pathway. For example, in earlier studies of the auditory cortex, it has been postulated that an intensity-dependent increase of inhibition can result in nonmonotonic response-level functions (Ojima and Murakami, 2002; Sutter and Loftus, 2003). These studies were based on extracellular recordings of spike responses under two-tone stimuli or intracellular recordings of tone-evoked membrane potential responses, both of which detect inhibition in an indirect way. More recent whole-cell recording studies have revealed that excitatory inputs to intensity-selective auditory cortical neurons are already well tuned, indicating that intensity selectivity in the cortex can be partially inherited from previous stages, although local cortical inhibition can further sharpen the selectivity (Wu et al., 2006; Tan et al., 2007; Zhang et al., 2011b). Thus, it is worthwhile to examine the first stage where the selectivity is initially generated.
The receptive field maps of intensity-selective pyramidal cells in this study look different from the circumscribed, O-shaped response maps often observed in the cortex (e.g., Phillips and Kelly, 1989; Wu et al., 2006, Sadagopan and Wang, 2008), which have perfect intensity selectivity. The DCN receptive field maps are not completely enclosed, and intensity tuning functions at off-CF can be in fact monotonic. Possibly, along the ascending auditory pathway, recruitments of inhibition (e.g., lateral inhibition) at each higher processing stage progressively sculpt the receptive field map, in particular by silencing CF and off-CF responses at high intensity levels (Sutter and Loftus, 2003; Sadagopan and Wang, 2008; de la Rocha et al., 2008; Wu et al., 2011; Zhang et al., 2011b). This may eventually lead to the emergence of O-shaped receptive field maps in the cortex.
Synaptic mechanisms underlying intensity selectivity
To convert monotonic response-level functions to nonmonotonic functions, synaptic inhibition is generally thought to be required (Shamma, 1985; Phillips, 1988; Calford and Semple, 1995; Ding and Voigt, 1997). Previously, three different scenarios have been postulated with regard to how the excitatory–inhibitory interplay might generate intensity tuning. First, based on the observation that inhibitory type II neurons exhibit a higher intensity threshold than their principal neuron counterparts (Voigt and Young, 1980; Spirou et al., 1999), it is suggested that a lower intensity threshold of excitation than inhibition may result in the highest E/I ratio and the strongest spike response at the excitatory intensity threshold (Nelken and Young, 1994; Young and Davis, 2001). Our data however indicate that in both intensity-selective and intensity-nonselective neurons, the intensity threshold of excitation is ∼10 dB lower than that of inhibition. Therefore, a threshold difference alone does not necessarily produce intensity tuning. Second, it has been found for intensity-selective cortical neurons that the onset latency of inhibition relative to excitation shortens with intensity increments (Wu et al., 2006). Modeling work further suggests that this intensity-dependent decrease in the temporal delay of inhibition can sufficiently produce intensity-tuned output responses (Wu et al., 2006). In the DCN, we found that the relative delay of inhibition was only slightly reduced with intensity increments and that it was not different between intensity-selective and intensity-nonselective neurons (Fig. 5C). In addition, because DCN responses are much more sustained than cortical responses (Wehr and Zador, 2003; Zhang et al., 2003; Tan et al., 2004), any small variations of relative inhibitory delay would not affect the overall response level significantly. Third, a theoretic study of thalamocortical networks has demonstrated that under certain network properties unbalanced recruitments of excitation and inhibition can occur such that inhibition grows slower than excitation with intensity increments (de la Rocha et al., 2008). Under such unbalanced excitation and inhibition, the integrated output response peaks well before inhibition reaches maximum. Consistent with this postulate, unbalanced recruitments of excitation and inhibition occur in the DCN and play a key role in generating intensity selectivity.
Potential underlying circuits
How can the unbalanced recruitments of excitation and inhibition be achieved? Previously, a circuit with broad lateral inhibition was proposed for cortical neurons to result in a slower recruitment of inhibition than excitation as the number of inhibitory neurons activated gradually increases with increasing intensity (Sutter and Loftus, 2003; de la Rocha et al., 2008). This mechanism requires that the receptive fields of inhibitory neurons be relatively broad and be spectrally offset but have overlapping regions. Here in the DCN, we notice that the speed of recruitment of inhibition does not differ between intensity-selective and intensity-nonselective neurons (Fig. 4F). It appears that it is the recruitment pattern of excitation that makes the difference.
The two different excitatory tuning patterns observed in intensity-nonselective and intensity-selective neurons were reminiscent of the two types of intensity-dependent responses of auditory nerve fibers reported previously: fast saturating and slow saturating (Sachs and Abbas, 1974; Winter et al., 1990). This raises an interesting possibility that the different excitatory tuning patterns may be conveyed from the two different types of auditory nerve inputs, as it has been well documented that pyramidal neurons receive direct auditory nerve inputs (Oertel and Young, 2004). In addition, we found in two intracellularly recorded type II neurons that they received slowly saturating excitatory input similar to intensity-nonselective pyramidal neurons (Fig. 6A). This observation suggests that the slow-saturating inhibition to pyramidal neurons could be relayed by inhibitory type II neurons innervated by slow-saturating auditory nerve fibers. Based on the above evidence, we propose a simple model that can explain DCN intensity selectivity (Fig. 6B). In this model, the intensity tuning properties of excitation and inhibition are simply inherited from auditory nerve fibers. The intensity-selective pyramidal neuron may receive fast-saturating excitatory input from fast-saturating auditory nerve fibers (Input A), and slow-saturating inhibitory input from inhibitory neurons driven by slow-saturating auditory nerve input (Input B). Through such a feedforward inhibitory circuit likely mediated by type II/vertical cells (Voigt and Young, 1980; Oertel and Young, 2004; Kuo et al., 2012), the threshold of inhibition becomes higher than excitation due to the spike thresholding effect on inhibitory neuron responses, while the intensity tuning function of inhibition remains slow-saturating. In addition, the low spontaneous activity of type II neurons (Fig. 2G) may also contribute to their higher intensity threshold, since relatively stronger auditory nerve input is required to drive these cells compared with pyramidal neurons. By this analogous differential circuit, the pyramidal neuron can extract the difference between the intensity tuning patterns of Input A and Input B to create intensity-selective output responses. The overall structure of our proposed circuit is in line with the one previously proposed for type IV neurons (Voigt and Young, 1980; Spirou et al., 1999), except that differential AN fiber inputs are incorporated. It is unknown why AN fibers have different saturation slopes. Is it due to distinct properties of different types of spiral ganglion cells? How specific is the connectivity between different types of DCN neurons and different types of AN fibers? The proposed synaptic circuitry mechanism for the generation of intensity selectivity raises many interesting questions for future investigations.
This work was supported by grants to L.I.Z. from the United States National Institutes of Health/National Institute on Deafness and Other Communication Disorders (R01DC008983), and the David and Lucile Packard Foundation (Packard Fellowships for Science and Engineering). H.W.T was supported by a grant from the United States National Institutes of Health (R01EY019049).
- Correspondence should be addressed to L.I. Zhang, Keck School of Medicine, University of Southern California, 1501 San Pablo Street, Los Angeles, CA 90033.