Early experience shapes sensory representations in a critical period of heightened plasticity. This adaptive process is thought to involve both Hebbian and homeostatic synaptic plasticity. Although Hebbian plasticity has been investigated as a mechanism for cortical map reorganization, less is known about the contribution of homeostatic plasticity. We investigated the role of homeostatic synaptic plasticity in the development and refinement of frequency representations in the primary auditory cortex using the tumor necrosis factor-α (TNF-α) knockout (KO), a mutant mouse with impaired homeostatic but normal Hebbian plasticity. Our results indicate that these mice develop weaker tonal responses and incomplete frequency representations. Rearing in a single-frequency revealed a normal expansion of cortical representations in KO mice. However, TNF-α KOs lacked homeostatic adjustments of cortical responses following exposure to multiple frequencies. Specifically, while this sensory over-stimulation resulted in competitive refinement of frequency tuning in wild-type controls, it broadened frequency tuning in TNF-α KOs. Our results suggest that homeostatic plasticity plays an important role in gain control and competitive interaction in sensory cortical development.
Rodent auditory cortex undergoes rapid maturation during early postnatal development, as manifested by the emergence and refinement of cortical sound representations (Zhang et al. 2001; Chang et al. 2005; de Villers-Sidani et al. 2007; Insanally et al. 2009). This process is shaped by acoustic experience in a “critical period” of heightened plasticity (de Villers-Sidani et al. 2007). Recent studies indicate that the auditory cortex is sensitive to different sound features across developmental stages within the critical period (Insanally et al. 2009; Popescu and Polley 2010). For example, early critical period experience shapes the cortical frequency map (de Villers-Sidani et al. 2007; Insanally et al. 2009), whereas later critical period experience shapes frequency modulation selectivity (Insanally, Kover et al. 2009, Insanally, Albanna et al. 2010). In addition, the characteristics of developmental plasticity depend on the properties of the acoustic input (Chang and Merzenich 2003; Zhou et al. 2008). While exposure to a pulsed tone repeated at an ethological rate results in enlarged representation of the tone, exposure to the same tone repeated at a higher or lower rate does not (Kim and Bao 2009). Early experience also alters sound perception and perceptual behaviors in ways consistent with the reorganized sound representation in the auditory cortex (Han et al. 2007). Thus, multifaceted auditory cortical plasticity may be a useful model to investigate molecular/cellular mechanisms of sensory development and pathologies of developmental sensory disorders.
Refinement of sensory representations during the critical period is believed to be mediated by experience-dependent synaptic plasticity (Dan and Poo 2006; Feldman 2009). Sensory experience is shown to engage at least 2 types of synaptic plasticity in sensory cortex: Hebbian and homeostatic synaptic plasticity (Desai et al. 2002; Fu et al. 2002; Heynen et al. 2003; Crozier et al. 2007; Goel and Lee 2007; Maffei and Turrigiano 2008). Hebbian plasticity, which includes long-term potentiation (LTP) and long-term depression (LTD), and spike-timing–dependent plasticity, rapidly alters the strength of individual synapses in an input-specific manner (Zhang et al. 1998; Abbott and Nelson 2000; Malenka and Bear 2004; Dan and Poo 2006). In contrast, homeostatic plasticity globally or locally adjusts synaptic strength onto the neuron following prolonged changes in neuronal activity level (Davis and Bezprozvanny 2001; Burrone and Murthy 2003; Turrigiano and Nelson 2004; Hou et al. 2008). An important difference between Hebbian and homeostatic plasticity is how they adjust synaptic strength when a neuron is over-stimulated. Hebbian plasticity stengthens excitatory synapses when pre- and postsynpatic neurons are co-activated and weakens excitory synapses when presynaptic neuron is activated alone. In the sensory cortex, repeated activation of a cortical neurons by a stimulus may engage Hebian plasticity to strengthen excitatory connections, resulting in enhanced cortical responses to the stimulus (Zhang et al. 2001). This change is, at least partly, mediated by enhanced excitatory responses (Froemke et al. 2007; Sun et al. 2010). Sensory deprivation may reduce cortical responses to the deprived sensory organ through Hebbian synaptic depression (Heynen et al. 2003). In contrast, homeostatic plasticity should weaken exciatory synpases and strengthen inhibitory synpases onto a neuron when the neuron is over-stimulated. Thus, throught homeostatic plasticity, repeated sensory stimulation could lead to weakened cortical responses (Condon and Weinberger 1991; Pienkowski and Eggermont 2012).
While Hebbian plasticity in sensory development has been investigated extensively (Feldman 2009), the role of homeostatic plasticity has been investigated only recently in the visual and somatosensory systems (Mrsic-Flogel et al. 2007). Experimental findings indicate that a form of homeostatic plasticity is involved in ocular dominance shifts during the critical period but not in adulthood (Kaneko et al. 2008; Ranson et al. 2012). Following monocular deprivation, Hebbian LTD causes a reduction in responses to stimulation of the deprived eye and subsequent homeostatic plasticity results in competitive enhancement of responses to the open eye (Frenkel and Bear 2004; Kaneko et al. 2008; Ranson et al. 2012). While these findings provide convincing evidence of a role for homeostatic plasticity in this particular experimental paradigm, it remains to be determined whether and how it may be involved in other forms of developmental plasticity both within the visual system and in other sensory systems.
Earlier studies have taken advantage of strains of mice that are deficient in homeostatic plasticity (Kaneko et al. 2008; Ranson et al. 2012). For example, the tumor necrosis factor-α (TNF-α) knockout (KO) mouse, which is deficient in homeostatic up-regulation of excitatory synaptic transmission and downregulation of inhibitory transmission in response to activity blockade (Stellwagen and Malenka 2006; Kaneko et al. 2008), exhibits normal monocular deprivation-induced loss of deprived-eye responses in the initial stages of ocular dominance shift but not the subsequent increase in open eye responses (Kaneko et al. 2008).
In the present study, we examined the role of homeostatic plasticity in the development and plasticity of sound representations in the TNF-α KO mouse. KO mice raised in the typical animal room environment had highly variable cortical frequency maps, often showing incomplete representation of the mouse-hearing frequency range. Although both WT and KO mice developed enlarged representations of a repeatedly exposed tone, they showed opposite effects as a result of multi-frequency exposure—narrowed frequency tuning in WT mice and broadened frequency tuning in KO mice. These results suggest that homeostatic plasticity may be involved in normal development and competitive refinement of acoustic representation in the primary auditory cortex.
Materials and Methods
All procedures used in this study were approved by the UC Berkeley Animal Care and Use Committee. Litters of juvenile TNF-α KO mice (KO) and corresponding C75Bl/6 wild-type mice (WT) from the Jackson Laboratory, together with the nursing females, were assigned to one of the 3 groups—a tone-exposure group, an enriched environment group, and a control group. The 2 experimental groups were repeatedly exposed to 1 s long trains of 6 tone pips (100 ms, 65 dB sound pressure level (SPL), 5-ms on and off cosine-squared ramps), with one train occurring every 2 s. The frequency of the tone pips within a train was the same, and was either fixed at 25 kHz for the tone-exposure group or randomly chosen from a continuum that ranged from 4 to 45 kHz for the enriched environment group. For this group, the acoustic power of the exposure sounds was uniformly distributed along the logarithmic frequency axis from 4 to 45 kHz (e.g. the power in the 4–8 kHz range is the same as in 8–16 kHz range). Sounds were generated with a National Instrument I/O card at a sampling rate of 200 kHz, amplified, and played through a calibrated speaker in a sound-attenuation chamber where the animals were housed. The sound exposure started on postnatal day 9 (P9) and ended immediately before electrophysiological examination was conducted, typically on P19 to P21. Both sound exposed groups were compared with the control group, which were maintained in regular animal rooms and were mapped during the same period as the 2 experimental groups. The acoustic environment of the animal room is dominated by ambient low-level noise. Constant, broadband sounds at such low levels cannot mask sensory input. In addition, unmodulated sounds are ineffective in shaping sensory representations (Kim and Bao 2009).
Electrophysiological Recording Procedure
The primary auditory cortex (AI) in naïve and sound-exposed KO and WT mice was mapped as previously described (Kim and Bao 2009). Mice were anesthetized with Ketamine (100 mg/kg, IP) and xylazine (10 mg/kg, IP), and placed on a homeothermic heating pad at 36.5°C (Harvard Apparatus) in a sound attenuation chamber. The head was secured with a custom head-holder that left the ears unobstructed. The right auditory cortex was exposed and kept under a layer of silicone oil to prevent desiccation. Multi-unit activity was evenly sampled from primary auditory cortex. AI in both WT and KO mice was found consistently underneath the caudal half of the temporal–parietal bone suture. It can be identified by its tonotopic orientation—higher frequencies are represented more rostrally and slightly more dorsally. Other auditory cortical fields have different tonotopic orientations (Guo et al. 2012). The border of AI was defined by unresponsive sites or sites whose CFs were incongruent with the AI tontopic gradient. Because KOs tended to have incomplete representations of low and high frequencies (Fig. 1), we carefully searched for those representations near the rostral and caudal ends of AI in both WTs and KOs, while maintaining the same sampling density. Typical sampling extent and density is shown in the map from animal number 3 in Figure 1A. Neural responses were recorded using tungsten microelectrodes (FHC) at a depth of 400–450 µm, presumably from the thalamorecipient layer. Responses to 25-ms tone pips of 41 frequencies (4–64 kHz, 0.1 octave spacing) and 8 sound pressure levels (10–80 dB, 10-dB steps) were recorded to reconstruct the frequency-intensity-receptive field. A Tucker-Davis Technologies coupler model electrostatic speaker was used to present all acoustic stimuli into the left ear (contralateral to the recorded cortical hemisphere). Each frequency × intensity combination was repeated 3 times.
The receptive fields and response properties were isolated using custom-made programs. First, the peri-stimulus time histogram (PSTH) was generated from responses to all 328 (41 frequencies × 8 intensities) tone pips, with 1-ms bin size (Fig. 4E). The mean firing rate was calculated for each bin and smoothed with a 5-point mean filter. The multiunit spontaneous firing rate was taken as the mean firing rate in the 50-ms window prior to stimulus onset. Peak latency was defined as the time to the peak PSTH response between 7 and 50 ms after the stimulus onset. The response window was defined as the period encompassing the PSTH peak, in which the mean firing rate in every bin was higher than baseline firing rate. The onset latency was defined at the onset of the response window. The tone-evoked response was measured as the maximum firing rate within the response window. Spikes that occurred within the response window were counted to reconstruct the receptive field.
The tuning curve contour was determined using a smoothing and thresholding algorithm. The response magnitude was plotted in the frequency–intensity space, and smoothed with a 3 × 3 mean filter (see Fig. 2B for examples). It was then thresholded at 28% of the maximum value of the smoothed response magnitude. Response areas smaller than 5 pixels were removed. The contour of the suprathreshold area was defined as the tuning curve. The raw responses in the suprathreshold area were defined as the isolated receptive field. The threshold of the neuron was the lowest sound level that elicited responses in the isolated receptive field. The characteristic frequency (CF) of a neuron was defined as the center of mass of the isolated receptive field for the 2 lowest suprathreshold sound levels. The center of mass CF was defined as (∑Ri× fi)/∑Ri, in which Ri is the response magnitude to the ith tone with frequency fi, and responses were collapsed across the 2 sound levels. The maximum RF response was the maximum number of spikes activated by a single frequency–intensity combination. The mean RF response was the mean number of spikes for all frequency-dB combinations within the receptive field. Since each frequency–intensity combination was repeated 3 times, the average of those 3 responses was taken. Tuning bandwidth (BW) was defined as the BW of the receptive field at the specified intensity. We quantified BW at 80, 70, and 60 dB, but not at lower SPLs because many neurons did not respond at those low levels. To more accurately quantify BW at low SPLs, we measured BW at 10, 20, and 30 dB above the threshold. The receptive field size was the number of frequency–intensity combinations within the receptive field.
Auditory cortical map was reconstructed by Voronoi tessellation of the AI space and assigning response properties of a recording site to the corresponding polygon.
Unless otherwise stated, statistical significance was determined with ANOVAs with post hoc Bonferroni's test.
Development of Cortical Frequency Representations Is Impaired in TNF-α KO Mice
We examined the development of sound representations in the primary auditory cortex of KO and WT mice by mapping frequency–intensity representations at 3 ages: postnatal day 15 (P15), P20, and P30. The basic characteristics of sound representations observed in AI of WT mice in the present study (including tonotopic organization, frequency range, tuning BW, intensity threshold, and response latency; see figures below) were consistent with those reported before (Guo et al. 2012). By P15, the WT mice had already developed finely topographic representations of nearly the full range of frequencies from 4 to 50 kHz persisting through P30 (Fig. 1A). In contrast, the frequency map in KO mice was more variable throughout the developmental window from P15 to P30 (Fig. 1A2). AI in KO mice generally represented a narrower frequency range than that in WTs (WT, n = 8, 2.9 ± 0.1 octaves; KO, n = 11, 1.8 ± 0.1 octaves; ANOVA, F1,17 = 35.07, P < 0.00002) often concentrated in the middle of the hearing range (Fig. 1). In P20 animals, there were more sites representing a middle frequency range of 8–30 kHz in KO mice than WT mice (WT, 75/130; KO, 166/176; χ2 (1) = 56.96, P < 0.0001; Fig. 1C).
Receptive field characterization indicated that tuning BW was not altered in KO (n = 5) compared with WT mice (n = 6; BW at 60–80 dB, 2-way repeated-measures ANOVA, genotype effect F1,9 = 0.16, P = 0.70, interaction F2,18, = 0.48, P = 0.63; BW10–30, genotype, F1,9 = 0.56, P = 0.47, interaction F2,18, = 0.39, P = 0.68; Fig. 2A–D). Although KO-receptive fields tended to have higher thresholds than those of WT, the effect was not significant (F1,9 = 3.47, P = 0.095; Fig. 2E). While no significant difference was found in spontaneous firing rate between WT and KO (F1,9 = 2.31, P = 0.16), the tone-evoked firing rate was significantly higher in WT than in KO mice (F1,9 = 5.36, P = 0.046; Fig. 2F).
TNF-α KO Mice Exhibit Single-Frequency Exposure-Induced Map Reorganization
We exposed both KO (n = 4) and WT (n = 4) mice to a 25-kHz tone repeated from P9 to P20, and examined sensory exposure-induced changes in cortical frequency representations. Exposure significantly increased the number of sites representing the frequency range of 25 kHz ± 0.3 octaves in both WT (naïve, 16/130; exposed, 68/163; χ2 (1) = 30.59, P < 0.0001) and KO (naïve, 43/176; exposed, 70/150; χ2 (1) = 17.68, P < 0.0001), indicating that KO mice undergo normal single-frequency exposure-induced map reorganization (Fig. 3). While single-frequency exposure did not alter the range of frequencies represented in AI of WT (2.9 ± 0.2 octaves, compared with naïve WT at 2.9 ± 0.1 octaves), it increased the AI frequency range in KO mice (2.7 ± 0.2 octaves, compared with naïve KO at 1.8 ± 0.1 octaves; ANOVA, F3,23 = 15.36, P < 0.0001; post hoc: naïve KOs vs. exposed KOs, P = 0.0002; naïve WTs vs. exposed WTs, P = 0.97; naïve WTs vs. exposed KOs., P = 0.46).
Single-frequency exposure had limited effects on receptive field and neuronal firing properties. Receptive field analysis indicates that single-frequency exposure did not alter overall tuning BWs of AI units (experience × genotype × intensity ANOVA with repeated measures for BW at 60–80 dB SPL, experience, F1,15 = 0.007, P = 0.93, genotype, F1,15 = 1.85, P = 0.19, interaction, F1,15 = 0.94, P = 0.35; ANOVA for BW10–30, experience, F1,15 = 3.29, P = 0.09, genotype, F1,15 = 3.22, P = 0.09, interaction, F1,15 = 0.003, P = 0.96; Fig. 4A,B). The receptive field size was not altered (genotype × experience ANOVA, genotype, F1,15 = 0.001, P = 0.98, experience, F1,15 = 0.18, P = 0.9; Fig. 4C).
We quantified mean and maximum response magnitude in the receptive field (for details, see Materials and Methods). Maximum response magnitude is the maximum number of spikes activated by any stimulus used to characterize the receptive field—that is, it measures response of a unit to its best stimulus. Mean response magnitude measures the overall responsiveness of the unit. The single-frequency exposure had different effects on KO and WT mice, enhancing responses in KO but not in WT (genotype × experience × type of response ANOVA, genotype × experience interaction, F1,32 = 4.15, P = 0.0498; Fig. 4D). We separated recorded units by their CFs—those with CFs ≥ 16 kHz and those with CFs < 16 kHz. Neurons with CFs ≥ 16 kHz were more likely to be activated by the 25-kHz exposure tone than neurons with CFs < 16 kHz (Fig. 3B). The maximum response magnitude was greater for tone-exposed KO mice than naïve KO mice only for units with high CFs (CFs ≥ 16 kHz) (F1, 7 = 10.23, P = 0.015), but not those with low CFs (CFs < 16) kHz (F1, 7 = 1.70, P = 0.233). There was no difference between naïve and tone-exposed KO mice in mean response magnitude in either of the CF groups (CFs ≥ 16 kHz: F1, 7 = 2.28, P = 0.174; CFs < 16 kHz: F1, 7 = 0.775, P = 0.408). We also separately analyzed high-CF and low-CF units recorded from naïve and tone-exposed WT mice, and found no significant difference in maximum or mean response magnitude (CFs ≥ 16 kHz: maximum response magnitude, F1, 8 = 0.891, P = 0.373, mean response magnitude, F1, 8 = 0.150, P = 0.709; CFs ≥ 16 kHz: maximum, F1, 8 = 0.169, P = 0.692, mean, F1, 8 = 0.024, P = 0.880).
We constructed PSTH with responses to all 328 (41 frequencies × 8 intensities) tone pips (Fig. 4E) to extract spontaneous and evoked firing rates, and onset and peak response latencies. Whereas the spontaneous firing rate was generally higher in WT versus KO mice (genotype × experience ANOVA, genotype effect, F1,17 = 5.264, P = 0.035; Fig. 4F), it was not altered by single-frequency exposure (experience effect, F1,17 = 0.048, P = 0.83). The tone-evoked firing rate was not different between WT and KO mice, nor between naïve and tone-exposed mice (genotype, F1,17 = 0.61, P = 0.45; experience, F1,17 = 0.47, P = 0.50; Fig. 4G). We also separately analyzed effects of tone-exposure for units with CFs ≥ 16 kHz and those with CFs < 16 kHz, but did not find statistically significant differences between naïve and tone-exposed mice, in either WT or KO group, for either spontaneous or tone-evoked firing rates (data not shown).
Single-frequency exposure delayed the onset and peak latencies of tone-evoked responses in WT but not in KO mice (onset latency: experience, F1,15 = 5.70, P = 0.031; post hoc WTs, P = 0.002; KOs, P = 0.98; Fig. 4H; peak latencies: experience, F1,15 = 10.87, P = 0.0049; post hoc, WT, P = 0.0009; KO, P = 0.33; Fig. 4I). A similar finding has been reported before (Engineer et al. 2004), but the neural mechanisms underlying the slower tone-evoked responses are unknown. Because each frequency–intensity combination was played only 3 times, we do not have enough data to reliably estimate onset or peak latency for individual tones.
Multi-Frequency Exposure Refines Frequency Tuning in WT and Broadens Tuning in KO Mice
To explore competitive interactions between different frequency inputs, we exposed WT (n = 4) and KO (n = 5) mice to an enriched multi-frequency tonal environment, in which the tone frequency was randomly chosen every 2 s from a uniform distribution ranging from 4 to 45 kHz and played in trains of 6 pips at a rate of 6 Hz. Like single-frequency exposure, our enriched environment manipulation altered the range of frequency representations (ANOVA group difference, F3,24 = 11.58, P < 0.0001; Fig. 5C,D). Post hoc pairwise tests showed that exposure to the enriched environment expanded the represented frequency range in KO mice (compared with naïve KO mice, P = 0.036), but not in WT mice (P = 0.18). Even after exposure, the frequency range represented by KO mice was still significantly narrower than the range represented by naïve WT mice (P = 0.017). However, after exposure, the representation range was no longer different between WT and KO mice (P = 0.36).
Exposure to the multi-frequency-enriched environment resulted in narrower frequency tuning in WT mice and broader tuning in KO mice measured at 60–80 dB sound intensity levels (repeated-measures ANOVA, genotype × experience interaction, F1,17 = 15.22, P = 0.0011; post hoc, exposed WT vs. all other groups, P < 0.03; exposed KO vs. all other groups, P < 0.05) and at 10–30 dB above the threshold (interaction, F1,17 = 14.03, P = 0.0016; post hoc, exposed KO vs. 2 WT groups, P < 0.01; Figs 5 and 6A,B). Consistent with the altered tuning BW, we also observed reduced receptive field sizes in WT but not in KO (genotype × experience interaction, F1,17 = 9.17, P = 0.0076; post hoc, exposed WT vs. naïve WT, P = 0.0048; exposed WT vs. exposed KO, P = 0.0061; Fig. 6C). The broadened tuning in exposed KO mice did not result in enlarged receptive field size, possibly because their threshold was slightly but not significantly increased. These results suggest that exposure to tones of different frequencies results in a winner-take-all type of competitive refinement of frequency representation in WT mice, which was impaired in KO mice lacking homeostatic plasticity.
KO Mice Are Impaired in Homeostatic Regulation of Cortical Responses
Overstimulation of AI with a wide range of frequencies resulted in a lower neuronal firing rate compared with naïve WT mice (spontaneous firing rate, F1,10 = 5.13, P = 0.047; tone-evoked, F1,10 = 5.71, P = 0.038; Fig. 6E,F), which may be considered a type of in vivo homeostatic regulation of neuronal activity by sensory experience (see Discussion). However, exposed KO mice showed a greater tone-evoked firing rate compared with naïve KO mice (tone-evoked, F1,8 = 5.54, P = 0.046; Fig. 6D). Spontaneous firing rate also trended higher in the exposed KO mice, although the effect was not significant (F1,8 = 1.99, P = 0.19; Fig. 6D). A comparison of mean and maximum responses in the receptive field confirmed the above observations, showing that exposure to the multi-frequency-enriched environment lowered AI responses in WT while increasing responses in KO mice (genotype × experience × response ANOVA, experience, F1,36 = 4.91, P = 0.033; Fig. 6F). Like single-frequency exposure, the multi-frequency exposure also delayed onset and peak latencies in WT but not in KO mice (onset latency: experience, F1,17 = 6.36, P = 0.022; post hoc WTs, P = 0.0091; KOs, P = 0.93; Fig. 6G; peak latencies: experience, F1,17 = 4.98, P = 0.039; post hoc, exposed WT vs. all other groups, P < 0.04; Fig. 6H). Instead of responding to overstimulation with a homeostatic decrease in neural activity, KO mice displayed an increase in responses indicating an absence of homeostatic processes.
We have compared the development and sound-induced reorganization of sound representations in AI of TNF-α KO mice and their WT controls. Our results indicate that, compared with WT mice, KO mice 1) develop more variable and incomplete frequency representations, 2) have weaker cortical responses to tones, 3) exhibit normal expansion of cortical representations in response to single-frequency, repeated tone exposure, 4) show enhanced cortical responses after sensory over-stimulation, 5) and do not show competitive refinement of frequency representations after exposure to multifrequency tones. These results suggest that TNF-α and its associated cellular processes are important in cortical response gain control and competitive refinement of cortical acoustic representations.
The mammalian auditory cortex evolved to be highly adaptive, such that it overrepresents prevalent and salient environmental sounds within the acoustic environment (Diamond and Weinberger 1986; Gonzalez-Lima and Scheich 1986; Ohl and Scheich 1996; Pantev et al. 1998; Edeline 1999; Gao and Suga 2000; Zhang et al. 2001; Syka 2002; Fritz et al. 2003; Mrsic-Flogel et al. 2003; Dean et al. 2005; Popescu and Polley 2010; Cohen et al. 2011; Takahashi et al. 2011). However, adaptation to some sounds may impair subsequent learning of new sounds in the future (Sarro and Sanes 2011). Therefore, it is important to strike a balance between plasticity and stability. The acoustic environment can be highly variable. For example, while the acoustic environment of a typical animal room may be considered impoverished, it can be dramatically enriched locally by conspecific vocalizations (Kim and Bao 2009; Grimsley et al. 2011). Species-specific vocalizations often occur in a high-frequency range, whereas sounds in the natural environment have more power in the lower frequency range (Liu et al. 2003; Kim and Bao 2009). In addition, animals are typically more sensitive to certain frequencies in their hearing range. Experimental evidence indicates that, in spite of the environmental variability and hearing constraints, the auditory cortex more or less consistently represents a large range of hearing frequencies (e.g. see WT cortical map in Fig. 1). This stability appears to break down in TNF-α KO mice, which display narrower ranges and more variability in their frequency representations. This impairment seems to be the result of impoverished sensory experience, as it is reversed by repeated acoustic exposure to single or multiple tones. It is conceivable that WT animals maintain stable acoustic representations in an impoverished sensory environment by enhancing input connectivity in the understimulated sensory pathways through homeostatic mechanisms. In animals with deficient homeostatic plasticity but normal Hebbian plasticity, cortical representations may be dictated to a greater degree by the highly variable acoustic environment, leading to impaired frequency representation as we observed in KO mice. It should be noted that no differences in retinotopy or basal visual response were observed in the primary visual cortex (VI) of the TNF-α KO (Kaneko et al. 2008). The different effects of TNF-α KO on basal stimulus representations in AI versus VI could be due to the fact that the visual stimulation is more uniform in the retinocentric visual space, whereas auditory stimulation is highly variable in the frequency space as we have discussed above.
The lack of homeostatic regulation in KO mice was also evident in the magnitude of cortical responses. Cortical responses to tone pips were weaker in naïve KO than in naïve WT mice, which again could be attributed to the acoustically impoverished housing environment. More telling evidence comes from the finding that, after repeated exposure to the multifrequency enriched environment, cortical responses in WT mice were reduced. Presumably, this occurs through homeostatic processes, whereas cortical responses in KO mice are enhanced through Hebbian plasticity. This is also consistent with the previous findings that TNF-α KO mice have normal LTP (Albensi and Mattson 2000; Stellwagen and Malenka 2006; Kaneko et al. 2008).
Refinement of neuronal connectivity requires competitive synaptic interactions, and theoretical considerations suggest that homeostatic plasticity may be involved in such interactions (Davis and Bezprozvanny 2001; Burrone and Murthy 2003; Turrigiano and Nelson 2004). Recent experimental studies examined competitive interactions in TNF-α KO mice by blocking the activity of a subset of sensory input through monocular deprivation. The results support a role of homeostatic plasticity underlying the competitive component of ocular dominance plasticity (Kaneko et al. 2008; Ranson et al. 2012). In the present study, we examined competitive interaction using the opposite sensory manipulation—by overstimulation of all sensory input asynchronously. In WT mice, exposure to a multi-frequency environment resulted in narrower frequency tuning, indicative of competitive refinement of sensory representations. In contrast, the identical sensory manipulation resulted in broadened tuning in mice lacking TNF-α, suggesting that a TNF-α–mediated process, presumably homeostatic plasticity, is required for the refinement of frequency selectivity observed in WT mice. Our results support a role of homeostatic plasticity in competitive refinement of sensory representations and neuronal circuits.
As we have discussed above, the more variable and incomplete frequency representations in the AI of naive KOs are likely due to impaired upregulation of sensory responses in the understimulated pathways. This is consistent with the findings that TNF-α is required for homeostatic upregulation of excitatory synapses and downregulation of inhibitory synapses in hippocampal and cortical slices (Stellwagen and Malenka 2006; Kaneko et al. 2008). Our observation of impaired competitive interaction in the overstimulated TNF-α KO mice suggests that over activation-induced homeostatic downregulation of excitatory synapses and/or upregulation of inhibitory synapses is disrupted in the KOs. Although electrophysiological studies indicate that TNF-α is not needed for homeostatic downregulation of excitatory synapses in hippocampal slices (Stellwagen and Malenka 2006), it is unclear whether homeostatic upregulation of inhibitory synapses is normal in TNF-α KO mice. Such a mechanism can be induced by sensory overstimulation, suppressing cortical responses (Knott et al. 2002). Furthermore, the underlying mechanisms may be different between in vivo homeostatic plasticity following sensory stimulation and in vitro homeostatic plasticity after neuronal stimulation. It remains to be determined whether TNF-α KO mice are impaired in homeostatic plasticity following sensory deprivation or overstimulation in vivo. While TNF-α KO mice are impaired in some forms of homeostatic plasticity and normal in one form of LTP (Stellwagen and Malenka 2006; Kaneko et al. 2008), they may well have other uncharacterized plasticity deficits that could lead to the observed impairments in the development and competitive refinement of sound representations in AI. Further characterization of different forms of synaptic plasticity in the TNF-α KO mice, and identification of the types of synaptic plasticity involved in experience-dependent cortical development and refinement, may shed new light on how TNF-α is involved in those processes.
In the present study, we focused on 2 cortical plasticity effects observed in WT mice: single-frequency exposure-induced expansion of cortical representations at the exposure frequency and multifrequency exposure-induced reduction of cortical response magnitude (Condon and Weinberger 1991; Zhang et al. 2001; Pienkowski and Eggermont 2012). Frequency map expansion appears to involve Hebbian-type, LTP-like potentiation of excitatory synapses (Froemke et al. 2007; Sun et al. 2010). The mechanism underlying sound exposure-induced response reduction is largely unknown. We considered it a type of homeostatic regulation of cortical activity (which is different from homeostatic synpatic plasticity) purely based on its impact at the level of cortical activity. In WT mice, repetitive stimulation with multifrequency tones is likely to increase the overall level of cortical activity. The observed reduction of sound-evoked responses should dampen activity levels, and counterbalance the overstimulation of the auditory cortical neurons. In the present study, we defined homeostatic regulation of cortical responses as a reduction of cortical responses induced by sensory overstimulation, or an enhancement of responses caused by sensory deprivation. We found that in KO mice single-frequency exposure induced frequency map changes, but multifrequency exposure did not cause–response reduction, suggesting a role of TNF-α in homeostatic regulation of cortical activity.
The normal single-frequency exposure-induced map expansion, impaired multi-frequency exposure-induced tuning refinement, and differentially regulated cortical response patterns observed in the TNF-α KO mice indicate that multiple cellular mechanisms are at work in shaping cortical sensory representations and response properties (Bear 2003; Burrone and Murthy 2003; Turrigiano and Nelson 2004; Dan and Poo 2006; Liu et al. 2007; Wu et al. 2008; Feldman 2009). Using genetic manipulations to target-specific cellular mechanisms, we may be able to dissect the circuits and cellular mechanisms involved in physiological and pathological plasticity. For example, both Hebbian plasticity-mediated sensory map changes and homeostatic plasticity-mediated changes in spontaneous firing rate are considered potential mechanisms underlying hearing loss-induced tinnitus and phantom pain (Eggermont 2006; Yang et al. 2011). Further studies using TNF-α KO mice may help clarify the specific roles of those cellular mechanisms in pathological phantom perception.
This study was supported by National Institute on Deafness and Other Communicative Disorders (DC009259). R.G. was supported by the National Science Foundation's Graduate Research Fellowship Program.