Thalamocortical function in developing sensory circuits

Abstract

Thalamocortical activity patterns, both spontaneous and evoked, undergo a dramatic shift in preparation for the onset of rich sensory experience (e.g. birth in humans; eye-opening in rodents). This change is the result of a switch from thalamocortical circuits tuned for transmission of spontaneous bursting in sense organs, to circuits capable of high resolution, active sensory processing. Early ‘pre-sensory’ tuning uses amplification generated by corticothalamic excitatory feedback and early-born subplate neurons to ensure transmission of bursts, at the expense of stimulus discrimination. The switch to sensory circuits is due, at least in part, to the coordinated remodeling of inhibitory circuits in thalamus and cortex. Appreciation of the distinct rules that govern early circuit function can, and should, inform translational studies of genetic and acquired developmental dysfunction.

INTRODUCTION

The development of function in sensory systems requires the convergence of two interdependent processes: (1) the sequential emergence of relative selectivity for specific stimuli by individual neurons (receptive fields); and (2) the maturation of network level dynamics that determine information flow, timing and modulation by behavioral state. The developmental timing and mechanisms of the former have been extensively studied[1-5]. However, the development of sensory processing is only partially explained by receptive field maturation[6]. Network level dynamics may fill this gap. Here we explore recent evidence that the network properties of central thalamocortical circuits are tuned to development, with each stage representing a ‘developmental niche’ in time, each with its own goals and brain network dynamics to achieve them. In other words, maturation of network properties should be understood not as a simple march to adulthood, with early networks modeled as degraded adult networks, but as a series of circuit configurations which balance receptive field formation with changing needs for sensory processing at each age.

We focus on perhaps the most dramatic shift in developmental niche: exit from the womb, which in humans and precocial animals brings unfiltered sensory experience as well as active sensing. Altricial animals make this transition post-natally, when eyes and ears open and they begin active-touch (e.g. whisking). Before this shift, in the ‘pre-sensory period’, all three non-chemical senses produce spatially-restricted, spontaneous (not sense-dependent), bursting activity in the sense organ (Figure 1). This activity is critical for the initial establishment of topography[7]. If early thalamocortical network properties are indeed ‘tuned’ for these pre-sensory activities, we would expect them to prioritize the location signal (i.e. presence or absence) of spontaneous bursts over any graded quality of the input signal at each locus. We propose they do this by maximizing amplification and synchronization of inputs at the expense of processing capacity. Likewise, the developmental switch to adult-like activities should be correlated to a loss of spontaneous bursting activity in the sense organs and the onset of active, exploratory sensing at the behavioral level[8].

Figure 1. Two modes of thalamocortical network function in sensory cortex during development.

(a) Cartoon quantification of cortical responses to brief sensory stimuli over development. Traces show whole-field light-flash responses in rat V1 (multi-unit activity (MUA)[58] and depth EEG[36]) and human preterm infant occipital cortex[36]. Pre-vision, light produces long lasting, large oscillatory responses. After the sensory switch, responses are short and smaller. Timelines below show change in evoked network properties as outlined in the text. Red box depicts timing of vision onset at eye-opening. Width of time courses metaphorically represents relative quantity. Overlain on the sparsity graph are calcium images of visual responses in unanesthetized ferret before and after the switch[**40]. (b) Depiction of spontaneous activity in cortex. Top traces are in vivo intracellular voltage in unanesthetized neonatal rat V1 before and after eye-opening[*28]. Below is an example depth EEG from mouse V1[27]. Cartoon quantifications show the disappearance of early activity patterns and emergence of adult ones, while below are shown quantitative characterization of background activity.

We review evidence for this hypothesis, first by describing the specific network properties that characterize early thalamocortical function and their transformation to adult-like network dynamics, and then reviewing current progress to define the circuit basis of early activity patterns and their maturation.

Bursting input to thalamocortex during pre-sensory period

The mechanisms and patterning of activity generation in the sense organs during development have been extensively studied in the visual system[9,10]. Before eye-opening, a succession of transitory circuit properties in retina lead to the spontaneous generation of activity waves. By sequentially activating local patches of ganglion cells on a background of silence, these waves provide the information needed to refine topography and eye-segregation in central structures, and also contribute to the initial generation of basic receptive field properties such as orientation selectivity[1].

The cochlea also produces highly localized bursting before hearing onset[11], but the somatosensory system differs. It generates both localized bursting, via myoclonic jerks during active sleep, as well as topographically unrestricted tonic firing due to volitional movements during waking. Such broad firing would be expected to disrupt the local synchrony required for topographic alignment [12]. This is prevented in vivo, because activation of forebrain regions by somatic stimuli during volitional (waking) movements is suppressed during the first two post-natal weeks by an inhibitory gate in the cuneate nucleus[**13,14]. Because of this gate, somatosensory activity only reaches early thalamus and cortex when it is produced by spontaneous myoclonic jerks, small muscle twitches, or when it is unexpected[15].

This tight control over incoming activity suggests that locally restricted bursting is essential for development of all three non-chemical senses. Bursting is so important, that when the developing brain cannot block other inputs during active sensing at the receptor level, it takes other active measures to do so.

Pre-sensory thalamocortical networks have a unique mode of function

Even a cursory glance at the EEG trace of a preterm infant or pre-eye-opening rat reveals thalamocortical dynamics grossly different from the adult ([16]; Figure 1). Despite diversity among ages, brain regions, and species, activity in the presensory ages is distinguished from the adult by three major characteristics that define a single thalamocortical mode of function: (a) the presence of (largely) thalamic oscillations triggered by spontaneous sensory input; (b) minimal ‘background’ activity resulting in a discontinuous EEG; (c) poor modulation of activity by arousal state. Recent reviews detail these early thalamocortical activities[7,16-18]; here we outline only key relevant points.

Oscillatory networks

Throughout the pre-sensory period, activity in the primary somatosensory (S1) and visual (V1) cortex of rodents is dominated by rhythmic activity ‘bursts’ that can be clustered into two dominant patterns: ‘early gamma oscillations’ (eGO) and ‘spindle-bursts’. eGO are rapid oscillations (30-50Hz) that engage only superficial layers of a single column. Spindle-bursts are slower(6-20Hz), but engage all layers and synchronize multiple columns. Both oscillations are triggered by brief input to thalamus—in S1, eGO are triggered by single whisker stimulation, and spindle-bursts by multiple whiskers—but neither are dependent on external stimulation. Instead they occur largely as a response to spontaneous activity in the sense organ as outlined above. Human infants display a prominent spontaneous oscillatory burst, the delta-brush, which is a likely homologue of the spindle-burst as it is triggered by touch, sound or light in electrodes overlying the appropriate sensory cortex[19].

Spindle-bursts and eGO are different expressions of the same oscillator located in relay thalamus, supported by several findings: Both patterns have current sources resembling that of a direct thalamic input, thalamic oscillations can match each frequency, thalamic lesions eliminate both oscillations, and GABA_A antagonists eliminate both patterns when infused in thalamus[20] but not in cortex[21]. A second potential contributor to the early oscillations is subplate, a transient population of early-born neurons that receive thalamic input and make synaptic connections with overlying cortex[22]. Recent work in ferrets[**23] shows that subplate neurons are the first cortical neurons to fire to auditory input, then become synchronized with overlying cortex in early oscillations. While clearly important for establishing thalamic input and the early oscillations[24], the exact role of subplate in generating early oscillations is unknown. Possibly they function as a (resonance) amplifier of the rhythmic thalamic input, particularly during very early development before thalamic ingrowth to layer 4 [25].

Lack of continuity & state modulation

Between bouts of oscillations the pre-sensory cortex is largely quiet (called ‘discontinuity’). In rodents, continuous background activity, which is the basis of commonly observed EEG patterns modulated by vigilance, begins at the same age that evoked eGO/spindle-bursts disappear[26,27]. This emergence of continuous, adult-like state modulation is driven by the rapid development of bistability in the network leading to the ‘active’ (up)-state[*28]. Active states emerge abruptly in rat V1 on P12-13, resulting in the strong modulation of background EEG by state at eye-opening (Figure 1, 2). Network primitives of the active-state appear 2-3 days before this, but these cannot be considered true active states as their depolarization is variable and unbalanced[*28-30]. Human infants undergo a similar shift to continuity, driven by slow-waves between the rapid rhythms[31], but acquisition is more gradual (perhaps because EEG integrates over larger areas) and state dependent. However, full continuity and sleep-states (but not ultradian rhythms) similar to adult are acquired within months after term[32] (Figure 2).

Figure 2. Timing of switch in thalamocortical processing mode across species and systems.

Aligned timeline for species and sensory systems in which the change in network dynamics has been examined. For animal models, each sensory system is shown separately, with species examined on individual lines. The main drives are shown at the top of the box with the ages of first observed cortical response for each species below. The vertical green box shows the beginning and end of the switch period with ages for each species bracketing it. Human infants are shown below for all sensory systems, which appear to develop responses on a similar time-course in anticipation of birth. References are as follows: Rat V1[36,60], mouse V1[27,61], ferret V1[62-64], rat S1[36], ferret A1[**23], human[36,41,65,66].

So far we have described differences in spontaneous and evoked activity that define pre-sensory ages and support their consideration as a singular developmental stage with network dynamics vastly different from the adult. There remain critical developmental changes that occur after sensory onset. Most notably, there is the experience-dependent development of cortical gamma-oscillations[20,*33,*34] and the later emergence of thalamocortical sleep-spindles[35]. Furthermore, there are layer-specific differences for state modulation of processing, particularly a delay in the most superficial layers[*34].

Pre-sensory booster circuits prioritize detection at the expense of discrimination

What processing features do early thalamocortical dynamics prioritize and what can’t they support? Pre-sensory processing differs from adult in three major ways: amplification, replay, and dense coding. While conceptually separable, each may arise from the same circuit changes and be co-dependent. For example, replay and dense firing contribute to the process of amplification. Evidence for amplification comes from V1, where the ratio of cortical to retinal firing in the pre-sensory period is ~30 times that of an older P13-15 pup, and even the smallest input generates a saturating response[36]. The role of amplification is likely to maximize the down-stream impact of even weak inputs to thalamus in order to support topographic synapse formation. Evidence for replay is best observed in S1, where single whisker deflection results in repetitive, rhythmic reactivation of local neurons suggesting an internally mediated replay of the recent experience[20]. This rapid replay induces potentiation at thalamocortical synapses. Evidence of dense coding comes from imaging in anesthetized mice, which showed high participation rates during early network events in V1 and S1 [*28,37,38]. Early dense firing results in hyper-synchronization on long (200+ms) but not short (10-200ms) timescales[39], suggesting that its role is to assure all potential neurons are active in order to maximize circuit formation. In V1, sparsification of L2/3 firing develops over a more extended time course than the loss of replay and amplification, suggesting different mechanisms.

The idea that these early circuit dynamics are actively detrimental to sensory processing is supported by recent work, and helps explain why the sensory switch is completed in advance of active sensing. In unanesthetized ferrets around eye-opening, dense firing and a propensity to trigger waves reduces visual discrimination[**40]. Sparsification of visual responses develops in parallel with direction-selectivity, suggesting a common mechanism, probably cortical inhibition. In rat V1, early bursting occludes graded responses to luminance, and reduces the temporal fidelity of the response[36]. In human cortex, early network dynamics obscure electrophysiological differentiation of painful and touch stimuli at the same location[41] (i.e. discrimination of topographically similar inputs) but not differentiation of different syllables[*42] (i.e. tonotopic/topographic discrimination).

These data suggest that early networks prioritize faithful transmission of certain information such as topography and frequency tuning, while minimizing the processing of amplitude and informational subdivisions of the receptive field such as direction selectivity or pain, in order to maximize appropriate circuit formation.

Synchronized, multi-level changes in inhibitory circuitry drive switch in network dynamics

Our understanding of the network mechanisms underlying pre-sensory network dynamics remains incomplete. The only clear in vivo evidence implicates thalamus as a critical locus. In the visual pathways, the relay thalamus and its corticothalamic inputs form a feedback excitation loop, creating a ‘booster’ circuit that amplifies retinal input as much as 5-fold[**43]. Subplate circuits may play a similar feedback amplification role through their interconnection with L4, or by participating in the corticothalamic loop[44]. Recurrent connectivity in L4, which amplifies input in adults, is probably not a factor in the neonate[*45]. Another potential contributor to early dynamics is the ability of the early network to drive plateau potentials--a maximal gain saturation--both in thalamus[**46] and cortex[*28].

There are a number of circuit changes that might contribute to the elimination of sensory bursting (Figure 3). The most promising is a massive and distributed reorganization of GABAergic circuitry which occurs simultaneously in thalamus and cortex. In V1 and S1 feedforward inhibition, which is the result of strong thalamic synapses onto parvalbumin interneurons, emerges in synchrony with the switch to sensory processing[17,36]. During prolonged spontaneous spindle-bursts, synchronized GABAergic currents are observed. A number of transient circuit configurations may support this early inhibition. L4 interneurons receive local AMPA-receptor connections, as well as NMDA-receptor-only ‘silent’ synapses from subplate and L5[47]. The latter likely contribute only during prolonged activation during spindle-bursts. A transitional interlaminar circuit involving L5 somatostatin interneurons is also present during the pre-sensory period[**48,49]. These neurons make feedback inhibition with L4 principal cells, but also receive input from thalamic axons. This early GABAergic circuitry restricts the spread of activity to the appropriate topographic area[50]. It may aid the synchronization of deep and superficial layers[29]. These early populations, subplate and L5 somatostatin neurons, are critical to the development of adult inhibitory circuits[22], by allowing thalamic axons to make strong synapses on L4 PV-interneurons at the switch.

Figure 3. Major thalamocortical circuit rearrangements associated with the switch in processing.

Simplified diagrams for four major identified circuit changes discussed. For each circuit the primary function at each age is shown with relevant connectivity changes below. References by panel: A[**43,67];B[20,51,67];C[22,**23];D[20,47-49].

Along with the development of feedforward inhibition in cortex, the corticothalamic excitatory feedback that comprises the pre-sensory booster is silenced by the rapid development of feedforward inhibition, likely from the thalamic reticular nucleus[**43]. Feedforward inhibition of retinal input to thalamus develops around the same time[51].

How three distant inhibitory circuits develop in such coordinated fashion is unknown. In cortex, synaptic strengthening on PV neurons requires NMDA-receptors and intact sensory input[**48,49], suggesting that an increase in activity at this time drives the inhibitory circuitry into new and more effective configurations.

After the switch to sensory processing, inhibitory development is by no means complete. Continuing maturation of multiple interneuron sub-types, particularly in superficial layers, contributes to the delayed sparsification and wave-suppression, as well as the eventual development of adult gamma-oscillations and critical period plasticity[52-54].

Conclusions and future directions

In this review, we have focused on recent evidence that the pre-sensory thalamocortex, roughly equivalent to the second half of gestation in humans, is specialized for the transmission of spontaneous activity in the sense organ at the expense of processing capacity. In addition to informing our basic understanding of the development of sensory processing and mechanisms of circuit formation, these findings should be applied to the design and interpretation of clinical and translational studies. We conclude with emphasis on two important points that impact clinical research.

First, the pre-sensory period must be considered largely non-continuous with the sensory period. This includes the common reification of frequency bands, such as gamma and delta, into discrete network functions[55]. Fetal frequencies do not cohere as in adults, nor do individual frequencies derive from the same circuits. Thus, comparison of frequency power or distribution across these age barriers is largely meaningless. This principle should also be kept in mind when interpreting derived measures of neural activity based on blood-flow, as ‘activation’ of a region may be a result of different circuits (and thus different information transfer) during the fetal period than even full-term neonates.

Second, the circuitry of the pre-sensory brain is an important potential window into the mechanisms of acquired and genetic dysfunction. The fact that early circuitry is so functionally different from adult means that adult circuits can be disrupted by transient early activity, which then normalizes around term, obscuring evidence of the original defect. For example, hypoxia-ischemia at P3 disrupts spindle-burst production during initial circuit formation, but activity appears to normalize over time, hiding the initial insult [*56]. Subplate circuits are disrupted in an autism model mouse[57], while knockout of FMRP1 causes hypo-excitability during the pre-sensory period, but hyper-excitability later[58,59].

We expect that being able to disrupt activity in specific cell types during early development will resolve the current ambiguities of early circuit function to allow for better understanding of early neural diseases and be able to tailor interventions to these unique ages.

HIGHLIGHTS.

• A switch in thalamocortical dynamics defines ‘pre-sensory’ and ‘sensory’ stages.

• Pre-sensory networks prioritize detection at the expense of discrimination.

• A thalamic booster circuit drives amplification in the pre-sensory brain.

• Coordinated maturation of inhibition accompanies switch to a sensory network.