Neural Circuits: How the Brain Introduces Different Scales of Temporal Processing

Abstract

A new study describes a novel passive integration mechanism of inhibition in auditory neurons in the dorsal nucleus of the lateral lemniscus that turns extremely well-timed synaptic events into a signal code that is 3 orders of magnitude slower.

Sensory information needs to be processed by the brain at different timescales simultaneously for a meaningful representation of the exterior environment. In the auditory world, time is one of the key variables and encompasses a wide range of processing speeds from microseconds up to tens of seconds. In a research article published in this issue of Current Biology, Ammer et al. [1] describe a cellular mechanism in neurons of the dorsal nucleus of the lateral lemniscus (DNLL) of the auditory brain stem that creates a neural code on a timescale on the order of tens of milliseconds from neural inputs that are some of the fastest and temporally precise in the entire brain.

In the auditory system, spatial information is not inherently contained in the information captured by the ears, as it is for example in the visual system, where the retina could be seen as a projection screen of the outside world. Because the main organizing principle in the auditory system is frequency not space, spatial information has to be computed in the brain by comparing time and intensity differences that a given sound creates between the two ears, thus re-creating representations of space computationally [2]. These time and intensity differences arise because, depending on the location from which a sound is emitted, it arrives a moment earlier and will be a louder at the ear closer to the sound source than at the more distant ear. While there are two separate neural pathways to compute the interaural time and intensity differences, both pathways critically rely on precise timing on the order of tens microseconds to hundreds of microseconds [3–5]. Given these requirements, it is not surprising that the lower auditory system devised a number of specializations to tackle the problem of high-speed synaptic transmission and extremely well-timed signal processing to detect these time and intensity differences, and deal with temporal codes that which are smaller than even the width of a single action potential [4, 6, 7]. While localization is achieved at the level of the superior olivary complex (SOC, green nuclei in figure 1) of the auditory brainstem, the auditory system also has to deal with signal detection and processing on much longer timescales: for example, speech recognition, envelope structure encoding of complex biologically relevant sounds, or echo detection/suppression. One prominent nucleus that has been shown to be instrumental for echo suppression is the DNLL [8–11]. Its likely functional role is to ensure correct localization of sounds in echoic environments including all indoor spaces, where large objects, walls, and ceilings reflect the sound. In such scenarios, the listener’s ears receive the initial direct wave front that was emitted by the actual sound source. Within a few milliseconds to tens of milliseconds after that, multiple copies of that sound –the echoes- arrive at the ears from multiple directions. The first sound localization circuit in the SOC cannot discriminate between sounds and echoes, which would lead to a confusion of the listener. Just one processing step further, at the level of the DNLL, this discrimination can be made. How is this achieved? DNLL neurons receive GABAergic inhibition from the contralateral DNLL through the commissure of Probst [12, 13], red arrow in figure 1B. This inhibition outlasts the duration of the sound stimulus by up to several tens of milliseconds, and suppresses neural activity for this time period in DNLL neurons – a duration closely matching the duration of typical echoes (figure 1C). This phenomenon is commonly known as the “precedence effect” [11].

Figure 1.

The circuit of the precedence effect inhibits DNLL neurons for the duration of typical echoes, thus removing DNLL input from the inferior colliculus for the same time period. A: Sound from the left hemisphere arrives at the left ear at a louder intensity than at the right ear due to the sound shadow created by the head. This intensity difference is computed and systematically represented in the SOC (green nuclei), specifically the LSO. The computation involves integration of excitation from the left ear (left green arrow) and inhibition from the right ear via an extra synapse at the MNTB (right green/red arrow). B: The computational output is relayed, among other targets, to the DNLL via an excitatory projection (green arrow). DNLL neurons are activated and in turn inhibit neurons in the opposite DNLL via a GABAergic inhibition (red arrow). This inhibitory projection has very long lasting properties that are created by a combination of spillover, asynchronous release, and passive membrane properties (Ammer et al, this issue). C: Due to the long lasting inhibition, echoes of the initial sound that arrive from various directions cannot excite DNLL neurons and thus these neurons are removed as inputs to IC for as long as the GABAergic inhibition is active. Abbreviations: IC: inferior colliculus; DNLL: dorsal nucleus of the lateral lemniscus; INLL: intermediate nucleus of the lateral lemniscus; VNLL: ventral nucleus of the lateral lemniscus; MNTB: medial nucleus of the trapezoid body; MSO: medial superior olive; LSO: lateral superior olive.

The complex circuit revolving around the precedence effect has been studied in-vivo, in-vitro, on a psychophysical level, has been modeled [8–11], and has even been recreated in the engineering of household robots [14], allowing them to localize accurately in indoor environments. However, one major gap in our understanding has been the question by which cellular or subcellular mechanisms DNLL neurons are suppressed for these long time periods. Ammer et al. show that this is done by a combination of two synaptic mechanisms (transmitter spillover and asynchronous vesicle release) and a novel passive integration mechanism of inhibition that substantially prolongs the hyperpolarizing effects of the GABAergic conductance in DNLL neurons. Transmitter spillover and asynchronous release have been described to prolong inhibitory and excitatory synaptic transmission in other studies [15–20]. By voltage- and current-clamping DNLL neurons, Ammer et al. show that both mechanisms are activated following high-frequency stimulation of the input fibers. The presynaptic firing rate of auditory neurons has been shown to code for stimulus intensity, while the suppression of signal transmission in the precedence effect has been shown to extend with increasing sound intensity as well [8–11]. More importantly, Ammer et al. discovered a novel mode of inhibitory integration that seems to tie the amplitude of an inhibitory conductance (GABA in this case) with the decay kinetics of the resulting hyperpolarization in the postsynaptic neuron. According to their findings, a larger inhibitory conductance causes a longer hyperpolarization of a DNLL postsynaptic cell, which cannot be explained by the much faster decay kinetics of the (pre)synaptic currents and the passive membrane properties of the DNLL neurons themselves. As a possible explanation, Ammer et al. offer a mechanism of synaptic saturation.

These three mechanisms together lead to a marked amplification of even small differences in the decay times of the initially much faster presynaptic currents in an activity-dependent manner.

As a downstream-result of that suppression, DNLL neurons are inhibited for prolonged periods of time following an auditory input, and temporarily do not tune the receptive field of their target neurons in the next higher brain nucleus, the inferior colliculus. Thus, the properties of this circuit are consistent with our experience in echoic environments: we hear echoes (for example in an empty room with hard surfaces) but we do not localize them.

It remains to be seen whether other factors could play a role in extending the timescales of a neuronal response Naturally, another very intriguing question is whether this described mechanism of passive integration is more generally exploited by other neuronal populations in different areas of the mammalian brain. This could hint at a common strategy of neuronal circuits to introduce different layers of time processing in their input-output functions.