Mixing audio and video in dendrites

Ever since Wilfrid Rall established once and for all the importance of dendritic structure for neuronal computation, a multitude of experiments and models have studied synaptic integration in dendrites at increasing levels of detail. Dendrites can perform a number of abstract computations such as multiplication, detection of spatiotemporal sequences in their inputs and even complex image processing (London & Häusser, 2005). Many of these computations can be explained directly from cable theory, and thus the passive electrotonic architecture of the cell, while ignoring non‐linear ion channels involved in the neuron's spiking mechanisms. For example, thinner dendrites have a higher input resistance leading to larger local voltage deflections in response to synaptic currents. Also, signals attenuate more strongly in thinner dendrites leading to a compartmentalisation of the neuron's input structure.

Based on cable theory, a large number of studies – many of them in slices – have resulted in compartmental models that accurately predict the spread of currents in a wide variety of dendritic trees. The success of these models is tribute to the importance of cable theory in understanding neuronal functioning. However, only rare instances exist where dendritic computation is directly understood with respect to its function for the behaviour of an animal. Only when neurons reside in the early sensory systems such as in the retina can synaptic inputs typically be associated with their behaviourally relevant functional counterparts. All the more intriguing is the Mauthner cell involved in the startle response of the goldfish (Preuss & Faber, 2003). Inputs from two different modalities impinge on the two separate dendrites of the same cell (the ventral dendrite (VD) carrying the visual channel and the lateral dendrite (LD) carrying the auditory channel respectively; see Fig. 1). These are then processed to elicit multimodal behavioural responses.

Figure 1. Reconstruction of the thick Mauthner cell dendrites (LD and VD).

Sketch illustrating the different signal decay of same voltage inputs in LD (audio) and VD (video). Lighter shading indicates signal attenuation beyond 50% (adapted from Medan et al. (2018).

In this issue of The Journal of Physiology, Medan et al. (2018) report on a study of the integration of combined sound and optic tectum stimulation on the dendrites of Mauthner cells. Using in vivo intracellular recordings, they elucidate much of the precise dendritic computation that occurs when these two sources of inputs are active. The dendrites of Mauthner cells are thick, allowing for multiple simultaneous dendritic recordings. In the recordings, both visual and auditory stimulation led to complex voltage traces of similar amplitudes at the respective locations of their synapses (VD and LD). Both signals invaded the opposing dendrites but the response to auditory stimulation travelled further into the VD than the response to visual stimulation travelled into the LD. Simple cable theory can account for much of this effect: The LD was shown to be thicker and shorter and exhibited more side branches (5–7), while the VD was, overall, thinner and longer and had fewer side branches (4–5). These morphological differences could directly lead to the effect observed here (see Fig. 1).

The thinner VD requires less synaptic current to elicit the same local voltage response because of its higher input resistance. However, this signal propagates less efficiently towards the soma. When the signal crosses over to the LD it is confronted with a thicker dendrite that makes it hard for the visual signal to further propagate without leaking through the membrane. Conversely, the thicker LD requires more synaptic current to elicit the same local voltage response because of its lower input resistance. But the auditory signal then travels more easily into the soma and further into the thinner VD, which is easier to fill.

The asymmetry observed in these experiments is therefore actually due to the difference in current at the synaptic inputs (less in VD, more in LD, leading to the same local voltage responses in both dendrites). This would not be the case if the currents were the same. From basic cable theory, the current transfer in a passive cable is known to be perfectly symmetrical independently of the branching structure of a dendrite (Koch, 1999). This means that injecting one current in VD and measuring voltage in LD should lead to the same voltage in the VD when injecting the same current in the LD. Interestingly, this is not what Medan et al. found in Mauthner cells. The corresponding current injections showed an additional asymmetry. Such an asymmetry can only result from active voltage‐gated ion channels. Indeed, using computational modelling, the authors found that an inhomogeneous distribution of active channels along the dendrites reproduced the precise signal propagation in VD and LD.

So it seems that mixing VD and LD inputs asymmetrically is important for the Mauthner's cell function in the startle response to save the fish from being the next bird's dinner.