Müllner, Roska
6 min
Sensory information from the periphery reaches the cortex via the thalamus, where the dorsolateral geniculate nucleus (dLGN) serves as the primary visual relay. While excitatory thalamocortical neurons (TCNs) receive specialized retinal inputs, GABAergic inhibitory interneurons in the dLGN were long hypothesized to act as multiplexors that broadly integrate many different retinal channels along their extensive dendrites. To test this assumption, the researchers combined targeted single-cell-initiated monosynaptic rabies tracing with in vivo two-photon calcium imaging in mice. This approach allowed them to map the presynaptic retinal ganglion cells (RGCs) of individual dLGN interneurons and simultaneously record the visual response properties of interneuron somata and dendrites.
Using automated machine-learning tools to virtually flatten retinal image stacks and analyze dendritic stratification within the inner plexiform layer, the authors tracked 638 RGCs presynaptic to dLGN interneurons. Contrary to the classical multiplexor hypothesis, specialization Z scores revealed that interneurons are not driven by random sampling. Instead, they display significant input specialization similar to TCNs. Interneurons receive inputs from a substantial fraction of motion-selective RGC types, including ON-OFF direction-selective (DS), ON DS, and JAM-B-expressing cells. Furthermore, contralateral retinal inputs to interneurons are heavily biased toward transient response profiles, whereas ipsilateral inputs favor sustained profiles.
In vivo two-photon calcium imaging demonstrated that dLGN interneurons are functionally specialized to encode diverse visual features rather than merely providing unselective inhibition. By examining mutant mice lacking retinal horizontal direction selectivity (FRMD7 mutants), the researchers established a causal link showing that horizontal direction selectivity in dLGN interneurons is inherited directly from their retinal inputs. Crucially, this functional feature selectivity is not restricted to the cell soma; it extends uniformly into the complex dendritic arbors, indicating that individual interneurons maintain coherent feature preferences across their entire cell-wide structure.
These findings fundamentally revise our understanding of early visual processing in the thalamus. Rather than acting as homogeneous local gain controllers or uniform multiplexors, dLGN interneurons function as feature-selective selectors. By receiving biased inputs from specific subsets of retinal ganglion cells, they are ideally equipped to mediate targeted, feature-specific feedforward inhibition at the very first central synapse of the mammalian visual pathway.
Inhibitory interneurons in the dorsolateral geniculate nucleus (dLGN) are situated at the first central synapse of the image-forming visual pathway, but little is known about their function. Given their anatomy, they are expected to be multiplexors, integrating many different retinal channels along their dendrites. Here, using targeted single-cell-initiated rabies tracing, we found that mouse dLGN interneurons exhibit a degree of retinal input specialization similar to thalamocortical neurons. Some are anatomically highly specialized, for example, toward motion-selective information. Two-photon calcium imaging performed in vivo revealed that interneurons are also functionally specialized. In mice lacking retinal horizontal direction selectivity, horizontal direction selectivity is reduced in interneurons, suggesting a causal link between input and functional specialization. Functional specialization is not only present at interneuron somata but also extends into their dendrites. Altogether, inhibitory interneurons globally display distinct visual features which reflect their retinal input specialization and are ideally suited to perform feature-selective inhibition.
Alex: That's precisely the challenge. The researchers used a technique involving a specially modified rabies virus. Normally, rabies spreads aggressively through the nervous system. But this modified version is engineered to enter just one genetically tagged cell and then jump only to its immediate neighbours — the cells directly connected to it. So you get a precise map of one cell's input network.
Sam: Like pulling on one thread to see exactly which other threads it's tied to.
Alex: That's a good way to put it. And alongside that, they used a technique called two-photon calcium imaging, which lets you watch neurons light up in real time as they respond to visual stimuli in a living mouse. So they could see not just the wiring, but also what each cell actually responds to.
Sam: What did they find?
Alex: They found that individual inhibitory interneurons were not collecting random mixtures of inputs. Instead, each cell was receiving signals predominantly from retinal ganglion cells of the same functional type — cells tuned to the same visual feature, like a particular direction of motion or a specific contrast pattern.
Sam: So rather than a mixing bowl, it's more like each interneuron has a specialised diet — only taking in one kind of visual information.
Alex: That's a clear way to put it. And what made the finding more compelling is that this selectivity wasn't just at the cell's core — it extended across the entire branching structure of the neuron. The dendrites, which are the tree-like arms that receive incoming signals, showed the same feature preference throughout.
Sam: So the whole cell is tuned consistently, not just parts of it?
Alex: Exactly. That's what the researchers call the selector model — the idea that these interneurons are structurally organised to amplify or gate specific visual features, rather than blend them. The wiring matches the function all the way through.
Sam: That raises an obvious question though — what are the limits of this study? How confident can we be in these conclusions?
Alex: That's worth being careful about. The imaging was spatially biased toward the outer layers of the dLGN, so cells deeper inside weren't as well represented. And when measuring the branching arms of these neurons, the researchers used straight-line estimates rather than tracing every curve and bend in three dimensions. That likely underestimates the true complexity of the dendritic structure.
Sam: So the picture is suggestive, but not complete.
Alex: Right. The study makes a strong case for the selector model, but the authors are clear that more work is needed — particularly to examine cells deeper in the structure and to build more detailed three-dimensional maps.
Sam: And if the selector model does hold up more broadly, what does that mean for how we understand visual processing?
Alex: It reframes the role of this relay station significantly. Rather than being a passive throughput — just forwarding signals to the cortex — the dLGN appears to be doing active, selective work. These interneurons may be shaping which features get emphasised before the signal even reaches conscious processing. That has implications for understanding conditions involving disrupted visual attention, and potentially for future approaches to treating sensory processing disorders — though that's speculative at this stage.
Sam: So a structure we thought was relatively simple turns out to be doing something considerably more organised.
Alex: That's the suggestion, yes. And it's a reminder that even the earliest stages of sensory processing involve more structure and specificity than we previously gave them credit for. Thanks for listening to ResearchPod.