Figure 2:

Towards a classification system based on thalamo-cortical motifs. In panels A and B, individual and idealized single thalamic neurons are shown to illustrate the notion of the motif being a single-cell attribute (A) Example of a well-characterized thalamocortical motif, involving the retino-geniculo-cortical pathway, with the set of computations required for retinal signal transmission. X and Y geniculate neurons of the cat receive retinal inputs with minimal convergence, and therefore show similar responses to those of the retina. An important feature of this system is that the retinogeniculate connections are relatively stable in adult animals, rendering geniculate neurons stably tuned to visual features. On the output side, because geniculate neurons providing predominantly driving excitatory inputs to visual cortex, cortical responses can be largely explained as weighted sums of the thalamic output (B)Two examples of inferred motifs within the mediodorsal thalamus of the mouse. These connections are derived from statistical dependencies between the mediodorsal neurons and prefrontal ones, which are recorded in a context-switching task (Rikhye et al., 2018). These statistical dependencies have also been tested through pathway-specific optogenetic manipulations. Mediodorsal neural types can be segregated based on inputs, which are explained by the degree of their cortical input convergence. Specifically, a subset of neurons encodes a set of task-relevant prefrontal cue-selective signals over a broad temporal scale (left), and another encoding the same type of task-relevant variable but on a shorter timescale (right). Given that these response profiles shift on a session-by-session basis depending on how the context is experimentally configured, a reasonable interpretation is that the cortico-thalamic connections are highly plastic. These same mediodorsal types also segregate based on output, with the high input convergence neurons exhibiting predominantly suppressive effects on prefrontal cortical activity, while the low input convergence neurons exhibiting predominantly modulatory excitatory effects. By modulation, we do not necessarily mean that the effect is implemented through a neuromodulator, but rather that it controls the gain of effective recurrent connections in the prefrontal cortex (see equations within the figure). (C) A putative classification space for thalamocortical motifs, with the relevant geniulate and mediodorsal types plotted within. The number and distribution of points within this space are currently unknown, but we expect that their pattern will inform function and comparisons across species.

Notation: the embedded equations describe the input-output transformations of the thalamocortical motifs. Lower boldface symbols denote vectors and upper boldface symbols denote matrices. r_cortex: cortical output, f(): cortical non-linearity, r^_cortex: recurrent drive, r_thal: thalamic input, g(): non-linearity over thalamic input for the middle motif, b: scaling parameter for the thalamic term in the last motif. The key idea for this formalism is that the thalamic input shows up as very different terms in the cortical computation performed. This is what is precisely meant by relay vs. non-relay functions of the thalamus.