How do neural circuits compute?
We are happy to welcome Matthew T. Soleiman, a student in the Graduate Program in Neuroscience and the Department of Biochemistry at the University of Washington, Seattle, WA, as a guest blogger on the PLOS Neuro Community! Matthew’s posts focuses on recent exciting findings on the cortical-thalamic-cortical circuit, with great insights from the authors.
The neuron has been considered the computational unit of the brain. But the brain is not just a clump of neurons. Rather, populations of neurons are connected and organized into specific circuits, each with a number of inputs and outputs. Modern techniques in manipulating and monitoring circuits have revealed them to be the computational units that transform input from the senses into behavioral output. The question then is—how do circuits compute? While Ralph Adolphs of Caltech estimates it should take about 50 years to have an answer, a new paper published in Neuron from the lab of Barry Connors at Brown University in Providence, Rhode Island, might help us get there a bit quicker.
The paper reports on a series of experiments probing the circuitry connecting the somatosensory cortex and the thalamus. Excitatory neurons of layer 6 of the cortex target two nuclei within the thalamus—the ventral posterior medial nucleus (VPm) and the thalamic reticular nucleus (TRN). The TRN in turn targets the VPm. Because the TRN is composed of inhibitory neurons, L6 cortical neurons can then directly excite and indirectly inhibit VPm neurons (see Figure). Complex enough? Oh, and the cortex receives a hefty input from the VPm.
“This is a system that has been known anatomically for ages,” says Connors. “Its general function is still unclear, and precisely how it works physiologically and dynamically has also not been clear.”
Upstream to downstream
With both direct and indirect routes of information from L6 neurons to VPm neurons, how is firing upstream transformed into firing downstream? The authors tested this question by using the now-standard technique of optogenetics, whereby neural circuits can not only be activated by light, but also at specific frequencies so as to try to mimic the natural firing of neurons. While recording from VPm neurons in brain slices, L6 neurons were activated at a low frequency (0.1Hz). VPm neurons were quickly excited, and then mainly quieted. However, at a higher frequency (10Hz), these cells increased their firing vigorously. The take-home message: the cortical-thalamic circuitry computed information and responded differentially depending on the upstream firing rate. What does this imply for the function of the circuitry?
“What we think is happening is that the cortex has the ability to both suppress and enhance its own information” suggests Shane Crandall, a postdoctoral fellow and the paper’s lead author.
So when L6 neurons are firing at a low level, their input is suppressed. But when they switch to a higher rate, it’s enhanced. Exactly what mechanisms could explain these dynamics?
Facilitating excitation and depressing inhibition
The experimenters went on to show that the dynamics could be explained by the synaptic properties of the circuitry. At first, inhibition dominates over excitation, leading to the overall inhibition of VPm neurons during low-frequency firing of L6 neurons. Yet, with their continued firing, the balance of excitation and inhibition shifts. The excitatory synapse connecting L6 and VPm neurons increases in strength with time, while the inhibitory synapse connecting TRN and VPm neurons decreases in strength. These are in fact common synaptic phenomena known as short-term facilitation and short-term depression.
Common circuit properties?
This specific cortical-thalamic system is just a piece of the vast circuitry that makes up the brain. But this direct-excitatory and indirect-inhibitory (aka feed-forward inhibition) structure is not unique.
“Nearly all thalamic-cortical and cortical-thalamic circuits have feed-forward inhibition” says Scott Cruikshank, assistant professor at Brown and the second author of the paper.
The circuit architecture can even be found in the hippocampus (Owen et al., 2013) and the midbrain (Stamatakis & Stuber, 2012; Watabe-Uchia et al., 2013).
The authors note that despite similarities, there are likely important differences.
“They’re not all going to work dynamically the same way,” adds Connors. “There’s going to be a wide range of manifestations of this”.
So what will determine the computational properties of a circuit? Connors identifies three crucial elements: the connectivity, the dynamic synaptic features, and the neurons’ intrinsic electrical properties.
Implications for optogenetics & behavior
The number of studies using optogenetics to find which circuits control which behaviors has exploded. Occasionally, an effect on behavior using one activation frequency is controlled for by using a lower frequency or a different pattern of activation. For example, Land et al. (2014) found that activating cortical neurons expressing the D1 receptor for dopamine at 20 Hz, but not 5 Hz, increased food intake. The findings from Crandall et al. might then help us understand why higher frequencies or specific activation patterns produce behavioral effects.
More often experimenters choose a single frequency to use, although the stimulation frequency is usually based, at least, on recordings in slices. Connors emphasizes the importance of trying to recreate the normal activity of neurons. Another issue discussed by the paper is the range of speeds that L6 neurons signal at. If the input to VPm from L6 neurons is not as synchronized as the artificial activation using optogenetics, then the suppression seen with low frequency may become even greater. Experimenters activating other circuits with neurons that signal at different speeds may then want to consider how optogenetic activation is different from the normal activity of neurons.
“In most applications, you’re still activating a bunch of them synchronously, and that’s quite different from the way most brain circuits work,” cautions Connors.
How well do these findings generalize beyond the slice? It’s worth noting that the frequencies used have been observed in the intact brain. While cortical neurons are known for being almost silent (see here for a recent example), they fire at much higher rates when an animal is stimulated (see here for an example). In humans, attention has been shown to control the intensity of activity of the lateral geniculate nucleus, another thalamic nucleus (O’Connor et al., 2002; Ling et al., 2015). Even though we cannot know if attention mimics the effects seen with sensory stimulation in rodents, at the very least, we know the thalamus is subject to higher-level control. It seems likely then that the circuit mechanisms identified by Crandall et al. are somewhat similar in humans.
Adolphs, R. (2015). The unsolved problems of neuroscience. Trends Cogn Sci 19(4):173-5. doi: 10.1016/j.tics.2015.01.007
Crandall, SR et al. (2015). A Corticothalamic Switch: Controlling the Thalamus with Dynamic Synapses. Neuron 86(3): 768-82. doi: 10.1016/j.neuron.2015.03.040
Kwegyir-Afful, EE & Simons, DJ. (2009). Subthreshold receptive field properties distinguish different classes of corticothalamic neurons in the somatosensory cortex. J Neurosci 29(4):964-72. doi: 10.1523/JNEUROSCI.3924-08.2009
Land, BB et al. (2014). Medial prefrontal D1 dopamine neurons control food intake. Nat Neurosci 17(2):248-53. doi: 10.1038/nn.3625
Ling, S et al. (2015). Attention alters orientation processing in the human lateral geniculate nucleus. Nat Neurosci 18: 496-98. doi: 10.1038/nn.3967
O’Connor, DH et al. (2002) Attention modulates responses in the human lateral geniculate nucleus. Nat Neurosci 5(11):1203-9. doi: 10.1038/nn957
O’Connor, DH et al. (2010). Neural activity in barrel cortex underlying vibrissa-based object localization in mice. Neuron 67(6):1048-61. doi: 10.1016/j.neuron.2010.08.026
Owen, SF et al. (2013). Oxytocin enhances hippocampal spike transmission by modulating fast-spiking interneurons. Nature 500(7463):458-62. doi: 10.1038/nature12330
Stamatakis, AM & Stuber, GD. (2012). Activation of lateral habenula inputs to the ventral midbrain promotes behavioral avoidance. Nat Neurosci 15(8):1105-7. doi: 10.1038/nn.3145
Watabe-Uchida, M et al. (2012). Whole-brain mapping of direct inputs to midbrain dopamine neurons. Neuron 74(5):858-73. doi: 10.1016/j.neuron.2012.03.017
Any views expressed are those of the author, and do not necessarily reflect those of PLOS.
Matthew Soleiman is currently finishing his graduate work at the University of Washington on the cell types and circuitry of the central amygdala. In parallel, he is working as a freelance science writer. You can find him on Twitter as @MatthewSoleiman, or contact him at firstname.lastname@example.org.