I was reading a text on graph theory applied to neuroscience, Networks of the Brain (Sporns, 2016), where in chapter 8 the author talks about spontaneous neural activities in the brain. The chapter also mentions up (high subthreshold membrane potential) states of neurons and that recurrent neural connections are likely to be the cause of synchronized up states of multiple neurons, his reasoning being that these up states last longer than what feedforward connections would be able to achieve.

As I understand it, though, the reason recurrent connections would be able to maintain a longer period of activations is because, for example, neuron A activates, further activating other neurons whose signals loop back to A, and repeat. However, up states are subthreshold membrane potentials, so if A sends a signal to other neurons and changes them into an up state, the other neurons would not fire back to neuron A and thus the loop gets discontinued. How would the recurrent mechanism work in this case?

Sporns, O. (2016). Networks of the Brain. MIT press


2 Answers 2


There's lots of spiking activity during UP states, by both excitatory and inhibitory neurons.

Neske et al (2015) measured firing rates during UP states in different cell types of layers 5 and 2/3. Among excitatory cells, the highest rates were among layer 5 pyramidal cells with a "bursting" spike-type (this spike type is associated with the big, thick-tufted layer 5 cells) at 15.8 Hz, followed by L5 regular-spiking (RS) cells (4.9 Hz) and L2/3 RS cells (1.7 Hz). Among inhibitory cells, parvalbumin (PV) cells in L5 and L2/3 fired at 30.6 and 26.5 hz, respectively; somatostatin (SOM) cells in L5 and L2/3 at 9.7 and 8.0 Hz, and L2/3 VIP and NPY cells at 5.9 and 1.6 hz.

Here are some pictures of pyramidal cell membrane potentials including spikes during these UP states from Neske et al 2015:

Pyramidal cell UP state firing

My colleagues and I have also studied evoked UP states in neocortical slices of auditory cortex, in Krause et al 2014. Like Neske et al, we found most spiking among excitatory neurons in layer 5. I think it's fairly safe to say that layer 5 is where the UP state is "generated":

UP state spiking by layer

In this figure, panel A, "early" refers to spikes fired before UP states are detected; these occur in layers 4 and 5, but are a very small fraction of the total spikes fired. Like for Neske et al, L5 cells fire the most spikes. Panel D is a peri-stimulus time histogram (PSTH) triggered on UP state detection: you'll see the L5 cells, in green, fire earlier in UP states than the other cell types.

UP state calcium imaging

In this paper, we also used calcium imaging to visualize spiking activity. These panels are derived from calcium imaging; the dots are cell bodies that showed a calcium signal (indicating spiking). One of the things we discovered is that UP states come in different "sizes" with a bit of a bimodal distribution. There are "small" UP states, like on the left, where all the spiking cells are in layer 5. There are also bigger "full-column" UP states, like on the right, where there are also a lot of spiking cells in the upper layers.

If you're recording just superficially in cortex from living animals, you're likely recording in layer 2/3. A lot of experiments were done that way, it's difficult to record the deeper layers in vivo. From that perspective, if you're recording mostly cells in layer 2/3 while all the spiking is happening in layer 5, it might be easy to describe UP states as "subthreshold" phenomena. Indeed, for many cells that may typically be the case. However, that doesn't mean their neighbors aren't spiking away to maintain the UP state!

Krause, B. M., Raz, A., Uhlrich, D. J., Smith, P. H., & Banks, M. I. (2014). Spiking in auditory cortex following thalamic stimulation is dominated by cortical network activity. Frontiers in systems neuroscience, 8, 170.

Neske, G. T., Patrick, S. L., & Connors, B. W. (2015). Contributions of diverse excitatory and inhibitory neurons to recurrent network activity in cerebral cortex. Journal of Neuroscience, 35(3), 1089-1105.

  • $\begingroup$ Comments are not for extended discussion; this conversation has been moved to chat. $\endgroup$
    – Bryan Krause
    Nov 1, 2022 at 14:09

Indeed neurons in UP state (high subthreshold membrane potential) would not fire action potential by its definition in this state, but the same neuron will tend to trigger firing action potential later on more probably compared to other subthreshold state. See Stern et al's 1997 paper "Spontaneous Subthreshold Membrane Potential Fluctuations and Action Potential Variability of Rat Corticostriatal and Striatal Neurons In Vivo" for reference.

All neurons showed spontaneous subthreshold membrane potential shifts from 7 to 32 mV in amplitude, fluctuating between a hyperpolarized down state and depolarized up state... noisy fluctuations in the up state can trigger action potentials in either cell type (Cowan and Wilson 1994; Wilson 1993; Wilson and Kawaguchi 1996), and thus the firing patterns of corticostriatal neurons and striatal neurons are largely shaped by the membrane potential shifts. Therefore it has been suggested that the membrane potential shifts of striatal neurons occur in the way they do mainly because they reflect the firing (and the membrane potential shifts) of the cortical cells that provide much of their excitatory input (Wilson 1993).

This basic background knowledge can also be cross referenced by wikipedia source about Subthreshold membrane potential oscillations.

In some types of neurons, the membrane potential can oscillate at specific frequencies. These oscillations can produce firing by joining with depolarizations.[5] Although subthreshold oscillations do not directly result in neuronal firing, they may facilitate synchronous activity of neighboring neurons. It may also facilitate computation, particularly processing of sensory signals.[5] All in all, although the subthreshold membrane potential oscillations do not produce action potentials by themselves, through summation, they are able to still impact action potential outcomes... In addition to neurons firing action potentials, they can also perform synchronized spiking or bursts. Subthreshold membrane potential oscillations do not create an action potential; however, neurons do experience bursting when they group together and create a synchronized potential by firing all at once, which is usually the result of these subthreshold potentials.[25]... These oscillations are present even before birth and also before a newborn opens its eyes, as they are forms of maturation and preparation of the human sensory cortex, which is a part of the cerebral cortex that is responsible for processing and encoding sensory information.[2] This subthreshold activity is responsible for shaping circuits for maturation and are especially distinct in the retina, in the form of retinal waves.[2]

Thus it's clear a recurrent network consisting mostly of UP states neurons can likely cause sustained high subthreshold membrane potentials with possible activation and spiking bursts. You may refer to Renart et al's 2003 paper Robust Spatial Working Memory through Homeostatic Synaptic Scaling in Heterogeneous Cortical Networks for a theoretical explanation via recurrent attractor model.

In most simulations, the excitatory recurrent connections were all to all, with a magnitude that decayed as a function of the difference between the cells' preferred cues. In some simulations (see below), the pyramidal cells were connected probabilistically and with random synaptic efficacies. In these simulations, the probability of connection decreased as the difference between the cells' preferred cues increased... A transient spatial input triggers a localized activity pattern in the network (Figure 1B). This bump state outlasts the stimulus, resulting in a persistent activity state sustained by excitatory synaptic reverberations. The ideal memory function of the network is realized when all cells are identical. In this case, when different stimuli are used in different trials, in any given trial the peak location of the persistent firing pattern (as measured with the population vector method [Georgopoulos et al. 1982]) remains close to the stimulus location throughout the memory period, so that the stimulus location can be accurately read out several seconds after the stimulus offset (Figure 1C).

Thus here Sporns discusses about recurrent topology of neural networks in the likely cerebral cortex, and specifically the synchronized active UP state is a prominent phenomenon in the prefrontal part (PFC) acting as working memory to maintain and update various short-term information to robustly achieve its executive-like functions such as sustained attentional control of goals and interconnecting of high level information collected from other areas of the cortex. PFC encodes information in highly synchronized and active states via sustained recurrent neural firing and thus is more flexible and rapidly updatable than via mere synaptic weight changes learned in other parts of the cortex over a comparatively much longer period, such as proved by the famous Stroop effect sustained active PFC recurrent stripes consisting mostly of level V pyramidal neurons can excite more difficult linguistic pathway on demand and fairly rapidly.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.