Synaptic transmission is the well-known signature of neuronal signaling with time-frames of information transfer down axons of 0.1–100 m/s, and time-delays for information to cross synapses (at 38°C) of around 150–300 μs (Sabatini and Regehr, 1999). However, some anomalies become immediately evident when comparing assemblies generated in two different, well-established in vitro preparations: first, coronal brain slices where assemblies are evoked using direct electrical stimulation to the cortex (Yuste et al., 1997; Zochowski et al., 2000; Petersen and Sakmann, 2001; Bourgeois et al., 2014), and secondly, a thalamocortical section (Agmon and Connors, 1991; Takashima et al., 2001; Llinas et al., 2002; Hill and Greenfield, 2014) that enables indirect, remote activation of the cortex region via neuronal innervation resulting from thalamic stimulation. Other studies have also investigated the downstream cortico-cortical connectivity elicited by exogenous activation of the lemniscal pathway, leading for example to active communication between primary somatosensory cortex and motor cortex (Ferezou et al., 2007) or for the purposes of general brain mapping in vivo (Lim et al., 2012).

Direct stimulation with a single electrical pulse evokes activity from an epicenter with fast and efficient recruitment of large numbers of neurons in near-synchronous fashion manifesting as a circular propagating wave of activity spreading outwards from the locus of stimulation (Lopes da Silva, 1991) as seen using VSDI, Figure 1A. This is a stereotypical activation pattern which has been reported in virtually all studied neuronal population systems using VSDI, both in the cortex and sub-cortical structures (Badin et al., 2016), and it is those dynamics which can be modulated with bioactive compounds. Signal propagation via action potentials traveling down axons and activating chemical transmitters at synapses with the neurons it contacts takes just over 1 ms. The speed of action potential propagation varies significantly across circuits in the brain as well as with the distance they travel, but even the slowest signals (traveling through unmyelinated axons) take 0.5 ms to travel 1 mm, while subsequent transmitter release and diffusion across the synaptic cleft is approximated to take just under 0.75 ms. The activation of synapses has been found to decay with time-scales that go from a few ms (e.g., for synapses rich in GABA_A and/or AMPA receptors) to 100 ms (for those most influenced by GABA_B or NMDA receptors). However, if this was the dominant mechanism at play, given the speed of transmission, the greatest activity would most probably be observed furthest from where the stimulus was received, i.e., at the spreading perimeter (Figure 2A), a configuration which conflicts with the data. Under normal conditions in direct stimulation paradigms, assemblies usually reach maximum lateral spread within 5–6 ms after stimulus delivery (Figure 1B), while by comparison, for remote thalamocortical stimulation (i.e., neuronal activation of cortical tissue), this occurs between 10 and 13 ms after stimulation, as seen in Figure 1C, is delivered to the cortex (i.e., with time delay corrected for impulse conduction time from thalamus to cortex – of the order of 4–5 ms) (Landisman and Connors, 2007); i.e., where assemblies are evoked in a more physiological manner than those triggered with direct electrical stimulation. This scenario suggests that other factors, in addition to traditional synaptic transmission, may be affected by this difference in stimulation paradigm; leading to the emergence of different profiles of activation dynamics and resulting time courses.

In terms of time, it takes some 300 ms for decay of assembly activity to fall to even 20% of its maximum strength (Chakraborty et al., 2007). The dye used to visualize voltage changes has a latency and decay time of the order of a millisecond or less (Tominaga et al., 2000): hence, the persistent activity observed must be a genuine physiological phenomenon. This continuation of activity over several hundred milliseconds could be attributable to the prolonged duration of signal decay time-scale of synapse operations that can last up to 100 ms (Hestrin et al., 1990). However, if synaptic transmission were the sole mechanism, the greatest activity would be observed furthest from where the stimulus was received, i.e., at the spreading perimeter (Figure 2A): yet this configuration conflicts with the experimental data (Figure 1). Alternatively, the typically lengthy time frame of an assembly could represent the summation of thousands of sequential synaptic connections whereby the cumbersome process of neurotransmission will impede the signal speed. However, this scenario too can potentially be discounted if spatial features are now considered. There are two possible scenarios which could account for the characteristic and extensive assembly spatial spread. One option is that, most probably at the locus of stimulation, a small proportion of directly activated neurons fire action potentials, which in turn propagate potentials affecting the resting membrane potential of target neurons with which they form synapses. In this case, the length-scale of interaction will be a function of synaptic connectivity, network coupling strength (balance between inhibition and excitation) and synaptic dynamics where the firing of a neuron at any given point will affect, above the reference threshold, neurons about 50% further than the connectivity length-scale (Jirsa and Haken, 1996). Typically (Perin et al., 2011), this metric implies a spatial spread that will still be, nonetheless, two to five times less than actually observed, and can thus be discounted.

The second and more plausible option as the underlying dominant process, and one in any case that is preferentially detectable with voltage sensitive dyes (Chen et al., 2012): a widespread sub-threshold depolarisation. If so, a second signaling mechanism is needed that could accommodate the extensive spatial spread. Just such a mechanism is also suggested by a discrepancy between predictions from synaptic transmission alone, and what is empirically observed in time frames (Figure 1C): whilst the transmission of a signal from thalamus to cortex, via classic synaptic transmission, takes only 5 ms to travel some 1–2 mm, a further 20 ms is required for the assembly to spread within the cortex, to its full extent.

Volume transmission enables interaction between neurons in a way that is much less specific and significantly slower, yet with the pay-off that it involves far more cells at any one time: it is considered a complementary counterpart to classic synaptic “wired” transmission (Agnati and Fuxe, 2014; Taber and Hurley, 2014). In fact, it has been known since the 1970s that classic transmitters such as dopamine can be released (Nedergaard et al., 1989) as can protein (Greenfield et al., 1980) from a part of the neuron dendrites that typically has a very different role. Normally, dendrites, traditionally regarded as being the target for incoming connections, can actually release substances in their own right, and do so independent of the action potentials generated at the cell body (Greenfield, 1985). Moreover, this dendritic release also affects a much wider area than standard synaptic transmission, is far less precise and uses different ionic mechanisms and cellular storage, – all suggesting a contrasting yet complementary modulatory process. In addition to neuro-active chemical release from dendrites, there are a wide range of afferent fibers originating from sub-cortical structures, specialized in releasing neuromodulatory chemicals, such as acetylcholine, dopamine, noradrenaline, and serotonin (to name but a few), which do not form traditional synapses, but instead are specialized in releasing large amounts of modulatory transmitters into the extracellular space for tonic influence of population activity (Seguela et al., 1987; Soghomonian et al., 1989; Descarries and Mechawar, 2008). Every neurotransmitter system potentially influences neuronal networks in a tonic manner (Zoli et al., 1999; Kullmann, 2000), i.e., via volume transmission, as would the presence of extra-synaptic glutamatergic and GABAergic receptors suggest (Belelli et al., 2009), and as has already been widely reported for monoaminergic and cholinergic systems (Beaudet and Descarries, 1978; Descarries and Mechawar, 2000). In addition to these well known transmitters, a wide range of other endogenous signaling molecules act exclusively via volume transmission: growth factors (Carro et al., 2000), hormones (Mody, 2008), and peptides (de Wied et al., 1993), confirming that volume transmission is an essential and effective mechanism of cell signaling, operating on both high time- and spatial-scales. But whilst synaptic transmission is too local and too fast, the passive diffusion of bioactive agents beyond the synapse, volume transmission, is too slow a mechanism for the generation of assemblies (Figure 2B), In order to counterbalance the slow speed of the extensive extra-synaptic outreach of volume transmission, what is required now to accomodate the characteristic dynamics of an assembly, is an ultra fast signaling system (Steriade, 2005).

Electrotonic spread via gap junctions (Anava et al., 2013; Agnati and Fuxe, 2014) is a widespread mode of close-range and high-speed neuronal signaling and has been reported to exist both in inhibitory and excitatory networks of coupled neurons (Tamas et al., 2000; Traub et al., 2001a). Gap junctions are a form of intercellular connection, where the trans-membrane pores formed by a congregation of connexin proteins in the membrane of two neighboring cells creates an open channel, such that the cytoplasm of two cells are effectively continuous, allowing free flow of ions and therefore: electrical pulses. In addition to neurons, connexin proteins are expressed in a range of cells involved in neuronal communication, including glial cells, where gap junctions also mediate substantial parts of their communication (Nagy and Rash, 2000; Orthmann-Murphy et al., 2008), allowing them to carry out essential roles in the maintenance of network health for appropriate functioning (Watkins and Maier, 2002). These low resistance connections have been mapped throughout the brain (Nagy et al., 2001), highlighting their ubiquitous expression and function throughout the CNS in inhibitory (Beierlein et al., 2000; Tamas et al., 2000; Krook-Magnuson and Huntsman, 2005; Fukuda et al., 2006) and excitatory neurons (Spray and Bennett, 1985; Traub et al., 2001b). Gap junctions have been found to play a key role in operating a range of functions including neuronal differentiation (Bani-Yaghoub et al., 1999), synaptogenesis and circuit formation (Connors et al., 1983; Peinado et al., 1993; Kandler and Katz, 1995), yet the role for which they are considered here as a key method of cell signaling in assembly generation is for their permissive capacity in transmembrane impulse propagation (Spray and Bennett, 1985; Bennett, 1997), allowing signaling independent of transmitter release (Eugenin et al., 2012; Leybaert and Sanderson, 2012; Belousov and Fontes, 2014). In accordance with these findings, it has been found that in neuronal networks, very fast oscillations (200 Hz), that could underlie the sustained activity seen here in assemblies, are mediated not via synapses but by these gap junctions (Draguhn et al., 1998). So if fast oscillations underlie the sustained activity seen here in assemblies (van den Pol, 2012), although they will take longer to effectively reach a maximum, once underway (Petersen et al., 2003a), the spread of activity will reach further than any signal from a synapse ever could. Hence, assemblies will be a very appropriate neuronal correlate for the space-time requirements for consciousness since, unlike localized neuronal circuitry, they are neither hard-wired in time, nor spatially restricted.

The existence of an electrochemical gradient across neuronal cell membranes generates a small electric field (Kandel et al., 2000), and the diffusion of ions contributes to the time-variability of this electric field, possibly representing per se a further form of signaling: electromagnetic transmission. Time-variable electric fields induce electromagnetic fields and vice versa (Jackson, 1962), thus allowing for electromagnetic energy to propagate directly via gap junctions and in the form of electromagnetic waves from the source of the signal, exciting neighboring neurons. Energy conservation will result in the intensity of the energy being transferred to decrease proportionally with the inverse square of the distance from the source. Integrated over the time window necessary to observe a depolarization in the receiving neurons, this mechanism predicts a drop in assembly-spread velocity proportional to the inverse of the distance. Wave-like patterns have been observed throughout the brain, though so far mostly attributed to feed-forward networks (Meijer and Coombes, 2014): we found that the velocity profile as a function of distance for assemblies is consistent with a spatiotemporal spread due to self-sustained electromagnetic waves (Figure 2C).

The mechanism by which radiation can excite neuronal activity, to the point of remotely triggering action potentials, is not unfamiliar (Huang et al., 2010); the most obvious candidate vehicle to host the magnetic field outside the neuron are glial cells, and/or the extracellular matrix: packed with ions, they are available in large quantities and continuously surrounding the network of neurons, providing a coherent medium where electromagnetic waves can propagate self-substainedly. Indeed, the cooperative action of neurons and glia has already been proven capable of influencing the timing of neuronal activity (Anastassiou et al., 2010).

In conclusion, disparate empirical findings both in vivo and in vitro could most readily be accommodated theoretically in the integration of three distinct signaling mechanisms over an epoch of some 250–300 ms (Figure 2D). As such, this approach to analyzing brain operations at the meso-scale would have the potential for a more accurate modeling of drug action and more generally a quantification of holistic brain states with a temporal and spatial resolution commensurate with neurophysiological and neurochemical events.