The starting point of the model is a contrast with artificial systems. A computer executes commands given from outside; the brain, by contrast, is driven by continuously generated desires rooted in biological needs and experience-dependent values. Tsubo organizes these drives under an Affective Valuation System, or AVS, which he divides into an innate Proto-AVS present from birth, involving survival circuits such as the hypothalamus, amygdala, and brainstem, and a higher-order AVS built through postnatal learning in regions including the orbitofrontal cortex and hippocampus. These valuation signals, the model argues, do not merely evaluate behavior after the fact; they actively initiate and steer the exploration of contextual information in the dorsolateral prefrontal cortex, where the DCEM is centered.

Technically, the DCEM is conceived as a dynamic contextual memory composed of logical Contextual Elements, or CEs, each implemented by multiple physical cortical columns. As a concrete structural example, the model assumes eight CE Categories, labeled Subject, Verb, Object, Complement, When, Where, Why, and How, each containing roughly 25 CEs, for about 200 logical CEs and roughly 2,000 physical columns throughout the system. Tsubo is careful to stress that these numbers are model assumptions introduced to describe the structure and operation of the DCEM, not established neuroanatomical constants. The contents of each CE are assumed to exist in distributed form as numerous Meaning Groups formed through experience within the parietal and temporal association cortices, which the model treats as functional, not strict anatomical, subdivisions handling perceptual and conceptual information respectively.

The model unfolds in two stages of recurrent loops. In the first stage, Layer 2/3 of the DCEM receives inputs from the association cortices, the AVS, and a thalamic population, while large Layer 5 pyramidal neurons project outputs widely back to the association cortices. This forms recurrent loops, PAC/TAC Layer 2/3 to DCEM Layer 2/3 to DCEM Layer 5 and back, that oscillate in the gamma band. Because Layer 2/3 neurons have inhibitory connections with neighboring columns, the formation of a new loop terminates previously active loops nearby, so the candidate set is continuously replaced as inputs change. The result is a state in which numerous asynchronous gamma-band candidate loops run in parallel, exploring diverse contents that could fill each CE, a picture consistent with experimental work linking gamma activity to transient information representation.

The second stage is where desire does its work. A specialized population of integrative neurons, hypothesized to exist in the mediodorsal thalamic nucleus and termed Contextual Element Decision Neurons, or CEDeNs, monitors the candidates. Each CEDeN receives Layer 5 inputs from the multiple physical columns constituting one logical CE, together with positive and negative valuation signals from the AVS. The initiation condition is expressed as a threshold equation in which the summed Layer 5 input plus positive valuation minus negative valuation, multiplied by a gain factor from the Global Neuromodulatory System, must exceed a threshold. In the asynchronous gamma state, the summed Layer 5 inputs alone are unlikely to reach threshold; only when sufficient positive valuation is added does the CEDeN fire and send synchronized feedback to the columns of that CE.

That synchronized feedback triggers a remarkable local circuit dynamic. When simultaneous input arrives at Layers 2 and 3, Layer 2 pyramidal neurons, which are smaller and have less membrane capacitance, depolarize faster and fire first, driving fast-spiking parvalbumin-positive inhibitory interneurons. This inhibition transiently suppresses the larger Layer 3 neurons, delaying their firing until the inhibitory effect decays and excitation reaccumulates. The Layer 3 delay propagates to Layer 5, so the synchronized Layer 5 outputs have a longer period than the gamma band, producing a coherent beta-band Contextual Element Loop. The beta periodicity, in other words, is not imposed from outside but emerges from the intrinsic temporal structure of the L2, PV-interneuron, L3, L5 sequence within each column, a mechanism the author derives from his previous biophysical studies of dendritic spines and neuronal computation.

Once formed, a beta-band CE Loop is relatively self-sustaining. Because the Layer 5 outputs of the columns are now synchronized, they sum temporally in the CEDeN, making maintenance less dependent on continuous positive valuation. But the loop is not permanent: when new sensory information, memories, behavioral outcomes, or language input shift the AVS evaluation in a negative direction, negative signals rise, the CEDeN’s integrated input falls below threshold, synchronized feedback stops, and the loop disappears. Meanwhile, the thalamic reticular nucleus and Layer 6 circuits gate sensory input to favor information relevant to the current context, and recurrent feedback from DCEM Layer 5 keeps reactivating latent representations, preparing the next candidates in parallel. A conscious state, in this framework, is the simultaneous combination of multiple CE Loops, and conscious state transitions are the formation, maintenance, disappearance, and replacement of individual loops.

The framework extends naturally to memory, language, and learning. Episodic memories reactivated through the hippocampus become part of the gamma candidate pool, and only contextual states stabilized as beta CE Loops for a sufficient period are predicted to proceed to memory consolidation in hippocampal circuits, a testable relationship. Language is positioned not as constituting consciousness but as an output format: the language network, including Broca’s area, serializes the nonverbal CE Loop structure into a temporally ordered linguistic sequence. Learning, meanwhile, is interpreted as the reuse of existing CE structures in new combinations, which the author suggests may underlie the rapid human ability to understand new concepts through metaphor and analogy, with long-term consolidation involving NMDA receptor calcium influx, CaMKII activation, and structural plasticity of dendritic spines.

The model is explicit about how it relates to existing theories of consciousness. The CEDeN threshold mechanism bears functional similarity to the ignition of Global Workspace Theory, but in DCEM it operates separately for each logical CE rather than broadcasting an entire representation globally. Unlike Integrated Information Theory, DCEM does not quantify the integration of consciousness but addresses the temporal state-transition mechanism of contextual selection and replacement. And while it shares with Predictive Processing the property that states update on the basis of external and internal information, DCEM uniquely introduces continuously provided desires and valuation as the driving conditions for context formation, proposing a specific circuit mechanism for how individual CEs are selected, synchronized, and terminated when valuation changes.

Crucially, the model is designed to be falsifiable. It predicts that preserving cortical gamma candidate activity while selectively suppressing the corresponding CEDeN should leave the candidates intact but selectively impair the transition to synchronized beta activity and CE Loop formation, and conversely that selective CEDeN activation should facilitate that transition without disrupting other coexisting candidates. It predicts that beta Loop maintenance should not require continuous positive valuation, that rising negative valuation should selectively terminate the corresponding loop, and that disrupting beta synchronization for one CE should selectively impair decoding of that CE’s information downstream. Testing these predictions will require laminar recordings, simultaneous measurement of the mediodorsal thalamus and cortex, and cell-type-selective manipulations. The author acknowledges that the existence of CEDeNs and several biophysical mechanisms, including membrane-potential formation from retained spine charge, remain hypotheses requiring direct experimental verification. If the principal predictions hold, the DCEM model could provide a new biophysical foundation linking the intrinsic properties of single neurons to the large-scale dynamics through which desire becomes context, and context becomes consciousness.

Subject of Research: A biophysical framework for conscious state transitions based on intrinsic neuronal properties: the dynamic contextual element memory (DCEM) model

Article Title: A biophysical framework for conscious state transitions based on intrinsic neuronal properties: the dynamic contextual element memory (DCEM) model

Article References: Tsubo, T. (2026). A biophysical framework for conscious state transitions based on intrinsic neuronal properties: the dynamic contextual element memory (DCEM) model. Frontiers in Systems Neuroscience, 20, Article 1894336. https://doi.org/10.3389/fnsys.2026.1894336

Image Credits: AI Generated

DOI: 10.3389/fnsys.2026.1894336

Keywords: biophysical, framework, conscious, state, transitions, based, intrinsic, neuronal, properties, dynamic, contextual, element