The Sanchez-Raghavendra-Chen Activation-Synthesis Model

The Activation-Synthesis model devised by Hobson and McCarley proposed that dreaming arises when the forebrain synthesizes essentially random signals generated by brainstem activation during REM sleep. We retain the model's core insight, that pontine mechanisms are central to REM, but depart from its emphasis on stochastic generation. We propose instead a neurophysiological cascade: a structured, stepwise sequence in which pontine activation is gated and amplified through the thalamus and distributed to cortical and limbic targets in a predictable order. We use the term cascade deliberately, because the model's explanatory value depends on the ordering of events, not simply their co-occurrence.

The signals originate in the subcoeruleus, which houses glutamatergic neurons that drive cortical activation and reduce muscle tone (Vetrivelan & Bandaru, 2025). The pons is the primary and necessary node for REM generation, but it is not the sole contributor: medullary regions, including the ventral and dorsomedial medulla, modulate REM and atonia (Weber et al., 2016; Stucynski et al., 2021), and forebrain circuits contribute to sleep-wake control (Xu et al., 2018). We therefore frame the pons as the initiating node of the cascade rather than an exclusive origin.

• Pontine onset and atonia. Subcoeruleus glutamatergic firing initiates cortical activation and, via descending projections, muscle atonia (Vetrivelan & Bandaru, 2025).

• Neurochemical shift. REM-promoting GABAergic and glutamatergic populations increase their activity while norepinephrine, histamine, and serotonin fall; orexinergic and cholinergic populations regulate the transition (Siegel, 2004; Van Dort et al., 2014).

• Thalamic amplification. The thalamus relays and amplifies pontine signals, projecting to sensory cortices where they are interpreted as external input (Nakajima & Halassa, 2017).

• Limbic projection. The thalamus also projects amplified signals to the amygdala and hippocampus, which we propose account for the emotional tone of dreams (Sanford et al., 2022; Ferrara et al., 2012). This thalamo-limbic route is the principal differentiator of our model.

• Prefrontal suppression. Prefrontal cortical activity decreases, associated with elevated acetylcholine and reduced norepinephrine (Hobson et al., 2002; Arnsten & Robbins, 2002). We propose this underlies dream incoherence and impaired reality-monitoring.

• Sensory synthesis. Heightened activity in visual and auditory cortices supplies the sensory material of the dream. Thalamic signaling via the lateral geniculate nucleus is transmitted to the occipital cortex, consistent with the classical order of PGO-wave propagation (Wang et al., 2022).

Because our central claim concerns the order in which regions are recruited, two limitations govern the entire design.

First, the BOLD signal is an indirect, slow proxy for neuronal activity, with a hemodynamic response that peaks several seconds after the underlying neural event and whose latency varies across brain regions. The cascade we propose unfolds on a scale of milliseconds to a few hundred milliseconds. Region-to-region differences in hemodynamic lag can therefore obscure, or even reverse, the apparent order of activation when inferred from BOLD alone. Magnitude comparisons can establish involvement, but not sequence.

Second, Dynamic Causal Modeling (DCM) is insufficient on its own to establish the linear temporal association our three-step relay (pons, thalamus, sensory/limbic cortex) requires. DCM estimates effective connectivity by fitting a generative model of how regions influence one another, and it can show that regions are coupled and compare competing network architectures. But when applied to BOLD data, it infers directionality from a hemodynamically filtered signal; a strong estimated coupling is consistent with several underlying neuronal orderings, and its directionality estimates are model-relative rather than an independent measurement of when each region actually fired. DCM can confirm that the nodes of the cascade communicate, but it cannot by itself prove they were recruited in the specific linear order the model predicts.

We therefore treat ordering as a question that must be answered with temporal resolution the BOLD signal does not possess, and design the study around convergent measures rather than a single method.

NREM dreaming is well documented (Siclari et al.; Nayak & Anilkumar, 2023), but a pontine-REM cascade does not straightforwardly explain it. To avoid over-claiming, this proposal restricts its primary test to REM dreaming, where the pontine mechanism is best supported. We treat NREM dreaming as outside the scope of the present design and note it as a direction for future work: if the cascade is REM-specific, an adequate account of NREM dreaming will require a distinct mechanism, such as slow-wave or spindle-associated activity (Schonauer & Pochlen, 2018). We flag this explicitly because earlier framings of the model conflated the two, and the methods below cannot adjudicate NREM dreaming.

This study is designed primarily as a secondary analysis of existing simultaneous EEG-fMRI sleep datasets, supplemented by limited prospective data collection only if required. All analyses are conducted in collaboration with a professional PI and an appropriate imaging lab. The design is built around triangulation: three convergent methods, each compensating for the others' weaknesses, so that no single method bears the full weight of the ordering claim.

If the cascade holds, we expect thalamic BOLD to be greater in REM than NREM; PGO-analog onset to precede thalamic activation, and thalamic activation to precede sensory-cortical activation, with consistent within-subject delays; directed connectivity from pontine and thalamic nodes to cortical nodes in the electrophysiological signal; and a better fit for a serial than a parallel network model. We further expect thalamo-limbic recruitment to track the emotional tone of reported dream content. Convergence across the three methods, rather than any single result, would constitute support for the ordering claim; divergence among them would appropriately weaken it.

• Human PGO detection. The timing anchor depends on a PGO-analog marker whose validity in humans is not fully established; results are contingent on the quality of this marker and are strongest where intracranial data exist.

• Hemodynamic lag. BOLD onset latencies are only interpretable as neuronal order after accounting for regional hemodynamic differences; residual confounding may remain.

• Dataset dependence. Secondary analysis is constrained by the staging quality, sampling rate, and regional coverage of available EEG-fMRI datasets.

• Scope. The design tests REM dreaming only; it cannot adjudicate NREM dreaming.

• DCM interpretability. Effective-connectivity estimates are model-relative and are used only as convergent evidence.