When a fearful face is flashed for a few dozen milliseconds and immediately masked, most observers report seeing nothing at all. Yet their amygdala responds, their skin conductance shifts, and their facial muscles faintly mirror the expression. This dissociation raises a fundamental question for affective neuroscience: how does the brain extract emotional significance from a stimulus it has not consciously perceived, and what anatomical machinery makes that possible?

For three decades, one candidate answer has centred on a structure long treated as a passive relay: the pulvinar, the largest nucleus of the primate thalamus. Together with the superior colliculus, it forms a proposed subcortical route that delivers coarse visual information to the amygdala without first passing through primary visual cortex, the canonical gateway to conscious sight. The proposal, popularised by Joseph LeDoux's 'low road' framework, has been both influential and persistently contested.

The debate matters well beyond anatomy. If emotional evaluation can begin before cortical analysis, then affect is not merely a downstream interpretation of perception but a parallel computation that shapes perception from the outset. That view sits comfortably with Damasio's somatic marker hypothesis. What follows examines the pathway itself, the evidence from cortically blind patients, and the hypothesis that an overweighted preconscious channel contributes to anxiety. Throughout, it distinguishes where the data are strong from where they remain thin.

The proposed route begins in the retina, where a subset of ganglion cells projects not to the lateral geniculate nucleus but to the superficial layers of the superior colliculus. These inputs favour low spatial frequencies, high temporal resolution, and luminance contrast over fine detail and colour. From the colliculus, projections reach the inferior and medial pulvinar. The medial pulvinar in turn sends efferents to the lateral and basolateral nuclei of the amygdala.

Tract-tracing work in primates established these connections piecemeal. Single-unit recordings in macaques have identified pulvinar neurons that respond selectively to faces and face-like configurations with short latencies, sometimes responding robustly to coarse, low-pass-filtered images. Such tuning is what one would expect from a system optimised for detection rather than identification. It registers that something biologically relevant is present before resolving exactly what it is.

In humans, the case has accumulated through converging methods. Diffusion tractography studies, including work by Rafal and colleagues and later by McFadyen, Garrido and colleagues, have reconstructed white-matter streamlines consistent with a colliculus–pulvinar–amygdala pathway. Intracranial recordings reported by Méndez-Bértolo and colleagues in 2016 found fear-selective amygdala responses emerging at roughly 74 milliseconds, driven by low spatial frequencies. That latency is difficult to reconcile with a complete ventral-stream cascade.

The 'quick and dirty' framing has nonetheless drawn sustained criticism. Pessoa and Adolphs argued that cortical routes can be fast enough to account for many rapid amygdala responses. They also noted that pulvinar activity is heavily modulated by cortical feedback, and that the structure is better understood as a hub coordinating attention and salience across cortical networks than as an isolated shortcut. On this many roads view, the subcortical route is one channel among several rather than a privileged emotional bypass.

What survives the critique is more modest but still significant. The subcortical route appears to transmit coarse signals about biologically salient configurations, such as widened eyes, looming shapes, and threatening postures. These signals can bias amygdala activity and orient attention before detailed analysis completes. LeDoux himself has since emphasised that such circuits detect and respond to threat rather than generate the experience of fear. The pathway computes a priority signal, which downstream systems may or may not elaborate into an emotion.

Takeaway

The pulvinar route is best understood not as a shortcut to feeling but as an early priority signal. It is a coarse bet about relevance that the rest of the brain is free to confirm, refine, or discard.

The most compelling evidence for subcortical sufficiency comes from patients with damage to primary visual cortex. Such lesions produce cortical blindness in the corresponding visual field, yet Weiskrantz and others documented that some patients can localise and discriminate stimuli they insist they cannot see. The phenomenon was termed blindsight. In 1999, de Gelder and colleagues extended it to emotion, showing that patient GY could identify the expression of faces presented to his blind hemifield at rates well above chance.

Neuroimaging subsequently linked this residual ability to the proposed pathway. Morris and colleagues found that unseen fearful faces in GY's blind field engaged the amygdala alongside the superior colliculus and pulvinar, with covariation among these regions. Pegna and colleagues later reported that patient TN, left fully cortically blind by bilateral occipital strokes, discriminated facial expressions above chance and showed amygdala responses to emotional faces despite reporting no conscious visual experience.

Behavioural and physiological measures reinforce the picture. Tamietto and colleagues recorded facial electromyography in patients with unilateral V1 damage. Unseen facial expressions and emotional body postures elicited congruent facial muscle activity, and these reactions emerged faster for unseen stimuli than for seen ones. Pupillary responses showed related effects. In Damasio's terms, the body was registering a somatic marker for an event that never reached awareness.

Structural findings close part of the loop. In 2012, Tamietto and colleagues reported evidence of colliculus–pulvinar–amygdala connectivity in GY, with differences in the lesioned hemisphere suggesting reorganisation after V1 loss. Lesion evidence points the same way. Ward and colleagues described a patient with unilateral pulvinar damage in whom threatening images in the contralesional field failed to receive the rapid prioritisation that threat normally commands.

These cases carry interpretive limits. Blindsight patients are rare, lesions are rarely clean, and long post-lesion intervals allow plasticity that may not reflect the intact brain. Residual islands of cortex and direct geniculate projections to extrastriate areas offer alternative explanations. Still, the convergence across behaviour, physiology, and anatomy makes a strong case. When V1 is absent, a subcortical channel can carry enough affective information to shape discrimination and bodily response.

Takeaway

Blindsight shows that the brain can respond emotionally to what it cannot report seeing. Conscious perception, then, is not the gateway to affect but one of its possible destinations.

If a preconscious channel can bias amygdala activity, an obvious question follows: what happens when that channel is overweighted? Anxiety disorders are characterised by attentional bias toward threat, heightened physiological reactivity, and often an experience of fear that arrives without an identifiable cause. Each of these features would be predicted by a system that flags potential danger before conscious evaluation can contextualise it.

Several findings support the link. Etkin and colleagues showed in 2004 that trait anxiety predicted basolateral amygdala responses to backward-masked fearful faces, a relationship not mirrored for consciously perceived faces. Behavioural work by Mogg, Bradley and colleagues found that anxious individuals show attentional capture by threat stimuli presented below awareness thresholds. Together, these results suggest the bias operates partly at a preattentive stage.

In post-traumatic stress disorder, Lanius and colleagues have proposed an 'innate alarm system' encompassing the superior colliculus, pulvinar, locus coeruleus, amygdala, and insula. Imaging studies using subliminal trauma-related cues report heightened engagement of several of these structures in PTSD, alongside weaker recruitment of prefrontal regions that normally contextualise threat. The clinical phenomenology fits: patients often describe being flooded before they know what triggered them.

The interpretation requires caution. Masking paradigms vary in how rigorously they establish unawareness, and anxious participants may differ in detection thresholds rather than in subcortical gain. Correlational imaging cannot isolate the colliculus–pulvinar route from fast cortical contributions. A constructionist reading in the tradition of Lisa Feldman Barrett offers another possibility. Heightened amygdala responses may reflect a predictive brain that has learned to assign high prior probability to threat, rather than an autonomous low-level alarm running hot.

These accounts need not be mutually exclusive. Predictive models can treat subcortical signals as fast sensory evidence whose influence depends on learned precision weighting. On that view, anxiety may involve both amplified early salience signals and priors that treat them as reliable. Intervention could then target two levels. Exposure and attention-training approaches aim to recalibrate early salience, though attention bias modification has shown mixed efficacy so far. Interoceptive and cognitive work addresses how those signals are interpreted once they arrive.

Takeaway

In anxiety, the problem may not be that the alarm sounds too early but that the brain has learned to trust it too much. That distinction changes what treatment should aim to recalibrate.

The pulvinar emerges from this evidence not as a dedicated fear shortcut but as a node in a distributed system for prioritising biologically significant input. Its subcortical inputs carry coarse, rapid signals that can bias amygdala activity, orient attention, and trigger bodily responses before conscious perception consolidates.

Blindsight establishes that this channel is sufficient for rudimentary affective discrimination when visual cortex is unavailable. Whether it is necessary in the intact brain, and how much it contributes relative to fast cortical routes, remains genuinely unresolved. Anxiety research suggests that individual differences in preconscious threat processing are clinically meaningful, though the precise locus of those differences is still debated.

For emotional intelligence, the implication is clarifying. Much of affective life begins before awareness, as Damasio long argued. The relevant skill may lie less in preventing early signals than in learning how much weight to give them. Unlike subcortical anatomy, that capacity for weighting appears to remain substantially plastic.