Ventromedial prefrontal cortex (VM)
The lesion site the somatic-marker-hypothesis was built to explain, and — under a different name and a different research programme — one of the visceromotor-areas Seth & Friston place atop the interoceptive hierarchy. Those two facts have never been put together in this wiki’s sources.
The syndrome
Bilateral VM damage produces the pattern Bechara & Damasio (2005) spent fifteen years characterizing: severe impairment of personal and social decision-making — difficulty planning a workday, choosing friends, partners, activities — with choices that lead to financial loss, loss of social standing, loss of family and friends, and a striking failure to learn from previous mistakes. The choices are “remarkably different from the kinds of choices they were known to make in the pre-morbid period.”
What makes it a puzzle rather than a deficit: problem-solving in laboratory settings remains largely normal. Intellect is normal on conventional neuropsychological testing (Bechara et al. 1998; Damasio et al. 1990; Eslinger & Damasio 1985), as are language, working memory, and attention. The disturbance cannot be explained by missing knowledge or general intellectual compromise.
What is impaired alongside decision-making is the ability to express emotion and experience feelings in appropriate situations. That co-occurrence — normal intellect, abnormal feeling, abnormal deciding — is the observation the somatic marker hypothesis was invented to explain.
The lineage runs through Phineas Gage, and Bechara & Damasio note that similar frontal cases had appeared and been ignored (Ackerly & Benton 1948; Brickner 1932; Welt 1888) until Eslinger & Damasio’s (1985) modern counterpart, patient EVR, revived interest.
Role in the somatic marker architecture
The VM cortex is the trigger structure for somatic states from secondary inducers — thoughts and memories — as the amygdala is for primary ones.
Mechanically it is a convergence–divergence zone: neuron ensembles holding records of temporal conjunctions of activity across sensory cortices and limbic structures. It couples (a) a category of event held in high-order association cortices, to (b) the effector structures that execute a somatic state, and to (c) the neural patterns of the non-conscious (brainstem/PBN) or conscious (insula/SII, SI) feeling of that state. In one line: it couples knowledge of events to “what it feels like” to be in a given situation.
It can also couple to covert effectors only, producing a somatic bias with no felt state — the anticipatory SCRs of the pre-hunch period in the iowa-gambling-task, which steer the choice “without any awareness of why the choice was made.”
Where “convergence zone” comes from. Bechara & Damasio (2005) use the term without the architecture behind it. The architecture is in Damasio (1996), and it constrains what the VM cortex can be: a convergence zone holds no representations at all — not the facts, not the emotional state — only the potential to reactivate an emotion by acting on the appropriate cortical or subcortical structures. The VM’s job is a linkage, “a memory in fact,” between the disposition for an aspect of a situation and the disposition for the emotion previously paired with it. Content lives where it always lived. See convergence-zones, which also explains why this makes the predictive-coding parallel below more precise and less flattering than the wiki had it.
The 1996 network, and what changed
Damasio (1996) specifies the system in three parts, which is worth recording because two of the three moved by 2005:
- Ventromedial frontal cortices containing the convergence zones — links between dispositions representing categorized situations and dispositions representing their associated somatic states. (Unchanged.)
- Central autonomic effectors, “for example the amygdala” — activating somatic responses in viscera, vascular bed, endocrine system, non-specific neurotransmitter systems. (In 2005 the amygdala is upstream of VM, not downstream of it. See amygdala.)
- Somatosensory cortices — insula, SII, SI — and their interlocking projections, especially in the non-dominant hemisphere, receiving signals from the soma or from VM prescribing an as-if pattern. (The laterality claim is dropped by 2005, which lateralizes the VM instead.)
Basal ganglia are floated as a possible fourth, mediating VM responses via somatomotor structures (Tranel & Damasio 1993).
The anatomical warrant offered in 1996 is the same one the 2005 gradient rests on and is stated more modestly: these cortices receive projections from all sensory modalities, are the only known frontal source of projections to central autonomic control structures (Nauta 1971), have extensive bidirectional connections with hippocampus and amygdala, and sit beside orbitofrontal secondary association areas for taste and olfaction (Rolls). No posterior→anterior gradient is proposed. That is a 2005 addition.
The Nauta citation is doing less work than Nauta did. Dunn et al. (2006) identify Nauta (1971) as the closest antecedent of the whole framework, not merely a source for its wiring. Nauta argued from those same reciprocal connections that a key role of the frontal lobes is to integrate feedback from the internal milieu with neocortical reports of the world — and that damage produces an “interoceptive agnosia,” an impairment in exactly that integration. He went further: comparing strategic alternatives is “in the final analysis… one between the affective responses evoked by each of the various alternatives,” and interoceptive information establishes “a temporal sequence of affective reference points serving as navigational markers.” Dunn et al.: “This account, predating the SMH by over 20 years, seems to have come to remarkably similar conclusions.”
So Damasio cites Nauta for the projections and not for the theory those projections were adduced to support. The wiki records the asymmetry because it is checkable in both texts and neither source explains it — not as a priority claim, which these sources cannot settle. Dunn et al. themselves conclude the SMH does contribute something Nauta did not: a clearer neural implementation, hormonal as well as visceral routes, the as-if loop specified, and a task. Worth noting for this wiki specifically that Nauta’s disconnection studies in animals (Divac et al. 1975) “generally offered little support for the model” — the idea was old, and its first tests failed too.
The scope disclaimer, which the wiki should keep. Damasio (1996) states outright that the hypothesis concerns the ventromedial sector, does not necessarily apply to prefrontal cortex as a whole, and should not be seen as an attempt to unify frontal lobe functions under a single mechanism — this large and parcellated sector in all likelihood accomplishes several separate albeit cooperative functions. Patients with non-ventromedial frontal lesions may or may not be accountable by it. That discipline is not audible in the 2005 paper, and it is the right frame for the “hold this at arm’s length” section below.
”The VM patient” is at least two patients
The most practically useful thing the Bechara, Damasio & Damasio (2000) ingest adds, and it is a caution on every group-level result on this page.
All nine VM patients in the delay-task experiment failed the gambling task. They split on working memory, and the split tracked how far posteriorly the lesion extended (Fig. 7):
| lesion | delay tasks | motor impulsiveness / perseveration | |
|---|---|---|---|
| anterior-confined (n = 4) | maximal overlap anterior, basal forebrain spared | normal | absent |
| posteriorly-extending (n = 5) | spans the mesial orbital surface into the posterior sector where basal forebrain structures lie | impaired | may be present |
Posterior extension is the same variable that predicts the other co-morbid deficits in this material. Tranel et al. (1996), cited in the gradient section below, report that patients whose lesions extend posteriorly into ACC and basal forebrain fail Pavlovian conditioning as well as gambling, where anterior patients condition normally. Same anatomical margin, three different co-morbidities.
Two consequences the wiki should carry:
- Group-level VM findings are averages over a mixture. A “VM group” assembled without regard to posterior extent contains patients whose deficit is decision-specific and patients whose deficit is broader, and the proportion is unreported in most of this literature.
- It gives the “five rival explanations” list on iowa-gambling-task an anatomical handle. Working memory, response inhibition and perseveration are all present in one subgroup and absent in the other — which means the rival accounts are not competing for the whole deficit so much as describing a subgroup. Nobody has run the analysis that would settle this, and n = 4 and n = 5 could not settle it.
Decision-making and working memory, dissociated asymmetrically
The other result the 2000 paper supplies first-hand: 21 controls, 9 bilateral VM, 10 dorsolateral patients on the delay tasks and the gambling task in one session. Anatomically a double dissociation — anterior VM impairs deciding with delay performance intact; right dorsolateral impairs delay performance while gambling stays advantageous. Functionally it is not symmetric: right-DL patients scored low-normal on gambling, so decision-making is degraded (not abolished) by a working-memory deficit while the reverse does not hold.
This is the same concession Dunn et al. (2006) extract as revision 2 below — DLPFC is involved, the framework absorbs it via attention/working-memory allocation, and the price is that its predictions become hard to distinguish from non-somatic theories. Worth recording that the Iowa group reported the dependency themselves, in 1998 and again in 2000, as a finding rather than a concession. See delayed-response-task.
The proposed hierarchical organization
The paper’s most ambitious and least constrained claim: four gradients laid on a single posterior→anterior axis.
| axis | posterior/caudal VM | anterior/rostral VM |
|---|---|---|
| time | outcomes near in time | outcomes far in time |
| abstractness | concrete/tangible (a loved one, food, physical harm) | abstract (money → credit → a grade → a diploma) |
| probability | predictable, sure (100% Pavlovian CS) | improbable (10–50% gambling task punishment) |
| valence (lateral, not A–P) | left VM ↔ positive | right VM ↔ negative |
The anatomical warrant: posterior VM (e.g. BA 25) connects directly to brainstem effector, neurotransmitter, and sensory nuclei, and to the insula/SII/SI structures holding conscious somatic representations (Ongur & Price 2000). Anterior VM connections to those structures are indirect. So posterior coupling is fast, effortless, strong; anterior coupling is slow, effortful, weak.
The evolutionary warrant: the major human advance in frontal lobe size, complexity and connectivity is concentrated in BA 10, the frontal pole (Semendeferi et al. 2001), not in posterior VM (Semendeferi et al. 2002). So the capacity to be moved by the distant, the abstract, and the improbable is the recent addition — and its weakness is structural, not incidental.
Supporting lesion evidence, such as it is:
- Time: VM patients with rostral-sparing-posterior lesions show “myopia” for remote-future consequences (Damasio 1994).
- Abstractness: those patients make choices leading, several steps down the line, to financial and relational ruin, but “never engage in actions that immediately lead to physical harm to themselves or to others.” Childhood VM lesions impair moral/ethical judgment specifically (Anderson et al. 1999); moral-dilemma imaging activates the frontal pole (Greene et al. 2001).
- Probability: VM patients generate anticipatory somatic responses under 100%-certain Pavlovian conditioning (Bechara et al. 1999) but not during the gambling task’s 10–50% punishment (Bechara et al. 1996). Patients whose lesions extend posteriorly into ACC and basal forebrain fail at both (Tranel et al. 1996).
- Valence: right VM damage produces myopia for negative consequences; left VM damage does not, or does less (Manes et al. 2002; Tranel et al. 2002).
Hold this at arm’s length
The gradients are not dissociated. Four axes, one anatomical dimension, and no study testing more than one at a time. In practice the axes correlate — immediate outcomes tend to be more concrete and more certain — so it is not clear the cited data could distinguish them even in principle. The wiki should record the hierarchy as a proposal with anatomical motivation, not as a finding.
The valence axis is half-tested. Only the negative half has support. Bechara & Damasio concede it: “It remains to be seen whether the impairment in the left VM patients relates to ‘myopia’ for future positive consequences.”
That half-tested claim also sits beside Craig’s forebrain emotional asymmetry (right AIC ↔ sympathetic/arousal/withdrawal/aversive; left AIC ↔ parasympathetic/affiliative/appetitive), which runs in the same direction from entirely different premises — autonomic innervation asymmetry and brain energy optimization rather than lesion myopia. Two independent routes to right-negative/left-positive is worth noticing. It is not corroboration in any strong sense: the regions differ (VM vs AIC), and neither author cites the other. Recorded as a convergence to watch.
The other name for this region
Here is the connection the sources do not make. Seth & Friston (2016) identify the visceromotor-areas — agranular AIC, ACC, subgenual cingulate, and OFC — as the top of the interoceptive hierarchy, and argue from their agranularity (no layer IV) that they are sources of descending prediction rather than targets of ascending prediction error.
So the same cortex is:
- for Bechara & Damasio, the structure that couples event knowledge to stored somatic patterns and triggers states from thoughts;
- for Seth & Friston, a source of descending interoceptive predictions.
These are close enough to be worth stating plainly: triggering a somatic state from a thought and issuing a descending interoceptive prediction may be the same operation described in two vocabularies thirty years apart. That reading is the wiki’s, not either paper’s, and it should not be promoted past a conjecture — the territories are not identical (the somatic marker literature’s “VM” is medial and orbital; Seth & Friston’s visceromotor set is defined cytoarchitecturally), and neither source engages the other. But it is the most promising unexplored link in this material, and it is what would place the somatic marker hypothesis on feedforward-vs-predictive-interoception, where it currently has no position.
The 1996 ingest gives this conjecture a middle term, and the verdict is mixed. Convergence zones are content-free re-activation devices — structurally generative models, which is the strongest form the parallel takes. But their re-activation is explicitly retrodictive: an attempt to reconstitute the somatic state that belonged to an earlier conjunction, with no comparator, no prediction error, and no precision. Reconstituting a past state and predicting the next one look identical in prose and are not the same computation. So the conjecture survives with a sharper edge: Damasio built the right architecture pointed the wrong way in time. Developed on convergence-zones.
A naming caution. “VM cortex,” “VMPFC,” and “OFC” are used loosely and inconsistently across this wiki’s sources, and lesion extent in the Iowa patient group (Fig. 1 shows overlap of 4+ patients across medial and orbital territory) is not identical to any cytoarchitectural definition. Bechara & Damasio themselves distinguish lateral OFC/DLPFC — where somatic states bias thought and working memory — from the medial VM that triggers states. Treat cross-source claims about “the VM” with suspicion.
The region survives its critics — with revisions
The half of the somatic marker framework that Dunn et al. (2006) uphold. Their verdict on the neural substrate is that it is “reasonably well supported”: lesions to VM, amygdala, insula and somatosensory cortex do impair the iowa-gambling-task, and these regions do activate in decision-making and body-state imaging. That is “a clear advance on earlier models of decision-making.” Three revisions follow, and the wiki should carry them here rather than on the study page.
1. It is right-lateralized, and it extends past Damasio’s VM. Tranel et al. (2002) contrasted bilateral, left-only and right-only VM lesions: the bilateral and right-only groups showed impaired everyday decision-making and marked IGT deficits; the left-only group was low-normal. Clark et al. (2003), designed to control the laterality confound, found deficit severity correlated with right- but not left-lesion volume (n = 21 right, 20 left, 20 controls) — and region-of-interest analysis implicated right middle frontal gyrus, superior frontal gyrus, and medial PFC. That is territory outside the VM as Damasio defines it. Note this is orthogonal to the valence axis above: Tranel’s laterality is about which hemisphere carries the deficit, the gradient table’s is about which carries positive vs. negative states, and only one of them has independent anatomical motivation.
2. DLPFC is involved, and the absorption costs something. IGT performance is worse with damage extending into DLPFC (Bechara et al. 1998b), correlates with DLPFC resting-state activity (Adinoff et al. 2003), and large PFC lesions impair the task where discrete ones may not. The framework can absorb this — markers work partly by allocating attentional and working-memory resources, so DLPFC involvement is predicted rather than embarrassing. Dunn et al. accept the move and name the price: it “does make the predictions of the SMH difficult to distinguish from other, non-somatic theories.”
3. Two findings cut against VM centrality specifically. Manes et al. (2002) divided patients by lesion type and found those with discrete OFC lesions (including VM) performed in the normal range on the IGT and two other decision measures — only the large dorsal-plus-ventral lesion group was impaired. (Dunn et al. flag the confound themselves: the OFC group’s lesions were predominantly left-sided, the large group’s predominantly right.) And Fukui et al. (2005), using event-related fMRI to isolate the risk-anticipation period, found risk anticipation activated the medial frontal gyrus and, “surprisingly,” produced no orbitofrontal difference — with OFC signal dropout from susceptibility artefact as a partial excuse (Cusack et al. 2005).
And the naming problem is worse than this page had it. Dunn et al. spend a paragraph on it before reviewing any lesion data, which is the right order. Damasio’s VM spans BA 10, 11, 12 (medial aspects), lower 24, 25 and 32; other groups use “OFC” for overlapping-but-different territory extending onto the lateral orbital surface. Öngur & Price (2000) argue the region contains two distinct networks: an orbital network for sensory integration (BA 11, 13, 12/47) and a medial visceromotor network (BA 9, 10, 11, 13, 14, 24, 25, 32). Dunn et al. adopt Damasio’s term while noting “the lack of specificity of this term” — the same compromise this page makes.
That last point is not just housekeeping. Öngur & Price’s medial visceromotor network is close kin to the visceromotor-areas Seth & Friston place atop the interoceptive hierarchy — and Bechara & Damasio cite Öngur & Price for the connectivity underwriting their posterior→anterior gradient. So three literatures are carving the same cortex with the same word, none citing the others’ use of it. The conjecture above (that triggering a somatic state from a thought and issuing a descending interoceptive prediction may be one operation in two vocabularies) now has a third source using the same anatomical division. Still a conjecture; better furnished than it was.