NOTE-332: Damasio — The somatic marker hypothesis and the possible functions of the prefrontal cortex

Contribution

A compact statement, by its author, of a hypothesis already set out in Descartes’ Error (1994) and earlier reviews (the text “follows closely, in form and substance, several reviews”, p. 1413). What it adds as a paper is the explicit list of assumptions and operations (§2), a named network (§3), an evolutionary argument for why the markers are somatic (§4), and a summary of the first three lines of evidence (§5). There is no new experiment here.

Key insight

Good decisions under uncertainty need something besides knowledge and logic. Patients with ventromedial prefrontal damage keep their IQ, factual and social knowledge, language, attention, working memory and performance on the Wisconsin Card Sorting Test, and still decide badly in personal and social life. What they lose along with good decisions is emotional responsiveness. Damasio’s account is that past experience leaves a link between a kind of situation and the body state it produced. Re-enacting that body state when an option is imagined “marks” the option as good or bad. The mark either is felt (an alarm or incentive signal) or acts covertly as a bias, and so cuts the option space down before cost-benefit reasoning starts. Without the marks, “options and outcomes become virtually equalized” (p. 1415), and choosing is slow, error-prone, or random and impulsive.

Assumptions

• Scope. The hypothesis is about ventromedial prefrontal patients and the normal function their deficit reveals. It is “not … a general theory for how prefrontal cortices work” (p. 1414). Patients with other frontal lesions “may or not be accountable by” it.
• Taken for granted. “An operative self and consciousness, the basis for neither of which I will discuss” (p. 1414).
• Four background assumptions (§2a):
  1. Reasoning and decision depend on many levels of operation, some conscious (“minded”) and some not. Minded operations depend on sensory images built in early sensory cortices.
  2. All mind operations depend on support processes such as attention and working memory.
  3. Knowledge of situations, actors, options and outcomes is stored in “dispositional” form (coded, implicit, non-topographic) in higher-order cortices and some subcortical nuclei, per the convergence-zone framework (Damasio 1989a, b; LIT-385 is 1989b). It is made explicit in motor responses (some of which constitute emotions) and in images.
  4. Knowledge divides into (A) knowledge about bioregulatory processes and body states, including emotions; (B) knowledge about entities, facts, actions and stories; (C) the links between B items and A items formed in individual experience; and (D) categorisations of A, B and C.
• Emotion is body-state representation. Emotion is “most importantly, though not solely” expressed “through changes in the representation of body state”, and represented in the brain as “transient changes in the activity pattern of somatosensory structures” (p. 1414). “Somatic” covers musculoskeletal, visceral and internal-milieu components, and a somatic signal need not originate in the body.

Key results

The paper’s results are its proposed operations, and the evidence it reports from earlier studies.

• Proposed operations (§2b).
  • (i) Ventromedial prefrontal cortex holds dispositional links between a class of situation and the emotional state that went with it. The link holds neither the facts nor the emotion, only “the potential to reactivate an emotion” (p. 1414).
  • (ii) When a situation recurs, facts are recalled as images and, “nearly so” simultaneously, the ventromedial link activates the emotional apparatus (e.g. the amygdala). Reactivation goes by a “body loop” (the body changes and the change is sensed) or an “as if body loop” (somatosensory cortex adopts the pattern directly, the body bypassed). Damasio believes the body loop came first but that the “as if” loop is now used more often (p. 1415). Either may be overt or covert.
  • (iii) The somatosensory pattern, co-displayed with the imagined scenario, “marks the scenario as good or bad”. Overt, it is an alarm or incentive signal. Covert, it is a biasing signal, for instance through dopamine (p. 1415).
  • (iv) Markers also boost attention and working memory and may assist response inhibition.
  • (v) Markers make logical reasoning tractable by letting some option-outcome pairs be “rapidly rejected or endorsed”.
• Network (§3). Ventromedial convergence zones; central autonomic effectors (amygdala, hypothalamus, brainstem nuclei); somatosensory cortices (insula, SII, SI), “especially in the non-dominant hemisphere”; possibly basal ganglia. The anatomical case is that ventromedial cortex receives all sensory modalities, is “the only known source” of frontal projections to central autonomic control (Nauta 1971), and is reciprocally linked to hippocampus and amygdala. Basic unconditioned stimuli use a separate thalamus-to-amygdala route, and the formulation “predicts a dissociation” between responses to complex and to basic stimuli (p. 1416).
• Why somatic (§4). Personal and social situations are bound to reward, punishment and homeostatic regulation, which are represented somatically. A survival-guidance system was “pre-adapted” for abstract decisions (p. 1417). Only the situation-to-body-state links need to be stored, because body states “can be re-enacted on demand”.
• Evidence (§5), all summarised from other papers, without statistics.
  • Emotive pictures (Damasio et al. 1990, 1991; Tranel 1994; Tranel et al. 1995). Normal controls and non-frontal lesion patients give skin conductance responses (SCRs) to emotionally charged pictures and few to neutral ones. Bilateral ventromedial patients give none to the charged pictures, though their SCRs to startle and deep breaths are normal.
  • Gambling task (Bechara et al. 1994). Four decks; A and B pay $100acardwithlargeoccasionalpenalties(net-$250 per ten cards); C and D pay $50withsmallones(net+$250 per ten cards); 100 selections, a $2000 loan. 44 controls (21 women, 23 men, aged 20–79) against 6 ventromedial patients (4 men, 2 women, aged 43–84). Controls drift to C and D and, “about halfway through”, keep to them. Patients keep choosing A and B and lose the loan. High-IQ controls asked to track the penalties aloud could not compute the decks’ net yields. A reversed task, with punishment up front and unpredictable reward, leaves the patients unchanged, which the paper takes against both hypersensitivity to reward and insensitivity to punishment (Anderson et al. 1996, an abstract).
  • SCRs during the task (Bechara et al. 1996). In the roughly four seconds before a choice, controls develop large anticipatory SCRs before choosing from the bad decks, rising over the task, and not before the good decks. Ventromedial patients show none, though their SCRs to actual losses are normal.
  • Timing of knowledge. “Recent studies in our laboratory suggest” that controls produce the anticipatory SCRs “long before they have … any notion whatsoever” of which decks are good (p. 1419). This is the result published in the 1997 report (LIT-383); here it is one sentence.
• Discussion. To Everitt: in adults the “as if” loop is the usual mode, so changes in the periphery should have relatively little effect on emotion, feeling and decision (paraphrased: the line is cut off in the scan); the evidence from cord damage and peripheral neuropathy is left open. To Bishop: a failure to inhibit an early-rewarded response is “conceivable”, but the striking fact is that the patients do not correct when they realise their strategy is losing; most patients who fail the task “acquire classical conditioning normally”.

Claims

id	claim	strength	support
C1	Ventromedial prefrontal patients decide badly in personal and social life despite normal intellect, knowledge, language, attention and working memory	moderate	clinical observation and cited neuropsychology (Damasio 1979, 1994; Saver & Damasio 1991); no data in this paper
C2	The same patients lack SCRs to emotionally charged pictures while their autonomic responsiveness to startle is normal	moderate	cited studies (Damasio et al. 1990, 1991; Tranel 1994); group sizes and statistics not given here
C3	Ventromedial patients choose the disadvantageous decks of the gambling task and lose; controls learn the advantageous ones	moderate	Bechara et al. 1994, 44 controls against 6 patients; no statistics given here
C4	Controls develop anticipatory SCRs before risky choices; patients do not, though they respond to losses	moderate	Bechara et al. 1996, summarised; no numbers here
C5	Re-enacted body states mark imagined outcomes and so constrain reasoning (the somatic marker mechanism)	assertion	the hypothesis itself; argued from C1–C4 by inference to the best explanation, which the paper does not test against alternatives beyond the reward and punishment sensitivity accounts
C6	The “as if body loop” has largely superseded the body loop in adults	assertion	”I believe” (p. 1415); no evidence offered, and the reply to Everitt defers the question to further study
C7	Anticipatory SCRs precede any conscious notion of which decks are good	weak here	one sentence citing unpublished “recent studies”; the evidence is the 1997 report
C8	The patients’ deficit is not hypersensitivity to reward or insensitivity to punishment	weak	a reversed task reported only as a conference abstract (Anderson et al. 1996)

Concepts

• Somatic state. An umbrella term for emotional changes, read as representations of body state (musculoskeletal, visceral, internal milieu), which need not originate in the body.
• Somatic marker. A somatic state, re-enacted from a learned link, co-displayed with an imagined option-outcome scenario so that it qualifies it as good or bad.
• Dispositional. “Coded, implicit and non-topographically organized” (p. 1414): a record that can reconstruct a pattern elsewhere but does not hold it. From the convergence-zone framework.
• Body loop / “as if body loop”. Re-enactment through an actual change in the body, sensed back; or directly in somatosensory cortex, with the body bypassed.
• Overt / covert. A marker that is attended and felt (a feeling), or one that biases processing without awareness. Feelings are body states or their surrogates “made conscious”; unattended ones still act (p. 1416).
• Ventromedial patients. Bilateral lesions of the ventral and medial prefrontal cortex with the “index condition”: acquired defects of real-life decision making.

Connections

• Damasio’s 1989 convergence-zone proposal (LIT-385). Cited as 1989b for the dispositional and convergence-zone framework. The ventromedial links here are convergence zones that record conjunctions of situation and body state. This paper is the clearest statement I could read of what that framework is for; the 1989 paper itself could not be reached.
• The 1997 Science report (LIT-383). The “recent studies” of the last paragraph of §5. Its note checks how far its statistics carry the claim this paper previews.
• Damasio’s later work. His later papers on feeling, homeostasis and consciousness are being filed in this record at the same time; how far they keep the body-state reading of emotion and the “as if” loop stated here is for their readings to say. This paper mentions homeostasis only in passing, in §4, where personal and social situations are said to be linked to reward, punishment and homeostatic regulation.
• Seth (LIT-135, NOTE-169). Seth’s interoceptive inference in the service of allostasis puts body-state signals at the centre of consciousness; this paper puts them at the centre of decision and sets consciousness aside. The two meet in the claim that body-state representations, not the body, carry the signal: the “as if” loop is a representation-level mechanism of the kind a predictive-processing account would give.
• Roberts (LIT-207, NOTE-157). Roberts lists “affect with somatic expression” among states out of bounds to systems without a body. The “as if body loop” is a case where the somatic state is produced with the body bypassed, so an argument from this hypothesis to “affect needs a body” would have to rest on the developmental claim (the body loop came first), not on the adult mechanism.
• Cross-modal fluency (LIT-190, NOTE-146). A felt processing quality enters judgements about its source. Overt somatic markers are the same shape of claim with a body state as the feeling.
• Not held in either record. Descartes’ Error (1994); Bechara et al. 1994 (Cognition 50:7–12) and 1996 (Cerebral Cortex 6:215–225), the primary reports of the gambling task and its SCRs; Shallice & Burgess (1993), whose marker proposal §4 compares. The later critiques of the hypothesis and of the gambling-task evidence, Maia & McClelland (2004, PNAS, DOI-10.1073/pnas.0406666101) among them, are not held either and were not read here.

Bearing on the record

• No THEORY bears on it. The record holds no account of emotion or decision. A THEORY drawn from this work would be “body-state signals re-enacted from ventromedial prefrontal links bias decisions under uncertainty before and alongside reasoning”. It should not be sourced from this paper alone: this paper reports the evidence secondhand, and the timing claim it most needs is in the 1997 report, whose statistics are weaker than its abstract. A THEORY would also have to read the critiques named above.
• Topic. The work’s primary subject is emotion and its role in decision, which the new emotion-and-affect topic now holds. cognition covers reasoning and decision; neuroscience the lesion and autonomic evidence. Nothing else is left unsaid.
• No ML instruction. The paper says nothing about machines or learning systems. Nothing here belongs in the anthology.

Limitations

• Evidence secondhand. Every result is summarised from other papers, with no group statistics, effect sizes or tests.
• Small samples. The gambling comparison is six patients against 44 controls, and the patients are older (43–84 against 20–79).
• Alternatives treated briefly. Response inhibition is set aside as “implausible as the sole source of the defect” for reasons of task complexity, not by an experiment reported here. Reward hypersensitivity and punishment insensitivity are set aside on a conference abstract. The alternative that controls first reason and then respond somatically is found “less plausible” on evolutionary grounds and on the unpublished timing data (p. 1419).
• Mechanism unobserved. No experiment here measures the ventromedial link, the “as if” loop or the dopamine route. SCRs index a somatic state; they do not show that it marks options or that it causes the choice.
• Body loop untested. The author’s reply to Everitt leaves the role of peripheral signals open.

Open questions

• Is the anticipatory SCR a cause of advantageous choice, a correlate, or a consequence of knowledge participants have but are not asked about? The 1997 report is the test the paper points to; see its note.
• Do body-loop and “as if” loop separate experimentally, for instance in spinal-cord injury or peripheral autonomic failure, as the reply to Everitt suggests?
• Is the predicted dissociation between complex (cortical) and basic (thalamo-amygdala) triggers of markers found?
• Is the deficit specific to uncertain, experience-dependent choice, or does it reduce to reversal learning, as Bishop’s question implies?

Corrections

• none (there was no seed)
• Authorship. JSTOR’s record lists “Antonio R. Damasio, B. J. Everitt and D. Bishop”; Everitt and Bishop are discussants, not authors. Crossref and the article’s own byline give Damasio alone.
• Decks C and D. §5(b) says decks C and D “are also equivalent in terms of overall net loss” (p. 1418); by the paper’s own figures they are equivalent in net gain (+$250 per ten cards).