Emotion and Stress

Play a rat a tone that has repeatedly ended in a mild footshock, and two things happen when it hears the tone alone: it stops moving, and its arterial pressure rises.

Neither response is the feeling of fear. Both can be measured, and with them you can ask which parts of the brain learn that a tone predicts harm and which turn that knowledge into a reaction. A second system, a hormone loop from the hypothalamus to the adrenal glands, works on a slower clock.

Learning what predicts harm

In Pavlovian threat conditioning, an initially innocuous conditioned stimulus (CS), such as a tone, is paired with an aversive unconditioned stimulus (US), such as a brief footshock. The US produces an unlearned reaction. After pairing, the CS alone produces a conditioned response (CR): in rats, freezing (a defensive behaviour in which the animal holds still except for breathing) and autonomic changes such as a rise in arterial pressure.

A rat that freezes to a tone after shocks might simply be jumpy. The test of association is the unpaired control: a second group hears the same tones and receives the same shocks, but the two never predict each other. Only the difference between the paired and unpaired groups is learning about the tone. Keep that subtraction in mind; you will use it in the worked problem.

The circuit for this learning runs through the amygdala, a cluster of nuclei (a nucleus here is a compact group of neuron cell bodies in the brain) deep in the temporal lobe. In the rat, auditory information reaches the lateral amygdala, which acts as the input stage, and the central amygdala sends the outputs to the hypothalamus and brainstem. The damage that matters in the studies below is bilateral: both amygdalae.

An amygdala input that changes, and two outputs

If the lateral amygdala stores the tone-shock association, its response to the tone should change with pairing. Rogan, Stäubli and LeDoux (1997) recorded tone-evoked field potentials in the rat lateral amygdala and found that conditioning changed them in the same direction as long-term potentiation (LTP), a lasting increase in synaptic strength after strong activity (see synaptic plasticity). The changes were absent in rats given the tones and shocks unpaired. An LTP-like change that tracks the contingency fits plasticity as part of the learning; it does not show that this one change is all of it.

Now the outputs. LeDoux, Iwata, Cicchetti and Reis (1988) asked whether one route below the central amygdala carried both freezing and the pressure rise. Lateral-hypothalamic lesions impaired the conditioned pressure increase and spared freezing. Lesions of the caudal periaqueductal gray (the central gray, the sheath of neurons around the midbrain’s fluid canal) disrupted freezing and spared the pressure response. The hypothalamus’s sympathetic outputs are covered in the hypothalamus and the autonomic nervous system.

Lesion method matters. An electrolytic lesion passes current and destroys everything near the tip, including fibres of passage, axons that only travel through the region. Ibotenic acid, an excitotoxin, kills local neurons and spares those fibres, which is how the 1988 study made its caudal central gray lesions. A deficit after an excitotoxic lesion can be pinned on the neurons there; after an electrolytic one it might come from cut axons going elsewhere.

Worked problem. These numbers are invented, laid out in the design of the 1988 study; they are not its data. Rats were tested with the tone alone after conditioning. Freezing is the percentage of a 60-s tone spent frozen; pressure is the rise in mean arterial pressure in mmHg.

Group	Freezing (%)	Pressure rise (mmHg)
Intact, paired	60	+20
Intact, unpaired	8	+3
Lateral hypothalamic lesion, paired	57	+4
Caudal PAG lesion, paired	9	+19
Central amygdala lesion, paired	6	+3

Assume the lesions do not change the unpaired baseline and that the measures add linearly.

Step 1: the learned response. Subtract unpaired from paired in intact rats.

learned freezing = 60 − 8 = 52 percentage points; learned pressure = 20 − 3 = 17 mmHg

Step 2: each lesion as a share of what was learned. Subtract the baseline from each lesion group and divide by the learned response.

hypothalamus: freezing (57 − 8)/52 = 49/52 = 94%; pressure (4 − 3)/17 = 1/17 = 6%

PAG: freezing (9 − 8)/52 = 1/52 = 2%; pressure (19 − 3)/17 = 16/17 = 94%

central amygdala: freezing (6 − 8)/52, below baseline, so about 0%; pressure (3 − 3)/17 = 0%

Step 3: read it. Each output lesion removes one response and keeps the other: a double dissociation, two separate routes below one amygdala node. The amygdala lesion removes both, as a shared source for the two routes predicts. The result localizes the outputs of a defensive response; it says nothing yet about a feeling.

Group (all tested with the tone alone)

Intact, paired Intact, unpaired Lateral hypothalamic lesion, paired Caudal PAG lesion, paired Central amygdala lesion, paired

View

Raw values Share of the learned response

Freezing

—

Pressure rise

—

—

—

In this sketch: invented numbers laid out in the design of LeDoux and colleagues (1988), not its data; rats; the unpaired baseline is assumed unchanged by lesions and the measures are assumed to add linearly. No lesion combinations are offered because the lesson gives none. LH = lateral hypothalamus, PAG = caudal periaqueductal gray, CeA = central amygdala. “Spared” and “impaired” here mean at least or under half of the learned response, a sketch convention.

Each output lesion removes one response and keeps the other, a double dissociation below one amygdala node; the unpaired group is the subtraction that isolates learning about the tone; the result localizes outputs of a defensive response and says nothing about a feeling.

In mice, Kim and colleagues (2016) identified genetically distinct basolateral-amygdala excitatory populations linked to behaviours of opposite valence; testing such circuits by driving and silencing defined projections is the subject of designing a circuit experiment.

Lesions in rats and patients, and the CO2 exception

The amygdala is not the only structure in threat learning. Phillips and LeDoux (1992) conditioned rats to a tone in a distinctive chamber. Amygdala lesions disrupted conditioning to both the tone and the chamber; hippocampal lesions disrupted conditioning to the chamber (the context) while sparing conditioning to the tone. In this task the hippocampus was needed for context. That does not mean the hippocampus never takes part in cue learning.

Patients test what rats cannot report. Bechara and colleagues (1995) studied people with naturally occurring bilateral damage. A patient with selective amygdala damage did not acquire conditioned autonomic responses, yet learned the declarative facts (facts one can state) about which stimulus went with which. A patient with selective hippocampal damage acquired the conditioning but not the facts, and a patient with damage to both acquired neither. Three people are a double dissociation, not a population estimate; the kinds of memory involved return in memory systems.

Patient SM has bilateral amygdala damage from Urbach-Wiethe disease, a rare genetic condition. Feinstein and colleagues (2011) found that she reported at most minimal fear in the external-threat situations they tested, while her capacity for other basic emotions was not abolished. Then Feinstein and colleagues (2013) changed the threat. Three women with bilateral amygdala damage from Urbach-Wiethe disease, SM among them, inhaled a single breath of 35% CO2. All three panicked, against 3 of 12 neurologically intact comparison women on the first inhalation, and their anticipatory autonomic responses before the challenge were deficient. Fear was rated on a 0 to 100 visual analogue scale.

Design problem

You are reviewing the CO2 result. SM reported little fear to external threats, yet all three amygdala patients panicked after one breath of 35% CO2, against 3 of 12 comparison women. Interpret the result, and design the follow-up you would want before generalizing it.

One way to do it

Interpretation: the amygdala is not needed for every fear. A threat signalled from inside the body (rising blood CO2) may engage circuits outside the amygdala, for instance brainstem regions that sense CO2; that is a hypothesis this study does not test.

Limits: three patients whose damage is thought to have arisen during development, from one rare disease, give no population panic rate and may have compensated over years. The physiological records came from only two patients (heart rate) and one (skin conductance), so the autonomic conclusions rest on fewer people than the ratings.

Follow-up: keep an air inhalation as a sham, blind participants and raters to which gas is given, and randomize order. Record autonomic measures and fear ratings separately for every participant, so the reported feeling and the body’s response can dissociate. Add patients with adult-onset bilateral amygdala damage to address compensation, and image the brainstem during the challenge to test the alternative-circuit hypothesis.

The HPA loop: feedback and rhythm

The slower system is the HPA axis: hypothalamus, pituitary, adrenal. Neurons of the paraventricular nucleus (PVN) of the hypothalamus synthesize corticotropin-releasing hormone (CRH), a major activator of the axis. CRH travels a short distance in the hypophyseal portal vessels to corticotrophs of the anterior pituitary and stimulates release of ACTH (adrenocorticotropic hormone). Circulating ACTH stimulates glucocorticoid synthesis and secretion in the adrenal cortex: cortisol in humans, corticosterone in rats and mice. CRH and ACTH act through G-protein-coupled receptors; the machinery is in G proteins and their receptors and cAMP and PKA.

Keep the adrenal’s two parts apart. Adrenaline comes from the adrenal medulla, driven within seconds by sympathetic nerves; cortisol comes from the cortex, driven by a blood-borne hormone.

Glucocorticoids regulate their own secretion through negative feedback on brain regions and the anterior pituitary. Rodent studies distinguish two timescales: a fast, nongenomic feedback over seconds to minutes, and a delayed, genomic feedback over hours to days. The genomic arm runs through glucocorticoid and mineralocorticoid receptors, which can regulate transcription after moving to the nucleus (see nuclear receptors).

Cortisol is not released only during stress. Under basal conditions glucocorticoids are secreted in a circadian rhythm, highest before the active phase (morning in humans, around dark onset in nocturnal rats), and an ultradian rhythm of roughly hourly pulses. Walker and colleagues (2012) asked whether each pulse needs a hypothalamic command. In conscious male Sprague-Dawley rats, they infused CRH at a constant 0.5 µg/h for 6 h at the circadian trough, when endogenous CRH is minimal and corticosterone is not pulsatile, sampling blood every 5 min. Corticosterone still oscillated, at a mean 0.89 pulses per hour (individual range 0.79 to 0.93). So pulsatile CRH was not required in this preparation.

Turn the frequency into a period, treating the output as an ideal oscillator:

period = 1/f = 1/(0.89 per h) = 1.124 h = 67.4 min; cycles in 6 h = 0.89 × 6 = 5.34

That is model arithmetic, not a count of observed peaks, and a rat rate, not a human constant. Timing can matter for the brain: in 15 healthy men, Kalafatakis and colleagues (2018) found that the same daily cortisol dose given in different patterns changed fMRI responses to emotional faces, although momentary mood did not differ between the two subcutaneous patterns.

What a freezing rat can tell us

A rat cannot tell you how it feels. Freezing and a pressure rise are defensive responses, and a single defensive measure is not an unambiguous measure of fear. Researchers disagree on what follows. Fanselow and Pennington (2018) argue that fear is best treated as an integrated autonomic, behavioural and cognitive-emotional response, so animal circuits model fear itself. LeDoux has argued for separating defensive survival circuits from the conscious feeling of fear, which on that view needs other circuitry. The disagreement is about what the measurements license.

The rat work can locate necessary steps for a measured response under defined conditions. It cannot alone say what the animal experiences, or that a human feeling uses the same route. Cain (2023) points to differences in tasks and outcome measures across species as a limit on translation. The human cases show why both kinds of evidence are needed: SM’s minimal fear to external threats and her panic to CO2 came from one person tested two ways, so the difference lay in the threat, not the patient.