Loci of plasticity
In the Weinberger model (), excitation caused by the tone and the direct or indirect effects of the shock converge at three loci. Tone information ascends the lemniscal auditory pathway from the cochlea through the ventral medial geniculate body (MGv) to reach A1. Tone information also reaches the non-lemniscal medial (magnocellular) division of the medial geniculate body/posterior intralaminar complex (MGm), where it converges with nociceptive information from the shock that ascends the spinothalamic pathway, facilitating the response of the MGm to the CS tone on subsequent trials. The MGm projects mainly to apical dendrites of pyramidal cells in layer I of A1, where its facilitated discharges converge with the excitatory effects of the immediately preceding tone on pyramidal cells. (It has recently been shown that transmission from MGm to A1 occurs through giant, rapidly conducting axons and therefore might increase pyramidal excitation before input from the MGv reaches A1 (ref.
).) This convergence produces short-term RF plasticity that is sufficient for short-term memory but is too weak to induce enduring plasticity. However, the facilitated MGm response is also projected to the cholinergic nucleus basalis
(NB), through the lateral and central nuclei of the amygdala, where it causes an increased release of ACh in the auditory cortex (and other areas). ACh, acting at muscarinic receptors in A1, converges with cortical excitation from the effects of the tone (through the direct MGv and indirect MGm paths), producing long-term plasticity. Responses to the CS tone are thereby strengthened, and increased responses to this frequency successfully compete with inputs from other frequencies, producing a shift in tuning (for further details, see refs
This model was quickly shown to be wrong in its assumptions that transmission of CS information to the cortex involved no plasticity in the MGv. In fact, the MGv develops highly specific, but short-lasting, RF plasticity51
. There has also been experimental support for the model. For example, in the human imaging study of Morris and colleagues46
summarized above, frequency-specific plasticity was found not only in A1 but also in the medialgeniculate nuclei (MGv and MGm could not be separated), amygdala, basal forebrain and orbitofrontal cortex. All of these loci of plasticity, except the orbitofrontal cortex, were predicted by the model. However, the purpose of this section is not to defend the model, which has been evaluated elsewhere50
, but rather to use it as a point of departure in considering general findings and the recently formulated model of Suga and Ma.
Suga and colleagues have extended the domain of inquiry to the corticofugal system, specifically to the projections of the auditory cortex to the central nucleus of the inferior colliculus, and also to the somatosensory cortex, in the bat. Consequent to tone–shock pairing, they have reported CS-specific tuning shifts in both the auditory cortex and the inferior colliculus. Moreover, they report that collicular tuning shifts develop before auditory cortical shifts, although they disappear within about one hour whereas cortical shifts last at least 24 hours. Inactivation of the primary somatosensory cortex is reported to prevent cortical and collicular RF plasticity25,52
. As reviewed above, the studies do not include validation of behavioural learning.
posits that first, the auditory and somatosensory cortices receive tone and shock (nociceptive) information, respectively. Then the CS and US information converge either in association cortex, which then projects to the amygdala, or in the amygdala itself, through separate relays in association cortex. The amygdala then effects the release of ACh into the cortex from the NB. Finally, the resultant auditory cortical plasticity produces tuning shifts in the colliculus and these enter into a positive feedback loop with A1 to strengthen what would otherwise be weak plasticity. (As noted above, the collicular shifts are reported to develop rapidly while the cortical shifts develop slowly25
, which seems incompatible with the model’s principle that cortical plasticity induces collicular shifts.) Termination of the positive feedback loop, which ends the short-lived collicular plasticity, is hypothesized to be caused by inhibition from the thalamic reticular nucleus (TRS) (presumably at the level of the medial geniculate nucleus53,54
), which receives cholinergic input from the NB (). As the NB initiates and continues to promote plasticity in A1, it is not clear why its effect on the TRS should not simultaneously block ascending auditory input from the CS and break the positive feedback loop at the start of conditioning.
In considering active sites of plasticity, we begin with the auditory thalamus. As noted, the MGv does develop RF plasticity but it dissipates within an hour51
, indicating that although the MGv could participate in the induction of cortical plasticity, it cannot be responsible for its consolidation or long-term retention. MGv plasticity is more consistent with Suga’s model of time-limited subcortical auditory plasticity. However, incompatible with the Suga view of slowly developing cortical plasticity is the fact that RF plasticity in A1 develops rapidly, within only five training trials18
The MGm develops RF plasticity immediately after learning and for at least one hour (the longest period tested). However, it cannot simply project its plasticity to a ‘passive’ A1 because MGm RFs are much more complex, multipeaked and broadly tuned than those of auditory cortical cells55-57
. Therefore, long-term, specific plasticity in A1 is not merely a reflection of plasticity in the subcortical auditory system but probably reflects processes in the cortex.
The Suga model ignores the MGm, its intrinsic associative plasticity and its influences on both A1 and the lateral amygdala (LA), but the following findings directly implicate the MGm: acoustic and nociceptive information converge directly in the MGm58,59
; associative learning is accompanied by the development of plasticity in the MGm60-66
, which is long-lasting63
and is evident as CS-specific RF plasticity after conditioning56,67
; the MGm holds an associative memory trace after CS offset during conditioning68
; analogues of learning show that stimulation of the MGm induces long-term potentiation in A1 (ref.
) and tone paired with stimulation of the MGm induces heterosynaptic long-term potentiation in A1 (ref.
) and behavioural conditioning71
; lesions of the MGm interfere with auditory input to the amygdala during conditioning72-74
; the MGm develops synaptic plasticity during conditioning and does so with a shorter latency than does the amygdala65
; and fear conditioning produces increased presynaptic release of transmitter (glutamate) in MGm cells that project to the LA75
. Selective lesions of the MGm should impair cortical RF plasticity, although this has not been tested.
The two models postulate very different roles for the amygdala. Suga’s model holds it to be either the first (and only) site of convergence of the CS and US, each relayed from separate association cortices, or the recipient of plasticity from one part of the association cortex that was the site of such convergence. The Weinberger model treats the amygdala as part of the associative machinery but not as the prime site of CS–US association. The evidence indicates that it would be premature to assign a primary function for learning to the amygdala76,77
, particularly in light of the finding that destruction of the basolateral amygdala does not prevent fear conditioning77
but does impair unconditioned freezing, which is the behavioural assay on which the amygdala hypothesis is largely based78
The relative roles of the MGm and the LA remain unresolved. For example, recent studies report that plasticity in the MGm is dependent on the amygdala, although there are no reciprocal geniculo-amygdala projections79,80
. However, these studies inactivated the amygdala with muscimol, which has physiological effects for several millimetres around the injection site81
. On the other hand, in an appetitive task, the MGm develops strong plasticity in waking and continues to express it during paradoxical sleep, whereas the basolateral amygdala (BLA) exhibits weaker plasticity and does not show plasticity during paradoxical sleep82
. The authors conclude that the amygdala is more involved in strong emotional states, such as in aversive conditioning, whereas the MGm signals the importance of the CS for both aversive and appetitive conditioning. They also suggest that plasticity first develops in the MGm and then the results of this plasticity are sent to the lateral amygdala, which adds its own plasticity concerning the strength of motivation and/or the sign of emotion. This conception is compatible with the view that MGm plasticity affects A1 through its monosynaptic projections to the upper lamina. Given that RF and map plasticity develop in appetitive23,35,38
as well as aversive learning, the MGm might be more generally tied to cortical plasticity than is the amygdala.
The Suga model postulates that A1 is essential for fear conditioning because it provides auditory input to the amygdala. However, bilateral destruction of A1 does not impair fear conditioning to a tone83
and ablation of A1 does not prevent auditory stimuli from accessing the amygdala84
. The Weinberger model hypothesizes that specific memory traces in A1 are not tied directly to immediate fear behaviours but serve a flexible function that can promote adaptive behaviour in unforeseen future situations.
The Suga model postulates that the somatosensory cortex is essential for the formation of both specific plasticity in A1 and behavioural fear conditioning, because it is claimed to provide nociceptive input to the amygdala, either directly or indirectly through association cortices (). This conclusion is based on disruption of tone–shock tuning shifts following inactivation of the primary somatosensory cortex by muscimol. However, the adjacent A1 is well within the domain of diffusion of muscimol81
. More importantly, complete decortication does not preclude tone-shock fear conditioning in the rat85
, or auditory–auditory associations in humans88
. These findings also show that association cortices, which are hypothesized to be essential for fear conditioning as either direct or indirect conduits of tone and shock information to the amygdala, cannot fulfill that role.
Although the two models differ on a number of crucial points, the Suga model accepts the Weinberger model’s postulated role of the cholinergic NB, the system that we now address.
The NB cholinergic system
Several lines of research implicate ACh in learning-induced RF plasticity. However, other neuromodulators affect the function of A1. For example, noradrenaline alters tuning89,90
, serotonin can regulate intensity-dependent response functions91
and its levels increase in A1 during initial stages of avoidance learning92
, and dopamine is involved in the increased representation of a tone paired with stimulation of the VTA reward system38
The NB is the main source of cortical ACh93,94
and there is extensive evidence for the importance of ACh and of the NB in particular in many aspects of learning (reviewed in ref.
). Most relevant here, iontophoretic application of cholinergic agents to A1 acts through muscarinic receptors to produce long-lasting modification of frequency tuning
; pairing a tone with ion-tophoretic application of muscarinic agonists induces pairing-specific, atropine-sensitive shifts of tuning97
; and stimulation of the NB produces atropine-sensitive, persistent modification of evoked responses in A1 (refs
) and facilitates the responses of A1 to tones100-102
. Moreover, cells in the NB develop increased discharges to the CS+ during tone–shock conditioning before the development of neuronal plasticity in A1 (ref.
). Stimulation of the NB or treatment with ACh promotes tone–shock pairing-induced tuning shifts in A1, whereas cholinergic antagonists or lesions of the NB have the opposite effect in animals104-106
. Finally, NB neurons that project to A1 selectively increase transcription of the gene for choline acetyltransferase, which synthesizes ACh, during tone–shock conditioning, indicating that acoustic learning engages specific cholinergic subcellular mechanisms108
If learning-induced plasticity in A1 develops through engagement of the NB, then NB stimulation should be able to substitute for a standard reinforcer
, such as food or shock, although no motivational reinforcement would be involved; NB stimulation itself is apparently not itself rewarding or punishing, as it is not part of any known motivational system109-111
. It seems to act as an effective but neutral cortical activation mechanism112,113
that is ‘downstream’ of any motivational system. The NB cholinergic system can induce RF plasticity with the same characteristics as learning-induced RF plasticity. Pairing a tone with NB stimulation for only 30 trials induces CS-specific associative RF plasticity114
, as does two-tone discrimination115
, and this plasticity consolidates over 24 hours116
(). NB-induced RF plasticity depends on the engagement of muscarinic receptors in A1 (ref.
). Also, the representation of a tone that is paired with NB stimulation is increased in the A1 threshold frequency map118,119
CS-specific tuning plasticity induced by pairing a tone with stimulation of the nucleus basalis (NB), the major supplier of cortical acetylcholine
Although these findings show that the NB/ACh system can induce the same A1 plasticity that develops during learning, they do not speak directly to the issue of learning and memory. McLin et al.120,121
asked whether NB mechanisms are sufficient to produce a predicted specific behavioural memory. Rats received NB stimulation paired with a 6-kHz tone; a control group received unpaired stimulation. After training, they were tested in the absence of any NB stimulation. The specificity of behavioural effects was assessed by recording heart rate and respiration, using the well-established metric of the stimulus generalization gradient which is obtained when subjects trained with one stimulus are subsequently tested with many stimuli. If paired NB stimulation induces associative memory for the training tone, then this tone (6 kHz) should later elicit the largest behavioural responses of all tones tested.
Tone–NB pairing did induce CS-specific behavioural memory — the CS frequency of 6 kHz elicited the strongest cardiac and respiratory responses of any test frequency (). The subjects behaved as though they had learned that 6 kHz had acquired increased behavioural significance through a learning experience. The findings meet the dual criteria of associativity and specificity that have long been accepted as sufficient to allow memory to be inferred from behavioural change. Pairing induced another form of highly specific plasticity in A1, an increase in the power of high-frequency gamma waves in the electroencephalogram (EEG), which have been linked to memory formation122
. These findings indicate that pairing a tone with NB stimulation not only can induce cortical plasticity but also is sufficient for the formation of specific auditory associative memory. Overall, the results of NB studies support the hypothesis that this system is sufficient to be normally engaged by sensory stimuli and to produce both specific memory traces in A1 and specific behavioural memory.
Induction of specific behavioural memory by pairing a tone with stimulation of the nucleus basalis (NB)