Science - USA (2021-10-29)

INSIGHTS | PERSPECTIVES

science.org SCIENCE

By Jiyun N. Shin1,2, Guy Doron3,4,5,

Matthew E. Larkum4,5

M

emories are generally considered

to be embodied in the connection

strengths between neurons. Finding

the exact synaptic connections re-

sponsible for memories out of hun-

dreds of trillions at first appears

impossible. But new research may be taking

us a step closer. Specific brain areas under

the neocortex are known to be required for

memory formation (including the hippo-

campus and basal ganglia), while long-term

memory relating to facts and knowledge (se-

mantic memory) is thought to reside in the

neocortex ( 1 ). However, the precise location

of semantic memory is still unknown, pre-

venting researchers from understanding how

and where memories are formed.

Accumulating evidence in rodents sug-

gests that neocortical layer 1, the outermost

layer of the brain, might be a key site for

long-term plasticity ( 2 ) and learning ( 3 , 4 ).

Common to all these studies is the observa-

tion that subcortical memory structures pre-

dominantly target this layer. For instance,

higher-order thalamic input targets layer 1

in the sensory cortex, where it has also been

associated with learning ( 5 ) and enhanced

activity in the tuft dendrites of pyramidal

neurons, the principal excitatory neurons of

the neocortex ( 2 ). Similarly, it was shown that

projections from the amygdala to layer 1 are

crucial for fear memory formation ( 6 ). More

recently, it was shown that the hippocampus

also targets neocortical layer 1 via parahippo-

campal structures and is crucial for associa-

tive learning ( 7 ). This leads to the hypothesis

that layer 1 is the locus of semantic memory

formation and storage in the neocortex,

where the synapses encoding memories are

likely to be found.

Layer 1 of the neocortex is enigmatic, and

there is still no consensus about its function.

It stands out as the only layer almost devoid

of cell bodies, which makes it even more

surprising that it is the target of so many

long-range neuronal axon fibers. These fibers

tend to come from higher cortical areas and

higher-order thalamic nuclei and are known

to convey context-related feedback informa-

tion ( 8 ) and correlate with learning ( 5 , 9 , 10 ).

Layer 1 consists of a dense complex of over-

laying dendrites made up of the tops of pyra-

midal neurons (tuft dendrites). Thus, it is this

part of the pyramidal neurons that receives

feedback information, whereas the bottom

of these neurons primarily receives feature-

related, feed-forward information. Therefore,

the pyramidal neuron is ideally positioned to

link the two information streams ( 8 ).

The question remains: Why would memory-

related structures target context in layer 1

rather than features encoded in the lower

layers? Adding memory to an existing cogni-

tive theory describing the role of cortical den-

drites in perceptual experience may provide

the answer ( 11 ). This theory of perception

suggests a functional definition of “feature”

and associated “context,” where a feature is

defined as the primary output of a cortical

module (“column”), and context is defined as

relevant information from across the brain

that arrives in layer 1 of that column. Thus,

for an “orange column” responding during

a perceptual experience, “shape” would be a

part of the context for this column. This per-

ceptual theory explains how the properties of

pyramidal neurons bind context to features.

Linking this theory to memory could ex-

plain why and how memory structures in-

fluence layer 1 of the neocortex. There are

two possible scenarios: the memory content

scenario, in which memory structures pro-

jecting to layer 1 directly provide additional

context; or the memory switch scenario, in

which memory structures modulate or gate

other context-related inputs to layer 1. In

both scenarios, experiences would involve

the short-term association of context with

features, and memory would involve encod-

ing and stabilization of this association. For

example, recognizing a tiger should evoke

distributed neuronal activity around the cor-

tex according to the features accessible via

the senses and the context derived from pre-

viously learned associations (see the figure).

Seeing a tiger for the first time would involve

temporary associations that would need to be

formed or stabilized.

This implies that associative memory is the

convergence of different types of information

from various cortical and subcortical sources

to layer 1 and is therefore intrinsically dis-

tributed, even in a given cortical area. It is

possible that the different memory-related

brain structures projecting to layer 1 provide

different criteria for stabilization of context

association (novelty detection, importance,

emotional valence, or deviation from expec-

tation, etc.). Overall, semantic memory is hy-

pothesized to be the long-term association of

different contexts with particular features in

neocortical layer 1.

The two scenarios entail certain predic-

tions. The memory content scenario pre-

dicts that layer 1 inputs constitute memory

content themselves, show long-term plastic-

ity, and require the source structure to be

activated during retrieval. Evidence for the

memory content scenario has been observed

directly for thalamic projections to layer 1

( 5 ). Conversely, the memory switch scenario

explains how memory retrieval could occur

without the source structure. A recent study

showed that learning to associate a stimulus

with a reward requires hippocampal input

to layer 1 via the medial temporal lobe struc-

tures but could subsequently be retrieved

without this input ( 7 ).

It is possible that the two scenarios could

operate in parallel. For example, it is well

established that the hippocampus mediates

spatiotemporal context that is an important

aspect of episodic autobiographical memory

( 1 ). Thus, inputs to layer 1 could provide both

a criterion for consolidating associated con-

text (memory switch scenario) and a con-

stitutive part of the context itself (memory

content scenario), providing the autobio-

graphical component by linking time and lo-

cation to memories.

The precise mechanisms of memory

stabilization in layer 1 remain unknown.

Stabilization may involve the up- or down-

regulation of synaptic inputs to the densely

packed tuft dendrites of pyramidal neurons

in layer 1 ( 9 ), or possibly more-complex cir-

cuit refinements involving heterosynaptic

plasticity induction including other cortical

layers ( 2 ). There is also evidence that local

inhibition mechanisms in layer 1 can shape

and modulate long-range contextual inputs

( 5 ). Layer 1 inhibitory neurons are optimally

positioned to determine synaptic plasticity

via calcium-dependent signaling in the py-

ramidal cell tuft dendrites. Indeed, several

classes of inhibitory neurons reside in or in-

fluence layer 1 ( 8 ). These inhibitory sources

(^1) New York University Comprehensive Epilepsy Center,
New York, NY, USA.^2 Department of Neurology, New York
University School of Medicine, New York, NY, USA.^3 Scientific
and Competitive Intelligence, R&D, Pharmaceuticals, Bayer
AG, Berlin, Germany.^4 Institute for Biology, Humboldt
University of Berlin, Berlin, Germany.^5 NeuroCure Cluster
of Excellence, Charité – Universitätsmedizin Berlin, Berlin,
Germany. Email: [email protected];
[email protected]
NEUROSCIENCE
Memories off the top of your head
Cortical layer 1 has a special role in long-term memory
538 29 OCTOBER 2021 • VOL 374 ISSUE 6567