RES EARCH

◥ in this brain area. Using retrograde and

RESEARCH ARTICLE SUMMARY anterograde tracing methods, we found that
the perirhinal cortex (the last station in the
NEUROSCIENCE medial-temporal loop projecting to S1) pre-

Perirhinal input to neocortical layer 1

dominantly targets layer 1 (L1), suggesting that
important events relating to memory forma-

controls learning
tion occur in neocortical L1. We tested this with
targeted chemogenetic suppression of perirhi-
nal input to L1 above the stimulated S1 region.
Guy Doron*†, Jiyun N. Shin†, Naoya Takahashi, Moritz Drüke, Christina Bocklisch, Salina Skenderi, Notably, this very specific and localized mani-
Lisa de Mont, Maria Toumazou, Julia Ledderose, Michael Brecht, Richard Naud, Matthew E. Larkum* pulation was sufficient to disrupt learning. The
effect was learning-specific and had no influ-
ence in expert animals, which demonstrated
INTRODUCTION: Arguably one of the biggest mys- training rodents to associate the direct elec- that the perirhinal input did not alter the
teries in neuroscience is how the brain stores long- trical microstimulation of the cortex with a ability to perceive and behaviorally report the
term memories. Since the 1950s, it has been well reward. This also allowed us to specifically stimulus, per se. We found that the perirhinal
established that long-term memories reside in examine the contribution of circuit elements cortex signaled information related to success-
the neocortex but that their formation is depen- in the defined anatomical location. Rodents ful behavior during learning, gated the evolu-
dent on the hippocampus and medial-temporal learned to behaviorally report the microstim- tion of distinct firing, and enhanced burst
lobe structures. It is therefore remarkable that ulation within only a few trials and improved responses in 11% of layer 5 (L5) pyramidal neu-

Downloaded from http://science.sciencemag.org/ on December 17, 2020

we still do not know what cellular mechanisms over a 3-day period, during which we examined rons in S1 (40% of neurons had reduced firing
underlie long-term memory storage in the neo- the evolution of learned neuronal responses responses and 49% showed no change in fir-
cortex or exactly where they are located. in the stimulated area. We hypothesized that ing). Apical dendritic excitability was corre-
the influence of the hippocampus on mem- spondingly enhanced in a similar proportion of
RATIONALE: The primary challenge for inves- ory formation in the neocortex occurs at the L5 pyramidal neurons. This suggested that the
tigating the neural circuit underlying memory interface between these two structures. mechanism for memory formation had a den-
formation in the neocortex is the distributed dritic origin. Consistent with this hypothesis,
nature of the resulting memory trace through- RESULTS: We first confirmed that learning to we found that both activation of g-aminobutyric
out the cortex. We therefore used a behavioral associate microstimulation of the primary acid type B (GABAB) receptors—which disrupt
paradigm dependent on both the hippocam- somatosensory cortex (S1) with a reward de- apical dendritic calcium (Ca2+) activity—and
pus and neocortex that enabled us to generate pends on hippocampal activity using suppres- activation of dendritic-targeting, somatostatin-
memory traces in a specific cortical location by sion of action potentials (APs) with lidocaine positive interneurons impaired memory for-
mation similarly to suppressing perirhinal
A Neocortex B C D input to L1. Finally, we found that after learn-
µStim ? MTL
ing
ing the microstimulation detection task, evok-
Perirhinal cortex arn ing a single burst of APs (but not the same
Le
Performance

number of low-frequency spikes) in a single L5

Reward pyramidal neuron could trigger behavior.
Entorhinal cortex
MTL
L1 CONCLUSION: We found that medial-temporal
Hippocampus PRh L5 input to neocortical L1 gates the evolution of
Learning Time
MTL Sensory cortex specific firing responses in subpopulations of L5
pyramidal neurons including up- and down-
E No learning F Learning G Long-term memory regulated firing patterns and an elevation in
burstiness by means of a mechanism that is
Burst modulation Burst modulation Nanostim.
ON most likely related to apical dendritic activity.
Burst
OFF After learning, these neocortical responses be-
come independent of the medial-temporal in-
NR fluence but continue to evoke behavior with bursts
conveying higher saliency. We conclude that
S1 S1 S1 L1 is the locus for hippocampal-dependent as-
Gating Gating
? L1 L1 L1 sociative learning in the neocortex, where
OFF ON memory engrams are established in subsets
L5 L5 L5 of pyramidal neurons by enhancing the sen-
sitivity of tuft dendrites to contextual inputs
MTL MTL

Probing the influence of the medial-temporal lobe on memory formation. (A) The perirhinal cortex
MTL
and driving burst firing.
▪
The list of author affiliations is available in the full article online.
(PRh), part of the medial-temporal lobe (MTL) structures (blue box), targets L1 in S1. (B) We performed *Corresponding author. Email: [email protected]
microstimulation (mStim) of L5 in the sensory cortex while chemogenetically suppressing the axonal projection (M.E.L.); [email protected] (G.D.)
†These authors contributed equally to this work.
to cortical L1 (red) from PRh (blue arrow). (C) Rodents learned to associate mStim with water reward.
Cite this article as G. Doron et al., Science 370, eaaz3136
(D to F) Learning was suppressed by blocking MTL input [(D) and (E)] that otherwise evoked dendritic (2020). DOI: 10.1126/science.aaz3136
activity in L5 pyramidal neurons [(F), orange tuft], which allowed them to associate contextual input to L1
with the mStim. (G) More than 3 days of learning led to subpopulations of firing responses in L5 cells in S1 READ THE FULL ARTICLE AT
such that evoking a burst but not low-frequency spikes in single L5 cells retrieved learned behavior. https://doi.org/10.1126/science.aaz3136

Doron et al., Science 370, 1435 (2020) 18 December 2020 1 of 1

RES EARCH

◥ microstimulation before its association with

RESEARCH ARTICLE reward, but they were significantly modu-
lated after the first training session (Fig. 1C,
NEUROSCIENCE bottom). Providing the reward alone (not

Perirhinal input to neocortical layer 1

coupled with microstimulation) did not evoke
such axonal Ca2+ activity in perirhinal neurons.

controls learning Learning is dependent on perirhinal input

to neocortical L1
Guy Doron1*†, Jiyun N. Shin1†, Naoya Takahashi1‡, Moritz Drüke1, Christina Bocklisch1, Next, we aimed to examine whether perirhi-
Salina Skenderi1, Lisa de Mont1, Maria Toumazou1, Julia Ledderose1, Michael Brecht2,3, nal input to L1 is crucial for memory forma-
Richard Naud4,5, Matthew E. Larkum1,3* tion in the neocortex. To specifically inhibit
perirhinal axonal activity in L1 of S1 during
Hippocampal output influences memory formation in the neocortex, but this process is poorly microstimulation learning, we chemogeneti-
understood because the precise anatomical location and the underlying cellular mechanisms remain cally down-regulated synaptic transmission (14)
elusive. Here, we show that perirhinal input, predominantly to sensory cortical layer 1 (L1), controls at the axon terminals of perirhinal long-range
hippocampal-dependent associative learning in rodents. This process was marked by the emergence of projecting neurons without affecting their in-
distinct firing responses in defined subpopulations of layer 5 (L5) pyramidal neurons whose tuft fluence on other interconnected areas. We ex-
dendrites receive perirhinal inputs in L1. Learning correlated with burst firing and the enhancement of pressed hM4Di receptors [inhibitory designer
dendritic excitability, and it was suppressed by disruption of dendritic activity. Furthermore, bursts, but receptors exclusively activated by a designer

Downloaded from http://science.sciencemag.org/ on December 17, 2020

not regular spike trains, were sufficient to retrieve learned behavior. We conclude that hippocampal drug, DREADD (15)] in the perirhinal cortex of
information arriving at L5 tuft dendrites in neocortical L1 mediates memory formation in the neocortex. mice (Fig. 1D). The axon terminals in L1 of S1
were inhibited by the application of exogenous
ligand clozapine-N-oxide (CNO) (10 mM), in-

H
ippocampal-cortical dialogue mediates firms that the perirhinal cortex is the last station jected into L1 above the stimulated region
the stabilization of memory traces (1, 2). in the medial-temporal loop projecting to S1 in (Fig. 1D; see Materials and methods). Ex vivo
However, the distributed nature of long- rodents (Fig. 1A, right, and fig. S1) (2, 10–12). experiments validated that hM4Di activation
term memory representation in the neo- with CNO efficiently reduced light-evoked ex-
cortex has challenged research into Cortical microstimulation detection task citatory postsynaptic currents (EPSCs) (Fig.
the underlying mechanisms of this process. We used a fast-learning, associative, and cortex- 1E and fig. S4). We quantified the spread of
In particular, how the medial-temporal sys- dependent task (13). Rodents were trained to CNO in S1 by injecting the dye Chicago Sky
tem modulates primary sensory cortices dur- report short (200 ms) direct electrical micro- Blue at the site of CNO application. Con-
ing learning remains largely unknown. For stimulation pulses in layer 5 (L5) of S1 (Fig. 1B) volving perirhinal axonal density with CNO
hippocampal-independent learning paradigms, where the microstimulation detection thresh- spread showed that the inhibitory action of
there is converging evidence that suggests old is lowest (13). During training, animals hM4Di/CNO was predominantly in L1—where
that neocortical layer 1 (L1) is a locus for plas- received a block (five repetitions) of micro- apical tuft dendrites of pyramidal neurons are
ticity (3–6) involving activity in the distal tuft stimulation paired with reward (sweetened located—with far less influence on the deeper
dendrites of pyramidal neurons that inner- water) regardless of their licking responses layers containing the perisomatic dendrites
vate L1 (4, 7–9). We hypothesized that inputs to the stimulus. After a brief training period (Fig. 1F and figs. S4 and S5).
from the medial-temporal system also influ- (one to two blocks of pairings), learning was Specifically blocking perirhinal cortical in-
ence neocortical L1 to control memory forma- tested by making the reward available only if put to L1 of S1 severely impaired learning dur-
tion during hippocampal-dependent tasks. the animal actively licked within a response ing the first training session (Fig. 1, G and
We investigated this in the primary somato- window of 100 to 1200 ms after microstimu- H) and resembled the effect of blocking the
sensory cortex (S1) of rodents. Application of lation onset (which we refer to as a hit; Fig. 1B). hippocampus and/or perirhinal cortex (fig. S2).
the tracer fast blue to L1 of S1 retrogradely Both mice and rats learned this task rapidly We quantified learning as the cumulative dif-
labeled presynaptic neurons in the deep layers in the first session and became experts after ference between the number of successful and
of the perirhinal cortex (Fig. 1A, bottom left). a median of three sessions (one session per failed licking responses to microstimulation—
Conversely, expression of adeno-associated day, see Materials and methods). Ipsilateral i.e., S (hits − misses)—over the whole session
virus (AAV) containing channelrhodopsin (ChR2) injections of lidocaine in the hippocampus pre- (see Materials and methods). Mice for which
with enhanced yellow fluorescent protein (EYFP) vented learning of the task (fig. S2). Making the influence of perirhinal axons to L1 of S1
(AAV-ChR2-EYFP) injected into the deep layers behavior contingent on microstimulation of was suppressed could not learn to associate
of the perirhinal cortex densely labeled axons S1 allowed us to define the area of interest and the water reward with the microstimulation
in L1 of S1 (Fig. 1A, bottom right), which con- temporal window. Moreover, it allowed us to over the first session but rather randomly
precisely target the perirhinal projection to licked in ~50% of the trials (normalized learn-
ing score: control, 0.5 ± 0.06, n = 20 mice ver-
1
Institute for Biology, Humboldt-Universität zu Berlin, D-10117 L1 of S1. To confirm the involvement of the
Berlin, Germany. 2Bernstein Center for Computational
Neuroscience, Humboldt-Universität zu Berlin, D-10115 Berlin,
perirhinal cortex in learning of microstimu- sus hM4Di/+CNO, 0.04 ± 0.1, n = 12 mice;
Germany. 3NeuroCure Cluster, Charité - Universitätsmedizin lation and reward association, we measured Kruskal-Wallis test, P < 0.001; post hoc Dunn-
Berlin, D-10117 Berlin, Germany. 4University of Ottawa Brain and perirhinal axonal Ca2+ activity using two-photon Sidàk test against control, P = 0.02). The effect
Mind Institute, Department of Cellular and Molecular Medicine,
imaging of perirhinal axons in L1 of S1 before of blocking perirhinal input to L1 (or blocking
University of Ottawa, Ottawa, ON K1H 8M5, Canada. 5Department
of Physics, University of Ottawa, Ottawa, ON K1N 6N5, Canada. and after a single session of microstimulation the hippocampus) also did not completely pre-
*Corresponding author. Email: [email protected] training (Fig. 1C, top, and fig. S3). We expressed vent transient associations between the micro-
(M.E.L.); [email protected] (G.D.) GCaMP6s in the perirhinal cortex by means stimulation and reward and contrasted with
†These authors contributed equally to this work.
‡Present address: University of Bordeaux, CNRS, Interdisciplinary of viral (AAV) injection >3 weeks previously. untrained mice (i.e., mice that had never experi-
Institute for Neuroscience, IINS, UMR 5297, 33000 Bordeaux, France. Perirhinal axons responded only weakly to enced a reward paired with microstimulation)

Doron et al., Science 370, eaaz3136 (2020) 18 December 2020 1 of 10

RES EARCH | R E S E A R C H A R T I C L E

A B µStim C mice could not be suppressed by inhibiting the

Fast Blue AAV perirhinal influence on S1 by means of CNO
Neocortex GCaMP6s
S1 (expert hM4Di+/CNO+, 0.8 ± 0.08; Wilcoxon
S1 signed-rank test, P = 1; Fig. 1H). Inhibiting
PRh Perirhinal cortex PRh perirhinal activity with lidocaine in expert
AAV.EYFP rats did not affect the animal’s performance
PRh S1 response After (fig. S2).
L1 Entorhinal cortex window To distinguish the contribution of the peri-
L2/3
L4 µStim Reward rhinal projection to L1 in S1 during micro-
L5 Reward only
Fast Blue Hippocampus Before stimulation learning, we explored the influence
L6 Licks
of the higher-order somatosensory thalamic
YFP 1s 2 ΔF/F area—the posterior medial nucleus (POm)—
µStim 0.5 s
that also projects to L1 in S1 and has also been
D E CNO
10 pA
G 100 Ctrl implicated in different behavioral paradigms
Inhibition of
(4, 6, 17). To examine the influence of POm

Σ (hits - misses)
PRh projection 20 ms
50
Baseline +CNO input in the microstimulation task, we ex-
EPSC (pA)

0
CNO -4 0 hM4Di/+CNO pressed hM4Di receptors in POm using the
AAV -8 Gpr26-cre transgenic mouse line (18, 19). Sup-
hM4Di S1 -50
µStim -12 Untrained pression of this projection from POm did
-16 not significantly impair the behavioral re-

Downloaded from http://science.sciencemag.org/ on December 17, 2020

*** 0 20 40 60 80 100
port of microstimulation (POm-hM4Di, 0.3 ±
Trial #
PRh F H 0.2, n = 7 mice; Wilcoxon rank-sum test, P =
1
Norm. learning score

*** * 0.1; fig. S6).

Norm. Intensity (%)

CNO L1 L2/3 L4 L5 L6
0.5
PRh L1 CNO spread
n.s. Perirhinal input to L1 during learning
80
L5 PRh axons 0 modulates activity and enhances burst firing
in L5 pyramidal neurons
40 -0.5
What information does the perirhinal cortex
DREADD effect
0 -1 i/
convey to S1 during microstimulation learn-
200 600 1000 ned Ctrl 4Di/ i/ 4D ing? We first examined the activity of deep-
Depth (µm) n trai h M h M 4D hM O
N
U NO CNO +C
+C - layer neurons in the perirhinal cortex, which
Learning Experts we previously identified as the source of pro-
jection to L1 of S1 (Fig. 1A and fig. S1). We used
Fig. 1. Hippocampus via perirhinal cortex projects to neocortical L1 and is necessary for learning a juxtacellular single-cell recordings during learn-
microstimulation task. (A) Perirhinal (PRh) projection to S1. (Lower left) Retrograde tracing. PRh deep-layer ing (Fig. 2A, left; sessions one to three). The
neurons were labeled with Fast Blue after application to L1 of S1. (Lower right) Anterograde tracing. firing rate of perirhinal neurons significantly
ChR2-EYFP–labeled axons of PRh project to L1 of S1. (Right) Simplified connectivity map between the increased during hit trials but decreased dur-
neocortex, perirhinal cortex, entorhinal cortex, and hippocampus (12). Investigated connection indicated by ing miss trials (n = 6 rats; firing rate: miss, −8 ±
green arrow. (B) (Upper) Schematic of microstimulation (mStim) detection task. Tungsten electrode was 11%, n = 174 trials from n = 28 neurons,
placed in L5 of S1. (Lower) Animals were trained to lick within a 1.1-s window after mStim. (C) (Upper) Wilcoxon signed-rank test, P = 0.01; hit, 31 ± 11%,
Schematic of two-photon axonal imaging of GCaMP6s-expressing PRh axons in L1 of S1. (Lower) Ca2+ n = 287 trials from n = 28 neurons, Wilcoxon
transients of PRh axons in response to mStim before and after the first mStim training session and in response signed-rank test, P < 0.001; Fig. 2, B to D). The
to reward alone. (D) Schematic of chemogenetic silencing of PRh axons in L1 of S1 during mStim task. responses in the perirhinal cortex were also
AAV.hM4Di was injected to PRh, and CNO was applied locally to L1 of S1 before starting the training session. marked by a significant increase in burst rate
Inset shows an enlarged view of the CNO injection site in S1. (E) Chemogenetic suppression of synaptic compared with baseline only during hit trials
transmission ex vivo. Average EPSC amplitudes at the cell body of L5 pyramidal neurons after blue light (blue bar) (burst rate: hit, 56 ± 22%, Wilcoxon signed-
on L1 before and after CNO application. Wilcoxon signed-rank test; ***P < 0.001; n = 13 cells. (F) Spatial rank test. P < 0.001; miss, 14 ± 21%, Wilcoxon
profile of DREADD effect estimated by the spread of dye and the fluorescence intensity of PRh-YFP axons signed-rank test, P = 0.3), although the relative
as a function of cortical depth. (G) Cumulative difference between the number of hit and miss trials change of burst rate between miss and hit trials
[S (hits − misses)] in the first training session. Light lines represent individual mice and bold lines was not significant (burst rate miss versus hit,
represent the means ± SEM. Ctrl, control. (H) Normalized learning score of untrained (n = 5), control Wilcoxon rank-sum test, P = 0.1).
(n = 20), and mice under PRh axonal suppression during learning (hM4Di/+CNO, n = 12). Kruskal-Wallis Next, we investigated whether the increased
test, P < 0.001; post hoc Dunn-Sidàk test against control, *P < 0.05, ***P < 0.001. CNO did not affect activity in the perirhinal cortex modulated the
the learning score of expert mice expressing hM4Di receptors in PRh (experts). Wilcoxon signed-rank test, response of neurons in S1 while the animal
n.s., not significant; P = 1. Each dot corresponds to individual mice and black lines represents means. learned to associate the microstimulation with
reward. We performed juxtacellular record-
that hardly ever responded to microstimula- versus hM4Di+/CNO−, n = 3 mice; Kruskal- ings from L5 excitatory neurons in S1 during
tion (Fig. 1, G and H; untrained, −0.5 ± 0.05, n = Wallis test, P = 0.1; fig. S6) (16). After three to microstimulation learning (Fig. 2A, right, and
5 mice; Kruskal-Wallis test, P < 0.001; post hoc four sessions, the performance of control mice Fig. 2E). We reasoned that, because the micro-
Dunn-Sidàk test against control, P < 0.001). plateaued at improved performance levels, stimulation electrode was placed in L5, the
CNO alone without expression of hM4Di or and these animals were considered experts in stimulation affected L5 neurons the most.
hM4Di expression alone without CNO appli- the task (expert hM4Di+/CNO−, 0.9 ± 0.04, We confirmed ex vivo that perirhinal inputs
cation had no effect on learning (hM4Di−/CNO−, n = 3 mice; Wilcoxon rank-sum test, P = 0.009). arriving in L1 target the distal dendrites of
n = 10 mice versus hM4Di−/CNO+, n = 7 mice After this point, the performance of expert L5 pyramidal neurons (fig. S4). L5 pyramidal

Doron et al., Science 370, eaaz3136 (2020) 18 December 2020 2 of 10

RES EARCH | R E S E A R C H A R T I C L E

A Juxtacellular
responses to microstimulation, averaging over
Juxtacellular the whole population we observed no overall
PRh µStim S1 change in firing in response to microstimula-
S1 tion (control baseline 2.4 ± 0.4 Hz versus post-
CNO CNO stimulus 3.3 ± 0.7 Hz, n = 42 cells from 5 mice;
rf Wilcoxon signed-rank test, P = 0.7; hM4Di/
CNO baseline 1.8 ± 0.4 versus 1.6 ± 0.4 Hz, n =
AAV.hM4Di 41 cells from 5 mice; Wilcoxon signed-rank test,
PRh P = 0.3; fig. S7).
Firing patterns of L5 pyramidal neurons
B PRh Hit E Control S1 Hit H hM4Di/CNO S1 Hit are profoundly modulated by apical inputs,
Bursts such that coincidence of somatic and apical
dendritic inputs induces trains of high-frequency
action potentials (APs) (bursts) (20–23). Burst
firing also correlates with perceptual detec-
tion and depends on the activation of their
apical dendrites that project to L1 (24) where
0.5 mV 0.5 mV the perirhinal inputs arrive. We therefore ex-
0.5 s 0.5 s amined burst firing of the same populations
C F I of L5 neurons in control and hM4Di/CNO–

Downloaded from http://science.sciencemag.org/ on December 17, 2020

PRh Miss Firing rate Firing rate treated mice. Learning to report microstimu-
37 36 5 lation significantly modulated the burst rate
in about half of the L5 pyramidal neurons in
control mice (47%, n = 18 of 37 cells; Fig. 2G)
Cell #

Modulated

Modulated
0
compared with a significantly lower fraction
(22%, n = 8 of 36 cells; Fig. 2J) in hM4Di/CNO–
65%

33%
treated mice (control versus hM4Di/CNO, chi-
-5
1
-1 0 1 2
1
-1 0 1 2 square test, P = 0.02). Contrary to average firing
Time (s) Time (s) rate, we found a significant increase in average
D
G J burst rate over the population of control L5
Miss Burst rate Burst rate pyramidal neurons during learning (control:
***
Relative change (%)

80 Hit 37 36 3
baseline 0.1 ± 0.03 Hz versus poststimulus
60
*** 0.2 ± 0.07 Hz; Wilcoxon signed-rank test, P =
***
Cell #

z-score

40 0.03), which was abolished by chemogenetic

0
Modulated

Modulated

20 suppression of perirhinal input to L1 of S1

0 * 47% (fig. S7D; hM4Di/CNO: baseline 0.1 ± 0.04 Hz
22%
1 1 -3 versus poststimulus 0.1 ± 0.03 Hz; Wilcoxon
-20 -1 0 1 2 -1 0 1 2 signed-rank test, P = 0.8).
Firing Burst Time (s) Time (s)
Emergence of three distinct classes of cell
Fig. 2. Enhanced perirhinal activity in hit trials modulates neocortical activity during learning. firing after learning
(A) Experimental paradigm: single-cell, juxtacellular recordings from deep-layer pyramidal neurons in PRh (left) and
After 3 days, the animals became experts at
L5 pyramidal neurons in S1 (right). Red indicates CNO injection site in L1 of S1. rf, rhinal fissure. (B and C) High-pass
reporting the microstimulation (i.e., behavior
filtered recording traces from an example PRh neuron during a hit and a miss trial, respectively. Gray boxes
no longer improved and perirhinal input no
indicate the period of microstimulation (mStim). (D) Relative changes in firing rate and burst rate during miss
longer affected the behavioral response to
and hit trials from PRh neurons. *P < 0.05; ***P < 0.001. (E and H) High-pass filtered recordings from an
microstimulation; Fig. 1H). Juxtacellular re-
example neuron in S1 during a hit trial in control and hM4Di/CNO mice, respectively. Gray boxes indicate the
cordings from 273 L5 pyramidal neurons in
period of mStim, and red arrows indicate bursts. (F and I) (Left) Heatmap of z-score for normalized firing
rates of individual L5 neurons in S1 of control (n = 37 cells) and hM4Di/CNO mice (n = 36 cells), respectively.
expert animals revealed three distinct catego-
ries of firing responses to reward-associated
z-scores for cells with no spikes during the prestimulus period could not be computed (control n = 5 cells;
hM4Di/CNO n = 5 cells) and were therefore not included in this analysis. (Right) Fraction of significantly
microstimulation (Fig. 3, A and C; see fig. S8
for examples; see Materials and methods for
modulated neurons in control and hM4Di/CNO S1, respectively (see Materials and methods for classification
criteria). (G and J) Same as (F) and (I), but burst rate for the same cells. Cell number is constant, and cells with
classification criteria). In 11% of cells in expert
animals, we observed a significant increase in
no burst events during prestimulus period are marked with zero z-scores (green).
firing rate (n = 30 of 273 cells, D = 21.7 ± 7.5 Hz
from baseline; Wilcoxon signed-rank test, P <
neurons responded heterogeneously during tially decreased in firing. Overall, 65% of the 0.001) briefly after microstimulation presen-
successful reports of microstimulation (i.e., neurons were significantly modulated in firing tation (L5ON cells). The increase in firing rate
hits; Fig. 2, E and F). A few neurons responded relative to baseline (n = 24 of 37 cells; see Ma- in L5ON cells was also marked by a strong
with a large and sustained increase in firing, terials and methods for classification). Sup- increase in burst rate (D = 2.3 ± 0.5 Hz from
whereas others did not respond or even de- pressing perirhinal input to L1 reduced the baseline; Wilcoxon signed-rank test, P < 0.001;
creased in firing. In still others, where the fraction of significantly modulated neurons to Fig. 3C and fig. S8, pink). In a separate sub-
firing rate initially increased, this was often only 33% (Fig. 2, H and I; n = 13 of 36 cells; chi- population consisting of 40% of neurons, there
followed by inhibition compared with base- square test against control, P = 0.007; see also was a significant decrease in firing (n = 108
line and vice versa in some neurons that ini- fig. S7). Although there was a rich repertoire of of 273 cells, D = −6.6 ± 0.5 Hz from baseline;

Doron et al., Science 370, eaaz3136 (2020) 18 December 2020 3 of 10

RES EARCH | R E S E A R C H A R T I C L E

Wilcoxon signed-rank test, P < 0.001) imme-

A Experts B Untrained C Expert Soma diately after stimulus presentation (L5OFF cells).
268 62 5 Almost half of the recorded neurons (49%)
ON 11% showed no response to microstimulation (n =

Firing rate (Hz)

4

Bust rate (Hz)

135 of 273 cells) (L5NR cells). The average base-

z-score
OFF 40%
Cell #

0 20 line activity of L5 neurons in untrained rats

2
was significantly lower than that of experts
NR 49% (firing rate: untrained 1.2 ± 0.1 versus ex-
1 1 -5 0 0 perts 4.8 ± 0.3 Hz; Wilcoxon rank-sum test, P <
-1 0 1 2 -1 0 1 2 -1 0 1 2
0.001; burst rate: untrained 0.1 ± 0.02 Hz ver-
Time (s) Time (s) Time (s)
sus experts 0.5 ± 0.07 Hz; Wilcoxon rank-sum
D test, P < 0.001; fig. S9). However, most L5 py-
ramidal neurons in untrained rats were non-
E F G responsive to microstimulation delivered at
S1
180 ΔF/F Expert Dendrites the same intensity (95% L5NR, n = 63 of 66 cells)
4
60 ON 10% with a very small population positively mod-
Focal
Hit trial #

dF/F0 (%)
120 ulated although not significantly at the popula-
plane 20 µm 40 OFF 37%
0
20 tion level (5% L5ON, n = 3 of 66 cells, D = 11.9 ±
µStim 60
0 7.6 Hz from baseline; Wilcoxon signed-rank test,
NR 53% P = 0.3; fig. S9). The lack of response to micro-

Downloaded from http://science.sciencemag.org/ on December 17, 2020

1 -20
0.1 -1 0 1 2 stimulation in untrained animals shows that
1s -1 0 1 2
100 µm µStim Time after µStim (s) the direct effect of microstimulation on cell
Time (s)
firing was very modest and highlights the con-
clusion that the development of stereotypical
H I 80 Ctrl and bursty firing responses in L5 pyramidal
WT neurons is induced by learning.
Σ (hits - misses)

Baclofen 40
µStim Baclofen
Learning modulates apical dendritic activity
L1 0 in L5 pyramidal neurons
The emergence of a strongly bursty subpop-
S1 -40 SST ulation of L5 pyramidal neurons (L5ON; Fig.
L5
3C) after learning suggests that learning might
-80
0 20 40 60 80 100 enhance dendritic activity because dendritic
SST::ChR2 Trial # activity causes burst firing in L5 pyramidal
J neurons (20, 21, 25). We therefore examined
1 Ca2+-dependent activity in the apical dendrites
Norm. learning score

465 nm
L1 of L5 neurons in S1 using two-photon micros-
SST+
0.5 * copy in expert mice. We expressed GCaMP6s
in Rbp4-Cre transgenic mice (26) expressing
L5 0 *** Cre-recombinase in L5 cortical neurons and
imaged at a depth of 154 ± 2.1 mm—the region
-0.5 of the apical dendrite known for initiation of
dendritic Ca2+ spikes (21) (Fig. 3, D and E).
-1
Ctrl Baclofen SST Ca2+ transients measured from 1 s before and
2 s after microstimulation presentation (Fig.
Fig. 3. Emergence of distinct populations after learning depends on apical dendritic activity of L5 3F) in 318 dendrites (n = 4 mice) revealed three
pyramidal neurons. (A and B) z-score normalized firing rate PSTHs of L5 pyramidal neurons in S1 of expert populations with distinct fluorescence profiles
(n = 268 cells) and untrained rats (n = 62 cells), respectively. z-scores for cells with no spikes during (Fig. 3G). Dendritic responses closely resem-
prestimulus period could not be computed (expert n = 5 cells; untrained n = 4 cells) and therefore were not bled the classes of somatic responses, with 10%
included in this analysis. (C) (Left) Average PSTHs of L5ON, L5OFF, and L5NR neurons (total n = 273 cells) of dendrites showing substantial increases in
in expert rats. Gray box indicates microstimulation (mStim). (Right) The fraction of L5ON, L5OFF, and L5NR neurons. fluorescence after microstimulation (ON den-
(D) Two-photon Ca2+ imaging from the apical dendrites of L5 pyramidal neurons in Rbp4-cre transgenic drites), 37% of dendrites showing reduced Ca2+
mice during the mStim task. (E) Horizontal imaging plane (upper) (~150 mm from pia) and average Ca2+ fluorescence (OFF dendrites), and the rest
responses (lower) for all trials from an example dendrite marked with a yellow arrow. (F) Ca2+ responses (53%) remaining nonresponsive to microstim-
in an apical dendrite marked in (E) during 180 hit trials. (G) (Left) Average peri-stimulus Ca2+ responses ulation (NR dendrites).
in ON, OFF, and NR dendrites (total n = 318 dendrites) during hit trials. Gray box indicates mStim. (Right) The emergence of three distinct populations
Fraction of ON, OFF, and NR dendrites. (H) Experimental designs for manipulating dendritic activity during of dendritic Ca2+ responses in the same pro-
mStim task. (Upper) Baclofen application in superficial layers of S1 in wild-type mice. (Lower) Optogenetic portions as the populations of somatic firing
activation of SST+ interneurons during mStim task in SST::ChR2 transgenic mice. The surface of the classes suggests a tight coupling of dendritic
craniotomy was illuminated with blue light (465 nm) (see Materials and methods). (I) Cumulative learning Ca2+ with somatic activity consistent with pre-
curve of control mice (n = 20, black), baclofen-treated mice (n = 6, blue), and SST::ChR2 mice (n = 6, vious studies (20, 27–29). However, was the
purple) during the first session. (J) Normalized learning score of control, baclofen-treated, and SST::ChR2 learning dependent on the dendritic Ca2+?
mice after the first session. Each dot corresponds to an individual mouse, and the black line indicates We tested this by activating g-aminobutyric
the mean. Wilcoxon rank-sum test against control, ***P < 0.001; *P = 0.01. acid type B (GABAB) receptors with baclofen

Doron et al., Science 370, eaaz3136 (2020) 18 December 2020 4 of 10

RES EARCH | R E S E A R C H A R T I C L E

injected locally to the superficial layers of S1 a behavioral response in naive animals (naive, high-frequency burst (average 106.2 ± 7.6 Hz;
during learning (24, 30–33) (Fig. 3H, upper). 7.5%, n = 3 of 40 trials from n = 17 cells versus n = 28 cells) compared with catch trials, where
This approach has previously been shown to learned, 33.5%, n = 76 of 227 trials from n = 28 no current was injected into the neuron (burst
suppress dendritic Ca2+ activity both in vitro cells; chi-square test, P = 0.0009; 13). The ability hit rate, 31.8 ± 4.1% versus catch rate, 25.6 ±
and in vivo (24, 30–33) (for a discussion of to detect the activity of a single neuron oc- 3.5%; n = 28 cells; one-sided paired t test, P =
the limitations of this approach, see fig. S10). curred only when the AP firing consisted of a 0.02; Fig. 4, C to E, top row). Stimulating the
Baclofen disrupted learning relative to con-
trol mice (baclofen 0.2 ± 0.05, n = 6 mice ver-
sus control 0.5 ± 0.06, n = 20 mice; Wilcoxon
rank-sum test, P = 0.01; Fig. 3, I and J) to an A Experts
B response
extent similar to the down-regulation of peri- µStim training > 3 days Nanostim
window
rhinal inputs to L1 described earlier (Fig. 1H). Learning period
µStim 40%
We reasoned that if dendritic activity is nec- S1 L5 pyr
burst nanostim 20%
essary for learning, activating the endogenous Burst regular nanostim 20%

Experts
cortical circuitry responsible for direct sup- ?
catch 20%
pression of dendritic activity should result in
Single cell Rew. available
a similar learning deficit. We activated ChR2- ?
1s
expressing somatostatin-positive (SST) neurons, Regular spikes
which suppress dendritic activity (5, 6, 8, 34, 35) C D response E
during learning (Fig. 3H, lower). Activation of window

Downloaded from http://science.sciencemag.org/ on December 17, 2020

Burst (80 Hz) 100 39%

Hit rate (%)

42

ty
SST interneurons during training completely 80

ni
U
abolished the ability of mice to learn to as-
sociate microstimulation with water reward 1 mV 1 mV 40
50 ms 25 ms 0 0
(SST −0.4 ± 0.07, n = 6 mice versus control 42 100 23% P= 0.02
Regular (30 Hz) 0
0.5 ± 0.06, n = 20 mice; Wilcoxon rank-sum

Hit rate (%)

test, P < 0.001; Fig. 3, I and J). At present, there

Hit rate (%)

80
Trial #
is no method to specifically block dendritic 0 0
1 mV
Ca2+ actively [for example, SST neurons do not 100 ms 41 100 17% 40
Catch
exclusively target the apical dendrites of L5
0 P= 0.2
pyramidal neurons (36)]. Nevertheless, these

Hit rate (%)

manipulations, together with the observation 1 mV 0 0
80
that dendritic Ca2+ correlates with learning- 100 ms 42 100 87%
induced changes in firing activity, suggest that Microstim
40
a dendritic mechanism underlies the micro-
P< 0.001
stimulation learning and would also explain 0 0 0
-1 0 1 0 40 80
the importance of medial-temporal input to L1 500 ms
Time (s) False positives (%)
and the increased burstiness of L5 neurons.

Burst firing in single pyramidal neurons F Learning No learning Expert

retrieves learned behavior S1 S1
PRh PRh S1 PRh
What is the role of bursting during neocor- Long-range L1
(context)
tical memory processing? Although it has long ON OFF
been suggested that bursting might be a hall- Gating
mark of the neural code (37, 38), and down- Feed-forward
stream mechanisms for decoding high-frequency (sensory) L5
firing are known to exist (39–41), it has never
been conclusively demonstrated that bursts µStim µStim µStim
are a critical unit of neural information in the
mammalian brain. Here, we tested the hypothe-
sis that the firing pattern of L5 pyramidal neu- Fig. 4. Burst firing in single L5 pyramidal neurons is more salient than regular firing after learning.
rons in S1 influenced the behavior of the animal (A) Experimental paradigm single-cell nanostimulation. (Upper) Single-cell stimulation was performed after
in the microstimulation detection task (Fig. 4A). >3 days of microstimulation (mStim) training (i.e., in expert animals). (Lower) Hypothesis: Burst, but not
We first trained rats to respond expertly to mi- regular, firing induces learned behavior (i.e., licking to mStim). (B) Four types of stimulations were randomly
crostimulation and then manipulated the firing presented: 40% of mStim trials, 20% of burst (80 to 120 Hz, pink) nanostimulation trials, 20% of regular
of single L5 neurons in S1 using juxtacellular nanostimulation trials, and 20% of catch trials with no current injection. Rew., reward. (C) Examples of
stimulation, termed nanostimulation (Fig. 4B) juxtacellular responses recorded simultaneously during the four types of stimulations. Arrows indicate
(42). Consistent with previous studies (13, 43), stimulation artifacts. (D) (Left) All trials showing the spiking (black) and behavioral (red) responses to the
rats could detect brief stimulation of a single four types of stimulation for one example L5 pyramidal neuron. Black ticks indicate spikes, and red ticks
L5 neuron (nanostimulation trials) interleaved indicate licking. Gray box indicates mStim. (Right) Proportion of trials with hits. (E) Hit rate as a function of
with microstimulation trials (Fig. 4, C and D). A false-positive (catch) trials (n = 28 cells). Red circles indicate the example cell shown in (D). Statistical
previous study has shown that nanostimula- significance was tested with a one-sided paired t test. Dotted line indicates the unity line. (F) Gating theory of
tion is only effective in expert rats that have memory formation in the neocortex. PRh inputs to L1 gate long-range inputs that modulate the firing
learned to reliably respond to weak microstim- mode of L5 pyramidal neurons in S1 resulting in learning. Top row shows three behavioral conditions. Bottom
ulation currents and is ineffective in producing row shows stylized representative L5 pyramidal responses.

Doron et al., Science 370, eaaz3136 (2020) 18 December 2020 5 of 10

RES EARCH | R E S E A R C H A R T I C L E

same set of neurons to fire a similar number of serving as the neural signature of memory in and maintained with isoflurane at 5 and 2%,
regular APs (average 35.1 ± 2.4 Hz) failed to the neocortex. respectively, for mice. Rats were under anes-
induce behavioral report of nanostimulation thesia with ketamine/xylazine (100 mg kg−1 /
compared with catch trials (regular hit rate, Materials and methods 5 mg kg−1 for rats, i.p.). Animals were kept on
27.9 ± 3.9% versus catch rate, 25.6 ± 3.5%; one- Animals a thermal blanket during the entire surgery
sided paired t test, P = 0.2; Fig. 4, C to E, All experiments and procedures were approved and recovery. Animals were placed in a ste-
second row). For these experiments, we made and conducted in accordance with the guide- reotaxic frame, and craniotomies were per-
no attempt to target and stimulate a priori lines given by Landesamt für Gesundheit und formed on the basis of stereotaxic coordinates:
L5ON cells to elicit a behavioral report and Soziales Berlin (G0278/16). The following ani- AP −1.8 mm, ML ± 4.1 mm, dorsal-ventral
found that behavior was also elicited by nano- mal lines were used in this study: C57BL/6J (DV) −4.2 mm from bregma for mice; AP
stimulation of NR and/or OFF neurons (fig. wild-type mice, Gpr26-cre transgenic mice (18, 19), −3.8 mm, ML ± 6.0 mm, DV −7.0 mm from
S11). One possible explanation is that L5ON SST::ChR2 transgenic mice [SST-IRES-Cre mice bregma for rats. Injections were carried out
neurons, which were defined in expert ani- (JAX stock #018973) crossed with Ai32 mice (JAX using graduated pipettes broken back to a tip
mals, established the smallest population (en- stock #024109)] (52), Rbp4-cre transgenic mice diameter of 10 to 15 mm, at a rate of ~25 nl/
gram) that was sufficient for behavior, but a (26), and Wistar rats (Charles River). Male ani- min for a total volume of 50 to 70 nl (mouse)
larger number of neurons could nevertheless mals were used except for two Rbp4-cre female or 100 nl (rat). Incubation time was at least
influence behavior. This raises the possibility mice for two-photon dendritic imaging. The 3 weeks before transcradial perfusion or ex
that engram neurons are marked by their animals were housed in reversed 12-hour light- vivo experiment.
bursting behavior in addition to their overall dark cycles (light on between 21:00 and 09:00), AAV1.hSyn.hChR2(H134R)-EYFP.WPRE.hGH–
firing rate. We show that burst firing increases and all the behavioral experiments were per- containing acute brain sections were imaged

Downloaded from http://science.sciencemag.org/ on December 17, 2020

the saliency of cortical neurons and that learn- formed during the dark period of the cycle. using an Olympus BX51 Microscope with a 4×
ing disproportionately enhances burst firing objective. Fluorescence intensity was quanti-
behavior. Retrograde tracing and analysis fied with ImageJ software by plotting a line
Wistar rats and C57BL/6J mice (>2-weeks-old) profile across the cortical layers that calculates
Discussion were anesthetized under ketamine/xylazine the brightness value. The average gray value
We found that the perirhinal projection to L1 [65 mg kg−1 / 10 mg kg−1 for mice, 100 mg kg−1 / of each image was normalized by subtracting
of neocortex is crucial for learning a cortical- 5 mg kg−1 for rats; intraperitoneal (i.p.) injec- the minimum intensity of each section (base-
and hippocampal-dependent microstimulation tion] before being head-fixed at the stereotaxic line) and dividing by the maximum intensity
detection task. This involves the emergence of frame. Animals were kept on a thermal blanket of each section.
distinct neuronal subpopulations marked by during the entire surgery and recovery. After
burst firing correlated with an increase in their incision of the scalp, skulls were cleaned with Ex vivo electrophysiology
dendritic Ca+2 activity. We have shown previ- 70% ethanol, and a craniotomy over the barrel After 3 to 4 weeks of virus expression, sagittal
ously that apical dendritic activity in L5 neu- field of S1 [mice: 1 mm by 1 mm, centered at or coronal slices (300-mm-thick) were prepared
rons is causally related to an (expert) rodent’s anterior-posterior axis (AP) −1 mm, medial- from 35- to 50-day-old C57BL/6J mice. Whole-
ability to detect a sensory stimulus (24, 44). lateral axis (ML) 3.25 mm; rats: 2 mm by 2 mm, cell patch-clamp recordings were performed
Several other studies have demonstrated that centered at AP −2.5 mm, ML 5.5 mm] was from visually identified L5 pyramidal neurons
dendritic activity is generated by long-range made on the basis of stereotaxic coordinates. and L1 interneurons using infrared Dodt-
cortico-cortical input to L1 (7, 9) that is cru- Retrograde tracer Fast Blue (1% in dH2O, Poly- gradient contrast video microscopy. The extra-
cial for learning (34, 45, 46). The fact that sciences) was soaked in a sterile piece of tissue cellular solution contained 125 mM NaCl,
perirhinal input to neocortical L1 is also cru- and then applied onto the surface of S1 for 25 mM NaHCO3, 25 mM Glucose, 3 mM KCl,
cial for memory formation (Fig. 4F, No Learn- 5 min for mice and 10 min for rats. Seven days 1.25 mM NaH2PO4, 2 mM CaCl2, and 1 mM
ing) but not for sensory perception (Fig. 4F, later, animals were perfused transcardially MgCl2 (pH 7.4) at ~33°C. The intracellular so-
Expert) suggests that perirhinal input to the with phosphate-buffered saline (PBS) followed lution contained 115 mM K+-gluconate, 20 mM
apical dendrites serves as a gating signal for by 4% paraformaldehyde (PFA), and the brain KCl, 2 mM Mg-ATP, 2 mM Na2-ATP, 10 mM Na2-
the enhancement of these long-range, cortico- was extracted. Next, 100-mm-thick (mouse) or phosphocreatine, 0.3 mM GTP, 10 mM HEPES,
cortical contextual inputs (Fig. 4F, Learning). 150-mm-thick (rat) coronal brain sections were and 0.05 mM Alexa 594 and biocytin (0.2%)
Moreover, converging evidence implies that made using a vibratome (Leica VT1000S, Leica (pH 7.2). Whole-cell voltage recordings were
perceptual detection involves the reintegration Biosystems) and mounted on glass slides with performed from the soma (4 to 6 megohms)
of sensory-driven, feedforward, and cortically synthetic mounting medium (ROTI Mount, using a Multiclamp 700b (Molecular devices)
driven contextual information arriving in the Carl Roth). Fast Blue–labeled cells in the pe- amplifier. Data were acquired with an ITC-18
basal and apical compartments of L5 cortical rirhinal cortex were detected manually under board and analyzed using Igor software. Opto-
pyramidal neurons, respectively (7, 9, 24, 47–50). an epifluorescence microscope (Leica DMI4000 genetic synaptic stimulation was performed
Our data therefore imply that medial-temporal– B, Leica Biosystems). These areas were iden- via a light-emitting diode (LED) (470 nm) (2-ms
dependent learning predominantly involves tified on the basis of stereotaxic coordinates pulses) located in L1 around the tuft dendrite.
the plasticity of internally generated feedback (53, 54). The number of counted cells in each To activate hM4Di receptor, clozapine-N-oxide
signals, which would have ramifications not coronal section was normalized by the total (CNO) (Tocris Bioscience) was bath applied (final
only for our understanding of the brain but number of counted cells per brain. concentration, 10 mM).
also for the principles of learning in artificial
recurrent neural networks (51). We show that Anterograde tracing and analysis Chemogenetic manipulation of perirhinal
the medial-temporal input to the neocortex For anterograde tracing and optogenetic ex vivo axonal activity
enables memory formation by means of a pro- experiments, AAV1.hSyn.hChR2(H134R)-EYFP. Mice (>4-weeks-old) were anesthetized with
cess in L1 that likely enhances dendritic Ca+2 WPRE.hGH (Penn Vector Core) was injected in ketamine/xylazine (65 mg kg−1 / 10 mg kg−1, i.p.)
activity of distinct L5 excitatory subpopulations the perirhinal cortex of >2-week-old C57BL/6J and kept on a thermal blanket during entire sur-
and promotes burst firing in these, thereby mice and Wistar rats. Anesthesia was induced gery and recovery. Lidocaine (1%, w/v, AlleMan

Doron et al., Science 370, eaaz3136 (2020) 18 December 2020 6 of 10

RES EARCH | R E S E A R C H A R T I C L E

Pharma) was injected around the surgical site defined as 250 mm above and 750 mm below and coupled with a drop of water reward (pair-
before the scalp incision. The periosteum was the rhinal fissure on the basis of stereotaxic ing period). After five pairing trials, testing
removed, and a small craniotomy was made coordinates (53). Expression rate was calcu- periods began where animals were rewarded
on the injection sites. Injection coordinates lated by measuring the area under the curve only if they licked the licking port within 100
for the perirhinal cortex were AP −1.8 mm, ML within the perirhinal cortex or outside the to 1200 ms (response window) after stimulus
±4.2 mm, DV −4.2 mm, and for POm were AP perirhinal cortex divided by the total area onset. Tongue lick responses were detected with
−2 mm, ML +1.2 mm, DV −3.0 mm from bregma. under the curve. piezo-based sensors (mice) or beam breaker
AAV1/2-hSyn1-hM4D(Gi)-mCherry-WPRE- (rats). The time of the first lick after stimulus
hGHp(A) (Viral Vector Facility of the University Headpost implant and head-restraint habituation onset was taken as the reaction time. Trials
of Zürich) was injected bilaterally except for A lightweight aluminum headpost (mouse) or with reaction times between 0.1 and 1.2 s from
n = 2 mice with unilateral injection in the peri- a metal bolt (rat) was implanted on the skull of the stimulus onset were counted as hits, and
rhinal cortex of the right hemisphere (0.15 to the animal under ketamine/xylazine anesthesia trials with no lick or reaction times longer than
0.20 ml per side). The behavioral training started (65 mg kg−1 / 10 mg kg−1 for mice, 100 mg kg−1 / 1.2 s were counted as misses. When animals
after >2 weeks of expression. 5 mg kg−1 for rats, i.p.). For mice used in chemo- licked prematurely within 0.1 s from stimulus
To activate hM4Di receptor, CNO dissolved genetic experiments, the implantation was per- onset, that trial was aborted. To encourage
in extracellular solution (final concentration formed >10 days after viral injection. After the animals to use a nonconservative response
10 mM) was back-loaded in a glass pipette and acalp and periosteum were removed, a thin criterion, we only mildly punished licks in the
applied into L1 (150 mm) of S1 by gentle pressure layer of light-curing adhesives (OptiBond, Kerr interstimulus interval with an additional 1.5-s
at least 20 min before the microstimulation and Charisma, Kulzer) was applied to the skull. delay to the next stimulus presentation. Once
training. CNO was applied into two adjacent A headpost was fixed on the skull on the left animals reached 80% hit rate, pulse intensity

Downloaded from http://science.sciencemag.org/ on December 17, 2020

sites (150 ml each) of the craniotomy to max- hemisphere with a dental cement (Paladur, was gradually decreased during and over the
imize the CNO diffusion area. Heraeus Kulzer). sessions until it reached the target intensity
Head-restraint habituation began >3 days of 10 mA (mice) or 5 mA (rats). The number of
Estimation of DREADD effect on cortical layers after the headpost implantation. Habituation total trials were not fixed per session, but a
First to quantify the spread of CNO across cortical time at the first day was 5 min and then grad- session was terminated either when animals
layers, 0.5% Chicago Sky Blue 6B (Sigma-Aldrich) ually increased each day until the animal sat stopped licking to free water presentation or
diluted in ringer solution was back loaded in calmly for 1 hour. Animals were water restricted when they reached the intensity threshold,
a glass pipette and applied with gentle pressure from the second day (1 ml per day) of the habi- which was defined as the lowest intensity with
in L1 (150 mm) of S1, where CNO was applied. tuation and then trained to receive the sac- hit rate ≥80%. Control animals reached to the
Then mice were euthanized, and the brain was charin (Sigma-Aldrich) water (0.5% for mice target intensity within 3 to 5 days of training.
extracted before postincubation in 4% PFA for and 0.1% for rats) from the licking port from We define expert animals as those that per-
>24 hours. Coronal brain sections were made the fourth day of the habituation. Licking was formed the task with ≥80% hit rate at the
using a vibratome (Leica VT1000S, Leica Bio- monitored using a piezo-based sensor attached target intensity. Untrained animals received
systems) and mounted on glass slides with 4′,6- to the licking port. Weight and health of the microstimulation at the target intensity (10 mA
diamidino-2-phenylindole (DAPI)–containing animal were monitored daily. Habituation for for mice and 5 mA for rats) without any pair-
mounting medium (ROTI Mount FluorCare head restraint and licking typically took 5 days. ing trial.
DAPI). Then brain sections were imaged under Two to 3 days before the microstimulation
an epifluorescence microscope (Leica DMI4000 training or/and juxtacellular recording, 1.5 mm Behavioral quantification
B, Leica Biosystems). by 1.5 mm craniotomy was made on the right The performance of the animals during micro-
Fluorescence intensity was quantified with barrel cortex centered at AP 1.25 mm and ML stimulationd detection task was quantified by
ImageJ software by plotting a line profile across 3.75 mm from bregma for mice and 2 mm by computing cumulative difference of the number
the cortical layers with a bin size of 2.594 mm. 2 mm craniotomy centered at AP 2.5 mm and of hit trials and the number of miss trials. We
The average gray value of each image was then ML 5.5 mm from bregma was made for rats. For define the cumulative value at the last trials as
normalized by subtracting the minimum in- juxtacellular recordings in the rat perirhinal learning score (i.e., number of total hit trials
tensity of each section and dividing by the cortex, 2 mm by 2 mm craniotomy was made minus number of total miss trials). Because
maximum intensity of each section. Average in- on AP 4.5 mm and ML 5.0 mm from bregma. the learning score compares the number of hit
tensity of Chicago Sky Blue was convolved with Then a recording chamber was implanted for trials and the number of miss trials, it is de-
the average intensity of EYFP in perirhinal axons chronic access to this region. The dura was left pendent on the total number of trials. As de-
in S1 measured from anterograde tracing to intact and the craniotomy was covered with sili- scribed above the total number of trials per
calculate the estimated effect of DREADD man- con (Kwik-Cast, World Precision Instruments). session was not fixed. To compensate for the
ipulation as a function of cortical depths. Then various total number of trials per animal, we
layer boundaries were marked on the basis of Microstimulation detection task normalized the learning score with the total
work by Lefort et al. (55). Animals were trained to perform the microstim- number of trials, resulting in a value ranging
ulation task as described elsewhere (13, 43). from −1 to 1. Negative values indicate a higher
Quantification of DREADD expression Briefly, animals were trained to respond with number of miss trials than hit trials, and posi-
hM4Di-mCherry fluorescence intensity was tongue lick to a 200-ms train of microstimula- tive values indicate that the hit number was
quantified with ImageJ software by plotting tion pulses applied to barrel cortex (40 cathodal greater than the number of miss trials.
a line profile, across 1000 mm above and below pulses at 200Hz, 0.3 ms pulse duration) through
the rhinal fissure with a bin size of 2.594 mm. a tungsten microelectrode (Microprobes) in Pharmacology
The average gray value of each image was then depth of ~700 mm (mice) or ~1500 mm (rats) To inhibit neuronal activity in the hippocampus
normalized by subtracting the minimum in- from pia. Interstimulus interval was randomly and perirhinal cortex, a glass pipette was filled
tensity of each section and dividing by the distributed by Poisson delay with time con- with lidocaine (AlleMan Pharma; 1%, w/v) and
intensity at the rhinal fissure of each section. stant 3 s. In the first session, initial intensity inserted into the ipsilateral hippocampus or
The boundaries of the perirhinal cortex were of 160-mA pulses were injected into the cortex perirhinal cortex on the basis of stereotaxic

Doron et al., Science 370, eaaz3136 (2020) 18 December 2020 7 of 10

RES EARCH | R E S E A R C H A R T I C L E

coordinates (53, 54). The drug (150 nl) was the time window (event/s). For the analysis in following imaging experiment began 4 weeks
injected by a gentle pressure at least 20 min Fig. 2, the time window between 1 and 0 s before after the virus injection.
before the training onset. To inhibit Ca2+ spikes the stimulus ([−1, 0] s) was used to calculate the Imaging from behaving mice was performed
in apical dendrites, 100 mM baclofen was ap- baseline activity and 0.2 to 2.5 s ([+0.2, +2.5] s) with a resonant-scanning two-photon micro-
plied 3 min before the training at a depth of after the stimulus was used to calculate post- scope (Thorlabs) equipped with GaAsP pho-
150 mm in S1. Injections (150 nl each) were made stimulus activity. Spikes could not be reliably tomultiplier tubes (Hamamatsu Photonics).
every 20 min throughout the training. detected during the microstimulation period GCaMP6s was excited at 940 nm (typically 30
([0, +0.2] s) because of electric artifacts caused to 40 mW at the sample) with a Ti:Sapphire
Optogenetic manipulations by microstimulation. For firing rate and burst laser (Mai Tai eHP Deep See, Spectra-Physics)
For optogenetic activation of SST neurons, rate change analysis in perirhinal recordings, and imaged through a 16×, 0.8 NA water im-
SST::ChR2 transgenic mice expressing ChR2 the difference between prestimulus frequency mersion objective (Nikon). Full-frame images
in SST-positive cells were used. Photostimula- and poststimulus frequency was divided by (256 pixels by 256 pixels, 175 mm by 175 mm)
tion light (465 nm, 2 mW; Doric Lenses Inc.) average prestimulus frequency. were acquired from apical dendrites of L5
with 500-ms pulse starting at 300 ms before For z-score normalization, peri-stimulus time neurons expressing GCaMP6s at a depth of
stimulus onset was delivered via the optic fiber histograms (PSTHs) were computed for each 150 to 200 mm at 58.6 Hz using ScanImage
placed above the craniotomy. To prevent the cell with the bin size of 50 ms, and the stationary 4.1 software (Vidrio Technologies). Tungsten
mice from responding to the light rather than rate and standard deviation were computed on electrodes for microstimulation were inserted
to microstimulation, the recording chamber the basis of the baseline PSTHs. The z-score of through the access port on the chronic glass
was covered with a black rubber to prevent cells with no spikes during the baseline period window.
light leakage from photostimulation into the could not be computed and not shown in Figs. 2 For axonal calcium imaging, AAV2/1-Syn-

Downloaded from http://science.sciencemag.org/ on December 17, 2020

animals’ eyes. and 3. The z-score of burst rate was computed in GCaMP6s-WPRE (Addgene) was into the bi-
the same manner but using PSTHs based on lateral perirhinal cortex of adult C57BL/6J mice
In vivo juxtacellular recording and analysis burst rate instead of firing rate. Absolute PSTHs on the basis of stereotaxic coordinates (AP
After head-restraint habituation, juxtacellular in the poststimulus period were used to define −1.8 mm, ML ±4.2 mm, and DV −4.2 mm from
recordings were performed from deep-layer significantly modulated cells in Fig. 2. For multi- bregma). Three weeks after the injection, a
neurons from S1 and the perirhinal cortex in ple comparisons, P value of 0.05 was corrected 3-mm chronic glass window was implanted
awake head-fixed animals during microstimu- by Bonferroni method and converted to z-score over the left S1. Images (512 pixels by 512 pix-
lation detection task. For experiments in Fig. 2, (z = 3.267 for 46 bins). Bins with the absolute els, 133 mm by 133 mm) were acquired from
control mice that did not express hM4Di re- z-score higher than the Bonferroni corrected GCaMP6s-expressing axonal fibers in L1 (at
ceptors were also treated with CNO in L1 of z-score were considered as significantly mod- a depth of 30 to 40 mm) at 30.3 Hz. For each
S1 to exclude the indirect effect of CNO on ulated bins. Cells with at least one significantly mouse, imaging was performed in the same
neuronal activity (16). The glass pipette (4 to modulated bin were defined as modulated cells. field of view with the same laser power through-
8 megohms) for juxtacellular recording dur- For burst rate, cells having spikes but no bursts out the experimental period.
ing microstimulation task was filled with extra- during the baseline period were classified as All analysis for Ca2+ imaging was performed
cellular solution containing: 135 mM NaCl, nonmodulated cells. using custom-written scripts in MATLAB
5.4 mM KCl, 1.0 mM MgCl2, 1.8 mM CaCl2, For the classification of L5 neurons in expert (MathWorks). Horizontal and vertical drifts
and 5 mM HEPES (pH 7.2). The juxtacellular animals (Fig. 3), PSTHs were calculated for of imaging frames caused by animal motion
signal was amplified and low-pass filtered at each cell by averaging spikes in time bins of were corrected by registering each frame to a
3 kHz by a patch-clamp amplifier (NPI) and 100 ms for times within 2 s of hit-trials. For reference image on the basis of whole-frame
sampled at 25 kHz by a Power1401 data acqui- each cell, the stationary rate and standard de- cross-correlation. The reference image was gen-
sition interface under the control of Spike2 viation (SD) were computed on the basis of the erated by averaging any given consecutive
software (CED). For perirhinal recording in PSTHs in the period [−2, 0] s. Cells were clas- 100 frames in which motion drifts were min-
rats, the pipette was inserted with 17° toward sified to ON cell or OFF cell if PSTHs in the imal. Regions of interest (ROIs) for apical
lateral and 50° toward anterior. The mean period [0.3, 0.4] s was either more than three dendrites of L5 neurons were manually se-
depth in juxtacellular recordings are as fol- SDs above the stationary rate, or less than lected with the help of average intensity and
lows: S1 in mice, 1145.0 ± 19.15 mm; S1 in rats, three SDs below, respectively. Other cells were standard deviation projections across movie
1462 ± 25.41 mm; perirhinal cortex in rats, classified as NR cells. frames. For each ROI, pixel values inside the
6339.64 ± 122.07 mm, which is likely an over- ROI were averaged to obtain the time series
estimate of the true depth owing to oblique Two-photon Ca2+ imaging and analysis of Ca2+ fluorescence. The extracted signals were
penetrations and dimpling. For dendritic Ca2+ imaging, AAV2/1-Syn-Flex- corrected for neuropil contamination by sub-
Recorded neurons were separated into puta- GCaMP6s-WPRE.SV40 (Addgene) was injected tracting the local, peri-dendritic neuropil sig-
tive fast-spiking (FS) interneurons and regular- through a glass pipette (tip diameter, 5 to 10 mm) nals. For axonal imaging, axonal Ca2+ signals
spiking (RS) pyramidal neurons on the basis of into the left S1 barrel cortex of adult Rbp4-Cre were extracted from field ROIs (~100 mm by
spike half-width and firing rate. Cells with spike mice (#031125-UCD, MMRRC) on the basis of 100 mm). Fluorescence change (DF/F0) was cal-
half-width lower than 0.5 ms and firing rate stereotaxic coordinates (AP −1.5 mm and ML culated as (F − F0)/F0, where F0 was the baseline
higher than 8 Hz were classified to FS. Only RS 3.2 mm from bregma). A single injection (100 nl) fluorescence value in the ROI throughout the
were used for further analysis. was made at 700-mm-deep from the pial sur- whole imaging session.
All the cells and trials recorded over days face. Three weeks after the injection, a 3-mm For the classification of dendrites, PSTHs
were pooled together for comparing activity craniotomy was made over the injection site were calculated for each dendrite by averaging
(firing rate or burst rate) changes during micro- and sealed with a 3-mm glass coverslip (#1) with Ca2+ responses during hit trials. For each den-
stimulation trials. Bursts were identified as at cyanoacrylate glue. A lightweight headpost was drite, the baseline fluctuation—i.e., SD—was
least two spikes with an interspike interval of fixed on the skull in the right hemisphere with computed from the PSTH in the period [−1, 0]
≤15 ms. Burst rates were calculated by dividing light-curing adhesives and a dental cement. s. Dendrites were classified to ON dendrites or
the number of total burst events by the length of Habituation of mice to head restraint and the OFF dendrites if the PSTHs in the period [0, 2]

Doron et al., Science 370, eaaz3136 (2020) 18 December 2020 8 of 10

RES EARCH | R E S E A R C H A R T I C L E

s were either more than three SDs above the by post hoc Dunn-Sidàk test, when applicable. 22. P. Schwindt, W. Crill, Mechanisms underlying burst and
baseline, or less than three SDs below, respec- For comparison of hit rates and false-positive regular spiking evoked by dendritic depolarization in
layer 5 cortical pyramidal neurons. J. Neurophysiol. 81,
tively. Other dendrites were classified as NR rates in nanostimulation experiment, one-sided 1341–1354 (1999). doi: 10.1152/jn.1999.81.3.1341;
dendrites. t test was performed. No statistical tests were pmid: 10085360
run to predetermine sample size, and blinding 23. M. E. Larkum, W. Senn, H. R. Lüscher, Top-down dendritic
Single-neuron stimulation detection task and randomization were not performed.
input increases the gain of layer 5 pyramidal neurons.
Cereb. Cortex 14, 1059–1070 (2004). doi: 10.1093/cercor/
Once rats performed at current intensities be- bhh065; pmid: 15115747
low 5 mA on 2 consecutive days, we switched to RE FERENCES AND NOTES
24. N. Takahashi, T. G. Oertner, P. Hegemann, M. E. Larkum, Active
cortical dendrites modulate perception. Science 354, 1587–1590
single-cell stimulation experiments, as previ- 1. H. Eichenbaum, A cortical-hippocampal system for declarative (2016). doi: 10.1126/science.aah6066; pmid: 28008068
ously described (13, 43). Briefly, the animals memory. Nat. Rev. Neurosci. 1, 41–50 (2000). doi: 10.1038/ 25. S. R. Williams, G. J. Stuart, Mechanisms and consequences of
35036213; pmid: 11252767
were head fixed during the task, and waited action potential burst firing in rat neocortical pyramidal
2. E. T. Rolls, A computational theory of episodic memory neurons. J. Physiol. 521, 467–482 (1999). doi: 10.1111/
for the microstimulation or nanostimulation formation in the hippocampus. Behav. Brain Res. 215, 180–196 j.1469-7793.1999.00467.x; pmid: 10581316
detection task to begin, which occurred when (2010). doi: 10.1016/j.bbr.2010.03.027; pmid: 20307583 26. C. R. Gerfen, R. Paletzki, N. Heintz, GENSAT BAC cre-
a neuron was found. The glass pipette for juxta- 3. J. J. Letzkus et al., A disinhibitory microcircuit for associative recombinase driver lines to study the functional organization
fear learning in the auditory cortex. Nature 480, 331–335 of cerebral cortical and basal ganglia circuits. Neuron 80,
cellular single-cell stimulation and recording (2011). doi: 10.1038/nature10674; pmid: 22158104 1368–1383 (2013). doi: 10.1016/j.neuron.2013.10.016;
was glued to a tungsten microelectrode used 4. F. Gambino et al., Sensory-evoked LTP driven by dendritic pmid: 24360541
for microstimulation at a distance of ~70 mm, plateau potentials in vivo. Nature 515, 116–119 (2014). 27. F. Helmchen, K. Svoboda, W. Denk, D. W. Tank, In vivo
doi: 10.1038/nature13664; pmid: 25174710
as described elsewhere (13, 43). The glass pi- 5. E. Abs et al., Learning-Related Plasticity in Dendrite-Targeting
dendritic calcium dynamics in deep-layer cortical pyramidal
neurons. Nat. Neurosci. 2, 989–996 (1999). doi: 10.1038/
pette was filled with intracellular solution con- Layer 1 Interneurons. Neuron 100, 684–699.e6 (2018). 14788; pmid: 10526338
taining 135 mM K-gluconate, 10 mM HEPES, doi: 10.1016/j.neuron.2018.09.001; pmid: 30269988

Downloaded from http://science.sciencemag.org/ on December 17, 2020

28. L. Beaulieu-Laroche, E. H. S. Toloza, N. J. Brown, M. T. Harnett,
6. L. E. Williams, A. Holtmaat, Higher-Order Thalamocortical Widespread and Highly Correlated Somato-dendritic Activity
10 mM Na2-phosphocreatine, 4 mM KCl, 4 mM Inputs Gate Synaptic Long-Term Potentiation via in Cortical Layer 5 Neurons. Neuron 103, 235–241.e4 (2019).
MgATP, and 0.3 mM Na3GTP (pH 7.2). Record- Disinhibition. Neuron 101, 91–102.e4 (2019). doi: 10.1016/ doi: 10.1016/j.neuron.2019.05.014; pmid: 31178115
ing depth was 1902.0.6 ± 63.12 mm, which is j.neuron.2018.10.049; pmid: 30472077 29. V. Francioni, Z. Padamsey, N. L. Rochefort, High and
7. N. L. Xu et al., Nonlinear dendritic integration of sensory and asymmetric somato-dendritic coupling of V1 layer 5 neurons
likely an overestimate of the true depth owing motor input during an active sensing task. Nature 492, independent of visual stimulation and locomotion.
to oblique penetrations and dimpling. 247–251 (2012). doi: 10.1038/nature11601; pmid: 23143335 eLife 8, e49145 (2019). doi: 10.7554/eLife.49145;
During single-cell stimulation trials, a fixed 8. J. Cichon, W.-B. Gan, Branch-specific dendritic Ca2+ spikes pmid: 31880536
cause persistent synaptic plasticity. Nature 520, 180–185
duration square-wave current pulse was in- 30. E. Pérez-Garci, M. Gassmann, B. Bettler, M. E. Larkum,
(2015). doi: 10.1038/nature14251; pmid: 25822789 The GABAB1b isoform mediates long-lasting inhibition of
jected into a neuron through a glass pipette. 9. S. Manita et al., A Top-Down Cortical Circuit for Accurate dendritic Ca2+ spikes in layer 5 somatosensory pyramidal
Every stimulation sequence contained each Sensory Perception. Neuron 86, 1304–1316 (2015). neurons. Neuron 50, 603–616 (2006). doi: 10.1016/
step exactly once, whereas their order was doi: 10.1016/j.neuron.2015.05.006; pmid: 26004915 j.neuron.2006.04.019; pmid: 16701210
10. M. P. Witter, H. J. Groenewegen, Connections of the 31. E. Pérez-Garci, M. E. Larkum, T. Nevian, Inhibition of dendritic
varied pseudorandomly from trial to trial. To parahippocampal cortex in the cat. III. Cortical and thalamic Ca2+ spikes by GABAB receptors in cortical pyramidal neurons
induce a regular spike pattern, we used a single efferents. J. Comp. Neurol. 252, 1–31 (1986). doi: 10.1002/ is mediated by a direct Gi/o-bg-subunit interaction with Cav1
100 ms DC current step. To elicit a burst-like cne.902520102; pmid: 3793972 channels. J. Physiol. 591, 1599–1612 (2013). doi: 10.1113/
11. K. L. Agster, R. D. Burwell, Cortical efferents of the perirhinal, jphysiol.2012.245464; pmid: 23184512
spike pattern, brief stimulation duration of postrhinal, and entorhinal cortices of the rat. Hippocampus 19, 32. L. M. Palmer et al., The cellular basis of GABAB-mediated
25 ms was used, followed by 1175-ms inhibi- 1159–1186 (2009). doi: 10.1002/hipo.20578; pmid: 19360714 interhemispheric inhibition. Science 335, 989–993 (2012).
tion at current intensities of 50% used in the 12. L. R. Squire, C. E. L. Stark, R. E. Clark, The medial temporal doi: 10.1126/science.1217276; pmid: 22363012
lobe. Annu. Rev. Neurosci. 27, 279–306 (2004). doi: 10.1146/ 33. M. Suzuki, M. E. Larkum, Dendritic calcium spikes are clearly
nanostimulation, to prevent any further spikes annurev.neuro.27.070203.144130; pmid: 15217334 detectable at the cortical surface. Nat. Commun. 8, 276 (2017).
during the stimulation trial. Single-cell stimu- 13. A. R. Houweling, M. Brecht, Behavioural report of single neuron doi: 10.1038/s41467-017-00282-4; pmid: 28819259
lation trials, catch trials without current injection, stimulation in somatosensory cortex. Nature 451, 65–68 34. H. Makino, T. Komiyama, Learning enhances the relative
(2008). doi: 10.1038/nature06447; pmid: 18094684 impact of top-down processing in the visual cortex.
and microstimulation trials were pseudoran- 14. T. J. Stachniak, A. Ghosh, S. M. Sternson, Chemogenetic Nat. Neurosci. 18, 1116–1122 (2015). doi: 10.1038/nn.4061;
domly interleaved in a series of six trials in- synaptic silencing of neural circuits localizes a pmid: 26167904
cluding three microstimulation trials, two hypothalamus→midbrain pathway for feeding behavior. 35. S. X. Chen, A. N. Kim, A. J. Peters, T. Komiyama, Subtype-
Neuron 82, 797–808 (2014). doi: 10.1016/
single-cell stimulation trials (each of different specific plasticity of inhibitory circuits in motor cortex
j.neuron.2014.04.008; pmid: 24768300 during motor learning. Nat. Neurosci. 18, 1109–1115 (2015).
duration), and one catch trial. All trials were 15. B. N. Armbruster, X. Li, M. H. Pausch, S. Herlitze, B. L. Roth, doi: 10.1038/nn.4049; pmid: 26098758
presented at random intervals (Poisson pro- Evolving the lock to fit the key to create a family of G protein- 36. Y. Wang et al., Anatomical, physiological and molecular
cess, mean 3 s). Microstimulation currents were coupled receptors potently activated by an inert ligand. properties of Martinotti cells in the somatosensory cortex of
Proc. Natl. Acad. Sci. U.S.A. 104, 5163–5168 (2007).
adjusted [range 3 to 8 mA, mean 4.2 ± 1.1 mA doi: 10.1073/pnas.0700293104; pmid: 17360345
the juvenile rat. J. Physiol. 561, 65–90 (2004). doi: 10.1113/
jphysiol.2004.073353; pmid: 15331670
(standard deviation)] such that animals per- 16. J. L. Gomez et al., Chemogenetics revealed: DREADD occupancy 37. F. Crick, Function of the thalamic reticular complex: The
formed close to the detection threshold, which and activation via converted clozapine. Science 357, 503–507 searchlight hypothesis. Proc. Natl. Acad. Sci. U.S.A. 81, 4586–4590
(2017). doi: 10.1126/science.aan2475; pmid: 28774929 (1984). doi: 10.1073/pnas.81.14.4586; pmid: 6589612
resulted in an average microstimulation hit 17. N. J. Audette, S. M. Bernhard, A. Ray, L. T. Stewart, A. L. Barth, 38. J. E. Lisman, Bursts as a unit of neural information: Making
rate of 90%. Only cells with all three types of Rapid Plasticity of Higher-Order Thalamocortical Inputs unreliable synapses reliable. Trends Neurosci. 20, 38–43
stimulation trials were included. during Sensory Learning. Neuron 103, 277–291.e4 (2019). (1997). doi: 10.1016/S0166-2236(96)10070-9; pmid: 9004418
doi: 10.1016/j.neuron.2019.04.037; pmid: 31151774 39. H. Markram, Y. Wang, M. Tsodyks, Differential signaling via
18. S. El-Boustani et al., Anatomically and functionally distinct the same axon of neocortical pyramidal neurons. Proc. Natl.
Statistics thalamocortical inputs to primary and secondary mouse Acad. Sci. U.S.A. 95, 5323–5328 (1998). doi: 10.1073/
All statistics were performed using MATLAB whisker somatosensory cortices. Nat. Commun. 11, 3342 pnas.95.9.5323; pmid: 9560274
(2020). doi: 10.1038/s41467-020-17087-7 40. E. M. Izhikevich, N. S. Desai, E. C. Walcott, F. C. Hoppensteadt,
(MathWorks). Unless otherwise stated, all val- 19. T. A. Zolnik et al., Layer 6b Is Driven by Intracortical Long- Bursts as a unit of neural information: Selective communication
ues are indicated as means ± standard error Range Projection Neurons. Cell Rep. 30, 3492–3505.e5 via resonance. Trends Neurosci. 26, 161–167 (2003).
of the mean (SEM). Statistical significance was (2020). doi: 10.1016/j.celrep.2020.02.044; pmid: 32160552 doi: 10.1016/S0166-2236(03)00034-1; pmid: 12591219
20. M. E. Larkum, J. J. Zhu, B. Sakmann, A new cellular mechanism 41. R. Naud, H. Sprekeler, Sparse bursts optimize information
determined using Wilcoxon signed-rank test
for coupling inputs arriving at different cortical layers. Nature transmission in a multiplexed neural code. Proc. Natl.
within a group and Wilcoxon rank-sum test 398, 338–341 (1999). doi: 10.1038/18686; pmid: 10192334 Acad. Sci. U.S.A. 115, E6329–E6338 (2018). doi: 10.1073/
between a groups at a significance level of 0.05. 21. M. E. Larkum, J. J. Zhu, Signaling of layer 1 and whisker-evoked pnas.1720995115; pmid: 29934400
For comparing proportions, chi-square test Ca2+ and Na+ action potentials in distal and terminal dendrites 42. A. R. Houweling, G. Doron, B. C. Voigt, L. J. Herfst, M. Brecht,
of rat neocortical pyramidal neurons in vitro and in vivo. Nanostimulation: Manipulation of Single Neuron Activity by
was performed. For nonparametric multiple J. Neurosci. 22, 6991–7005 (2002). doi: 10.1523/JNEUROSCI.22- Juxtacellular Current Injection. J. Neurophysiol. 103,
comparisons, Kruskal-Wallis test was followed 16-06991.2002; pmid: 12177197 1696–1704 (2010). doi: 10.1152/jn.00421.2009

Doron et al., Science 370, eaaz3136 (2020) 18 December 2020 9 of 10

RES EARCH | R E S E A R C H A R T I C L E

43. G. Doron, M. von Heimendahl, P. Schlattmann, A. R. Houweling, 51. G. Kreiman, T. Serre, Beyond the feedforward sweep: Feedback 785907/HBP SGA1, SGA2, SGA3, and 670118/ERC ActiveCortex),
M. Brecht, Spiking irregularity and frequency modulate computations in the visual cortex. Ann. N. Y. Acad. Sci. 1464, Einstein Foundation Berlin EVF-2017-363, and EMBO (ALTF
the behavioral report of single-neuron stimulation. Neuron 81, 222–241 (2020). doi: 10.1111/nyas.14320; pmid: 32112444 75-2014). R.N. was supported by the Natural Sciences and
653–663 (2014). doi: 10.1016/j.neuron.2013.11.032; 52. L. Madisen et al., A toolbox of Cre-dependent optogenetic Engineering Research Council of Canada (NSERC) (RGPIN-2017-
pmid: 24507196 transgenic mice for light-induced activation and silencing. 06872) and by the Canadian Institutes of Health Research (CIHR)
44. N. Takahashi et al., Active dendritic currents gate Nat. Neurosci. 15, 793–802 (2012). doi: 10.1038/nn.3078; (RN383647 – 418955). Author contributions: G.D., J.N.S., and
descending cortical outputs in perception. Nat. Neurosci. 23, pmid: 22446880 M.E.L. planned and designed the experiments and wrote the
1277–1285 (2020). doi: 10.1038/s41593-020-0677-8; 53. K. Franklin, G. Paxinos, The Mouse Brain in Stereotaxic manuscript. G.D., J.N.S., N.T., S.S., L.d.M., M.D., and C.B.
pmid: 32747790 Coordinates (Academic Press, 2001). performed the experiments. J.L. and M.T. performed the histology.
45. D. Miyamoto et al., Top-down cortical input during NREM sleep 54. G. Paxinos, C. Watson, The Rat Brain in Stereotaxic Coordinates G.D., J.N.S., N.T., R.N., and C.B. analyzed the data. M.E.L., G.D.,
consolidates perceptual memory. Science 352, 1315–1318 (Academic Press, 1998). J.N.S., N.T., C.B., M.B., and R.N. developed data analysis
(2016). doi: 10.1126/science.aaf0902; pmid: 27229145 55. S. Lefort, C. Tomm, J. C. Floyd Sarria, C. C. H. Petersen, methodologies, consulted on experimental design, and reviewed
46. A. Payeur, J. Guerguiev, F. Zenke, B. A. Richards, R. Naud, The excitatory neuronal network of the C2 barrel column and edited the manuscript. All authors approved the final version
Burst-dependent synaptic plasticity can coordinate learning in in mouse primary somatosensory cortex. Neuron 61, 301–316 of the paper. Competing interests: The authors declare no
hierarchical circuits. bioRxiv 2020.03.30.015511 [Preprint]. (2009). doi: 10.1016/j.neuron.2008.12.020; pmid: 19186171 competing interests. Data and materials availability: All data are
14 September 2020. https://doi.org/10.1101/2020.03.30.015511. 56. G. Doron et al., Perirhinal input to neocortical layer 1 available from a GIN repository (56). Customized MATLAB and
47. M. Larkum, A cellular mechanism for cortical associations: An controls learning, G-Node (2020); https://doi.org/10.12751/ Python codes are available on request.
organizing principle for the cerebral cortex. Trends Neurosci. 36, g-node.6weafc
141–151 (2013). doi: 10.1016/j.tins.2012.11.006; pmid: 23273272
SUPPLEMENTARY MATERIALS
48. D. J. Felleman, D. C. Van Essen, Distributed hierarchical ACKN OWLED GMEN TS
processing in the primate cerebral cortex. Cereb. Cortex 1, science.sciencemag.org/content/370/6523/eaaz3136/suppl/DC1
The authors thank M. von Heimendahl for preliminary discussions
1–47 (1991). doi: 10.1093/cercor/1.1.1; pmid: 1822724 Figs. S1 to S11
and analysis; N. Nitzan for assisting in vitro experiment; and J. Aru,
49. G. N. Ranganathan et al., Active dendritic integration and References (57, 58)
W. Senn, B. Mensh, and members of the Larkum laboratory for
mixed neocortical network representations during an adaptive MDAR Reproducibility Checklist
helpful discussions. We thank SciDraw (L. Petrucco) for the
sensing behavior. Nat. Neurosci. 21, 1583–1590 (2018). rodent-head schematic used in the summary figure and Fig. 4. View/request a protocol for this paper from Bio-protocol.

Downloaded from http://science.sciencemag.org/ on December 17, 2020

doi: 10.1038/s41593-018-0254-6; pmid: 30349100 Funding: The research was supported by the German Research
50. M. Suzuki, M. E. Larkum, General Anesthesia Decouples Foundation (LA 3442/2-1, EXC 257 NeuroCure, and project number 30 August 2019; resubmitted 27 August 2020
Cortical Pyramidal Neurons. Cell 180, 666–676.e13 (2020). 327654276 – SFB 1315), European Union Horizon 2020 Research Accepted 23 October 2020
doi: 10.1016/j.cell.2020.01.024; pmid: 32084339 and Innovation Programme (under grant agreement numbers 10.1126/science.aaz3136

Doron et al., Science 370, eaaz3136 (2020) 18 December 2020 10 of 10

Perirhinal input to neocortical layer 1 controls learning
Guy Doron, Jiyun N. Shin, Naoya Takahashi, Moritz Drüke, Christina Bocklisch, Salina Skenderi, Lisa de Mont, Maria
Toumazou, Julia Ledderose, Michael Brecht, Richard Naud and Matthew E. Larkum

Science 370 (6523), eaaz3136.

DOI: 10.1126/science.aaz3136

Memory consolidation in the neocortex

Information transfer between brain structures located in the medial-temporal lobe and the neocortex is essential
for learning. However, the neuronal underpinnings of this transfer are unknown. Doron et al. found that neurons located

Downloaded from http://science.sciencemag.org/ on December 17, 2020

in the deep layers of the perirhinal cortex exhibit increased firing after microstimulation upon learning (see the
Perspective by Donato). Learning was associated with the emergence of a small population of neurons in layer 5 of the
somatosensory cortex that increased bursting upon stimulation. This increase in bursting was accompanied by an
increase in dendritic activity, and silencing the perirhinal cortex to layer 1 projection effectively disrupted learning and its
physiological correlates. During learning, perirhinal inputs thus act as a gate for the enhancement of cortico-cortical
inputs, which are necessary for stimulus detection and are strengthened during learning.
Science, this issue p. eaaz3136; see also p. 1410

ARTICLE TOOLS http://science.sciencemag.org/content/370/6523/eaaz3136

SUPPLEMENTARY http://science.sciencemag.org/content/suppl/2020/12/16/370.6523.eaaz3136.DC1
MATERIALS

RELATED http://science.sciencemag.org/content/sci/370/6523/1410.full
CONTENT

REFERENCES This article cites 55 articles, 10 of which you can access for free
http://science.sciencemag.org/content/370/6523/eaaz3136#BIBL

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Use of this article is subject to the Terms of Service

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement of
Science, 1200 New York Avenue NW, Washington, DC 20005. The title Science is a registered trademark of AAAS.
Copyright © 2020 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of
Science. No claim to original U.S. Government Works