It is hypothesized that a memory is encoded in a subset ofneurons, which is activated by physiological input derived from a corresponding event, calledengram. The c-fos-TetTag system has been used to prove the engram theory, by manipulation ofcells that showed activity-dependent gene expression driven by learning. However, the physicaland basic activities that occur during learning and post learning in that group of cells, is stillpoorly understood. Here we show that the engram cells exhibit a remarkable synchronousactivity representing the contextual memory in the form of several ensembles in engram cells.These ensembles carry on their activity not only during learning but also during post-learningsleep and retrieval sessions, in contrast to non-engram cells. A compatible imaging system wasestablished to observe the neuronal activity of ~1000 CA1 neurons and the labelled engramcells; through a photoconvertible fluorescent protein Kikume Green Red (KikGR). The neuronalactivity of hippocampal CA1 neurons was observed, through Ca2+ influx with GCaMP7 in freely-moving animals by miniature head-mount fluorescent microscopy. Our advanced imagingsystem of engram cells and non-engram cells provides deeper insights into the dynamics of theneural activity during contextual memory processing. Engram cells exhibit highly repetitiveactivity corresponding to remarkable synchrony during novel context exposure. Populationvector distance (PVD) analysis indicates that total activity patterns of engram cells are stableand consistent across other sessions; sleep (NREM and REM) and retrieval, not only duringlearning. Furthermore, Non-negative Matrix Factorization (NMF) analysis extracted characteristicensembles activity that were constructed by subgroup of engram cells, representing thepersistent synchronous activity even during post-learning sleep (NREM and REM) sessions andretrieval session, but not in a distinct context. In contrast, these features were not seen in non-engram cells. These results suggest that there are several fundamental characteristics of theengram cells that give them superiority in encoding the ongoing event and consolidating thepast ones.

Ø Activity pattern extraction (Non-negative Matrix Factorization)

Ø Dot product between patterns (Pattern matching)To compare the patterns extracted across different sessions, cosine similarity between

2 vectors is used to quantify the similarity of 2 vectors. Pattern pair with > 0.6 isdefined as a significant pair in this work.

Ø Matching score (MS)To compare the activity of engram cells and non-engram cells across different sessions, we calculated a matching score across

sessions, which implies to what extent a pattern in the first session to be overlapped withany of the patterns extracted in the second session and vice versa.

Ø Population vector distance (PVD)The population vector distance is defined by the Mahalanobis distance, which implies the distance between 2 groups of

neuronal activity vector normalized by their variance.

Optogenetic activation of c-fos-Tet-tagged cells in CA1 induces recall of contextual memory②

P < 0.01, one-way ANOVA; *P < 0.05, **P < 0.01, Scheffe’s post-hoc test

Repetitive activity is a characteristic feature of Engram cells during

contextual learning③

Ensemble activity patterns are preferentially reactivated in Engram cells during both resting and retrieval sessions④

5. Ratio of sum of all correlations within the first 60s

Integrating Neuronal Ensembles Constitute and Replay Hippocampal Engram○Khaled Ghandour1,2,3, ○Noriaki Ohkawa1,2,3, ○Chi Chung Alan Fung2,4, Hirotaka Asai1,2, Yoshito Saitoh1,2,3, Takashi Takekawa2,5, Reiko Okubo-Suzuki1,2,

Shingo Soya6, Hirofumi Nishizono7, Mina Matsuo7, Masaaki Sato4,8,9, Masamichi Ohkura8,9, Junichi Nakai8,9, Yasunori Hayashi4,9,10, Takeshi Sakurai6, Makoto Osanai11,12, Tomoki Fukai2,4, Kaoru Inokuchi1,2

①

②

③

②

ROIｓ=Engram

c-fos promoter tTA TRE KikGR

Lentivirus

OFF Dox

tTA

DoubleTg mice

Thy1 gene G-CaMP7 2A RFP Thy1 gene

c-fos::tTA system with lentivirus to label engram cellsOhkawa et al., Cell Reports (2015) Vol. 11 p261-

・ Labeling of engram

Nat. Neurosci., 2013 Vol. 16 p264-

5 mm 5 mm

KiKGR (G)CA1 KiKGR (R) Nucleus MergeKiKGR (G)

control

365 nm light

Before After 365 nmUnder nVista

・ For Ca2+ imagingThy1::G-CaMP7 Tg established byDrs Masaaki Sato and

Yasunori Hayashi

CaMKII KikGRLentivirus

ContextApre-exposure

Context AImmediate shock

1 day

ON-Dox

2 day

s

~1 hr

c-fos::tTA Tg

Memory test inNew Cage

Optical stimulation ofCA1 cell ensemble

Freezing↑

LVinjection

FreezingFreezingFreezingFreezing(FreezingFreezing(FreezingFreezing(Freezing

(FreezingFreezing

(FreezingFreezing(FreezingFreezing(FreezingFreezing

(

Freezing

(

FreezingFreezing

(

FreezingFreezing

(

Laser OFF Laser ONLaserON/ OFF

2 days

Habituation inNew Cage with bedding

Memory test inContext A

3-5 hr

ContextBpre-exposure

Context AImmediate shock

1 day

ON-Dox

2 day

s

~1 hr

c-fos::tTA Tg

Memory test inNew Cage

Optical stimulation ofCA1 cell ensemble

Freezing↑

LVinjection

FreezingFreezingFreezingFreezing(FreezingFreezing(FreezingFreezing(Freezing

(FreezingFreezing

(FreezingFreezing(FreezingFreezing(FreezingFreezing

(

Freezing

(

FreezingFreezing

(

FreezingFreezing

(

Laser OFF Laser ONLaserON/ OFF

2 days

Habituation inNew Cage with bedding

Memory test inContextB

3-5 hr

Contextpre-exposure IS Test

Context B

Context A

Context A

Context A

Context B

Context A

Unpaired

Paired

0

10

20

30

40

50

Free

zing

(%)

**

UnpairedPaired

1. Context pre-exposure facilitation effect on immediate shock deficit procedure (CPFE procedure)

3. Labeling efficacy of c-fos-Tet Tag system

Ohkawa et al., Cell Reports (2015) Vol. 11 p261-

** P < 0.002, Welch’s t test, two-tailed

1-2 min, P > 0.82, one-way ANOVA

3-4 min, P < 0.01, one-way ANOVA; * P < 0.05, ** P < 0.01, Scheffe’s post-hoc test

## P < 0.01, Paired t test, two-tailed P > 0.52, one-way ANOVA

100 μm

Test 1

Test 2

Test 1 Test 2

4. Sum of all correlations within the first 60s (NV#69)

N =10, * P < 0.05, Wilcoxon signed-ranks test (two-tailed)

3. Analysis of Correlation Matrices Repetition (NV#69)

1. Ca2+ signal of Engram and Non-Engram during learning

2. Wave form of Ca2+ event

④

③

KiKGR=Engram

①

KiKGR

④

3. Pattern #5 and 9 of Engram in Session A of NV#196 5. Matching score of Engram and Non-engramcells relative to session A

1. Schedule Ca2+ imaging duringcontext memory processing

2. Dot product between patterns of different sessions extracted by NMF analysis (NV#196)

4. Engram patterns appeared in learning session are significantly highly reactive in both of either NREM or REM sleep and retrieval session compared with Non-engram patterns

Engram Non-engram

Paired

Unpaired

2. Engram of a context information is actually formed in CA1

●We declare no Conflict of Interest (COI).

EEG

EMG

Non-REM

REM

N=6, * P < 0.05; ** P < 0.01, Paired t test (two-tailed)

EYFP / Paired ChR2 / UnpairedChR2 / Paired

50 μm

Engram cellsNon-engram cells

Sum

of O

verla

ps o

f Cor

rela

tions

for a

Giv

en R

efer

ence

tim

e

0

1

2

3

Reference Time [s]0 10 20 30 40 50 60

4

5

Time Pattern # Time

Synchronyǁ

Synchronous activity

(C)Patternsǁ

Cell ensemble

(B)

・ ・ ・Pattern #1Pattern #2Pattern #3

Neu

ron

#

Neuron #1 2 3 4

・ ・ ・・ ・ ・

1 2 34

Data (D)Non-negativeMatrixFactorization(NMF)

N =6, * P < 0.05; ** P<0.01; # P<0.05 between Engram and Non-engram, Paired t test (one-tailed)

Engr

amN

on-e

ngra

m

1

21

41

401

1

101

201

301

Cel

l num

ber

Cel

l num

ber

0

2

4

6

8

10

0 60 120 180 240 300 360Time [s]

(NV#69)

(NV#69)

Engram cells Non-engram cells

Ca2+

sig

nal

0 10 20 30 40 50 60Time [s]

0 10 20 30 40 50 60Time [s]

0

8

The number of patterns are determined by AICc.

(= AICc)The data matrix (D) is approximatedby two matrices of Patterns (B) andSynchrony (C).

0

10

20

30

40

1-2 min 3-4 min4Hz Light

EYFP / PairedChR2 / PairedChR2 / Unpaired

Free

zing

(%)

***##

50

0

5

10

15

EYFPPaired

ChR2Paired

ChR2Unpaired%

of E

YFP

or C

hR2+

cel

ls /

DA

PI

n.s.20

Engram cells

0

10

20

30

40

50

Com

pare

d Ti

me

[s]

Reference Time [s]0 10 20 30 40 50 60

60

0

0.02

0.04

0.06

-0.04

-0.02

0

3

1

Engramcells

Non-engramcells

*

Ratio

of s

um o

f all

corr

elat

ions

with

inth

e fir

st 6

0s 2

iB iCiC

iDiD

iD

iEiE

iEiE

iBiB iB

iA iA iA iA

iC iC

iD

11

6

1

1161

1581

0

0.2

0.4

0.6

0.8

1

11

6

1

16

11

6

1

16

11

6

1

13

7

1

19

11

6

1

16

11

6

1

16

13

7

1

19

13

7

1

19

13

7

1

19

13

7

1

19

13

7

1

19

13

7

1

19

13

7

1

19

iFiF

iFiF

iD

13

7

1

19

1581 1581 1581 1581

1161 16 1161 16 1161 16 1161 16

1161 1161

iB

iA

iC

1161 16 1161 16

iE1371 19

iF

Com

pare

d Se

ssio

n

Reference SessionCBA D E

C

B

A

D

E

F

F0

0.2

0.4

0.6

0.8

1

iB iCiC

iDiD

iD

iEiE

iE

iBiB iB

iA iA iA iA

iC iC

iD

65331 97

39

20

1

58

iE

39201 58

37191 55

i Fi F

iF

iB

iA

iC

iD

43

22

1

iF

37191 55

i F

iE39201 58

39

20

1

58

39

20

1

58

37

19

1

55

65331 97 65331 97 65331 97 65331 97

39201 58 39201 58 39201 58

43

22

1

43

22

1

43

22

1

43

22

1

39

20

1

58

39

20

1

58

39

20

1

58

37

19

1

55

37

19

1

55

39

20

1

58

39201 58 39201 58 39201 58

Pattern #5 in Session A of Animal ID#196 Session A

Pattern #9 in Session A of Animal ID#19615 30 45 60

Neuron ID#In

tens

ityIn

tens

ityIn

tens

ityTime [s]

Overlay of Pattern #5 and #9

0

1

2

0

1

2

0

1

2

0 70 140 210 350280

0 70 140 210 350280

0 70 140 210 350280

15 30 45 60

15 30 45 60

Rela

tive

Mag

nitu

de

01234

Rela

tive

Mag

nitu

de

01234

Rela

tive

Mag

nitu

de

01234

Mat

chin

g Sc

ore

(MS)

Rel

ativ

e to

Sess

ion

A

EngramNon-engram

SessionsCBA D E

00.20.40.60.8

1

F

#

n.s.

Learning(A)

REM(D)

Retrieval(E)

20 40 60 80

1000

Patterns of Learning (Session A) [%]

*

**

n.s.n.s.

Time

Learning

Replay Retrieval

Ø Contextual memory in the hippocampus is represented as distinct subsets of synchronous activity (defined by Ca2+ transients) that comprise several ensembles of engram cells.

Ø In contrast to non-engram cells, these ensembles maintain their activity not only during learning but also during post-learning sleep and retrieval sessions.

Non-engram cells

Reference Time [s]0 10 20 30 40 50 60

0

0.02

0.04

0.06

-0.04

-0.02

Engram cells

0

10

20

30

40

50

Com

pare

d Ti

me

[s]

Reference Time [s]0 10 20 30 40 50 60

60

1Dept Biochem, Grad Sch Med Pharm, Univ Toyama, Japan, 2CREST, JST, 3PRESTO, JST, 4RIKEN BSI, 5Fac Informatics, Kogakuin Univ, 6 WPI-IIIS, Univ Tsukuba, 7 Life Sci Res Cent, Univ Toyama, 8Grad Sch Sci Eng, Saitama Univ, 9Brain Body Sys Sci Inst, Saitama Univ, 10Grad Sch Med, Kyoto Univ, 11Tohoku Univ Grad Sch Med, 12Grad Sch Biomed Eng, Tohoku Univ

Com

pare

d Se

ssio

n

Reference SessionCBA D E

C

B

A

D

E

F

F

Com

pare

d Se

ssio

n

Reference SessionCBA D E

C

B

A

D

E

F

F0

0.2

0.4

0.6

0.8

1

iB iCiC

iDiD

iD

iEiE

iEiE

iBiB iB

iA iA iA iA

iC iC

iD

11

6

1

1161

1581

0

0.2

0.4

0.6

0.8

1

11

6

1

16

11

6

1

16

11

6

1

13

7

1

19

11

6

1

16

11

6

1

16

13

7

1

19

13

7

1

19

13

7

1

19

13

7

1

19

13

7

1

19

13

7

1

19

13

7

1

19

iFiF

iFiF

iD

13

7

1

19

1581 1581 1581 1581

1161 16 1161 16 1161 16 1161 16

1161 1161

iB

iA

iC

1161 16 1161 16

iE1371 19

iF

N =6, ** P < 0.01; *** P<0.001; # P<0.05 between Engram and Non-engram, Paired t test (one-tailed)

Poul

atio

n Ve

ctor

Dis

tanc

e (P

VD)

Rela

tive

to S

essi

on A

0

1

2

3

4

Diffe

renc

e in

PVD

bet

wee

nEn

gram

and

Non

-eng

ram

cel

ls

0

0.5

1

1.5

2

SessionsCBA D E F

SessionsCBA D E F

*** *** *** ** **

#EngramNon-engram

Engram cells original data maintain higher MS across sessions compared to shuffled data

Mat

chin

g Sc

ore

(MS)

Rel

ativ

e to

Sess

ion

A

SessionsCBA D E

0

0.2

0.4

0.6

0.8

1

F

Original data of engramShuffled data of engram

Diff

eren

ce in

MS

betw

een

Orig

inal

and

Shu

ffled

dat

a

SessionsCBA D E

0

0.2

0.4

0.6

0.8

1

F

* * *n.s. n.s.

-0.2

-0.4

N =6, * P < 0.05 between Engram original and shuffled data, Paired t test (one-tailed)

⑤

PVD is more consistent in Engram cells than Non-engram cells across sessions⑥

Free

zing

(%)

***

0

10

20

30

40

50

EYFPPaired

ChR2Paired

ChR2Unpaired

60