7
ARTICLES Compartmentalized dendritic plasticity and input feature storage in neurons Attila Losonczy 1 *, Judit K. Makara 1 * & Jeffrey C. Magee 1 Although information storage in the central nervous system is thought to be primarily mediated by various forms of synaptic plasticity, other mechanisms, such as modifications in membrane excitability, are available. Local dendritic spikes are nonlinear voltage events that are initiated within dendritic branches by spatially clustered and temporally synchronous synaptic input. That local spikes selectively respond only to appropriately correlated input allows them to function as input feature detectors and potentially as powerful information storage mechanisms. However, it is currently unknown whether any effective form of local dendritic spike plasticity exists. Here we show that the coupling between local dendritic spikes and the soma of rat hippocampal CA1 pyramidal neurons can be modified in a branch-specific manner through an N-methyl-D-aspartate receptor (NMDAR)-dependent regulation of dendritic Kv4.2 potassium channels. These data suggest that compartmentalized changes in branch excitability could store multiple complex features of synaptic input, such as their spatio-temporal correlation. We propose that this ‘branch strength potentiation’ represents a previously unknown form of information storage that is distinct from that produced by changes in synaptic efficacy both at the mechanistic level and in the type of information stored. A distinctive morphological feature of neurons is the large number of dendritic branches that compartmentalize their primary input regions. Generally, these branches contain voltage-dependent ion channels, the activation of which produce local membrane potential nonlinearities (Na 1 , Ca 21 or NMDA spikes) that markedly shape incoming synaptic input and action potential output 1–9 . These chan- nels are modifiable by physiologically relevant stimuli and by beha- viour 10,11 . Recent reports suggest that channel modifications can take place within restricted dendritic regions, raising the possibility of a compartmentalized regulation of local dendritic nonlinearities 12–14 . Such plasticity might function as a powerful mechanism for storing particular input features encoded in highly correlated patterns. Variable spike strength and propagation We used multi-site two-photon glutamate uncaging to deliver spatially clustered and temporally synchronous input patterns required to evoke local dendritic spikes in basal and radial oblique dendrites of rat hippocampal CA1 pyramidal neurons (Fig. 1a, b) 4 . These dendritic spikes produce a nonlinear depolarization at the soma consisting of a fast Na 1 -channel-dependent component and a slow NMDA-receptor component (Fig. 1a, b) 4 . We observed that the strength of the Na 1 spike (amplitude and rate of rise (dV/dt)) within a branch was rela- tively invariant between trials and independent of input configuration or strength (Supplementary Fig. 2a, b) 4 . There was, however, a remark- able variation of spike strength between branches (Fig. 1a, b). We proposed that differences in the propagation of dendritic spikes within branches might underlie this variability. Therefore, we examined the within-branch propagation by measuring Ca 21 transients associated with dendritic spikes at various distances from the input location (Fig. 1c–e). In most cases, spike-associated Ca 21 signals propagating towards the soma decreased markedly within 20–40 mm of the input site (fluorescence change (DF/F), from 109 6 5% to 24 6 2%, n 5 65 of 76 dendrites; Fig. 1c, e), whereas those propagating towards the sealed end attenuated much less over a similar distance (DF/F, from 111 6 11% to 100 6 8%, n 5 21; Fig. 1e and Supplementary Fig. 3a). Notably, dendritic spikes propagated towards the soma much more robustly in a minority of branches (DF/F, 104 6 12% to 92 6 6%, n 5 11; Fig. 1d, e). In these dendrites, spike-associated Ca 21 signals did not attenuate over the ,60 mm of forward propagation measured. As expected, the amplitude and dV/dt of the fast spike component was directly proportional to propagation strength (strong, 5.06 6 0.36 mV, 5.43 6 0.78 V s 21 , n 5 11; weak, 1.15 6 0.06 mV, 0.45 6 0.05 V s 21 , n 5 65, P , 0.001; Fig. 1f and Supplementary Fig. 3b), whereas there was no difference in local spike threshold measured at the soma (strong, 3.34 6 0.35 mV, n 5 7; weak, 3.34 6 0.24 mV, n 5 34, P 5 0.551) or in slow component amplitudes at threshold level (Supplementary Fig. 3c). These data indicate that variability in branch spike forward propagation produces heterogeneity in the coupling between various dendrites and the soma/axonal output region 15 . To determine whether there were indeed two distinct populations of branch strengths, we focused on the spike dV/dt as a measure of propagation strength 4,6 using data from a larger population that included branches without simultaneous Ca 21 recordings (n 5 479). Data were better fit by a double gaussian function with peaks that were an order of magnitude apart (Fig. 1g, first peak, 0.37 6 0.001 V s 21 ; second peak, 3.49 6 0.11 V s 21 ) than by a single gaussian function (P , 0.001, Shapiro–Wilk test). Thus, there are two distinct populations of dendrites with most (78%) exerting weak coupling strength (weak dendrites) and a minority (22%) with a ten- fold stronger coupling (strong dendrites). Action potential output depends on spike strength The amplitudes of weak dendritic spikes are quite small at the soma (spike amplitude, 1.52 6 0.05 mV; peak amplitude near threshold, 4.38 6 0.20 mV, n 5 188) with maximum depolarization at the slow NMDAR-dependent peak, even when two separate weak dendrites are simultaneously activated (Supplementary Fig. 3d, e). Because of this, weak dendritic spikes produce imprecisely timed action potential 1 Howard Hughes Medical Institute, Janelia Farm Research Campus, 19700 Helix Dr Ashburn, Virginia 20147, USA. *These authors contributed equally to this work. Vol 452 | 27 March 2008 | doi:10.1038/nature06725 436 Nature Publishing Group ©2008

Compartmentalized dendritic plasticity and input feature storage in neurons

Embed Size (px)

Citation preview

ARTICLES

Compartmentalized dendritic plasticityand input feature storage in neuronsAttila Losonczy1*, Judit K. Makara1* & Jeffrey C. Magee1

Although information storage in the central nervous system is thought to be primarily mediated by various forms of synapticplasticity, other mechanisms, such as modifications in membrane excitability, are available. Local dendritic spikes arenonlinear voltage events that are initiated within dendritic branches by spatially clustered and temporally synchronoussynaptic input. That local spikes selectively respond only to appropriately correlated input allows them to function as inputfeature detectors and potentially as powerful information storage mechanisms. However, it is currently unknown whetherany effective form of local dendritic spike plasticity exists. Here we show that the coupling between local dendritic spikes andthe soma of rat hippocampal CA1 pyramidal neurons can be modified in a branch-specific manner through anN-methyl-D-aspartate receptor (NMDAR)-dependent regulation of dendritic Kv4.2 potassium channels. These data suggestthat compartmentalized changes in branch excitability could store multiple complex features of synaptic input, such as theirspatio-temporal correlation. We propose that this ‘branch strength potentiation’ represents a previously unknown form ofinformation storage that is distinct from that produced by changes in synaptic efficacy both at the mechanistic level and inthe type of information stored.

A distinctive morphological feature of neurons is the large number ofdendritic branches that compartmentalize their primary inputregions. Generally, these branches contain voltage-dependent ionchannels, the activation of which produce local membrane potentialnonlinearities (Na1, Ca21 or NMDA spikes) that markedly shapeincoming synaptic input and action potential output1–9. These chan-nels are modifiable by physiologically relevant stimuli and by beha-viour10,11. Recent reports suggest that channel modifications can takeplace within restricted dendritic regions, raising the possibility of acompartmentalized regulation of local dendritic nonlinearities12–14.Such plasticity might function as a powerful mechanism for storingparticular input features encoded in highly correlated patterns.

Variable spike strength and propagation

We used multi-site two-photon glutamate uncaging to deliver spatiallyclustered and temporally synchronous input patterns required toevoke local dendritic spikes in basal and radial oblique dendrites ofrat hippocampal CA1 pyramidal neurons (Fig. 1a, b)4. These dendriticspikes produce a nonlinear depolarization at the soma consisting of afast Na1-channel-dependent component and a slow NMDA-receptorcomponent (Fig. 1a, b)4. We observed that the strength of the Na1

spike (amplitude and rate of rise (dV/dt)) within a branch was rela-tively invariant between trials and independent of input configurationor strength (Supplementary Fig. 2a, b)4. There was, however, a remark-able variation of spike strength between branches (Fig. 1a, b). Weproposed that differences in the propagation of dendritic spikes withinbranches might underlie this variability. Therefore, we examined thewithin-branch propagation by measuring Ca21 transients associatedwith dendritic spikes at various distances from the input location(Fig. 1c–e). In most cases, spike-associated Ca21 signals propagatingtowards the soma decreased markedly within 20–40mm of the inputsite (fluorescence change (DF/F), from 109 6 5% to 24 6 2%, n 5 65of 76 dendrites; Fig. 1c, e), whereas those propagating towards thesealed end attenuated much less over a similar distance (DF/F, from

111 6 11% to 100 6 8%, n 5 21; Fig. 1e and Supplementary Fig. 3a).Notably, dendritic spikes propagated towards the soma much morerobustly in a minority of branches (DF/F, 104 6 12% to 92 6 6%,n 5 11; Fig. 1d, e). In these dendrites, spike-associated Ca21 signalsdid not attenuate over the ,60mm of forward propagation measured.As expected, the amplitude and dV/dt of the fast spike component wasdirectly proportional to propagation strength (strong, 5.06 6 0.36 mV,5.43 6 0.78 V s21, n 5 11; weak, 1.15 6 0.06 mV, 0.45 6 0.05 V s21,n 5 65, P , 0.001; Fig. 1f and Supplementary Fig. 3b), whereas therewas no difference in local spike threshold measured at the soma(strong, 3.34 6 0.35 mV, n 5 7; weak, 3.34 6 0.24 mV, n 5 34,P 5 0.551) or in slow component amplitudes at threshold level(Supplementary Fig. 3c). These data indicate that variability in branchspike forward propagation produces heterogeneity in the couplingbetween various dendrites and the soma/axonal output region15.

To determine whether there were indeed two distinct populationsof branch strengths, we focused on the spike dV/dt as a measure ofpropagation strength4,6 using data from a larger population thatincluded branches without simultaneous Ca21 recordings(n 5 479). Data were better fit by a double gaussian function withpeaks that were an order of magnitude apart (Fig. 1g, first peak,0.37 6 0.001 V s21; second peak, 3.49 6 0.11 V s21) than by a singlegaussian function (P , 0.001, Shapiro–Wilk test). Thus, there are twodistinct populations of dendrites with most (78%) exerting weakcoupling strength (weak dendrites) and a minority (22%) with a ten-fold stronger coupling (strong dendrites).

Action potential output depends on spike strength

The amplitudes of weak dendritic spikes are quite small at the soma(spike amplitude, 1.52 6 0.05 mV; peak amplitude near threshold,4.38 6 0.20 mV, n 5 188) with maximum depolarization at the slowNMDAR-dependent peak, even when two separate weak dendritesare simultaneously activated (Supplementary Fig. 3d, e). Because ofthis, weak dendritic spikes produce imprecisely timed action potential

1Howard Hughes Medical Institute, Janelia Farm Research Campus, 19700 Helix Dr Ashburn, Virginia 20147, USA.*These authors contributed equally to this work.

Vol 452 | 27 March 2008 | doi:10.1038/nature06725

436Nature Publishing Group©2008

output with variable latency (latency and jitter, 11.64 6 1.28 ms and3.51 6 0.85 ms, respectively, n 5 24, Fig. 1h). In contrast, the fast spikecomponent of strong branches is usually large enough to evoke actionpotential output (spike, 4.95 6 0.26 mV; peak, 7.81 6 0.56 mV,n 5 35). The action potentials initiated by strong dendritic spikesrequired much fewer inputs (,55% of weak dendrite inputs, Supple-mentary Fig. 3f) and showed a latency and jitter an order of magnitudelower than that observed for weak dendritic spikes (latency, 1.94 6

0.19 ms; jitter, 0.13 6 0.04 ms, n 5 13; Fig. 1h). Thus, local spikes fromstrongly coupled dendrites are capable of evoking an invariant, preciseaction potential output for a much reduced input level.

Dendritic morphology and spike strength

We next examined whether the pattern of weak and strong spikesvaried with distance of the branch from the soma. Using the abovebimodal distribution and the dV/dt versus DF/F data (Fig. 1f) toestablish a criterion, we separated dendritic spikes into a strong group(dV/dt . 2 V s21) and a weak group (dV/dt , 2 V s21). Althoughboth weak and strong branches could be observed among basal andradial oblique dendrites originating within ,80mm of the soma, theproportion of strong spikes decreased for radial oblique dendritesbifurcating from the apical trunk .40mm from the soma (Fig. 1i).A similar pattern was observed during dendritic recordings. Becauseof this relationship, the remainder of our analysis was limited to thisperisomatic region (basal or radial oblique bifurcating from the trunkwithin ,40mm from the soma). The infrequent observation of strongNa1 spikes in distal oblique compartments does not, however, meanthat branch strength is invariant in this region, as distal branchstrength may be expressed through a different nonlinear mechanismthan the Na1 spike. We next plotted spike dV/dt versus distance of

input location along dendrites for this proximal population(Supplementary Fig. 4a). Data points for weak branches varied as afunction of distance (l 5 69mm), whereas spikes from strongbranches did not.

Saltatory-like spike propagation

Proximal dendrites are rarely single branches (,11% here), ratherthey divide further, forming complex dendritic families that consistof a primary branch, terminal dendrites and occasionally intermediatesegment(s). By systematically testing the proximal dendritic families(Fig. 2a and Supplementary Fig. 4b), we discovered that branch coup-ling strength was hierarchically structured such that the majority ofprimary branches were strong (75%, 88 out of 117 branches), whereasmuch fewer terminal branches showed strong spikes (18%, 35 out of194). To quantify this hierarchical relationship, we calculated thewithin-family spike ratio, defined as dV/dt of the higher order daugh-ter (D) branch divided by dV/dt of the lower order parent (P) branch(Fig. 2b and Supplementary Fig. 4c). The distribution of spike ratio infamilies with strong parent branches demonstrated that by 15–40mmdistal to the branch point, daughter branches were already eitherextremely weak (spike ratio, 0.18 6 0.01, n 5 90) or slightly strongerthan the parent branch to which they were connected (spike ratio,1.17 6 0.05, n 5 30, P , 0.001) (Fig. 2b and Supplementary Fig. 4c, d).Consistent with this, two nearly coincident but separable spike dV/dtcomponents were usually found in the total fast spike of strong daugh-ter branches (dV/dtS), with one small component corresponding tothe daughter branch itself (dV/dtD) and a second dV/dt componentapproximately the same size as the strong parent branch spike(dV/dtP; Fig. 2c). In several cases, two daughter branches bifurcatingfrom a common parent branch showed highly divergent spike ratios,

Wea

k

32

1

46 7

5

20 mV25 ms

ms

h

ms

12

8

4

0

Str

ong

5

4

3

2

1

0

Str

ong

Wea

k

a

2 V s–1

dV/dt

2 m

V

FastSlow

c

Cou

nt

d-spike dV/dt at soma (V s–1)

AP output timing

Latency Jitter

0

40

80

120

160

0 4 6 8 10 12142

1st peak = 0.37 ± 0.001 V s–1

g

200 ms

Input

9 µm

22 µm

37 µm

dV/dt

2 mV

5 ms

1 V

s–1

Vm

gluEPSP

4 ms

2 V s–1

0.5 V s–1dV/dt

gluEPSP

4 msdV/dt

Spike Spike

Fast

Slow

b

2nd peak = 3.49 ± 0.11 V s–1

Input

16 µm

36 µm200 ms

5 ms

2 mV

2 V

s–1

dV/dtVm

eInput site

Distance from input site (µm)

∆F/F

(%)

604020040 20

To somaTo end

0

40

80

120

160

60 100500

V s

–1

∆F/F (%)

f

0

4

8

12

150

100%∆F/F

100%∆F/F

Str

ong

Wea

k

05

101520

Distance (µm from soma)

Basal Apical

dV

/dt

(V s

–1)

WeakStrong

i

20 0

024

0 20 40 60 80 100 120 140

0 20 40 60 80 100 120 140

Dendritic recording

Somatic recording

soma

soma

20 µm

∆F/F ∆F/F

Strong branch Weak branch

d

5 µm

5 µm

20 µm

dV

/dt

(V s

–1)

2 mV

5 µm

Figure 1 | Two populations of dendritic spikestrengths. a, Left, image stack. The input regionis expanded with numbered uncaging locationsindicated. Right: membrane potential (Vm, top)and dV/dt (bottom) traces of a strong dendriticspike. Black traces, subthreshold gluEPSPs; redtraces, dendritic spike. b, Weak branch. c, d, Spike-associated Ca21 signals in weakly (c) and strongly(d) forward propagating dendrites. Distancesfrom input site (dots and orange line) areindicated. e, Forward propagating Ca21 signalsversus distance from input site in weak (bluedots) and strong (red dots) dendrites and forspikes propagating towards the sealed end inweak (black) and strong (red triangles) branches.f, Spike dV/dt versus spike Ca21 signals (at20–40mm proximal to input). g, Distribution ofspike dV/dt fit by a double gaussian function (d-spike, dendritic spike). h, Action potential (AP)output evoked by weak (upper left) or strong(upper right) dendritic spikes (four trials each).Action potential latency (lower left) and jitter(lower right) for weak and strong branches isshown. i, Spike dV/dt versus dendrite distancefrom soma in somatic and dendritic recordings.

NATURE | Vol 452 | 27 March 2008 ARTICLES

437Nature Publishing Group©2008

with one being strong and the other weak (Fig. 2d, e and Supplemen-tary Fig. 4e). In these cases, the dV/dtD component from the stronglycoupled daughter branch was nearly twice that of the weakly coupledsister branch (strong, 0.89 6 0.08 V s21; weak, 0.51 6 0.05 V s21,n 5 9, P , 0.001; Fig. 2f). Ca21 imaging confirmed that only strongdaughter branches fired local spikes capable of propagating throughthe branch point into the strong proximal compartment (Fig. 2d, e, g).The strength of the daughter branch spike was independent of anyparticular synaptic property of the initiating input (SupplementaryFig. 2b, c). Thus, local spikes within daughter branches act as variable,all-or-none triggers of stronger parent branches, with only a minorityof daughter branches possessing adequate excitability to supporteffective propagation. These data indicate that basal and proximalradial oblique dendrites form a distinct, extensively branching periso-matic integration zone with a large number of terminal compartments(,70% of total)16 whose internal properties can be independentlyregulated to shape the impact of their synaptic input (which originatesprimarily from subregions CA3a and b)17,18.

Branch strength potentiation

The variable pattern of branch coupling strengths observed withindendritic arborizations might reflect a natural regulatory processgoverning the excitability of local branches. If this is the case, weshould be able to alter experimentally the coupling strength of selec-tive branches. We tested this hypothesis on weak daughter branches(15–40mm from branch point; dV/dtD, 0.72 6 0.05 V s21, n 5 36) thatoriginated from strong parent dendrites (dV/dtP, 5.31 6 0.59 V s21,n 5 36). We first tested whether input patterns designed to mimicthose found during exploratory sharp waves (SPWs)19 could inducea change in local spike propagation. More than 40 min of repeatedlocal spike initiation by itself caused only a minor change in thestrength of the weak daughter branch (dV/dtD, to 111 6 3%,

P , 0.01, n 5 7; Fig. 3A, D, E) and did not result in recruitment ofthe strong branch spike. However, when we combined the samerepeated local spike initiation with a transient application of the cho-linergic agonist carbachol (CCh, 5mM, 20 min), to mimic the neuro-modulatory state of exploratory behaviour20, the stimulus induced arobust, slowly developing enhancement of weak branch spike strengththat continued even after CCh washout (dV/dtD in CCh, to 158 6 9%of control, P , 0.001, n 5 13; Fig. 3B, D, E). In most cases the daughterspikes became effective enough to evoke the strong parent branchspike, leading to a large step increase in dV/dt and in spike ratio (9out of 13 experiments, dV/dtS, to 548 6 129%; spike ratio, from0.27 6 0.05 to 1.09 6 0.07; Fig. 3B, F–H). Dendritic spike amplitudewas also significantly enhanced (to 245 6 35%, n 5 13, P , 0.01). Wealso used a more general associative theta (h)-pairing protocol thatinvolved local branch spike initiation paired with two or three back-propagating action potentials (Fig. 3C). This induction protocol alsosuccessfully enhanced weak branch spike strength (dV/dtD, to188 6 13%, n 5 16, P , 0.001; Fig. 3C–E), often to a level where itrecruited the parent branch spike (11 out of 16 experiments, dV/dtS,to 1,198 6 294%; spike ratio, from 0.16 6 0.03 to 1.09 6 0.06, n 5 11;amplitude, to 409 6 72%, n 5 16; Fig. 3C, F–H). We named this formof dendritic plasticity ‘branch strength potentiation’ (BSP).

BSP was found to be branch specific, as both induction protocolsenhanced only those weak branches where local spikes were repeat-edly initiated, while leaving other unstimulated branches unaffected(dV/dtD in CCh: stimulated, to 148 6 13%, n 5 5, P , 0.05; unsti-mulated, to 103 6 13%, n 5 5, P 5 0.649; dV/dtD after pairing: sti-mulated, to 173 6 12%, n 5 4, P , 0.01; unstimulated, to 99 6 5%,n 5 7, P 5 0.839; Fig. 4a–c), including the strong parent branches towhich they were connected (dV/dtP in CCh, to 116 6 39%, n 5 4,P 5 0.446; h-pairing, to 111 6 11%, n 5 4, P 5 0.546). Also, bothlocal spikes and back-propagating action potentials were required

a

f

Cum

ulat

ive

pro

bab

ility

dV/dt (V s–1)

0

1.0

0.6

0.8

0.4

0.2

0 30108642 40

Primary

cb

Distance(µm from branch point)

2.0

1.6

1.2

0.8

0

Sp

ike

ratio

0 20 40

0.4

5 m

V

5 ms

2 V

s–1

Vm

dV/dt

d

20 µm

100%∆F/F

100 ms

Input

Propagated

P

e

∆F/F

D-1 D-2

Terminal

Vm

dV/dt

Input

Propagated

∆F/F

local∆F/FVm dV/dt 100%

∆F/F100 ms

5 mV

5 ms 1 V s

–1

1 V s

–1

5 ms

5 m

V

0 µm1 100%

∆F/F100 ms

( )Strong P Weak D

Strong D ( )

Σ

dV/dt

2 ms

2 mV

1 V

s–1

Subthreshold

P

D

Vm

P′

2 ms

0 20 4020Distance

(µm from branch point)

∆F/F

(%)

0

25

50

75

100

125 Local

Propagatedto soma

0

0.4

0.8

1.2

1.6

V s

–1

D branch pairs

Strong Weak

dV/dtD

g

5 mV

4 V s–1

Primary Terminal

5 ms

1 V s–1

2 mV

P D

( )Strong P Weak D

Strong D( )

dV/dt

Vm

{

{

Figure 2 | Relationship of dendritic morphology and branch strength.a, Membrane potential (Vm; top) and dV/dt (bottom) traces from a strongprimary and a connected weak terminal branch. Right: the cumulativeprobability of spike dV/dt for all primary and terminal branches is shown.Red, strong; blue, weak. b, Spike ratio for daughter (D) branches originatingfrom strong parent (P) versus distance of the input location on the daughter.c, Vm and dV/dt from a strong parent (left) and connected strong daughter(right) for suprathreshold (red) and just subthreshold (grey) input. ParentdV/dt has one component (P) whereas the strong daughter (S) has two

components (D and P). d, e, Basal arbor with marked region expanded ine showing outline of a parent (P) and two connected daughters (D-1 andD-2). Ovals indicate uncaging locations. Blue and red lines: line-scanlocations for spike-associated Ca21 signals propagating from daughters intothe parent branch. Orange line: input site line scans. Input onto D-1 branchinitiated a strong spike (left). Input onto D-2 branch initiated a weak spike(right). Only the spike evoked in D-1 produced a Ca21 signal in P. f, dV/dtD

of different sister branches with a common strong parent. g, Spike-associated Ca21 signals versus distance relative to branch point.

ARTICLES NATURE | Vol 452 | 27 March 2008

438Nature Publishing Group©2008

for potentiation induced by the h-pairing protocol, as the same repe-titive suprathreshold synchronous input alone or back-propagatingaction potentials paired with only a locally subthreshold (asynchron-ous) input pattern did not change spike strength (Fig. 4g and Supple-mentary Fig. 5a–c). This indicates that BSP is an associative form ofplasticity. BSP was not, however, dependent on the activation of aspecific set of dendritic spines, as the increase in spike strength couldstill be detected if naive spines close to the initiation site were activatedonce the potentiation fully developed (Supplementary Fig. 5d). Theeffect of CCh was prevented by atropine (5mM), indicating that mus-carinic acetylcholine receptors (mAChRs) were involved (dV/dtD, to102 6 16%, n 5 4, P 5 0.679; Fig. 4g), and both forms of plasticity wereblocked by the NMDAR antagonist D,L-2-amino-5-phoshonovalericacid (AP5, 50mM, dV/dtD for CCh in AP5, to 105 6 6%, n 5 4,P 5 0.398, Fig. 4d, f–g; dV/dtD for h-pairing in AP5, to 103 6 6,n 5 6, P 5 0.577, Fig. 4e–g). In summary, repetitive local spiking inweak branches when associated with either mAChR activation or back-propagating action potentials triggers an NMDAR-dependent signal-ling pathway that leads to a gradual enhancement of local spikepropagation that eventually becomes effective enough to activate themore powerful proximal dendrites of the dendritic family.

Voltage-gated ion channel mechanisms

One candidate mechanism for BSP is a long lasting modificationof A-type K1 currents that shape the propagation of bothback-propagating action potentials12,21 and dendritic spikes4, and

that are regulated both by long-term potentiation (LTP)-inducingprotocols12,14 and mAChR activation22. A-type K1 currents can berelatively specifically blocked by low concentrations of Ba21 (ref. 21).We therefore tested the effect of Ba21 (200–250 mM for 10–25 min)on spike propagation in weak daughter branches originating fromstrong parent dendrites (Fig. 5a). Ba21 caused a clear increase indV/dtD of the weak spike (to 156 6 9%, n 5 6, P , 0.001; Fig. 5a, c),and the strong parent branch spike was activated in all cases (dV/dtS, to931 6 102%, n 5 6; spike ratio: control, 0.13 6 0.02; Ba21, 1.16 6 0.06;Fig. 5a, d). Both effects reversed on Ba21 washout. This result indicatesthat A-type K1 currents may have an important role in suppressing thepropagation of dendritic spikes in weak branches.

In CA1 pyramidal neurons, A-type K1 currents are mediated mainlyby Kv4.2 K1 channel subunits23. We next performed experiments ontransgenic mice in which the Kv4.2 gene (also called Kcnd2) had beensilenced (Kv4.22/2)23,24. Consistent with a role of Kv4.2-mediated cur-rents in limiting spike propagation, dV/dtD in weak terminal brancheswas significantly higher in Kv4.22/2 mice (0.71 6 0.66 V s21, n 5 44)than in wild-type littermates (0.39 6 0.26 V s21, n 5 26, P , 0.001;Supplementary Fig. 6a, c), whereas the strong populations did notdiffer (Supplementary Fig. 6a, c). Branch strength heterogeneity andthe characteristic hierarchical organization of coupling strength inproximal dendritic families were, however, similar between Kv4.22/2

and wild-type mice (Supplementary Fig. 6a–d), suggesting thatcompensatory mechanisms partially counterbalanced the loss ofKv4.2 (refs 23, 24).

20 µm

P

D

200

160

120

Time (min)40200 3010–10 50

Per

cent

age

of c

ontr

ol

D E

θ-pairingCCh with d-spikesd-spikes

2.0

1.5

1.0

0.5

CCh with d-spikes

Ctrl d-spikes

V

s–1

d-spikesdV/dtD dV/dtD

G

b

V s

–1

θ

Time (min)

00.4

0.81.21.6

4

8

dV/dtD

dV/dtΣdV/dtPP

dV/dtP

2 m

V

2 V s

–1

pre-θ post-θ 20 min

dV/dtΣ

dV/dtD dV/dtDdV/dtD

5 ms

θ-pairing

20 m

V

50 ms

1 D branch spike+ 2–3 bAPs

~10”

15X totalSomatic current injection

Synch. uncaging on D branch }

Capost-θ 49 min

θ-pairing

carbachol with d-spikes

–20 20 40 600

0. 4

0.8

1.2

1.6

34

0

V s

–1

Time (min)

dV/dtP

CCh

dV/dtD

dV/dtΣ

Ba

5 mV

5 ms

dV/dtD 1 V

s–1

CCh 20 min

dV/dtΣ

CCh 58 min

dV/dtD dV/dtD

dV/dtP

P

2 V

s–1

bpre-CCh

d-spikes

Time (min)

0

0.4

0.8

1.2

1.66

8

V s

–1

–20 20 40 600

dV/dtD

dV/dtD

5 V

s–1

5 mV

P

dV/dtP

–1 min

2 mV

1 V

s–1

52 min

dV/dtD

5 ms

dV/dtD

Aa b

Ctrl CCh Ctrl θ

F

Time (min)

1.6

1.2

0.8

0.4

040200 30 5010–10

Sp

ike

ratio

CCh with d-spikesθ-pairing

0

2

4

6

815

V

s–1

dV/dtΣdV/dtDdV/dtP (V s–1)

CCh with d-spikes

12840

0.8

0.4

0

1.2

1.6

Hθ-pairing

Sp

ike

ratio

CChdV/dtΣdV/dtD

Ctrl Ctrl θ

****

D

DD

Vm

Vm

Vm

2 ms 2 ms

2 ms–20 20 40 600

c

2 V

s–1

CCh

θ

CCh

θ

CCh with d-spikesθ-pairing

Figure 3 | Branch strength potentiation. A, a, b, Vm and dV/dt from parent(P) and connected daughter (D) at indicated times after the start of localspike initiation protocol (a). The entire time course of the experiment(b). The dashed line indicates the parent branch dV/dt. B, a, b, Vm and dV/dtfrom a strong parent (P) and connected daughter before (pre-CCh), 20 minafter (CCh 20 min) and 58 min (CCh 58 min, red) after starting carbachol(5 mM) application (a). The entire time course of the experiment is shown(b). C, a–c, Input locations on parent (red) and daughter (blue) branches in afamily (a). b, Left: Vm (top) and schematic (bottom) showing h-pairingprotocol applied to the daughter branch. bAPs, back-propagating action

potentials. Middle: Vm and dV/dt of parent. Right: Vm and dV/dt of daughterbefore (pre-h), 20 min after (post-h 20 min) and 49 min (post-h 49 min, red)after h-pairing. c, Entire time course of experiment. D, Summary of baseline-normalized dV/dtD versus time for d-spikes alone, with carbachol andh-pairing. E, dV/dtD before and after spikes alone, CCh and h-pairing.F, Spike ratio versus time for individual experiments with CCh andh-pairing. Note that the parent branch spike was recruited at different timepoints. G, Maximal daughter branch spike dV/dt in control (dV/dtD) andafter CCh or h-pairing (dV/dtS) when parent branch was recruited. H, Spikeratio after CCh or h-pairing versus parent branch dV/dt.

NATURE | Vol 452 | 27 March 2008 ARTICLES

439Nature Publishing Group©2008

We next determined the effect of Ba21 on weak daughter branchesoriginating from strong parent dendrites in Kv4.22/2 and wild-typemice. Ba21 enhanced branch spike strength in wild-type mice similarlyas observed in rats (dV/dtD, to 205 6 19%, n 5 4, P , 0.05; dV/dtS, to2,550 6 1170%, n 5 3; Fig. 5c, d and Supplementary Fig. 6e), whereas ithad no significant effect on branches from Kv4.22/2 neurons (dV/dtD,to 111 6 15%, n 5 8, P 5 0.839; Fig. 5b–d), confirming the channelspecificity of low Ba21 concentrations. Finally, we tested the ability ofthe CCh and h-pairing protocols to induce BSP in Kv4.22/2 and wild-type mice. Notably, in Kv4.22/2 mice, neither induction protocolinfluenced dV/dt in weak daughter branches (CCh, to 109 6 8%,n 5 8, P 5 0.164, Fig. 5e, g; h-pairing, to 93 6 4%, n 5 7, P 5 0.572,Fig. 5f, h), whereas BSP was successfully induced in wild-type cellsboth by CCh (dV/dtD, to 190 6 24%, n 5 7, P , 0.01; Fig. 5g andSupplementary Fig. 6f) and h-pairing (dV/dtD, to 314 6 47%, n 5 6,P , 0.01; Fig. 5h and Supplementary Fig. 6g). These results stronglyimplicate downregulation of A-type K1 currents, mediated by Kv4.2subunits, as the underlying mechanism of BSP.

Discussion

In summary, branch coupling strength is not a static feature, as theassociative pairing of a highly synchronous input pattern with actionpotential output or mAChR activation leads to a long-term elevation ofcoupling strength of only the stimulated branch. Such a selectiveenhancement could be used to store the occurrence of this uniquelycorrelated input, as the subsequent arrival of only this specific spatio-temporal input pattern would evoke a precise, invariant form of action

Ctrl Post

4 V

s–1

Pθ-pairing + AP5

0.5 V s

–1

D post-θ 48 min

5 ms

epre-θ

240

200

160

120

8040200 30 5010–10

Time (min)

AP5

f

Per

cent

age

of c

trl

2.0

1.5

1.0

0.5

V s

–1

dV/dtDCCh with d-spikesθ-pairing

0

100

200

300

d-sp

ikes

Per

cent

age

of c

trl

g

+d-spikes

Asynch.

θ

+bAPs

1,200800

bdV/dtD

Stimulated branch

pre-θ

post-θ25 min

20µm

Unstimulated branch

5 ms

1 V s

–1

Ctrl 5min

Ctrl55 min

Unstimulated branch

2 ms

1 V s

–1

dV/dtD

pre-CCh CCh 66 min

Stimulated branch

pre-CCh CCh 57 min

a

20µm

dV/dtΣ

dV/dtD2,0001,000

200

100

0

c

d-spikes+ -

CCh θ

d-spikes+ -

θ

CCh

4 V s

–1

PCCh with d-spikes + AP5

pre-CCh

0.5 V

s–1

D CCh 51 min

5 ms

d

AP

5

CCh

Atrop

ine

* * ** ** *** *

D-2 D-1

D-1D-2

Per

cent

age

of c

trl dV/dtΣ

dV/dtD

AP

5

+d-spikes

AP5CCh

θ

Figure 4 | Specificity and associativity of branch strength potentiation.a, Stack image (top left) and outline of two terminal branches (bottom left,D-1 and D-2). Right: dV/dt from D-1 that received repeated local spikes inCCh (stimulated branch). D-2 received only CCh application (unstimulatedbranch). b, Similar to a but with h-pairing. c, Summary of branch selectivity.d, e, NMDAR antagonist blocked the effect of both CCh (d) and h-pairing(e) on BSP. f, Summary of baseline-normalized spike dV/dt versus time forCCh and h-pairing (left) and summary of dV/dtD before and after CCh orh-pairing in AP5 (right). g, Summary of the effect of different stimulationforms on daughter branch spike strength.

D P

a

c

2 V s

–1

1.6

1.2

0.8

0.4

0

Ctrl Ba2+ Wash

Rat +/+

V

s–1

dV/dtΣ

θ-pairingfe

0.8

0.4

0

20151050dV/dtP (V s–1)

1.2

Rat +/+ –/–

–/–

Ctrl Ba2+ Wash

Rat

b

20µm

P

D

Kv4.2–/–

dV/dtD

DControl

Ba2+

Wash

2 mV

2 ms

1 V

s–1

Control WashD

2 ms

0.5 V s

–1

2 mV

P

dV/dtP

2 V s

–1

P

dV/dtP

PD

20µm

d

Kv4.2 –/– Kv4.2 –/–

P

2 V s

–1

4 V s

–1

5 m

V

5 m

V

2 ms 4 msS

pik

e ra

tio (B

a2+)

1 V

s–1

1 V s

–1

CCh with d-spikes

Dpre-θ post-θ

50 minpre-CCh CCh 54 min

2 V

s–1

2 ms

dV/dtD

1 V

s–1dV/dtD

2 ms

Ba2+

dV/dtD

h

Ctrl θCtrl θ

V

s–1

2

1

0

–/– +/+

dV/dtD

Ctrl CCh Ctrl CCh

1.0

0.5

0

dV/dtDg

Vm

Vm Vm

Vm

2 ms 4 ms

dV/dtD

V

s–1

–/– +/+

dV/dtP dV/dtD

dV/dtP dV/dtD

Figure 5 | BSP is mediated by reduced A-type K1 channel function. a, Inrat, image showing input locations on strong parent (P) and weak daughter(D). Right: Vm (top) and dV/dt (bottom) from parent (P) and daughter (D)before (control), during (Ba21) and after (wash) Ba21 (200–250mM)application. b, Same as a but in Kv4.22/2 mice. c, Daughter spike dV/dtD

before, during and after Ba21 application in rat, Kv4.22/2 mice and wild-type littermate mice. d, Plot of spike ratio for Ba21 application versus parentdV/dt. e, Carbachol has no effect on branch strength in Kv4.22/2 mice.Traces from a strong parent (P) and connecting weak daughter (D) branchbefore (pre-CCh) and 54 min after (post-CCh 54 min) starting CChapplication. f, Same as for e but for h-pairing. g, h, Summary of D branchspike dV/dtD before and after CCh (g) or h-pairing (h) in Kv4.22/2 and wild-type mice.

ARTICLES NATURE | Vol 452 | 27 March 2008

440Nature Publishing Group©2008

potential output. BSP might, therefore, be considered input patternspecific, particularly because it will not affect unitary excitatory post-synaptic potential (EPSP) amplitudes4. Furthermore, BSP could inter-act with synaptic plasticity, as LTP of synaptic inputs onto a particularbranch25,26 would increase the probability of future local spike initiationand therefore BSP. The resulting compartmentalized downregulationof Kv4.2 function will also increase action potential back-propagationinto that branch, promoting LTP maintenance there12,23. Such a positivefeedback could elevate the retention time of stored information27.

Although we focused primarily on the regulation of local Na1

spike propagation within perisomatic dendrites of CA1 pyramidalneurons, other dendritic nonlinearities (for example, Ca21/NMDAplateaus) exist both in more distal regions of these cells as well as indendrites of other neurons11. Compartmentalized alterations in ionchannel function could also affect the strength of these dendriticspikes. Because of the ubiquity of modifiable active dendritic com-partments, BSP is potentially a general feature of neuronal informa-tion processing and storage. In the hippocampus, BSP may beparticularly applicable for the compressed sequence replay occurringduring slow-wave sleep (SWS)28–30. Repeated synchronous activationof a CA3 cell ensemble during exploratory SPWs19 could induce BSP,storing the specific feature represented by these neurons in enhancedbranch coupling strength (Supplementary Fig. 1). Subsequent syn-chronous ensemble activation during SWS would evoke strongdendritic spikes and the associated precise action potential outputcharacteristic of SWS. Consequently, the heterogeneous distributionof branch strengths observed in CA1 neurons may be an imprint ofalready stored sequence items onto the arborization.

METHODS SUMMARYAcute hippocampal slices were prepared from 8–12-week-old rats and mice

(Kv4.22/2 or wild-type littermates). Current-clamp whole-cell recordings were

performed at 33–36 uC from soma or proximal apical trunks of hippocampal

CA1 pyramidal neurons filled with Oregon green BAPTA-1 or Alexa 488. Two-

photon laser-scanning microscopy and two-photon glutamate uncaging wereperformed using a dual galvanometer-based scanning system with two

Ti:sapphire lasers4. For glutamate uncaging, MNI-glutamate was uncaged at

5–20 visually identified spines (0.2-ms exposure times, with 0.1–0.2-ms inter-

stimulus intervals between exposures at different spines). Repeated local spiking

protocols were given as trials of 3–10 local spikes every ,5–30 s with ,5-min

inter-trial intervals. In consecutive trials of an experiment, different sets of

spines were usually activated, by randomly picking the available spines within a

,10–20mm region of a dendritic segment in the same focal plane. The h-pairing

protocol consisted of locally suprathreshold uncaging paired with back-

propagating action potentials. That is, fifteen trains (five trains at 0.01 Hz

repeated three times with a 30-s interval) of two bursts (repeated at 5 Hz), with

each burst composed of a local spike (synchronous uncaging on a cluster of 10–15

spines with a 0.1-ms interval) paired with two–three back-propagating action

potentials at 100 Hz. The back-propagating action potentials were timed so that

the peaks of the action potentials were coincident with the peak of the uncaging

evoked EPSP (gluEPSP), as measured at the soma. Different spines were used for

the h-pairing induction and for testing the spikes before and after induction.

During h-pairing induction, different spines were usually used between consec-utive sets of trains. Distance measurements and anatomical evaluation were per-

formed on stacked images collected at the end of recordings. All data are given in

mean 6 s.e.m. In all figures, symbols with error bars indicate mean 6 s.e.m.;

asterisk, P , 0.05; double asterisk, P , 0.01; triple asterisk, P , 0.001.

Full Methods and any associated references are available in the online version ofthe paper at www.nature.com/nature.

Received 23 October 2007; accepted 24 January 2008.

1. Ariav, G., Polsky, A. & Schiller, J. Submillisecond precision of the input-outputtransformation function mediated by fast sodium dendritic spikes in basaldendrites of CA1 pyramidal neurons. J. Neurosci. 23, 7750–7758 (2003).

2. Larkum, M. E., Zhu, J. J. & Sakmann, B. Dendritic mechanisms underlying thecoupling of the dendritic with the axonal action potential initiation zone of adultrat layer 5 pyramidal neurons. J. Physiol. (Lond.) 533, 447–466 (2001).

3. Polsky, A., Mel, B. W. & Schiller, J. Computational subunits in thin dendrites ofpyramidal cells. Nature Neurosci. 7, 621–627 (2004).

4. Losonczy, A. & Magee, J. C. Integrative properties of radial oblique dendrites inhippocampal CA1 pyramidal neurons. Neuron 50, 291–307 (2006).

5. Gasparini, S., Migliore, M. & Magee, J. C. On the initiation and propagation ofdendritic spikes in CA1 pyramidal neurons. J. Neurosci. 24, 11046–11056 (2004).

6. Gasparini, S. & Magge, J. C. State-dependent dendritic computation inhippocampal CA1 pyramidal neurons. J. Neurosci. 26, 2088–2100 (2006).

7. Poirazi, P., Brannon, T. & Mel, B. W. Pyramidal neuron as two-layer neuralnetwork. Neuron 37, 989–999 (2003).

8. Poirazi, P. & Mel, B. W. Impact of active dendrites and structural plasticity on thememory capacity of neural tissue. Neuron 29, 779–796 (2001).

9. London, M. & Hausser, M. Dendritic computation. Annu. Rev. Neurosci. 28,503–532 (2005).

10. Zhang, W. & Linden, D. J. The other side of the engram: experience-driven changesin neuronal intrinsic excitability. Nature Rev. Neurosci. 4, 885–900 (2003).

11. Magee, J. C. & Johnston, D. Plasticity of dendritic function. Curr. Opin. Neurobiol.15, 334–342 (2006).

12. Frick, A., Magee, J. C. & Johnston, D. LTP is accompanied by an enhanced localexcitability of pyramidal neuron dendrites. Nature Neurosci. 7, 126–135 (2004).

13. Yasuda, R. et al. Supersensitive Ras activation in dendrites and spines revealed bytwo-photon fluorescent lifetime imaging. Nature Neurosci. 9, 283–291 (2006).

14. Kim, J., Jung, S. C., Clemens, C. M., Petralia, R. S. & Hoffman, D. A. Regulation ofdendritic excitability by activity-dependent trafficking of the A-type K1 channelsubunit Kv4.2 in hippocampal neurons. Neuron 54, 933–947 (2007).

15. Golding, N. L. & Spruston, N. Dendritic sodium spikes are variable triggers ofaxonal action potentials in hippocampal CA1 pyramidal neurons. Neuron 21,1189–1200 (1998).

16. Bannister, N. J. & Larkmann, A. U. Dendritic morphology of CA1 pyramidalneurones from the rat hippocampus: I. Branching patterns. J. Comp. Neurol. 360,150–160 (1995).

17. Ishizuka, N., Weber, J. & Amaral, D. Organization of intrahippocampal projectionsorginating from CA3 pyramidal cells of the rat. J. Comp. Neurol. 295, 580–623 (1990).

18. Csicsvari, J., Hirase, H., Czurko, A., Mamiya, A. & Buzsaki, G. Ensemble patterns ofhippocampal CA3–CA1 neurons during sharp wave-associated populationevents. Neuron 28, 585–594 (2000).

19. O’Neill, J., Senior, T. & Csicsvari, J. Place-selective firing of CA1 pyramidal cellsduring sharp wave/ripple network patterns in exploratory behavior. Neuron 49,143–156 (2006).

20. Hasselmo, M. E. & Giocomo, L. M. Cholinergic modulation of cortical function.J. Mol. Neurosci. 30, 133–135 (2006).

21. Gasparini, S., Losonczy, A., Chen, X., Johnston, D. & Magee, J. C. Associativepairing enhances action potential back-propagation in radial oblique branches ofCA1 pyramidal neurons. J. Physiol. (Lond.) 580, 787–800 (2007).

22. Hoffman, D. A. & Johnston, D. Neuromodulation of dendritic action potentials.J. Neurophysiol. 81, 408–411 (1999).

23. Chen, X. et al. Deletion of Kv4.2 gene eliminates dendritic A-type K1 current andenhances induction of long-term potentiation in hippocampal CA1 pyramidalneurons. J. Neurosci. 26, 12143–12151 (2006).

24. Guo, W. et al. Targeted deletion of Kv4.2 eliminates Ito,f and results in electricaland molecular remodeling, with no evidence of ventricular hypertrophy ormyocardial dysfunction. Circ. Res. 97, 1342–1350 (2005).

25. Govindarajan, A., Kelleher, R. J. & Tonegawa, S. Clustered plasticity model of long-term memory engrams. Nature Rev. Neurosci. 7, 575–583 (2006).

26. Harvey, C. D. & Svoboda, K. Spatially clustered and dynamic synaptic learningrules in pyramidal neuron dendrites. Nature 450, 1195–1200 (2007).

27. Fusi, S., Drew, P. J. & Abbott, L. F. Cascade models of synaptically storedmemories. Neuron 45, 599–611 (2005).

28. Wilson, M. A. & McNaughton, B. L. Reactivation of hippocampal ensemblememories during sleep. Science 265, 676–679 (1994).

29. Nadasdy, Z., Hirase, H., Czurko, A., Csicsvari, J. & Buzsaki, G. Replay and timecompression of recurring spike sequences in the hippocampus. J. Neurosci. 19,9497–9507 (1999).

30. Lee, A. K. & Wilson, M. A. Memory of sequential experience in the hippocampusduring slow wave sleep. Neuron 36, 1183–1194 (2002).

Supplementary Information is linked to the online version of the paper atwww.nature.com/nature.

Acknowledgements We thank G. Buzsaki, D. Johnston, J. Lisman and Z. Nusser fortheir comments on a previous version of the manuscript. We thank D. Johnston forthe Kv.4.2 mice and B. K. Andrasfalvy for valuable discussions and help withexperiments using Kv4.2 mice.

Author Contributions A.L. and J.K.M. performed and analysed the experiments,A.L., J.K.M. and J.C.M. designed the experiments, and all authors contributed to themanuscript preparation.

Author Information Reprints and permissions information is available atwww.nature.com/reprints. Correspondence and requests for materials should beaddressed to A.L. ([email protected]) or J.K.M.([email protected]).

NATURE | Vol 452 | 27 March 2008 ARTICLES

441Nature Publishing Group©2008

METHODSHippocampal slice preparation and patch-clamp recordings. Transverse slices

(400 or 300mm) from the middle to dorsal part of the hippocampus were pre-

pared from 8- to 12-week-old male Sprague–Dawley rats and mice (Kv4.22/2 or

wild-type littermates) as previously described4, according to methods approved

by the Janelia Farm Institutional Animal Care and Use Committee. Cells were

visualized using an Olympus BX-61 epifluorescent microscope equipped with

differential interference contrast optics under infrared illumination and a water

immersion lens (360, 0.9 NA, Olympus). The CA3 region was usually cut off to

decrease spontaneous activity. Experiments were performed at physiologicaltemperature (33–36 uC) in ACSF containing the following (in mM): NaCl 125,

KCl 3, NaHCO3 25, NaH2PO4 1.25, CaCl2 1.3, MgCl2 1, and glucose 25; and was

saturated with 95% O2 and 5% CO2. Current-clamp whole-cell recordings from

soma or proximal apical trunks of hippocampal CA1 pyramidal neurons were

performed using a Dagan BVC-700 amplifier in the active ‘bridge’ mode, filtered

at 1–3 kHz and digitized at 50 kHz. Patch pipettes had a resistance of 2–6 MVwhen filled with a solution containing (in mM): K-methylsulphate or

K-gluconate 120, KCl 20, HEPES 10, NaCl 4, Mg2ATP 4, Tris2GTP 0.3, phos-

phocreatine 14, pH 7.25. The series resistance was usually ,20 MV.

Kv4.2 mice and genotyping. Kv4.22/2 mice were generated in the 129/SvEv

background. Genotypes were confirmed by PCR results using Kv4.2-specific

primers (forward, 59-GTGGATGCCTGTTGCTTC-39; reverse, 59-CCCACAA-

GGCAGTTCTTTTA-39) and neo-specific primers (forward, 59-AGGATCTCC-

TGTCATCTCACCTTGCTCCTG-39; reverse, 59-AAGAACTCGTCAAGAAGG-

CGATAGAAGGCG-39)23,24.

Two-photon imaging and uncaging. Neurons were filled with 100–150mM

Oregon green BAPTA-1 (OGB-1, Molecular Probes) in K-gluconate- or

K-methylsulphate-based internal solution for at least 15 min before the start ofrecordings. For plasticity experiments, cells were loaded with 100mM Alexa 488

(Molecular Probes) in K-gluconate-based internal solution for at least 5 min.

The presentation of variable spatio-temporal input patterns was performed

using a dual-galvanometer-based scanning system (Prairie Technologies) to

multi-photon photo-release glutamate at multiple dendritic spines and simul-

taneously image local Ca21 signal in basal and radial oblique dendrites of CA1

neurons4. Ultra-fast, pulsed, laser light (Mira 900F or Chameleon Ultra II;

Coherent) at 920–930 nm was used to excite the OGB-1 and Alexa 488, whereas

720-nm ultra-fast, pulsed, laser light (Chameleon or Chameleon Ultra II;

Coherent) was used to photolyse MNI-caged L-glutamate (Tocris Cookson;

10 mM applied via broken pipette above slice). The intensity of each laser beam

was independently controlled with electro-optical modulators (Model 350-50,

Conoptics). For glutamate uncaging, MNI-glutamate was uncaged at 5–20 visu-

ally identified spines (0.2-ms exposure times, with 0.1–0.2-ms intervals between

exposures for galvanometer repositioning). To isolate better the fast spike com-

ponent from soma/axonal action potentials, the membrane potential was kept

slightly hyperpolarized (between 262 and 268 mV) from the resting potential

(259.2 6 0.2 mV, n 5 177). For input site Ca21 signals in Fig. 1e, recordingswere performed in the presence of NMDAR antagonist (D,L-AP5, 50 mM) to

isolate spike-associated Ca21 influx mediated by voltage-gated calcium chan-

nels. Local Ca21 signals in Fig. 2e, g were measured in D,L-AP5 (50mM) or

,10 mm from the input site on daughter branches to isolate spike-associated

Ca21 influx. No significant difference was found in the local Ca21 signals

between weak and strong spikes (weak, 100 6 5% DF/F, n 5 15; strong,

98 6 12% DF/F, n 5 8; P 5 0.42).

Repeated local spiking protocols were given as trials of 3–10 local spikes every

,5–30 s with ,5-min inter-trial intervals. In consecutive trials of an experiment,

different sets of spines were usually activated, by randomly picking the available

spines within a ,10–20mm region of a dendritic segment in the same focal plane.

The h-pairing protocol consisted of locally suprathreshold uncaging paired with

back-propagating action potentials (Fig. 3C). That is, fifteen trains (five trains at

0.01 Hz repeated three times with a 30-s interval) of two bursts (repeated at

5 Hz), with each burst composed of a local spike (synchronous uncaging on a

cluster of 10–15 spines with a 0.1-ms interval) paired with 2–3 back-propagating

action potentials at 100 Hz. The back-propagating action potentials were timed

so that the peaks of the action potentials were coincident with the peak of thegluEPSP, as measured at the soma. Different spines were used for the h-pairing

induction than those that were used for testing the spikes before and after

induction. Also, during h-pairing induction, different spines were usually used

between consecutive sets of trains.

Carbachol (CCh) was prepared freshly every day in 20 mM stock solution in

distilled water and diluted to the bath solution to reach its final concentration.

Application of CCh usually led to a rapid (within 2 min), ,2–6 mV depolariza-

tion as measured at the soma, that was compensated with hyperpolarizing cur-

rent to keep Vm constant. Only those experiments were included where

depolarization by CCh was .1.5 mV.

BaCl2 was directly dissolved into the external solution before use. Atropine-

sulphate (Sigma-Aldrich) and D,L-AP5 (Tocris Cookson) were prepared in

10 mM and 50 mM stock solution (respectively) in distilled water and stored

at 220 uC. In experiments with atropine and D,L-AP5, the MNI-glutamate solu-

tion also contained the inhibitor.

Analysis. Spike amplitude at the soma was calculated using the spike component

of the dV/dt as a measure of the spike duration (see black circles on the dV/dt

traces in Fig. 1a). These time points were used to determine the base and peak of

the voltage spike (see black circles on the gluEPSP traces in Fig. 1a). In initial

experiments examining the distribution of dendritic spikes, individual dendrites

were randomly selected, tested at one time and at one location, and are therefore

represented by a single dV/dt value (Fig. 1 and Supplementary Fig. 3, n 5 479

branches). In later experiments (Figs 2–5 and Supplementary Figs 2, 4–6), dend-

ritic families were usually tested at two or more locations. By dendritic record-

ings (Fig. 1i, n 5 44 branches), only oblique dendrites branching off the trunk

within 20mm from the pipette tip were tested.

The Shapiro–Wilk test revealed that spike dV/dt distribution differed from a

normal distribution (W 5 0.62, P , 0.001). Fitting the first visually identified

peak of the distribution by a single gaussian function in the form of

151*exp[2((x 2 374)/862)]2 revealed that a relatively large number of large-

amplitude events (,20% of total events) lay outside the mean 1 2 s.d. (mean:

0.37 V s21; s.d. 0.86 V s21) range, indicating more than one underlying distri-

bution. Thus, the data were least-squares fitted by the sum of two gaussian func-

tions in the form of: 146*exp[2((x 2 371)/830)]2 1 25*exp[2((x 2 3,489)/

3,183)]2. Although the overlap of the populations around their intersection cre-

ates some error in this determination, the fit result suggests two underlying

populations. Because of the need to use relatively small bin sizes (0.4 V s21), it

was necessary to increase the relative weight of the faster events by a factor of 3.

Fitting was done using the Levenberg–Marquardt algorithm to search for the

coefficient values that minimize chi-square (Origin, OriginLab Corp.).

The cumulative probability plot in Fig. 2a is presented only for primary

(n 5 117) and for terminal (n 5 194) dendrites that expressed dendritic spikes.

Primary branches where a spike could not be initiated (n 5 24), as well as single

branches (strong, 16 out of 45, ,36%; weak, 28 out of 45, ,62%; no spike, 1 out

of 45, 2%), are not included. Intermediate branches (n 5 48) showed transitional

distribution with ,63% of strong branches.

Distance measurements and anatomical evaluation of the dendritic arboriza-

tion were performed on stacked images collected at the end of recordings using

ImageJ (NIH). Distance was measured from the approximate midpoint of the

input site. Because we focused on functional properties of dendritic segments, we

applied a different categorization of dendrites than that used for classical ana-

tomical analysis. The apical trunk was not considered as primary dendrite. Also,

the first ‘stem’ segments originating from the soma (by basals, usually 10–30mm

long) or the apical trunk (by radial obliques, usually 5–20 mm long) and giving

rise to separate dendritic families were neglected because their spine density was

low and dendritic spikes could not be evoked here. For basal dendrites, the length

of the proximal spine-free ‘stem’ segment was considered as distance from the

soma, whereas for apical dendrites the distance of the point where the oblique

dendrite branched off from the trunk is plotted (Fig. 1i).

Significance of differences was tested with one sample or two sample (paired

or unpaired) Student’s t-test or ANOVA with Tukey’s honest significant differ-

ence test for post-hoc comparisons, and P , 0.05 was considered as significant.

The effect of repeated local spikes alone, CCh and h-pairing on dV/dtD (Fig. 3D)

was analysed with ANOVA with Tukey’s post-hoc test. Also, the effect on dV/dtD

in the experimental groups with CCh and h-pairing (Fig. 4g) was analysed with

separate ANOVAs with Tukey’s post-hoc test. The significance of the difference

of spike ratio values from 1 (Fig. 2b) was analysed using one-sample Student’s

t-test. All other comparisons were performed using two-sample paired or

unpaired Student’s t-test. Throughout the paper data are given in mean 6 s.e.m.

In all figures, symbols with error bars indicate mean 6 s.e.m.; asterisk, P , 0.05;

double asterisk, P , 0.01; triple asterisk, P , 0.001.

doi:10.1038/nature06725

Nature Publishing Group©2008