Beruflich Dokumente
Kultur Dokumente
neurons in mouse brain slices. We report a diversity of spine morphologies that argues against common categorization schemes
and establish a close link between compartmentalization and spine morphology, wherein spine neck width is the most critical
morphological parameter. We demonstrate that spine necks are plastic structures that become wider and shorter after long-term
potentiation. These morphological changes are predicted to lead to a substantial drop in spine head excitatory postsynaptic
potential (EPSP) while preserving overall biochemical compartmentalization.
Dendritic spines form the postsynaptic component of most excitatory living cultured mouse brain slices and computer simulations to directly
synapses, whose plasticity is essential for brain development and relate spine morphological measurements to functional assays. Our
higher brain functions1,2. In addition to the molecular composition experiments establish that spine morphology plays a determining role
of the synapse, the morphology of spines is thought to be critical in biochemical and electrical compartmentalization, which can vary
for synaptic function, as spine head size correlates with synaptic independently of each other. LTP leads to coordinated morphological
strength3,4 and undergoes changes during synaptic plasticity5–8. Even changes in spine heads and necks, which leave overall biochemical
so, our understanding of how spine structure shapes synapse function compartmentalization intact but are predicted to substantially affect
remains fragmentary. EPSPs in potentiated spines. Furthermore, our study argues against
It is well established that spines compartmentalize biochemical common categorization schemes of spine morphology and indicates
signals9. By contrast, the quantitative contribution of spine that stubby spines are substantially over-reported in the light micro-
morphology to compartmentalization is still unknown, and only scopic literature as a result of limited spatial resolution.
moderate correlations between spine neck length or head volume
and chemical diffusion have been reported9–11. It is an open ques- RESULTS
tion to what extent biochemical compartmentalization is determined Quantitative analysis of spine morphological parameters
primarily by spine geometry or by intracellular factors such as We imaged spines on secondary and tertiary dendritic branches of
organelles or protein assemblies. Concerning electrical compartmen- CA1 pyramidal neurons by STED microscopy in organotypic hip-
talization, it is not clear how electrical signals are transformed by pocampal slices cultured for 2 to 4 weeks. The images revealed a
the spine neck9,12–14. This is an important question because synaptic continuum of morphologies (Fig. 1a and Supplementary Fig. 1a,b).
strength may be adjusted through structural changes in spine necks, Spine neck widths were symmetrically distributed around a median
which has been a long-standing hypothesis15,16. of 147 nm, ranging from 51 to 279 nm (n = 309 spines, 15 slices,
An early electron microscopy (EM) study reported that the aver- 10 animals; Fig. 1a–c and Table 1), while median neck length was
age spine head becomes larger and the neck wider and shorter after 667 nm and median head width 519 nm (Fig. 1d,e and Table 1). The
the induction of long-term potentiation (LTP)17, which was corrobo- distributions of the morphological parameters appeared smooth and
rated more recently by work based on two-photon microscopy6,18,19. unimodal (Fig. 1c–e), and none of the morphological parameters
However, it is not known how these structural changes might affect were intercorrelated (Fig. 1f–h).
biochemical and electrical compartmentalization because two-photon We also imaged CA1 neurons in acute slices from 4- to 5-week-old
microscopy does not have sufficient spatial resolution to properly Thy1-YFP mice (n = 59 spines, 4 slices, 3 animals). The ranges and
resolve spines and EM cannot be combined with functional assays. distributions of morphological parameters were very similar between
Here we combined STED microscopy, FRAP experiments, two- the two preparations (Fig. 1c–h and Table 1). In acute brain slices the
photon glutamate uncaging and patch-clamp electrophysiology in median neck width was slightly larger than in organotypic slices (165 nm;
1Interdisciplinary
Institute for Neuroscience (IINS), University of Bordeaux, Bordeaux, France. 2UMR 5297, Centre National de la Recherche Scientifique (CNRS),
Bordeaux, France. 3Two-photon Imaging Center, Institute of Experimental Medicine of the Hungarian Academy of Sciences, Budapest, Hungary. Correspondence
should be addressed to U.V.N. (valentin.nagerl@u-bordeaux.fr).
Received 5 December 2013; accepted 25 February 2014; published online 23 March 2014; doi:10.1038/nn.3682
Frequency (%)
Frequency (%)
Frequency (%)
25 200 20 1,500
where R2 = 0.92 to 0.97 for organotypic data
nm
nm
15 800
nm
20 100
15 400
and R2 = 0.72 to 0.85 for acute slice data. 15 500
0 10 0
(c) Distribution of neck width measurements 10
10
(**P = 0.002, Mann-Whitney test). 5 5
5
(d) Distribution of neck length measurements
0 0 0
in organotypic cultures (n.s., P = 0.17, Mann- 0 50 100 150 200 250 300 0 600 1,200 1,800 2,400 0 400 800 1,200 1,600
Whitney test). (e) Head width distribution Neck width (20 nm bins) Neck length (200 nm bins) Head width (100 nm bins)
(n.s., P = 0.57, Mann-Whitney test). Insets
show median, interquartile and range; see also
f 1,600 Organotypic R2 = 0.02
g 2,500 R2 = 0.03
h 1,600 R2 = 0.0005
Acute R2 = 0.02 R2 = 0.21 R2 = 0.009
Table 1. (f–h) Linear correlations between neck and 2,000
1,500
and neck length and head width (h). (i) 800 800
Time-lapse imaging over 1 h (geometric mean 1,000
with 95% CI). (j–l) S.d. of the normalized changes 400
500 400
over 1 h (corresponding to the CV). (m) Variability
0 0 0
of morphological parameters over 1 h (one-way 0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 600 1,200 1,800 2,400
ANOVA *P = 0.03, followed by Hom-Sidak Neck width (nm) Neck width (nm) Neck length (nm)
multiple comparisons test, both P < 0.04;
Neck length
mean ± s.e.m.). L, length; W, width. (n) Spines
appear stubby in two-photon, but not in STED i 1,050 Head width
Neck width j k l m
mode (red arrowheads). (o) Degrading a spine
900
200 25 *
200 200 *
20
STED image digitally by convolving with a 600 150 150 150
CV (%)
450 15
200 nm Gaussian function makes the highlighted 100 100 100 10
300
spine (blue arrowhead) appear stubby. Res., 150 50 50 50 5
resolution. All scale bars represent 500 nm. 0 0 0 0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
e L
ec W
W
H ck
t (min) t (min) t (min) t (min)
N ad
k
e
Mann-Whitney test; P = 0.002; Fig. 1c), which
N
may be a genuine difference or may reflect n o
50 nm
Res.
a slight decrease in optical resolution due Two-photon STED
400 nm
Methods). In addition, there was a weak cor-
Res.
relation between neck width and neck length
in acute slices (R2 = 0.21; Fig. 1g). All remain-
ing experiments were carried out in organotypic cultures. especially spine neck length, is largely stable over periods of an hour
The morphological parameters head width, neck width and neck (Fig. 1m) and that repeated STED imaging did not induce visible
length did not undergo directional changes over time (Kruskal-Wallis photodamage.
test Phead = 0.97, Pneck-width = 0.99, Pneck-length = 0.999); n = 33 spines, Surprisingly, we observed few, if any, stubby spines (lacking an
2 slices, 2 animals; Fig. 1i–l), indicating that spine morphology, identifiable neck) in the STED images, which is in contrast to results
in the conventional light microscopic litera-
Table 1 Results
ture, which commonly reports high fractions
Interquartile
Organotypic slices Range Mean Median range n (spines/slices/animals) of stubby spines (up to 40%)21–23. However,
Neck width 51–279 150 147 124–173 309/15/10
direct comparison between STED and two-
Neck length 70–2,368 743 667 397–1,037 309/15/10 photon images showed that short-necked
Head width 190–1,482 575 519 385–730 307/15/10 spines frequently appear stubby in two-
Acute slices photon mode because of its lower optical
Neck width 59–292 167 165 147–188 59/4/3 resolution (Fig. 1n). Similarly, after spatially
Neck length 157–1,801 689 551 260–993 59/4/3 filtering the STED images to mimic the two-
Head width 262–1,104 583 564 397–756 59/4/3
photon case, short-necked spines wrongly
t (ms)
appear stubby (Fig. 1o). Moreover, given
τyfp 42–2259 518 399 206–676 148/15/12
τalexa 14–292 80 63 36–105 85/5/4 the limited z axis resolution of our STED
Rneck (MV)
approach, some spines might appear stubby
Rmorph 2.1–598 82 44 18–113 148/15/12 because of a projection artifact in the z axis,
Rτ-yfp 9.2–807 93 62 35–98 148/15/12 even if the resolution in x-y is high. Taken
Rτ-alexa 4.9–1,263 115 57 27–108 85/5/4 together, the STED images reveal structural
CV (%)
YFP
relative fluorescence of the line scan. (b) Variability of repeated FRAP 20
CV = 26%
measurements within spines. Horizontal scale bars, 250 ms; vertical
scale bar, 25% relative to intensity before bleaching (applies to both 10
YFP and Alexa data). (c) Mean CV (±s.d.) for YFP and for Alexa 488. 0
Alexa
Variability was not different between the two fluorophores (P = 0.15). CV = 22%
Al YFP
8
48
(d) Variability in compartmentalization between spines. The traces were
a
ex
recorded at the color-coded lines in the three spines. The bottom trace
(dark lilac) shows bleaching and recovery in the spine head, while the d
e 2,400 350
flat trace (light lilac) shows the fluorescence in the dendrite. Horizontal
2,000 300
scale bars, 500 nm for spines and 1 s for FRAP traces; vertical scale bar,
250
25% relative to intensity before bleaching. (e) Range and distribution of 376 ms 1,600
Alexa (ms)
yfp (ms)
all YFP and Alexa 488 FRAP measurements. Scatter plots show all data 200
1,200
points; box plots show the median, interquartiles and range. 150
856 ms 800
100
400
details and a diversity of spine shapes and sizes that validates previous 50
diffraction-limited resolution.
8
P
48
YF
a
© 2014 Nature America, Inc. All rights reserved.
ex
Al
Diffusional coupling of spines
To assess the degree of biochemical compartmentalization of spines, small differences in spine neck diameter have large effects on chemi-
we performed FRAP experiments using freely diffusible YFP (27 kDa) cal diffusion. Additionally, there was a positive correlation between
and Alexa 488 (0.64 kDa). The recovery time course of the FRAP head width and τyfp, described by a cubic fit (y = ax3+b) with a mod-
signal in individual spines could be well described by an exponential erate to strong correlation (R2 = 0.33; Fig. 3c), while neck length
function with a single time constant τ (Fig. 2a and Supplementary was a weaker predictor of τyfp (using a linear fit function; R2 = 0.18;
Fig. 2), where τ reflects the degree of biochemical compartmentaliza- Fig. 3d). Similarly, τalexa depended moderately to strongly on spine
tion. The coefficient of variation (CV) of repeated τ measurements in neck width (R2 = 0.31; n = 85; Fig. 3e,f), corroborating the finding
individual YFP-labeled spines was small (22 ± 6.7%; 6 repeated FRAP that biochemical compartmentalization depends sensitively on spine
measurements in 11 spines, mean ± s.d.; Fig. 2b,c), indicating that neck geometry.
the approach is well suited for reporting differences between spines. When normalizing the Alexa 488 diffusion data by the factor 6.3
In contrast, τyfp varied greatly from spine to spine (Fig. 2d), ranging (the ratio of medians τyfp/τalexa), we found no differences between
from 42 ms to 2,259 ms (median = 399 ms; n = 148 spines, 15 slices, the diffusion of YFP and Alexa 488 for given neck width bins (Sidak’s
12 animals; Fig. 2e and Table 1). Alexa 488 has a 4.8-fold smaller multiple comparisons: bin 50–70 nm, P = 0.95; bin 150–170 nm,
hydrodynamic radius than YFP (0.58 nm versus 2.8 nm)26,27 and P = 0.59; bin 170–190 nm, P = 0.79; all remaining bins, P > 0.99; Fig. 3g).
should diffuse accordingly faster if the diffusional milieu affects the In a linear regression analysis, neck length explained 18% of the
two molecules similarly28. Indeed, fluorescence recovery was much variation in τyfp (Fig. 3d), neck cross-sectional area 45% (R2 = 0.45)
faster for Alexa 488 (τalexa median = 63 ms; n = 85 spines, 5 slices, and head volume 27%, while the respective residual plots validated
4 animals; Fig. 2e) than for YFP. Like those for YFP, τalexa values the use of linear fits (Supplementary Fig. 3b–e). Overall, 60% of
differed greatly between spines, ranging from 14 ms to 292 ms the variation in τyfp could be explained by morphology (Fig. 3h).
(Fig. 2e), while varying little within spines (CV = 18 ± 4.7% (s.d.)) Taken together, the experiments established a strong link between
from 6 repeated FRAP measurements in 9 spines; Fig. 2b,c). spine morphology and biochemical compartmentalization, identify-
The ratio of the median time constants τyfp/τalexa was 6.3, close to ing spine neck width as the dominant parameter of the diffusional
the ratio of their hydrodynamic radii, suggesting that diffusion of barrier.
molecules in and out of spines is largely determined by their size28.
Estimating the electrical resistance of the spine neck
Spine morphology predicts synapse compartmentalization The electrical resistance of the spine neck (Rneck) can be estimated on
In contrast to previous studies, our super-resolution-based approach the basis of neck morphology or on spine head diffusion measure-
allows direct correlations between morphology and diffusion meas- ments using Ohm’s or Fick’s law, respectively9:
urements in single spines. The applied correlation tests are justified rL (2)
by a simple compartmental model in which the theoretically predicted Rmorph =
A
τ is given by tr D
VL (1) Rt = (3)
t calc = V
DA
where V is head volume, L is neck length, D is the diffusion coef- where ρ is resistivity of the cytoplasm, D the diffusion coefficient,
ficient, A is neck cross-sectional area, and assuming Dyfp = 16 µm2/s L spine neck length, A neck cross-sectional area, V head volume
(ref. 29) and Dalexa = 120 µm2/s (ref. 30). and, assuming ρ = 150 Ω cm (refs. 31,32), Dyfp = 16 µm2/s (ref. 29)
Plotting τyfp as a function of spine neck width revealed a strong and Dalexa = 120 µm2/s (ref. 30).
negative correlation, well described by an inverse square function From these equations, we calculated Rneck in three independent
based on equation (1) (y = ax−2 + b; R2 = 0.47; n = 148; Fig. 3a,b), ways, on the basis either of morphology or of diffusion measure-
indicating that neck width is a determining factor of τyfp and that ments of YFP and Alexa 488. This yielded highly consistent values
�yfp (ms)
traces; vertical scale bars, 20% relative to intensity before bleaching. 1,500
(b) τ plotted as a function of neck width. The inverse-square fit is shown 1,000
140 nm
(gray), with the 95% confidence interval (pink). (c) τ plotted as a function
of head width (cubic fit). (d) τ plotted as a function of neck length. 500
Linear regression with residuals between τ and neck cross-sectional area, 61 ms
0
head volume or neck length is shown in Supplementary Figure 3. (e) Two
0 50 100 150 200 250 300
examples of Alexa 488–labeled spines and recovery traces. Horizontal Neck width (nm)
scale bar, 250 ms; vertical scale bars, 20% relative to intensity before
bleaching. (f) τalexa as a function of neck width. (g) τyfp and τalexa plotted c 2,500 2
R = 0.33
d 2,500 R2 = 0.18
relative to binned neck widths (normalized bin heights not different;
Sidak’s multiple comparisons, all P > 0.59). The two axes are scaled by a 2,000 2,000
factor of 6.3, which equals the ratio of the median τyfp/τalexa. Bars indicate
�yfp (ms)
�yfp (ms)
1,500 1,500
mean ± s.d. (h) Calculated τ (equation (1)) plotted against measured τ.
1,000 1,000
Plotting the measured τ against the estimated Rneck revealed a mod- e 107 nm f 320 R2 = 0.31
erate correlation for a majority of spines. Interestingly, an iterative
approach found that the correlation was very weak in the population 240
of spines with high Rneck values, and that above around 80 MΩ the
�alexa (ms)
151 ms
linear regression line slope was not different from zero, both for YFP 160
(R2 = 0.28, P < 0.001 for Rneck < 84 MΩ, n = 100 spines; in contrast,
246 nm
R2 = 0.07, P = 0.07 for Rneck > 84 MΩ, n = 48 spines; Fig. 4c) and for 80
Alexa 488 (R2 = 0.30, P < 0.001 for Rneck < 80 MΩ, n = 59 spines; in
0
contrast, R2 = 0.14, P = 0.06 for Rneck > 80 MΩ, n = 26 spines; Fig. 4d). 31 ms
0 50 100 150 200 250 300
This decorrelation indicated that in highly compartmentalized spines, Neck width (nm)
biochemical compartmentalization and electrical neck resistance vary
g h 9,000 R2 = 0.60
independently of each other. 2,500 YFP 400
Alexa 488 7,500
2,000 300
Correlation between changes in morphology and diffusion 6,000
�alexa (ms)
�calc (ms)
�yfp (ms)
Rneck (MΩ)
(overall fit not shown, R2 = 0.32). For values of Rneck larger than 84 MΩ 100
40
(dotted line) the correlation was not significant (R2 = 0.07, P = 0.07,
n = 48 spines), while it was moderate below this value (R2 = 0.28, 10 20
P < 0.001, n = 100). (d) The same observation for τalexa, where the
correlation was not significant above 80 MΩ (dotted line; n = 26 spines; 1 0
P = 0.06, R2 = 0.14), while it was highly significant below, with a 0 300 600 900 1,200
ph
a
p
ex
yf
�
moderate correlation (n = 59 spines; P < 0.0001, R2 = 0.30).
or
Rneck (MΩ)
al
�
M
c 2,500
R2 = 0.07
d 350
R2 = 0.14
300
followed by Dunnett’s multiple comparisons test, all uLTP time 2,000
points after uncaging P < 0.001, all neighbor time points P > 0.79 and 250
�alexa (ms)
uLTP + Mg2+ time points P > 0.98; Fig. 6f). 1,500 200
�yfp (ms)
Concurrently, there was a highly significant decrease in neck length 1,000 150
by around 25% in potentiated spines (two-way ANOVA, effect before 100
500
versus after uncaging P = 0.07, effect of uLTP versus neighbor ver- 50
sus uLTP + Mg2+ spines P = 0.01; in Dunnett’s test, uLTP all time R2 = 0.28 R2 = 0.30
0 0
points after uncaging P ≤ 0.01, all neighbor time points P > 0.88 and
© 2014 Nature America, Inc. All rights reserved.
200 20
EPSPdend = (5)
1 + Gsyn Rspine
∆ � (%)
∆ � (%)
100
100 0
0 –20
0 where Rdend is the input resistance of the dendrite (assumed to be
50 MΩ; refs. 35,36), Esyn is the synaptic reversal potential relative
–100
–100 –40 to the resting membrane potential of −70 mV, Gsyn the synaptic
–40 –20 0 20
conductance and Rspine = Rneck + Rdend. Because the capacitance of
re
re
r
te
te
fo
fo
Af
Af
Be
the spine head and neck membrane is negligibly small (~0.01 pF), it
d Control e Control f 0 mV is not included in the model13.
300 P = 0.67 40 P = 0.16 20 P > 0.26
10 Figure 5 Changes in diffusion and spine neck width co-vary. (a) Change
∆ Neck width (%)
200 20
0 in τalexa after a 4-min depolarization to 0 mV. (b) Effect of depolarization
∆ � (%)
on the spine neck width. (c) Changes in τalexa as a function of neck width
∆ (%)
100 0 –10
Head width change (with 95% CI, pink). (d) τalexa of control cells, which were kept
–20
0 –20 Head length at −70 mV. (e) Neck width changes of control cells measured at two time
–30 Neck length points 5 min apart. (f) Depolarization did not lead to changes in head
–100 –40 –40 width, head length and neck length (P = 0.26 to 0.77). Before-and-after
plots depict mean ± 95% CI.
re
re
re
r
te
te
te
fo
fo
fo
Af
Af
Af
Be
Be
Be
uLTP
150
∆ uEPSC (%)
100 ** *
5 pA * * 300
–10
–2 min *** *** *** ***
+60 min 50 n.s. 200
20 ms –20
100
0 –30
** *
0 ***
Neighbor
uLTP ***
–50 –40 ***
Neighbor
–10 0 10 20 30 40 50 60 0 15 30 45 60 0 15 30 45 60
∆ (%)
20
0
d e 0
Before 60 min –50
Before 60 min
*** ***
–20 –100 *** *** ***
0 15 30 45 60 0 15 30 45 60
© 2014 Nature America, Inc. All rights reserved.
t (min) t (min)
j 150
�measured
k 300
100 �calc
200
∆ �calc (%)
50
∆ (%)
100
Figure 6 Structural neck and head plasticity during LTP. (a) Uncaging 0
EPSCs were measured before and up to 60 min after LTP induction. 0
–50
(b) Effect of the uncaging LTP protocol on EPSCs (mean change ± s.e.m.)
R2 = 0.49
from targeted spines, neighboring spines within 5 µm and spines –100 –100
subjected to the uLTP protocol in the presence of Mg 2+. Morphological –5 0 10 20 30 –100 0 100 200 300
changes during LTP were assessed in a separate set of neurons that were (uLTP) t (min) ∆ �measured (%)
not patch-clamped. (c) Effect of uLTP targeting the spine identified by the
l
0
arrowhead. (d) Another example uLTP-induced changes in morphology R2 = 0.54
–25
in a potentiated spine (boxed). (e) Zooms of the boxed spine in d. (f) Effect
∆ Rneck (%)
of LTP on head size. (g) Effect of uLTP on neck length. (h) Effect of uLTP on neck width. (i) Calculated changes –50
in Rneck and τ from the morphological measurements during LTP. (j) Calculated and observed changes in
τ after uLTP. (The corresponding morphological changes are depicted in Supplementary Fig. 5.) (k) Pairwise –75
comparison of changes in observed and calculated τ during LTP depicted in j. (l) Correlation of the initial
head volume change to the corresponding Rneck change (at 1 min; data from f–h). Time-lapse graphs display –100
mean ± s.e.m., and correlations plots show 95% CI (gray). N.s., not significant; *P < 0.05; **P < 0.01; 0 300 600 900 1,200
***P < 0.001 in Dunnett’s multiple comparisons test. Scale bars, 500 nm. ∆ Head volume (%)
The amplitudes of the EPSP in the spine head and the dendrite both To predict the effect of a 50% drop in Rneck, we plotted the expected
depend strongly and positively on Gsyn (Fig. 7a,b), but they depend on percentage changes of EPSPdend and EPSPspine as a function of initial
Rneck in opposite ways in that the EPSP in the dendrite gets attenuated, Rneck (Fig. 7d,e). Whereas EPSPdend increases merely by a few percent,
while in the spine it gets boosted, with increasing resistance (Fig. 7c). To EPSPspine is consistently reduced by 20–40% over a wide range of
illustrate this effect in relative terms, we plotted the ratios of the spine and Gsyn and initial Rneck values, indicating that physiological changes in
dendritic voltages as a function of Rneck (Fig. 7c). Notably, the ratios depend spine neck geometry can substantially suppress the boosting of spine
solely on Rneck and Rdend and are independent of Gsyn and Esyn (ref. 13). head voltages, but they increase dendritic voltages only modestly
a b c 25 1.0
d e 0
10 2.5 nS 60 60
∆ Rneck –50% ∆ Rneck –50%
EPSPspine /EPSPdend
EPSPdend /EPSPspine
0.1 nS 40 40
6 15 0.6 –20
30 30
4 10 0.4 –30
20 20
2 10 5 0.2 10 –40
0 0 0 0 0 –50
0 250 500 750 1,000 0 250 500 750 1,000 0 250 500 750 1,000 0 250 500 750 1,000 0 250 500 750 1,000
Rneck (MΩ) Rneck (MΩ) Rneck (MΩ) Rneck (MΩ) Rneck (MΩ)
Figure 7 Influence of Rneck on electrical signaling. (a) Simulation of the dependence of the dendritic EPSP on Rneck for three values of synaptic
conductance (Gsyn). The 56-MΩ median Rneck is depicted as a dotted line. (b) The corresponding EPSPs in the spine head as a function of Rneck.
(c) Ratios EPSPspine/EPSPdend and EPSPdend/EPSPspine relative to Rneck. The ratios are independent of Gsyn. (d) Percentage change in dendritic EPSP
caused by a 50% drop in Rneck, as a function of the initial Rneck. (e) The corresponding EPSP change in the spine head.
(Supplementary Fig. 6b). These simulated effects on EPSPs reflect half of all spines have Rneck values larger than 56 MΩ, with 5% having
solely the observed changes in spine neck geometry and do not take resistances larger than 500 MΩ. The previous discrepancies may be
into account any increases in synaptic conductance after LTP. due to relatively low numbers of experimental observations in some
reports and/or biases inherent to the methods that were used, which
DISCUSSION may have compressed biological variability.
We performed time-lapse imaging of dendritic spines in living brain
slices with a lateral resolution around 50 nm (ref. 37), in parallel with LTP leads to coordinated changes in spine heads and necks
functional assays. These studies effectively bridge the gap between EM Divergent observations regarding spine plasticity can be found in the
and conventional fluorescence microscopy. literature45. Whereas spine head enlargement after the induction of
LTP has been consistently reported5,6,19,34, much less is known about
Resolving live spine morphology changes occurring at the level of spine necks.
EM studies have provided exquisitely detailed and quantitative analyses In accord with our observations, an early EM study reported
of spine morphology in fixed samples24,38,39, but a comparable analysis changes in average spine neck geometry after tetanic stimulation17.
in live tissue has been lacking. Our STED images reveal a high degree of However, as EM cannot be used for longitudinal investigations, the
heterogeneity of spine sizes and morphologies, which agrees well with reported differences between the stimulated and control groups may
the previous EM work but argues against morphological categoriza- have been due to altered turnover of populations of spines of certain
tion schemes commonly used in the light microscopic literature40–42. sizes, as opposed to reflecting dynamic changes at the level of
Notably, our study indicates that stubby spines are greatly over- individual spines. A recent two-photon study reported changes in
reported in the light microscopic literature, owing to insufficient spine neck fluorescence, which were interpreted as shortening and
optical resolution. This conclusion is supported by EM studies that widening after LTP19; however, the quantification is problematic
© 2014 Nature America, Inc. All rights reserved.
typically observe only low fractions (a few percent) of stubby spines because of the limited optical resolution.
in adult tissue24,25,38,43. This is not merely a semantic issue, because Here we present direct evidence based on time-lapse STED imaging
spines with large heads and thin and short necks (for example, Fig. 1o) that spine necks are highly plastic structures, becoming substantially
represent completely different functional compartments than spines shorter and wider in a spine-specific LTP model. In addition, the
devoid of necks. analysis shows that structural changes in spine necks and heads occur
in a concerted fashion, with spine head enlargement scaling with neck
Morphology determines biochemical compartmentalization shortening and broadening.
Synaptic strength can be regulated independently of neighboring syn-
apses by way of compartmentalized signaling6,34, which is thought Functional impact of spine neck plasticity
to boost the computational power of neurons. We found that 60% Our data indicate that the observed morphological changes have
of the variation in biochemical compartmentalization across spines divergent consequences for biochemical and electrical compartmen-
could be accounted for by morphology and that changes in spine talization, which is theoretically possible given that τ depends on
structure strongly affected molecular diffusion. Neck width was the neck and spine head size (equation (1)), whereas Rneck depends only
most influential determinant of compartmentalization, potentially on neck size (equation (2)).
facilitating fast and cost-efficient regulation of the synaptic milieu. Whereas spine head enlargement increases biochemical compart-
However, we cannot rule out the possibility that intracellular factors, mentalization, the observed neck changes counteract this effect, so
such as organelles, co-vary with the morphology and also play a role. that diffusional recovery is largely preserved after LTP. Nevertheless,
Even so, nanoscale spine structure can be used as a reliable proxy for shorter and wider necks may facilitate access to the spine from the
synapse compartmentalization. dendritic side46, and chemical second messengers released into the
The finding that fluorescence recovery of Alexa 488 and YFP enlarged spine head might be more diluted and disperse faster into
largely followed basic diffusion theory suggests that the diffusion the dendrite, which may render spines less susceptible to subsequent
barrier holds for a wide range of small signaling molecules—for plasticity events.
example, Ca2+, cyclic AMP or inositol trisphosphate (IP3)—and In contrast to biochemical compartmentalization, our simulations
larger cytosolic proteins—for example, GTPases, actin or monomeric predict electrical signaling to be substantially altered by spine neck
calcium-calmodulin kinase II (CaMKII). Still, the slightly higher plasticity. While a 50% drop in Rneck is predicted to lead to only a
than expected median τ ratio (relative to the ratio of the hydro- slight increase in dendritic EPSPs, the amplitude of the EPSP in the
dynamic radius) may reflect a modest sieving effect that slows the spine would drop by 20% to 40% in most spines.
diffusion of molecules on the basis of their size. With substantially reduced boosting of the spine head voltage after
a sharp drop in Rneck, synapses will operate in a more linear regime:
Reliable estimates of spine neck resistance voltage-gated channels and NMDA receptors are less likely to become
As electrical measurements of Rneck are technically infeasible, we esti- activated47 and the voltage in the spine head is less likely to reach the
mated Rneck in three different indirect ways, using basic biophysical synaptic reversal potential and saturate the synaptic response48. Spines
equations and the morphological and diffusional data. The fact that with shorter and wider necks will be able to sustain stronger synaptic
the independent estimates are highly consistent suggests that the currents because the driving force will be effectively maintained even
measurements were robust and accurate. during large or repeated synaptic conductance increases. In this way,
Previous studies based on EM25 or diffusion measurements9 the observed neck changes may functionally disinhibit the synapse,
reported a range of 1 to 400 MΩ for Rneck, whereas a recent study which could contribute to synaptic weight changes during LTP44,49.
based on Ca2+ imaging estimated Rneck to be relatively large (500 MΩ) Taken together, our findings challenge the widespread notion that
and to vary little across spines, which suggests that morphology does spines primarily shape biochemical rather than electrical signaling at
not a play a major role44. Encompassing these values, our measure- synapses. Instead, they argue that both functions are distinctly shaped
ments revealed a broad distribution, indicating that, at any given time, and dynamically regulated by nanoscale spine morphology.
Methods 20. Urban, N.T., Willig, K.I., Hell, S.W. & Nägerl, U.V. STED nanoscopy of actin
dynamics in synapses deep inside living brain slices. Biophys. J. 101, 1277–1284
Methods and any associated references are available in the online (2011).
version of the paper. 21. Oray, S., Majewska, A. & Sur, M. Effects of synaptic activity on dendritic spine
motility of developing cortical layer v pyramidal neurons. Cereb. Cortex 16, 730–741
Note: Any Supplementary Information and Source Data files are available in the (2006).
online version of the paper. 22. Parnass, Z., Tashiro, A. & Yuste, R. Analysis of spine morphological plasticity in
developing hippocampal pyramidal neurons. Hippocampus 10, 561–568 (2000).
Acknowledgments 23. Peebles, C.L. et al. Arc regulates spine morphology and maintains network stability
in vivo. Proc. Natl. Acad. Sci. USA 107, 18173–18178 (2010).
We thank D. DiGregorio (Pasteur Institute) and all members of the Nägerl
24. Harris, K.M., Jensen, F.E. & Tsao, B. Three-dimensional structure of dendritic spines
laboratory for comments on the manuscript, and J. Angibaud for excellent and synapses in rat hippocampus (CA1) at postnatal day 15 and adult ages:
technical support. This work was supported by postdoctoral fellowships from implications for the maturation of synaptic physiology and long-term potentiation.
the Marie-Curie Program (grant 272351) and the European Molecular Biology J. Neurosci. 12, 2685–2705 (1992).
Organization (EMBO; grant 1518-2010) to J.T. and by grants from the French 25. Harris, K.M. & Stevens, J.K. Dendritic spines of CA 1 pyramidal cells in the rat
Institute of Health and Medical Research (INSERM), Agence Nationale de la hippocampus: serial electron microscopy with reference to their biophysical
Recherche (ANR), the Human Frontiers Science program (HFSP) and characteristics. J. Neurosci. 9, 2982–2997 (1989).
France-BioImaging (ANR-10-INSB-04) to U.V.N. 26. Heyman, N.S. & Burt, J.M. Hindered diffusion through an aqueous pore describes
invariant dye selectivity of Cx43 junctions. Biophys. J. 94, 840–854 (2008).
27. Kumar, M., Mommer, M.S. & Sourjik, V. Mobility of cytoplasmic, membrane, and
AUTHOR CONTRIBUTIONS DNA-binding proteins in Escherichia coli. Biophys. J. 98, 552–559 (2010).
J.T. and U.V.N. conceived and designed experiments. J.T. performed experimental 28. Einstein, A. Über die von der molekularkinetischen Theorie der Wärme geforderte
work. G.K. and B.R. developed and provided reagents. J.T. and U.V.N. analyzed Bewegung von in ruhenden Flüssigkeiten suspendierten Teilchen. Annalen der
data and performed modeling. J.T. and U.V.N. wrote the paper. J.T., G.K., B.R. and Physik 322, 549–560 (1905).
U.V.N. approved the paper. 29. Petrásek, Z. & Schwille, P. Precise measurement of diffusion coefficients using
scanning fluorescence correlation spectroscopy. Biophys. J. 94, 1437–1448 (2008).
30. Alevra, M., Schwartz, P. & Schild, D. Direct measurement of diffusion in olfactory
© 2014 Nature America, Inc. All rights reserved.
2.5 mM KCl, 1.25 mM NaH2PO4, 28 mM NaHCO3, 0.5 mM CaCl2, 1 mM ascor- two to three consecutive FRAP measurements were obtained and the average τ
bic acid, 7 mM glucose and 7 mM MgCl2. After cutting, slices were allowed to value used for further analysis, except for the depolarization experiments, where
recover for 1 h. Immediately before imaging, slices were placed on PLL-coated only one measurement was performed before the challenge. Alexa Fluor 488
glass coverslips, to which they would adhere, and transferred to the perfused (Invitrogen) was dissolved at 200 µM in intracellular solution containing 125 mM
imaging chamber. potassium gluconate, 5 mM KCl, 10 mM HEPES, 1 mM EGTA, 4 mM Mg-ATP,
0.3 mM Na-GTP and 10 mM sodium phosphocreatine. Cells not expressing YFP
STED microscope and imaging. Super-resolved images of spines were acquired were voltage-clamped at −70 mV in the whole-cell configuration and loaded
by STED microscopy52,53. The home-built STED microscope was described 5 min or more before experiments.
previously37. Briefly, it is constructed around an inverted confocal microscope
(DMI6000, Leica) using a 100×, 1.4 NA oil immersion objective (PL APO 100, Electrophysiology. Whole-cell patch clamp recordings were performed using the
Leica). It uses pulsed excitation at 488 nm (PicoQuant) and pulsed quenching at intracellular solution described above and glass capillaries with a tip resistance
594 nm. The quenching wavelength is derived from an optical parametric oscillator around 5 MΩ. Depolarizations consisted of a 4-minute-long voltage clamping
pumped at 796 nm by a Ti-sapphire femtosecond laser (MaiTai, SpectraPhysics). to 0 mV with 5 mM QX-314 in the patch pipette. Consecutive STED images
Emitted fluorescence is detected by an avalanche photodiode (Perkin Elmer). and FRAP measurements were obtained immediately before and around 2 min
Pixel dwell times were 40 to 50 µs for STED images, and pixel size was typically after 4 min of depolarization to 0 mV. Uncaging induced excitatory postsynaptic
20 nm by 20 nm. Typically three to five z-sections were acquired 400 nm to currents (uEPSCs) were recorded in the whole-cell mode with the addition of
500 nm apart. The effective lateral optical resolution of this setup is around 5 µM β-actin in the intracellular solution and with pipette tip resistances of
50 nm for work in live organotypic slices37. around 10 MΩ to reduce intracellular dialysis19. Synaptic uncaging currents
A second Ti-sapphire pulsed femtosecond laser (MaiTai, SpectraPhysics) were measured at a membrane potential of −70 mV. uEPSCs were averages of
in the infrared range was routed in and aligned with the excitation/quenching four repeats. Two uEPSC baseline time points were acquired in the presence of
beams. This beam was used for two-photon–mediated bleaching of YFP at MgCl2, and then MgCl2 was removed and 1 mM tetrodotoxin added during
900 nm and Alexa Fluor 488 at 800 nm, as well as for two-photon imaging at LTP induction. After induction of LTP, MgCl2 was added back to the solution
900 nm and two-photon glutamate uncaging at 740 nm. Image acquisition was and additional time points were acquired. Uncaging-induced LTP (uLTP) was
done using ImSpector software (http://www.max-planck-innovation.de/de/ induced by uncaging glutamate at 0.5 Hz for 1 min34. The caged glutamate com-
industrie/technologieangebote/software/). Experiments were performed in pound was a modified version of the commercial MNI-glutamate, which has
artificial cerebrospinal fluid containing 125 mM NaCl, 2.5 mM KCl, 1.3 mM a higher quantum efficiency. We used it at a concentration of 2 mM. The two-
MgCl2, 2 mM CaCl2, 26 mM NaHCO3, 1.25 mM NaH2PO4, 20 mM d-glucose, photon laser was tuned to 740 nm, and the pulses were 1 ms in length and
1 mM Trolox; 300 mOsm; pH 7.4. Perfusion rate was 2 ml/min and the delivered around 3 mW power to the back aperture of the objective. The uncaging
temperature 31.5 °C. Since our STED approach enhanced resolution only in pulses were controlled by a Pockels cell. For uncaging without electrophysiology,
the lateral plane, we limited our analysis to spines extending laterally from the as in the uncaging plus STED imaging experiments, before experiments a cell in
dendrite. For imaging of acute slices, the microscope objective was changed to the slice culture was patch-clamped and the optimal uncaging power determined
a glycerin objective (NA 1.3) equipped with a correction collar (Leica), which for the given conditions. The uncaging plus STED imaging experiments were
facilitates super-resolution imaging at tens of microns tissue depth, which is therefore performed in unperturbed neurons.
not possible with oil objectives because of spherical aberration54. However, During uLTP experiments, STED images were acquired at −5, 1, 15, 30, 45 and
because of the lower numerical aperture, the achievable resolution is slightly 60 min relative to LTP induction. During LTP + FRAP experiments, STED images
lower than with oil. were acquired at −5, 2, 15 and 30 min (the 1-min time point was not accessible
Geometric measurements of morphological parameters were done on raw owing to the time required for switching from uncaging to FRAP wavelengths
images of single sections in ImageJ. Spine neck widths were obtained from full on the laser).
width half-maximum (FWHM) measurements based on Gaussian fits of line
profile plots as in ref. 55. Each line profile was obtained from a 3-pixel-wide line, Simulations of EPSPs. EPSP amplitude as a function of Rneck was calculated
to avoid noisy single pixel outliers. Neck length was measured from the base of on the basis of a simplified equivalent circuit of an electrically passive spine
the parent dendrite to the base of the head, following the curvature of the spine using equations (4) and (5) above13. Assuming Gsyn to vary around 0.5 nS
neck. Head width was measured perpendicular to the spine neck axis, unless a (refs. 56,57), we calculated EPSPs for three different values of Gsyn (0.1, 0.5,
cup-like shape would suggest another location of the synapse than on the tip of 2.5 nS), lumping together the contributions from AMPA and NMDA receptors.