tent spines were formed. These findings suggest rapid changes in connectivity between motor centers and sensory cortex guide subse-
Perceptual learning and skill acquisition are associated with dra-
matic changes in the response properties of sensory cortex neu-
rons (Merzenich and Jenkins, 1994; Wiest et al., 2010). After
training, top-down influences from higher cortical areas, medi-
ated in part by layer 1 inputs, are capable of influencing func-
tional responses at the earliest stages of cortical processing, in a
context-dependent manner (Merzenich and Jenkins, 1994; Gil-
bert et al., 2009; Petreanu et al., 2012). Whether these changes in
function are accompanied by persistent structural changes lo-
calized within primary sensory cortex has not been addressed.
Previous studies have identified experience-dependent and cell-
tex (Trachtenberg et al., 2002; Holtmaat et al., 2006; Hofer et al.,
2009; Holtmaat and Svoboda, 2009; Keck et al., 2011; Moczulska et
plasticity triggered by sensory deprivation or enrichment. Little is
known about structural plasticity at connections that are likely in-
Layer (L) 2/3 pyramidal neurons in the barrel cortex receive
somatosensory input from L4 neurons. They are also modulated
by inputs from motor centers, including the primary motor cor-
tex and the medial subdivision of the posterior nucleus, both of
which impinge on their branches in L1 (Petreanu et al., 2009).
L2/3 pyramidal neurons are therefore a site of sensorimotor in-
tegration, and changes in synaptic input onto these cells may
ied the effect of learning an active tactile task on structural plas-
ticity of L2/3 pyramidal neurons.
Experiments were approved by the Janelia Farm Animal Care and Use
6Crl) were labeled with green fluorescent protein using in utero electro-
poration as described previously (Petreanu et al., 2009). Whiskers were
trimmed starting at 6–7 weeks of age, sparing rows C1–3 and D1–3.
Retrimming occurred every 3 d under brief isoflurane anesthesia. A tita-
nium headpost was attached to the skull to permit head fixation during
training (O’Connor et al., 2010b). A craniotomy (?2 mm in diameter)
was made over the left barrel cortex of 2-month-old mice as described
previously (Holtmaat et al., 2009). For ?10 d before training, mice were
limited to 1 ml/d of water. Mice of either sex were randomly assigned to
either vibrissa or auditory control training groups (Fig. 1A,B).
Behavioral training. Methods and apparatus for head-fixed object lo-
calization were as described previously (O’Connor et al., 2010a, b). The
stimulus object was a 1.59-mm-diameter pole (stainless steel dowel pin,
McMaster) coupled to a linear slider (Schneeberger), which moved the
pole in the anterior–posterior dimension (9.7 mm from the mouse mid-
line), driven by a stepper motor (Zaber).
Mice were initiated into the training environment on day 0, after the
water from a lickport.
For day 1 and onward, mice were trained in one session per day. On
days mice were also imaged under the two-photon microscope, the im-
aging session preceded training by at least 2 h. Thus, “4” days training
includes one shaping session and three discrimination sessions, “8” days
training includes the subsequent four discrimination sessions, etc. Mice
typically performed ?300 trials within a session (range, 290–400 trials).
Sessions typically lasted 40–60 min.
6078 • TheJournalofNeuroscience,April23,2014 • 34(17):6078–6083
2 months old
auditory control task
1 ml / day
C and D remain
New persistent spines per day (%)
New persistent spines per day (%)
New persistent spines per day (%)
Spines gained per day (%)
pre auditorypre vibrissa-
Kuhlman,O’Connoretal.•StructuralPlasticitywithintheBarrelCortexJ.Neurosci.,April23,2014 • 34(17):6078–6083 • 6079
Trials began with the pole descending (?0.5 s) into reach of the whis-
sound that signaled the start of the trial. After a 0.75 s grace period, the
mouse had until 2 s from the start of pole descent to either lick (“go”
response) or withhold a lick (“no-go” response). Correct no-go re-
sponses (“correct rejections”) were not rewarded, and incorrect no-go
responses (“misses”) were not punished. Correct go responses (“hits”)
were rewarded with a drop of water (?8 ?l). Incorrect go responses
(“false alarms”) triggered a 200 ms air puff and the start of a “time-out”
period in which the trial was paused for 5 s. If the mouse licked during
restarted. Each trial ended with the pole ascending (?0.75 s).
follows: (no. correct go trials ? no. correct no-go trials)/no. of trials,
using a 100-trial sliding average. When animals reached ?90% perfor-
mance or ?80% for 3 d, task difficulty was increased by moving the
The following offset schedule was used (in mm): 6.5, 4, 3.5, 3, 2.75, 2.5,
was never increased more than one time.
The auditory version of the task exploited the hissing sound made by
to control for sensory stimulation and stressors not associated with
whisker-dependent learning. The “go” stimulus was the sound of the
a correct rejection/false alarm response window of 2 s preceding the
auditory cue; the mouse had to withhold licking during this window to
avoid the punishment described above.
Intrinsic signal optical imaging. Intrinsic signal optical imaging
through the glass window was performed as described previously
(O’Connor et al., 2013). We localized 2–4 barrel columns within rows
C1–3 and D1–3 per mouse (Fig. 1D).
Two-photon imaging. Mice were lightly anesthetized using isoflurane
nm with a Ti:Sappire laser (Mai Tai, Spectra Physics) and imaged
through a 40? 0.8 NA objective (Olympus). To confirm correct barrel
column location, image stacks through layer 1 (10–80 ?m from the pia
which the shadow of blood vessels was evident (Fig. 1E). Dendrite
braches for longitudinal imaging were then located at a higher zoom
using a scan resolution of 0.1 ?m/pixel (Fig. 1E, inset). Image stacks
acquired for spine analysis consisted of sections (scan resolution: 0.07
?m/pixel) acquired using ScanImage software (Pologruto et al., 2003).
All dendrites were traced back to cell soma to ensure that soma were
located within rows C1–3 and D1–3. Soma location for the groups de-
fined in Fig 1B were as follows: auditory control, 2?C1, 2?C2, 1?C3,
2?D1, 1?D2; subsequent vibrissa training (same as auditory training
cells per mouse, dendritic length (microns), and spines per cell for the
three groups were (mean ? SEM) as follows: auditory control, 2.00 ?
0.41, 70.63 ? 9.21, 73.25 ? 8.35; subsequent vibrissa training, 1.75 ?
0.48, 76.22 ? 10.18, 79.29 ? 6.66; and vibrissa training, 2.17 ? 0.65,
55.75 ? 5.83, 59.92 ? 8.18.
Image analysis. Spine dynamics were manually annotated using the
Spine Analysis program included in ScanImage software, using criteria
previously described (Holtmaat et al., 2009). Spines were annotated and
classified as “Loss,” “Gain,” or “Stable.” A lost spine is a spine that was
present in the subsequent session. A gained spine is a spine that first
appeared in the scored session. A stable spine is a spine that was
session. “Gained” spines present for ?8 d (present for at least three
sessions) were classified as “New Persistent Spines.” Annotation was
done by an observer blind to sequence order after the sequential image
series of a given dendrite branch was randomly reversed at a probability
of 50%. Classification was corrected in reversed sequences after annota-
tion was completed.
To calculate percentage NPS/d for each neuron (Figs. 1H–J, 2a,b, and
by summing the number of all spines for the given neuron and dividing
the sum by the number of imaging sessions. Then for each imaging
number of spines as calculated above. Thus, for a given neuron, the
rons were imaged every 4 d, to report daily NPS formation the above
value was divided by 4. To report percentage NPS/d averaged across
training days (Fig. 1H–J), percentage NPS/d for each image session, as
calculated above, was averaged across the first 24 d of training for each
NPS/d across all control neurons, for a given imaging session. Similarly,
to calculate the fold difference in loss (vibrissa training/control; Fig. 2d),
the percentage loss/d value for each individual neuron in the vibrissa
training group was divided by the mean percentage loss/d across all con-
trol neurons, for a given imaging session. Relative spine intensity was
calculated as in Holtmaat et al. (2005), except that the background-
shaft pixel value rather than the summed shaft pixel value (Fig. 2c).
object localization task (vibrissa training; Fig. 1A,B). Mice were
required to move their whiskers to distinguish two locations of a
small pole with a subset (C and D rows) of their mystacial vibris-
sae (whiskers) for a water reward (O’Connor et al., 2010b). Task
difficulty was increased throughout training; when animals
version of the task defined by a particular offset between pole
positions, the offset was decreased (Fig. 1C).
We labeled L2/3 pyramidal neurons with GFP via in utero
electroporation (Petreanu et al., 2009). In naive young adult
a glass window (Holtmaat et al., 2009). Intrinsic signal optical
to task-relevant whiskers (D1-D3 and C1-C3; Fig. 1D) for imag-
ing. Approximately 3 weeks after window surgery, we used two-
over 1–2 months. First, mice were maintained in their home
cages and did not visit the training environment for 4–8 d while
we obtained baseline measures (two or three imaging sessions).
barrel cortex (O’Connor et al., 2010b; Guo et al., 2014). Alterna-
tory task and subsequently trained in the whisker task (n ? 7 cells, 555 spines, 3 animals),
6080 • J.Neurosci.,April23,2014 • 34(17):6078–6083 Kuhlman,O’Connoretal.•StructuralPlasticitywithintheBarrelCortex
tively, a different group of mice was trained on an auditory ver-
“auditory control”) and presumably did not engage the barrel
cortex but did expose the mice to the same environment and
stressors. A subset of auditory control mice were subsequently
trained in the object localization task (3 animals, “subsequent
vibrissa training”; Fig. 1B); dendritic spine data from these mice
appear only in Figures 1J and 3.
We found that training induced an increase in new spine for-
mation (Fig. 1G). A subset of newly produced spines form new
ent for ?8 days) (Holtmaat et al., 2005; Knott et al., 2006). The
fraction of NPS for individual neurons increased by 67% during
control auditory-trained mice (Fig. 1H; Mann–Whitney U test,
p ? 0.012). We also quantified NPS formation during the base-
line period before training (“previbrissae training”) and com-
pared these values with those obtained during vibrissa training,
pared with the baseline period (Fig. 1I; paired t test, p ? 0.001).
Spine dynamics can change in response to an enriched envi-
the enhanced spine growth we observed (Fig. 1G,H) was specific
(vibrissa-training / control)
Normalized spine density
Cumulative values of
spine loss (%)
Cumulative values of
NP spine gain (%)
NP spine gain, fold-difference
(vibrissa-training / control)
4812 16 20 24
4812 16 20 24
48 12 16 20 24
loss, last session
% of sample100
Relative spine intensity
for stable spines present for the entire experiment. Additionally, the intensity of NPS was significantly higher than spines lost in the next imagingsession (Kolmogorov–Smirnov, p ? 0.004),
Kuhlman,O’Connoretal.•StructuralPlasticitywithintheBarrelCortexJ.Neurosci.,April23,2014 • 34(17):6078–6083 • 6081
experienced the same environment and stressors as the vibrissa
trained group and yet developed fewer NPS. Furthermore, we
found NPS also increased in mice experiencing vibrissa training
after first learning the auditory task (53%; Fig. 1J; paired t test,
p ? 0.02). We did not detect a difference between the baseline
period of the vibrissa-trained group (Fig. 1I; “previbrissae train-
ing”) and the auditory control group (Fig. 1G,H; “au control,” t
test, p ? 0.186). Together, the vast majority (17 of 20) of imaged
L2/3 neurons showed an increase in NPS associated with tactile
learning (Fig. 1I,J).
We analyzed the time course of structural dynamics (Fig. 2).
We found that NPS accumulation, averaged across the 6 animals
8 d of training (Fig. 2a,b), while mice learned to associate pole
position with water reward in the simplest versions of the task
(Fig. 2g; median offset at the 8 day time-bin, 4.4 mm). To better
understand how NPS size compares with mature spine sizes, we
analyzed spine volumes of new persistent, stable, and lost spines.
The spine sizes in the NPS population were distinct from spines
that were lost in the next imaging session (Kolmogorov–Smir-
with lost spines, although it was 51% smaller than stable spines
present for the entire imaging experiment (Fig. 2c).
In addition to the rapid effect of training on spine formation,
our time course analysis revealed that spine loss was significantly
group (Fig. 2d,e; ANOVA, p ? 0.020). Consistent with these
results, we found that the net impact of training was a modest
increase in spine density (Fig. 2f). Thus, perceptual learning
specific to spines formed during training or applied also to
preexisting spines, we quantified spine survival percentages. We
found that perceptual learning stabilized both preexisting and
newly formed spines (Fig. 2h,i). These results distinguish the ef-
fect of reinforced sensorimotor learning in L1 branches of L2/3
spine elimination within L1 branches of L5 excitatory neurons is
increased (Yang et al., 2009).
Days to attain 80% correct
4 days training
8 days training12 days training
New persistent spines
per day (%)
New persistent spines
per day (%)
New persistent spines
per day (%)
48 12 162024
6082 • J.Neurosci.,April23,2014 • 34(17):6078–6083 Kuhlman,O’Connoretal.•StructuralPlasticitywithintheBarrelCortex
To further explore the behavioral relevance of increased NPS Download full-text
formation, we examined the relationship between performance
and NPS formation. When considering all whisker-trained ani-
mals, we noted that there was an animal-by-animal correlation
few days of training (Fig. 3A). Animals learned at different rates
less NPS at the 4 day time point. Thus, the initial training-
induced structural changes that occur on L1 branches of L2/3
excitatory neurons are predictive of subsequent performance.
Our results indicate that new spines are produced in layer 2/3
crimination training. We found that both preexisting spines and
newly formed spines in L2/3 neurons stabilize during perceptual
learning, resulting in a net increase in spine density. During
vibrissa-based object localization, L1 of primary somatosensory
cortex receives diverse, axon-specific information from motor
we observed within L1 could dramatically impact the manner in
which motor information is integrated and received within L2/3
neurons over the course of learning to facilitate progressive im-
provement of discriminative performance.
The rapid growth of persistent spines in response to behav-
ically during extinction training (Lai et al., 2012), and motor
cortex (Xu et al., 2009; Yang et al., 2009). Also in common with
our findings, previous studies have shown that learning-induced
et al., 2013). Thus, our findings establish that, similar to L5 neu-
Additionally, a recent study examining cocaine-conditioned
place preference learning in L5 neurons of the prefrontal cortex
reported that both preexisting and newly formed spines are sta-
However, in motor cortex, motor-skill learning leads to elimina-
tion of preexisting spines (Xu et al., 2009; Yang et al., 2009).
in a wide range of cortical regions and across many cortical be-
havioral paradigms, but the degree to which preexisting spines
are maintained differs across cortical areas and/or cell types.
Bock J, Gruss M, Becker S, Braun K (2005) Experience-induced changes of
dendritic spine densities in the prefrontal and sensory cortex: correlation
with developmental time windows. Cereb Cortex 15:802–808. CrossRef
Gilbert CD, Li W, Piech V (2009) Perceptual learning and adult cortical
plasticity. J Physiol 587:2743–2751. CrossRef Medline
Guo ZV, Li N, Huber D, Ophir E, Gutnisky D, Ting JT, Feng G, Svoboda K
(2014) Flow of cortical activity underlying a tactile decision in mice.
Neuron 81:179–194. CrossRef Medline
Hofer SB, Mrsic-Flogel TD, Bonhoeffer T, Hu ¨bener M (2009) Experience
leaves a lasting structural trace in cortical circuits. Nature 457:313–317.
Holtmaat A, Svoboda K (2009) Experience-dependent structural synaptic
plasticity in the mammalian brain. Nat Rev Neurosci 10:647–658.
Holtmaat AJ, Trachtenberg JT, Wilbrecht L, Shepherd GM, Zhang X, Knott
GW, Svoboda K (2005) Transient and persistent dendritic spines in the
neocortex in vivo. Neuron 45:279–291. CrossRef Medline
HoltmaatA,WilbrechtL,KnottGW,WelkerE,SvobodaK (2006) Experience-
Holtmaat A, Bonhoeffer T, Chow DK, Chuckowree J, De Paola V, Hofer SB,
Hu ¨bener M, Keck T, Knott G, Lee WC, Mostany R, Mrsic-Flogel TD,
Nedivi E, Portera-Cailliau C, Svoboda K, Trachtenberg JT, Wilbrecht L
(2009) Long-term, high-resolution imaging in the mouse neocortex
through a chronic cranial window. Nat Protoc 4:1128–1144. CrossRef
Keck T, Scheuss V, Jacobsen RI, Wierenga CJ, Eysel UT, Bonhoeffer T,
Hu ¨benerM (2011) Lossofsensoryinputcausesrapidstructuralchanges
of inhibitory neurons in adult mouse visual cortex. Neuron 71:869–882.
Knott GW, Holtmaat A, Wilbrecht L, Welker E, Svoboda K (2006) Spine
growth precedes synapse formation in the adult neocortex in vivo. Nat
Neurosci 9:1117–1124. CrossRef Medline
LaiCS,FrankeTF,GanWB (2012) Oppositeeffectsoffearconditioningand
extinction on dendritic spine remodelling. Nature 483:87–91. CrossRef
ListonC,CichonJM,JeanneteauF,JiaZ,ChaoMV,GanWB (2013) Circadian
glucocorticoid oscillations promote learning-dependent synapse formation
Merzenich MM, Jenkins WM (1994) Cortical representation of learned be-
haviors. Amsterdam: Elsevier.
Rumpel S (2013) Dynamics of dendritic spines in the mouse auditory
cortex during memory formation and memory recall. Proc Natl Acad Sci
U S A 110:18315–18320. CrossRef Medline
Mun ˜oz-Cuevas FJ, Athilingam J, Piscopo D, Wilbrecht L (2013) Cocaine-
induced structural plasticity in frontal cortex correlates with conditioned
place preference. Nat Neurosci 16:1367–1369. CrossRef Medline
O’Connor DH, Peron SP, Huber D, Svoboda K (2010a) Neural activity in
barrel cortex underlying vibrissa-based object localization in mice. Neu-
ron 67:1048–1061. CrossRef Medline
O’Connor DH, Clack NG, Huber D, Komiyama T, Myers EW, Svoboda K
30:1947–1967. CrossRef Medline
O’Connor DH, Hires SA, Guo ZV, Li N, Yu J, Sun QQ, Huber D, Svoboda K
sory touch. Nat Neurosci 16:958–965. CrossRef Medline
BL, Littman DR, Gan WB (2013) Microglia promote learning-dependent
synapse formation through brain-derived neurotrophic factor. Cell 155:
Petreanu L, Mao T, Sternson SM, Svoboda K (2009) The subcellular orga-
nization of neocortical excitatory connections. Nature 457:1142–1145.
SvobodaK (2012) Activityinmotor-sensoryprojectionsrevealsdistrib-
uted coding in somatosensation. Nature 489:299–303. CrossRef Medline
Pologruto TA, Sabatini BL, Svoboda K (2003) ScanImage: flexible software
for operating laser scanning microscopes. Biomed Eng Online 2:13.
Trachtenberg JT, Chen BE, Knott GW, Feng G, Sanes JR, Welker E, Svoboda
K (2002) Long-term in vivo imaging of experience-dependent synaptic
plasticity in adult cortex. Nature 420:788–794. CrossRef Medline
WiestMC,ThomsonE,PantojaJ,NicolelisMA (2010) ChangesinS1neural
responses during tactile discrimination learning. J Neurophysiol 104:
300–312. CrossRef Medline
Xu T, Yu X, Perlik AJ, Tobin WF, Zweig JA, Tennant K, Jones T, Zuo Y
ing motor memories. Nature 462:915–919. CrossRef Medline
Yang G, Pan F, Gan WB (2009) Stably maintained dendritic spines are as-
sociated with lifelong memories. Nature 462:920–924. CrossRef Medline
Kuhlman,O’Connoretal.•StructuralPlasticitywithintheBarrelCortexJ.Neurosci.,April23,2014 • 34(17):6078–6083 • 6083