Steady or changing? Long-term monitoring of neuronal population activity

TINS-980; No. of Pages 10

Review

Steady or changing? Long-term monitoring of neuronal population activity Henry Lu¨tcke, David J. Margolis, and Fritjof Helmchen Brain Research Institute, University of Zurich, Zurich, Switzerland

Stability and flexibility are both hallmarks of brain function that allow animals to thrive in ever-changing environments. Investigating how a balance between these opposing features is achieved with a dynamic array of cellular and molecular constituents requires long-term tracking of activity from individual neurons. Here, we review in vivo chronic extracellular recording studies and recent long-term two-photon calcium-imaging investigations that address the question of stability and plasticity of neuronal population activity in the mammalian brain. Overall, spiking activity is heterogeneously distributed among neurons in local populations and largely remains stable for individual cells over time. Tuning properties appear more flexible and may be adaptively stabilized, possibly by neuromodulators, to encode reliably and specifically salient stimuli or behaviors.

populations of neurons maintain stable activity patterns over extended time periods or adjust their processing capabilities in the face of changing environmental demands? Surprisingly, how and to what degree the stability of neuronal activity is balanced with plasticity in brain circuitry has been largely ignored in the past, mainly because addressing these challenges necessarily requires longitudinal activity measurements from the same neurons and populations under in vivo conditions, which has been difficult to achieve using classical neurophysiological techniques. Here, we review convergent evidence from chronic electrophysiological recordings and recent longitudinal two-photon calcium-imaging studies (see Glossary) that have tracked the activity of the same neurons and started to reveal how the trade-off between stability and plasticity in neuronal processing may be resolved.

The trade-off between stability and flexibility Information processing in the brain relies on coordinated activity within large networks of neurons. In particular, dynamic activity patterns in neuronal circuits of the cortex are thought to underlie sensory perception, encode longterm memories, and generate decisions as well as motor commands in mammals [1,2]. Neuronal circuits have to achieve an ongoing trade-off between stable operating regimes on the one hand and plastic adaptation to changing environmental demands on the other hand. Consider, for example, a hypothetical network of neurons specialized for recognizing the face of a close relative or friend. Over the course of weeks to months, these neurons presumably provide a stable representation that mediates recognition, while over even longer time periods, activity patterns need to adjust to changes in the person’s appearance, among other things. This achievement is not trivial because individual neurons remain integrated in cortical circuitry for several decades, whereas their molecular constituents, such as ion channels and neurotransmitter receptors, turn over at much higher rates [3]. Moreover, in vivo long-term structural imaging has shown that synaptic elements and, to a lesser extent axons and dendrites, can be highly dynamic even in adult animals [4]. These observations pose an elementary question: to what extent do individual neurons and

Longitudinal monitoring of neuronal activity To date, neurophysiologists have relied largely on extracellular recording of action potentials (APs) to link behavior and neuronal activity. With single- and multi-unit recordings, individual neurons or clusters of neurons can be isolated and stably recorded for several hours. Longterm measurements (over at least several days) from the same neuron or population of neurons are rarely performed because: (i) animals are sacrificed after the recording

Corresponding author: Lu¨tcke, H. ([email protected]). 0166-2236/$ – see front matter ß 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tins.2013.03.008

Glossary Calcium imaging: measurement of neuronal activity (mainly APs) by optically recording changes in intracellular calcium concentration using calciumsensitive fluorescent dyes. Calcium imaging is most frequently used for measuring the spiking activity of a local cortical microcircuit comprising tens to hundreds of neurons in anesthetized and awake animals. Extracellular recording: recording of APs from single neurons, or clusters of neurons, using electrodes implanted in the target brain area. Putatively different neurons may be distinguished by analyzing the AP waveform (spike sorting). Genetically encoded calcium indicator (GECI): fluorescent proteins (e.g., variants of GFP) that are linked to a calcium-sensor domain (e.g., calmodulin). Changes in calcium concentration lead to conformational rearrangements and corresponding changes in fluorescence properties that can be imaged, for example, by two-photon microscopy. GECIs can be stably expressed in neurons for long time periods, for example using viral techniques, thereby allowing chronic activity measurements from the exact same neurons. Two-photon microscopy: a nonlinear optical technique for minimally invasive imaging of fluorescently labeled cells or subcellular compartments deep inside scattering brain tissue, thereby allowing visualization of neuronal structure and function in the living animal (in vivo).

Trends in Neurosciences xx (2013) 1–10

1

TINS-980; No. of Pages 10

Review

Trends in Neurosciences xxx xxxx, Vol. xxx, No. x

session; (ii) the recording electrode is inserted de novo for subsequent recording sessions, making it unlikely that the exact same neurons can be found again; and (iii) chronically implanted electrodes undergo slight movements between recordings sessions, changing the cluster of isolated units. These technical limitations have mostly necessitated cross-sectional comparisons between groups of animals or at least different neurons, to assess changes in neural coding over time or across different experimental manipulations. Although technically difficult, longitudinal monitoring of the exact same populations of neurons offers several benefits and, therefore, has remained a major challenge that neuroscientists have continued to tackle in recent years. Longitudinal designs provide greater statistical efficiency, because between-neuron variances partial out (Box 1). Eliminating the variation between neurons

Box 1. Analysis approaches for chronic neuronal recordings Repeated measurements from identical neurons offer several advantages over traditional cross-sectional research designs. By sampling the same cells repeatedly, longitudinal approaches are statistically more powerful because they exclude time-invariant differences among neurons that remain unobserved in crosssectional designs. In statistical terms, repeated measurements allow for a partition of overall variation into ‘between-subject’ and ‘withinsubject’ variances s2 (subjects here are neurons) (Equation I): s 2error ¼ s 2between þ s 2within :

[I]

Intuitively, this partition corresponds to subtracting the response mean of each neuron across all conditions. If between-neuron variance is large, for example because neuronal responses are highly heterogeneous in the population, repeated measurements from the same cells under different conditions are more suited to reveal small, systematic changes due to treatment. In effect, each neuron becomes its own control condition. Repeated measurements also allow investigation of the temporal order of events, for example during processes that involve gradual changes, such as memory acquisition. Although long-term monitoring of neuronal activity offers significant advantages over traditional cross-sectional approaches, it also poses challenges for statistical data analysis, especially when recording (noisy) calcium signals from tens to hundreds of neurons. Correspondingly, several studies have described long-term (in)stability of neuronal responses in largely qualitative terms [22,27,71]. A simple quantification of response reliability can be obtained by comparing parameter variability for identical neurons across sessions with variability between different neurons [10]. Similarly, correlation analysis of neuronal responses from different sessions provides a standardized method for analyzing response stability in the framework of linear regression [42]. Alternatively, the stability of the relation between neural activity and behavior may be quantified by decoding aspects of behavior (e.g., movement parameters) from neuronal responses [41,58,72]. Decoding approaches use statistical learning algorithms to determine a mapping between neuronal activity and behavioral parameters. To test the generalizability of the derived model and to avoid overfitting, the mapping is subsequently applied to an independent test data set. Better-than-chance prediction on the test data indicates that behavioral parameters are presumably encoded in neuronal activity. Decoding approaches have the advantage that they can be easily applied to single-neuron as well as population data. Decoding approaches lend themselves to the analysis of response stability in two different ways. First, by calculating a different statistical model for each recording session, one can correlate decoding performance over time, thus asking to what extent a neuron encodes stimulus or behavior on different sessions (see Figure 3F in main text). Alternatively, the behavior-to-activity mapping derived on one session may be applied to data from a different session, thereby directly testing the stability of neuronal representations. 2

is particularly relevant, given the heterogeneous, lognormal distribution of activity levels that has been observed in the neocortex in vivo [5]. Another practical advantage of long-term measurements from the same neurons is simply that extended measurement time is available, permitting the collection of more trials per neuron (or more extensive sampling of stimulus space or different behavioral conditions), thereby increasing statistical power. Finally, tracking the activity of individual neurons over time can reveal fundamental aspects of neuronal circuit dynamics that remain inaccessible to single time point or interleaved cross-sectional sampling (Figure 1). Thus, whereas conventional recording methods may show identical distributions of neuronal activity on different sessions, chronic measurements can discern different underlying scenarios (Figure 1A,B). Furthermore, history effects influencing the response of individual neurons to particular manipulations, for example by specific reorganization (Figure 1C), can only be studied using longitudinal approaches. Longitudinal measurements thus provide several methodological advantages over conventional cross-sectional study designs and, in addition, have the potential to yield new insights into neuronal network dynamics during baseline conditions and following experimental manipulations. In the following sections, we review recent attempts to achieve chronic recordings from neurons and populations, using both electrophysiological and imaging approaches. Chronic extracellular recordings from the same neurons Chronically implanted extracellular recording electrodes (Figure 2A) typically move somewhat relative to the surrounding brain tissue between successive recording sessions, precluding straightforward matching of recorded spike trains across days. Nevertheless, stable functional characteristics of cells have been noted (e.g., place-field activity in hippocampus [6]), suggesting that identical neurons can be recorded. Other studies have analyzed the similarity of AP waveforms recorded across days, arguing that near-identical waveforms imply that the same neuron has been recorded [7–9]. This intuitive approach is prone to errors [10] because different neurons nearby may be mistakenly thought to be identical if they have similar AP waveforms (false positive). Furthermore, AP waveforms of the same neuron may conceivably change between recording sessions, leading to false negatives. Analogous concerns apply to similarity measures of particular waveform features [11]. The distribution of waveform similarities for neurons known to be not identical can be estimated by moving the recording electrode [10] or comparing neurons recorded on different electrodes [12]. Based on the derived distribution, the probability of two waveforms originating from different neurons can be calculated (Figure 2C). A very low probability suggests that the same neuron has been recorded on different sessions. This approach was used to identify and track convincingly the activity of individual neurons in macaque visual cortex [10] and motor cortex [12] for up to 100 days. Limitations of extracellular recordings Although recent chronic extracellular recording studies come very close to ‘proving’ that the same neurons have

TINS-980; No. of Pages 10

Review

Trends in Neurosciences xxx xxxx, Vol. xxx, No. x

(A)

Session 1

Session 2

Stable

Session 2

High

Low

Low Low (B)

Session 1

High

High

Random

Session 2

High

Low

High

Low Low (C)

Session 1

High

High

Session 2

Reorganized

High

Low

High

Low Low

Low

Session 1

High

TRENDS in Neurosciences

Figure 1. Steady or changing activity in individual neurons may occur for identical stable distributions of population activity. (A) Hypothetical local population of neurons with sparse (log-normal) activity distribution (color coded) measured at two different time points. Individual neurons preserve activity levels from one recording session to the next. Correspondingly, the scatter plot shows stable neuronal activity (all data points fall on the unity line). Scatter plot inset: histogram of activity levels for this population (broken line: underlying log-normal distribution). Note that, for clarity, population maps only show a subset of cells from the scatter plots. (B) Same plots as in (A) but with activity levels randomly shuffled among neurons between recording sessions. (C) Same plots as in (A) but with activity levels reorganized, such that neurons with large responses lose responsiveness and vice versa. Note that population distribution of activity is identical for both sessions in all panels. Thus, longitudinal measurements from the same neurons are required to distinguish stable and unstable scenarios.

indeed been identified across sessions, these techniques still suffer from several shortcomings. First, because extraordinary mechanical stability is required, chronic recordings have mainly been achieved several months after electrode implantation, possibly following stabilization of electrodes by glial scars [13]. Experiments are therefore time consuming and limited to larger animals, especially primates. Furthermore, the number of isolated neurons is usually small and the frequency of stable recordings drops quickly with longer recording intervals (e.g., half of isolated units lost over the first week [14]; but see [12]). Because extracellular recordings have a bias for highly active neurons, chronic recordings can only be obtained from neurons that remain in the same ‘activity regime’. Silent neurons that become responsive or neurons ‘dropping out’ of the representation will be missed. Finally, although neurons are commonly distinguished as ‘regular spiking’ or ‘fast spiking’ [15], extracellular recordings also fail to address the large heterogeneity of excitatory and inhibitory cell classes in the cortex. Chronic two-photon calcium imaging Two-photon imaging with genetically encoded calcium indicators (GECIs) (Figure 2B) has recently emerged as a powerful alternative for chronic activity measurements

from individual neurons and populations, addressing many of the shortcomings associated with long-term extracellular recordings (Table 1). Here, we provide a brief discussion of technical issues related to the use of GECIs for chronic activity measurements (see also [16,17]). Over the past decade, two-photon calcium imaging has become the method of choice for the comprehensive analysis of local neuronal population activity in the intact nervous system, with high spatial and temporal resolution (reviewed in [18–20]). Calcium imaging with small molecule indicators is now an established tool for studying in vivo functional organization of local neuronal circuits in different species [18,21]. Although small-molecule calcium indicators have excellent sensitivity, they are poorly suited for chronic activity measurements due to difficulties in relabeling the same neurons (but see [22]). By contrast, protein-based GECIs can be expressed in neurons for long time periods (Figure 2D) using genetic techniques. The general blueprint for GECIs comprises one or two fluorescent proteins coupled to a calcium-sensitive domain (e.g., calmodulin; Table 2) [17]. Changes in the intracellular calcium concentration cause conformational rearrangements of the protein, leading to a change in fluorescence properties that is observed with two-photon or wide-field fluorescence microscopy [23–25]. In the mammalian brain, GECIs have 3

TINS-980; No. of Pages 10

Review

Trends in Neurosciences xxx xxxx, Vol. xxx, No. x

(A)

(B)

Microelectrodes

Objecve

Guide tubes Cyano-acrylate

Cover glass

Cement Skull Dura Pia L2/3

L5 (C)

Session 1

Session 2

(D)

Cell 1

Session 1

Session 2

30

30 5

5 Cell 2

0.1 mV

2 3

2 3

0.5 ms

80 Number of AP pairs

Cell 3

Cell 3

Cell 30

Cell 30

10% ΔR/R 1s Cell 2 0

0

Cell 5

Cell 2

Cell 5

15 AP waveform distance TRENDS in Neurosciences

Figure 2. Chronic recording of neuronal activity: electrophysiology and imaging. (A) Schematic of a chronic extracellular multi-electrode recording. A recording chamber is attached to the skull using dental cement and microelectrodes are inserted into the brain via guide tubes. (B) Schematic of a chronic preparation for two-photon imaging in the mouse neocortex. A glass window is implanted on top of the dura and attached to the skull with dental cement. Chronic two-photon imaging of the same populations of neurons expressing a genetically encoded calcium indicator (GECI) can be performed using a high-numerical aperture objective. (C) Simulation of spikes recorded extracellularly from two different neurons with distinct spike waveforms. Action potential (AP) shape was modeled as sum of two gamma functions with different parameters for the two neurons (plus a random jitter applied to every AP). Gray: single APs, red: average of 100. Bottom: Similarity of AP waveforms was calculated for AP pairs from the same neuron on different recording sessions (black) and from different neurons on the same recording session (blue). Similarity measure: Euclidean distance of AP waveforms [10]. Note that spikes from the same neuron on different sessions (black) are significantly more similar than spikes from different neurons (blue). (D) Twophoton images of neuronal population in mouse barrel cortex expressing the GECI YC3.60. The same neurons were identified on different recording sessions (spanning 1 week). Scale bar = 20 mm. Bottom: Calcium transients evoked by whisker stimulation (0.5 s, onset marked by arrow) for four neurons from the population depicted above. Single-trial responses in gray; mean of 25 trials in red. Adapted, with permission, from [42] (D).

Table 1. Properties of chronic extracellular recordings and two-photon calcium imaginga Cell identification Number of neurons Time frame of chronic recording from same neurons Recovery time post-implant Temporal resolution Cortical recording depth Species a

Chronic extracellular Ambivalent 10 Days–weeks

Chronic two-photon Unequivocal 10 – 100 Weeks–months

Months

Days

<1 ms Mainly infragranular

>100 ms Mainly supragranular Large, mainly primate Rodents

Both techniques offer largely complementary approaches for long-term monitoring of neuronal population activity.

4

been expressed using viral delivery (e.g., adeno-associated viruses, AAV) [24,26] or in utero electroporation [27]. Furthermore, several transgenic mouse lines expressing GECIs have been created [25,28–31]. In all cases, longterm optical access to the brain is achieved by implanting a chronic cranial window [32] or through a delicately thinned skull [33,34]. GECIs are characterized by three important properties that govern their suitability for chronic in vivo imaging of neuronal activity. First, their sensitivity reflects how faithfully indicators report spiking activity. Initially introduced more than a decade ago [35,36], GECIs have been extensively optimized to report sensitively in vivo even low numbers of APs [24,26,37–39]. Second, the time course of elementary, single AP-evoked calcium transients is important because calcium responses can be approximated as a convolution of the underlying AP train with the elementary transient. This time course is affected by

TINS-980; No. of Pages 10

Review

Trends in Neurosciences xxx xxxx, Vol. xxx, No. x

Table 2. Examples of available and recently applied GECIs as well as summary of their use for (chronic) in vivo imaging of neural activity in the rodent neocortexa Indicator YC3.60 [30] YC-Nano140 [38] D3cpV [90] TN-XXL [27] GCaMP3 [26] GCaMP5s [39] GCaMP6s [46] G-GECOs [43] R-GECO1 [43] RCaMP1.07 [44] GCaMP6/GCaMP8 [45]

Fluorophore ECFP/cpV ECFP/cpV ECFP/cpV Citrine/CFP cpGFP cpGFP cpGFP cpGFP mApple mApple cpGFP

in vivo Application Barrel cortex Barrel cortex Barrel cortex Visual cortex Motor cortex Visual cortex Visual cortex – – – –

in vivo Sensitivity >1–2 AP [24] 1–2 AP [89] 1 AP [37] Visual-evoked transients >3 APs Visual-evoked transients 1 AP – – – –

Chronic imaging (days) >100 [22,42] – – 19 >100 – 27 – – – –

a

Note that estimates of in vivo sensitivity are based on different experimental conditions and not directly comparable between studies. Unless otherwise stated, references for characterization results are the same as for the indicator in column 1.

calcium-binding kinetics and concentration of the indicator. Finally, and critically for long-term monitoring of neuronal activity, GECIs should be expressed stably in neurons with little impact on cellular physiology. Stable expression at relatively low levels is important because GECIs act as calcium buffers, which may interfere with cellular processes depending on indicator concentration [40]. Reassuringly, current evidence indicates that the expression of GECIs using AAVs does not alter basic cellular physiology [26], long-term potentiation [41], or sensory map plasticity [42]. Optimization of GECIs has yielded a growing arsenal of constructs that have recently enabled systematic studies of neural circuit function (Table 2). The first indicators to successfully leave the ‘proof-ofconcept’ phase and be used for chronic monitoring of neuronal activity in the mouse brain were GCaMP3 and Yellow Cameleon 3.60 (YC3.60). The newest generation of GECIs further extends the range of available tools in terms of sensitivity, dynamic range, color, and decay time [38,39,43–46], and we expect to see biological applications of these novel constructs soon. Compared with chronic electrophysiological recordings, long-term imaging with GECIs, combined with the development of novel two-photon imaging approaches [47], offers huge potential for more refined, specific, and comprehensive neural circuit dissection, most of which has barely been addressed at this stage. We predict that many of the current limitations will be tackled systematically in the foreseeable future, including the identification of neuronal subtypes [48–50], imaging beyond the current depth limit of two-photon microscopy [51,52], increases in the temporal resolution of measurements [53], and the extension of technologies to nonrodent species, including primates [21]. It is expected that chronic two-photon calcium imaging will soon become the prime tool for analyzing the long-term stability and plasticity of neuronal population activity under behaviorally relevant conditions. Long-term stability of neuronal activity in motor cortex Early efforts to characterize the stability or variability of neuronal activity underlying behavior have focused on chronic extracellular recordings in primate motor cortex (e.g., [54]), mainly to develop neural prosthetic devices that translate cortical motor commands into actions

(brain–machine interfaces, BMI) [11,55–58]. Clearly, the stability of the relation between behavior and neuronal activity is important for BMI design. Early evidence suggested that the activity of individual neurons in primate motor cortex was highly variable, even during relatively stereotypical movements, whereas ensemble activity was more stable [55]. By contrast, chronic multi-unit recordings (for up to 2 days) during a ‘center-out’ reaching task [59] revealed high single-unit stability, including firing rates and directional tuning [60]. The remaining neuronal fluctuations in this task were explained by subtle behavioral changes, suggesting that apparent discrepancies between studies (see also [56,57]) are due to small variations in the movement itself, for example because of changes in posture or muscle activation patterns. Clearly, when the movement itself changes, neuronal activity will differ even if the underlying mapping between activity and behavior remains stable. Stability of activity patterns may also be influenced by the amount of task training, with highly trained monkeys showing more stable neuronal activity [60] compared with ‘naı¨ve’ subjects [55], or when the task demands change frequently [57]. Stabilization of activity patterns by motor learning The hypothesis that behavioral training stabilizes neuronal activity has been directly evaluated in two recent studies using chronic multi-unit recordings in monkeys controlling a neuroprosthetic device [58] and long-term two-photon calcium imaging in mice performing an object localization task [41], respectively (Figure 3). Ganguly and Carmena [58] trained monkeys to control a BMI based on multi-unit activity recorded in primary motor cortex over a time period of up to 19 days (Figure 3A). BMIs eliminate confounding variations in movement patterns because the recorded neuronal output is used directly for motor control. Interestingly, neuronal tuning (preferred movement direction) and ensemble activity patterns were labile over the first few days while the monkeys learned to control the BMI. As proficiency increased, tuning and population activity became highly reproducible across recording sessions (Figure 3B,C). Consolidation of activity was furthermore accompanied by increased mean firing rates in task-modulated neurons. Importantly, tuning stabilization of individual neurons and ensembles was closely 5

TINS-980; No. of Pages 10

Review

Trends in Neurosciences xxx xxxx, Vol. xxx, No. x

(B)

(A)

Day 3

Day 14

(C)

Day 18

15

Spikes

Neuron

BMI Predicted posion

Day 9–19

1

-1 1

Decoder -1

1 0°

(E)

(D)

360°

0°

360°

(F)

Head fixaon

Sessions

16x NA 0.8

Session 1

1

Two-photon microscope

2 3 4 5

Session 5

6

Water reward Nogo

Pole

Go

R Licking Whisking Touch Mixed

2

0.71 0.57 0.61 0.71

TRENDS in Neurosciences

Figure 3. Long-term stability of neuronal activity in motor cortex. (A) Schematic of a monkey performing a center-out task under brain control (BMI, brain machine interface). Cursor position on the screen (red dot) is determined from chronically recorded action potentials by a neural decoder (trained during manual control). (B) Colorcoded maps of preferred movement direction (PD) during brain control for an ensemble of 15 chronically recorded neurons (warmer colors indicate higher firing rates). As the monkey learnt to perform the brain control task, ensemble PD tuning became more stable, as indicated by higher map similarity from Day 14 to 18 (R = 0.9) compared with Day 3 to 14 (R = 0.5). (C) Overlay of PD tuning curves recorded during ten consecutive daily sessions of brain control for two units (top and bottom). Note the remarkable stability of PD tuning. (D) Schematic of a head-fixed mouse performing a whisker-based object-localization task while neuronal activity is imaged in motor cortex with a two-photon microscope. Depending on pole position, the mouse has to lick for a water reward (Go trial) or refrain from licking (No go). (E) Example neuronal population in whisker-related motor cortex labeled with the genetically encoded calcium indicator GCaMP3 and imaged chronically for approximately 1 week. Arrows mark three example neurons identified on all sessions. Scale bar = 100 mm. (F) Modulation of single neurons (columns) by different behaviors (licking, whisking, etc.) over time. Behavioral modulation was determined using a decision tree-based classification algorithm and quantified as correlation between data and model (R2). Mice improved in task performance from session 1 (‘naı¨ve’) to 6 (‘expert’). Note that most neurons do not consistently encode one behavioral category across sessions, but also rarely switch category. Bottom: Averaged classification across sessions. Adapted, with permission, from [58] (A–C) and [41] (D–F).

tracked by improvements in task performance. Proficiency in BMI control markedly dropped when small fractions of neurons were excluded during closed-loop brain control, suggesting that motor learning involves the consolidation and activation of highly specific ensembles of neurons. Stabilization of neuronal activity during learning has also been observed in the mouse whisker-related motor cortex using chronic two-photon calcium imaging [41]. When mice performed a whisker-based object-detection task while neuronal activity was monitored chronically in populations of layer 2/3 motor cortex neurons expressing GCaMP3 (Figure 3E), it was observed that individual neurons encoded different aspects of the behavior (e.g., whisking or touch) but did so relatively inconsistently across days, even in task-proficient animals (Figure 3F). Specifically, neurons usually contributed to one behavioral class but frequently dropped out of this representation on several sessions, indicating that a redundant pool of neurons encodes different behaviors. Learning was furthermore associated with stabilization of ensemble activity patterns across days as well as increased task-related neuronal activity. Together, these findings are consistent with the idea that motor cortex controls behavior with a redundant pool of flexible neurons that randomly and slowly change their tuning properties under baseline conditions (see also [61]). Redundancy ensures that changes in single neuron 6

properties do not affect behavior, such that different states of the system are behaviorally equivalent. Learning-related changes occur on top of random background fluctuations, by systematic changes in neuronal tuning properties. If considering learning as an error minimization process, random and noisy background fluctuations may be beneficial to avoid becoming trapped in local minima of the error landscape [61]. Long-term stability of neuronal activity in sensory cortex Neurons in sensory areas of the cortical sheath vary their firing rates in response to the presentation of selected stimuli; for example, a tone played at a particular frequency or a bar flashed at a certain orientation. To what extents are these tuning properties stable or fluctuate over time? Although this is an important question for theories of sensory coding, memory formation, or object invariance, it was explored rarely until the recent development of chronic imaging. Long-term recordings from sensory cortical neurons are advantageous not only for assessing the stability of tuning properties, but also for tracking changes in the activity of individual neurons during cortical reorganization (e.g., learning or plasticity). Contrary to classical cross-sectional approaches, chronic recordings from the same neurons during cortical map plasticity provide time courses of underlying gradual changes and offer insights

TINS-980; No. of Pages 10

Review

Trends in Neurosciences xxx xxxx, Vol. xxx, No. x

(A)

(B)

Day 3

Day 1

(C)

Day 5

8

0

V1

0 0

(D)

40 80 Time (s)

(E) Cell 1 C2 C1

6

0

40 80 Time (s)

Day 1 Day 3 Day 5

6 4 2

0

Day 1 C1

ΔR/R (%)

6

0

Day 4

0

40 80 Time (s)

Smulus orientaon (F)

Day 7

C2

Sm 1s

Cell 30

25% ΔR/R

S1

12 Amp. session N+1

ΔR/R (%)

6

0 0 C2

C1

C2

C1

Cell number

30

ΔR/R (%)

12

1

12 Amp. session N

C2

0

SI session N+1

C1

0.5 0

-0.5 -0.5

0 0.5 SI session N

TRENDS in Neurosciences

Figure 4. Stability of neuronal activity in mouse sensory cortex investigated by chronic two-photon calcium imaging. (A) Schematic of mouse brain with primary visual cortex V1 and stimulus presentation. Differently oriented moving gratings were displayed on a screen contralateral to the V1 region imaged by two-photon microscopy. (B) Average calcium transients evoked by visual stimuli with different orientations measured in the same neuron on three consecutive imaging sessions. Chronic two-photon calcium imaging was performed in mouse V1 after in utero electroporation of the genetically encoded calcium indicator (GECI) TN-XXL. Scale bar = 10 mm. (C) Overlay of orientation-tuning function for neuron shown in (B). This neuron consistently preferred gratings with top-left orientation, indicating that tuning is relatively stable over 1 week. (D) Schematic of mouse brain with whisker-related somatosensory (‘barrel’) cortex. Two whiskers contralateral to the imaged brain region were alternately deflected in repeated imaging sessions. (E) Calcium transients for two example neurons evoked by wiggling the C1 or C2 whisker (0.5-s stimulus indicated below; single trials in gray; mean of 25 trials in color). Bottom: Mean DR/R plot for the entire population of 38 neurons (same population as in Figure 2D). Note that only a few neurons respond robustly to whisker stimulation over consecutive imaging sessions, whereas most cells show consistently low responsiveness. (F) Top: Correlation of average calcium transient amplitudes for both whiskers on one imaging session with the previous session for the population shown in (E). Bottom: Correlation of whisker preference expressed as selectivity index (SI, the relative difference in average calcium transients for whiskers 1 and 2) on one imaging session with the previous session. Whereas average calcium transient amplitudes are stable over days, whisker preference is considerably more flexible (compare with Figure 1). Reproduced, with permission, from [27] (B,C); adapted, with permission, from [42] (D–F).

into the factors determining cellular changes, such as baseline activity levels or tuning properties. In the macaque primary visual cortex, chronic tetrode recordings have been used to isolate and convincingly identify single units over several days [10]. Orientationtuning functions of the same neurons were significantly more similar over days than for different neurons recorded on the same tetrode, indicating that orientation tuning is stable at least for a few days in the primate visual cortex. Recent attempts to characterize long-term neuronal stability and plasticity have relied on chronic two-photon calcium imaging. In a proof-of-principle study, Mank et al. [27] expressed the GECI TN-XXL in mouse visual cortex neurons by in utero electroporation and measured orientation tuning of the same neurons repeatedly over a period of up to 3 weeks (Figure 4A–C). Similar to [10], orientation-tuning curves of three selected neurons were largely stable over the recording period. Andermann et al. [22] followed individual visual cortex neurons expressing YC3.60 over several weeks while mice performed a visual discrimination task. In both of these studies, although tuning properties [27] or task-related activity [22] of

individual neurons could be recorded for weeks, sparse expression levels precluded a comprehensive quantification of neuronal responses within the local populations. Quantitative estimates of the distribution of baseline response stability in the population, as well as changes in neuronal properties during learning or plasticity, require dense, repeated sampling of cellular activity from a large fraction of the population, as is possible with GECI labeling using AAV delivery under the synapsin promoter [24,26]. We recently used this approach to measure the stability and plasticity of neuronal population activity in anesthetized mouse barrel cortex (Figure 4C,D) [42]. During whisker stimulation, responses among neurons were highly heterogeneous, with only a few neurons showing the highest levels of activity, consistent with other reports [62–64]. The corresponding log-normal distribution of activity levels, indicating ‘sparse’ coding [5,65], remained stable for at least 1 week for both individual neurons and populations. Thus, specific subnetworks of rare high-responsive and stimulus-driven neurons are stably embedded in a large pool of relatively weakly responsive cells [66,67]. Whereas individual neurons maintained a characteristic 7

TINS-980; No. of Pages 10

Review regime of responsiveness over weeks, tuning properties, measured as whisker selectivity, were more variable, even across imaging sessions separated by only a few days (Figure 4D). Although local populations showed on average a weak but consistent preference for one whisker, compatible with their anatomical position in the barrel map, these stable population-tuning properties were maintained by surprisingly variable single neurons [42]. In summary, these results indicate that substantial differences in the persistence of neuronal tuning exist for different aspects of neuronal selectivity. Whereas whisker preference in somatosensory cortex is based on reliable population-level representations with relatively flexible individual neurons [42], cells in the visual cortex show more stable orientation tuning [10,27]. A possible explanation for these differences could be the overall strength of tuning: whereas L2/3 neurons in barrel cortex show generally weak overall whisker tuning [62,64], visual cortex pyramidal neurons have clear orientation preferences [49]. Indeed, simulations demonstrate that tuning instability in the barrel cortex is largely explained by weak whisker preference [42]. Furthermore, recent time-lapse electrophysiological recordings of barrel cortex neurons indicate that the strong preference for contra- or ipsi-lateral whisker stimulation is also highly stable over several days [68]. Additional long-term recordings in visual and other neocortical areas, especially the auditory cortex, will be required to clarify this important question. Long-term stability of neuronal activity in hippocampus Pyramidal neurons in the rodent hippocampus fire APs in a location-specific manner [69]. They are ‘place cells’, which form a hippocampal map of the surroundings of the animal and are thought to contribute to spatial memory. The role of place cells as an integral part of spatial memory relies on their remarkable stability, revealed by chronic recordings from putatively stable single units for up to 8 months while rats explored a 6-arm radial maze [6]. Unfortunately, similar recording stability is difficult to achieve in other species, notably mice. Surprisingly, much shorter (<1 day) chronic recordings in mice spontaneously exploring a novel environment revealed that place fields were generally unstable between sessions (separated by several hours) [70]. Indeed, place-field stability strongly depended on task difficulty, with mice engaged in a demanding spatial task showing reliable and stable place field formation. Similarly, in mice running on a linear track, hippocampal neurons monitored with chronic two-photon calcium imaging showed clear location-specific activity, which was, however, unstable for the majority of cells across imaging sessions spanning several days to weeks [71,72]. Although only a small fraction of neurons (approximately 15%) retained stable place fields, spatial location could be reliably decoded from neuronal ensemble activity over several weeks, indicating that a stable representation of space is formed at the population level [72]. Place-field stability is enhanced by dopamine agonists and reduced by dopamine antagonists [70], suggesting that consolidation of neuronal representations in the hippocampus relies on active engagement with a novel spatial environment, which triggers attentional processes and, via neuromodulation, leads to 8

Trends in Neurosciences xxx xxxx, Vol. xxx, No. x

long-term synaptic changes. Similar mechanisms may underlie response stabilization in the neocortex [73–75], although the precise pathways and neuromodulators remain to be defined. Concluding remarks How stable is neuronal activity in the cortex? Initially, progress to answer this question was hindered by technical limitations of recording chronically from the same neurons, but recent advances in long-term extracellular recordings and two-photon calcium imaging have led to new insights that allow some tentative general conclusions. Concerning the overall level of activity in local cortical populations, there seems to be little doubt that the heterogeneous, lognormal distribution of spiking output in supragranular layers (reviewed in [5]) remains largely stable. Emerging evidence suggests that, in particular, the sensory cortex contains subnetworks of highly active neurons embedded in a sea of relatively unresponsive cells [63,76] and that this functional architecture is largely preserved over days to weeks [42]. The identity of these highly active neurons and their importance for sensory processing and behavior remains to be resolved. These findings are in agreement with evidence from in vitro studies showing that cultured hippocampal and neocortical neurons homeostatically scale synaptic efficiency to maintain activity levels near a defined set point [77,78]. Using chronic two-photon calcium imaging, we recently demonstrated that layer 2/3 neurons in the mouse barrel cortex similarly maintain a characteristic responsiveness to whisker stimulation, suggesting that homeostatic regulation of neuronal activity levels also occurs in vivo [42]. Similar evidence for in vivo homeostasis of neuronal activity has also been obtained in other systems, including visual cortex [79] and hippocampus [80]. Long-term stability of neuronal tuning and, therefore, the encoding of specific stimuli or behaviors remain more contentious, likely depending on various factors. Evidence from several species and systems indicates that learning and attention stabilizes neuronal tuning properties [41,58,70]. Interestingly, practicing a challenging motor skill also stabilizes dendritic spines of pyramidal neurons in mouse motor cortex [81,82]. In the hippocampus, stabilization of place-cell properties critically depends on dopamine [70]. Given that dopamine also enhances long-term potentiation in other brain areas, including the neocortex [83], and serves a critical function in reward-based learning [84], dopaminergic neuromodulatory influences may play an important role in long-term stabilization of neuronal function [74,85,86]. The precise mechanisms, pathways, and contributions by other neuromodulators remain to be identified. Further factors accounting for differences in the long-term stability of neuronal properties may be related to particularities of the recording techniques. Thus, compared with imaging studies, chronic extracellular recordings tend to report lower variability, possibly because they oversample highly active pyramidal neurons in infragranular cortical layers and the tuning properties of highly active cells are more stable [42]. Imaging of deep-layer neurons [51] or their subcellular compartments [87,88] could help to clarify this issue.

TINS-980; No. of Pages 10

Review Box 2. Outstanding questions  Does response stability differ between excitatory neurons and various classes of inhibitory interneuron? Likewise, do pyramidal cells projecting to different target regions, for example subcortical nuclei or association areas, show different degrees of response stability? These questions can now be addressed by combining chronic calcium imaging with transgenic [91–93] or post hoc labeling [48,49] techniques and tracing of projections.  How stable are specific aspects of neuronal activity, such as firing rate or spike timing [94], that cannot be resolved with current calcium-imaging methodologies? Addressing these issues will require high-speed microscopy [53] together with the development of even more sensitive GECIs that can report every AP. Alternatively, electrophysiological and optical techniques may be combined to achieve this goal, for example by repeated patchclamp recordings from fluorescently labeled neurons [68] or optogenetic excitation of specific cell types combined with extracellular recordings.  Does the observed day-to-day variability in neuronal activity reflect true internal variability of cortical circuits or rather ongoing changes in the environment, sensory stimulation, or animal behavior?  What are the mechanisms and neuromodulators mediating the effect of learning on stabilization of neuronal activity?  How is the trade-off between response stability and flexibility altered under pathological conditions, notably progressive neurodegenerative disorders?

In summary, chronic recording of neuronal activity with electrodes and, more recently, two-photon microscopes has just started to reveal some fundamental principles underlying the long-term stability or instability of response properties of individual neurons in the mammalian neocortex, highlighting in particular the role of learning in stabilization of activity. With the availability of optical techniques that have the power to unequivocally resolve hundreds of neurons in a cell-type specific manner across time, upcoming studies will soon be able to address many remaining questions in more detail (Box 2). Thus, our emerging view on the balance between stability and plasticity in neuronal function is bound to change and undergo further refinement in the years to come. Acknowledgments This work was supported by a grant from the Forschungskredit of the University of Zurich to H.L.; an AMBIZIONE grant from the Swiss National Science Foundation (SNSF) to D.J.M.; an SNSF grant to F.H. (3100A0-114624); the EU-FP7 program (PLASTICISE project 223524 to F.H., and BRAIN-I-NETS project 243914 to F.H.); and a grant from the Swiss SystemsX.ch initiative (project 2008/2011-Neurochoice) to F.H.

References 1 Hebb, D. (1949) The Organization of Behavior: A Neuropsychological Theory. John Wiley & Sons 2 Buzsa´ki, G. (2010) Neural syntax: cell assemblies, synapsembles, and readers. Neuron 68, 362–385 3 Ehlers, M.D. (2003) Activity level controls postsynaptic composition and signaling via the ubiquitin-proteasome system. Nat. Neurosci. 6, 231–242 4 Holtmaat, A. and Svoboda, K. (2009) Experience-dependent structural synaptic plasticity in the mammalian brain. Nat. Rev. Neurosci. 10, 647–658 5 Barth, A.L. and Poulet, J.F.A. (2012) Experimental evidence for sparse firing in the neocortex. Trends Neurosci. 35, 345–355 6 Thompson, L.T. and Best, P.J. (1990) Long-term stability of the placefield activity of single units recorded from the dorsal hippocampus of freely behaving rats. Brain Res. 509, 299–308

Trends in Neurosciences xxx xxxx, Vol. xxx, No. x

7 Greenberg, P.A. and Wilson, F.A.W. (2004) Functional stability of dorsolateral prefrontal neurons. J. Neurophysiol. 92, 1042–1055 8 Rousche, P.J. and Normann, R.A. (1998) Chronic recording capability of the Utah Intracortical Electrode Array in cat sensory cortex. J. Neurosci. Methods 82, 1–15 9 Schmitzer-Torbert, N. and Redish, A.D. (2004) Neuronal activity in the rodent dorsal striatum in sequential navigation: separation of spatial and reward responses on the multiple T task. J. Neurophysiol. 91, 2259–2272 10 Tolias, A.S. et al. (2007) Recording chronically from the same neurons in awake, behaving primates. J. Neurophysiol. 98, 3780–3790 11 Nicolelis, M.A.L. et al. (2003) Chronic, multisite, multielectrode recordings in macaque monkeys. Proc. Natl. Acad. Sci. U.S.A. 100, 11041–11046 12 Fraser, G.W. and Schwartz, A.B. (2012) Recording from the same neurons chronically in motor cortex. J. Neurophysiol. 107, 1970–1978 13 Jackson, A. and Fetz, E.E. (2007) Compact movable microwire array for long-term chronic unit recording in cerebral cortex of primates. J. Neurophysiol. 98, 3109–3118 14 Dickey, A.S. et al. (2009) Single-unit stability using chronically implanted multielectrode arrays. J. Neurophysiol. 102, 1331–1339 15 Connors, B.W. and Gutnick, M.J. (1990) Intrinsic firing patterns of diverse neocortical neurons. Trends Neurosci. 13, 99–104 16 Looger, L.L. and Griesbeck, O. (2012) Genetically encoded neural activity indicators. Curr. Opin. Neurobiol. 22, 18–23 17 Tian, L. et al. (2012) Neural activity imaging with genetically encoded calcium indicators. Prog. Brain Res. 196, 79–94 18 Grienberger, C. and Konnerth, A. (2012) Imaging calcium in neurons. Neuron 73, 862–885 19 Lu¨tcke, H. and Helmchen, F. (2011) Two-photon imaging and analysis of neural network dynamics. Rep. Prog. Phys. 74, 086602 20 Kerr, J.N.D. and Denk, W. (2008) Imaging in vivo: watching the brain in action. Nat. Rev. Neurosci. 9, 195–205 21 Nauhaus, I. et al. (2012) Orthogonal micro-organization of orientation and spatial frequency in primate primary visual cortex. Nat. Neurosci. 15, 1683–1690 22 Andermann, M.L. et al. (2010) Chronic cellular imaging of mouse visual cortex during operant behavior and passive viewing. Front. Cell. Neurosci. 4, 3 23 Minderer, M. et al. (2012) Chronic imaging of cortical sensory map dynamics using a genetically encoded calcium indicator. J. Physiol. 590, 99–107 24 Lu¨tcke, H. et al. (2010) Optical recording of neuronal activity with a genetically-encoded calcium indicator in anesthetized and freely moving mice. Front. Neural Circuits 4, 9 25 Hasan, M.T. et al. (2004) Functional fluorescent Ca2+ indicator proteins in transgenic mice under TET control. PLoS Biol. 2, e163 26 Tian, L. et al. (2009) Imaging neural activity in worms, flies and mice with improved GCaMP calcium indicators. Nat. Methods 6, 875–881 27 Mank, M. et al. (2008) A genetically encoded calcium indicator for chronic in vivo two-photon imaging. Nat. Methods 5, 805–811 28 Zariwala, H.A. et al. (2012) A Cre-dependent GCaMP3 reporter mouse for neuronal imaging in vivo. J. Neurosci. 32, 3131–3141 29 Heim, N. et al. (2007) Improved calcium imaging in transgenic mice expressing a troponin C-based biosensor. Nat. Methods 4, 127–129 30 Nagai, T. et al. (2004) Expanded dynamic range of fluorescent indicators for Ca(2+) by circularly permuted yellow fluorescent proteins. Proc. Natl. Acad. Sci. U.S.A. 101, 10554–10559 31 Chen, Q. et al. (2012) Imaging Neural Activity Using Thy1-GCaMP Transgenic Mice. Neuron 76, 297–308 32 Holtmaat, A. et al. (2009) Long-term, high-resolution imaging in the mouse neocortex through a chronic cranial window. Nat. Protoc. 4, 1128–1144 33 Drew, P.J. et al. (2010) Chronic optical access through a polished and reinforced thinned skull. Nat. Methods 7, 981–984 34 Grutzendler, J. et al. (2002) Long-term dendritic spine stability in the adult cortex. Nature 420, 812–816 35 Romoser, V.A. et al. (1997) Detection in living cells of Ca2+-dependent changes in the fluorescence emission of an indicator composed of two green fluorescent protein variants linked by a calmodulin-binding sequence. A new class of fluorescent indicators. J. Biol. Chem. 272, 13270–13274 36 Miyawaki, A. et al. (1997) Fluorescent indicators for Ca2+ based on green fluorescent proteins and calmodulin. Nature 388, 882–887 9

TINS-980; No. of Pages 10

Review 37 Wallace, D.J. et al. (2008) Single-spike detection in vitro and in vivo with a genetic Ca2+ sensor. Nat. Methods 5, 797–804 38 Horikawa, K. et al. (2010) Spontaneous network activity visualized by ultrasensitive Ca(2+) indicators, yellow Cameleon-Nano. Nat. Methods 7, 729–732 39 Akerboom, J. et al. (2012) Optimization of a GCaMP calcium indicator for neural activity imaging. J. Neurosci. 32, 13819–13840 40 Helmchen, F. and Tank, D.W. (2011) A single-compartment model of calcium dynamics in nerve terminals and dendrites. In Imaging in Neuroscience: A Laboratory Manual (Helmchen, F. and Konnerth, A., eds), pp. 355–368, Cold Spring Harbor Laboratory Press 41 Huber, D. et al. (2012) Multiple dynamic representations in the motor cortex during sensorimotor learning. Nature 484, 473–478 42 Margolis, D.J. et al. (2012) Reorganization of cortical population activity imaged throughout long-term sensory deprivation. Nat. Neurosci. 15, 1539–1546 43 Zhao, Y. et al. (2011) An expanded palette of genetically encoded Ca2+ indicators. Science 333, 1888–1891 44 Ohkura, M. et al. (2012) An improved genetically encoded red fluorescent Ca2+ indicator for detecting optically evoked action potentials. PLoS ONE 7, e39933 45 Ohkura, M. et al. (2012) Genetically encoded green fluorescent ca(2+) indicators with improved detectability for neuronal ca(2+) signals. PLoS ONE 7, e51286 46 Chen, T-W. et al. (2012) Reliable detection of single action potentials and synaptic calcium signals using improved genetically-encoded calcium indicators. In 2012 Neuroscience Meeting Planner, Program number 927.08/FFF77, Society for Neuroscience 47 Grewe, B.F. and Helmchen, F. (2009) Optical probing of neuronal ensemble activity. Curr. Opin. Neurobiol. 19, 520–539 48 Langer, D. and Helmchen, F. (2012) Post hoc immunostaining of GABAergic neuronal subtypes following in vivo two-photon calcium imaging in mouse neocortex. Pflugers Arch. 463, 339–354 49 Kerlin, A.M. et al. (2010) Broadly tuned response properties of diverse inhibitory neuron subtypes in mouse visual cortex. Neuron 67, 858– 871 50 Wilson, N.R. et al. (2012) Division and subtraction by distinct cortical inhibitory networks in vivo. Nature 488, 343–348 51 Mittmann, W. et al. (2011) Two-photon calcium imaging of evoked activity from L5 somatosensory neurons in vivo. Nat. Neurosci. 14, 1089–1093 52 Howard, S.S. et al. (2013) Frequency multiplexed in vivo multiphoton phosphorescence lifetime microscopy. Nat. Photonics 7, 33–37 53 Grewe, B.F. et al. (2010) High-speed in vivo calcium imaging reveals neuronal network activity with near-millisecond precision. Nat. Methods 7, 399–405 54 Schmidt, E.M. et al. (1976) Long-term chronic recording from cortical neurons. Exp. Neurol. 52, 496–506 55 Carmena, J.M. et al. (2005) Stable ensemble performance with singleneuron variability during reaching movements in primates. J. Neurosci. 25, 10712–10716 56 Musallam, S. et al. (2004) Cognitive control signals for neural prosthetics. Science 305, 258–262 57 Taylor, D.M. et al. (2002) Direct cortical control of 3D neuroprosthetic devices. Science 296, 1829–1832 58 Ganguly, K. and Carmena, J.M. (2009) Emergence of a stable cortical map for neuroprosthetic control. PLoS Biol. 7, e1000153 59 Georgopoulos, A.P. et al. (1982) On the relations between the direction of two-dimensional arm movements and cell discharge in primate motor cortex. J. Neurosci. 2, 1527–1537 60 Chestek, C.A. et al. (2007) Single-neuron stability during repeated reaching in macaque premotor cortex. J. Neurosci. 27, 10742–10750 61 Rokni, U. et al. (2007) Motor learning with unstable neural representations. Neuron 54, 653–666 62 Kerr, J.N.D. et al. (2007) Spatial organization of neuronal population responses in layer 2/3 of rat barrel cortex. J. Neurosci. 27, 13316–13328 63 Yassin, L. et al. (2010) An embedded subnetwork of highly active neurons in the neocortex. Neuron 68, 1043–1050 64 Sato, T.R. et al. (2007) The functional microarchitecture of the mouse barrel cortex. PLoS Biol. 5, e189 65 Hroma´dka, T. et al. (2008) Sparse representation of sounds in the unanesthetized auditory cortex. PLoS Biol. 6, e16

10

Trends in Neurosciences xxx xxxx, Vol. xxx, No. x

66 Shoham, S. et al. (2006) How silent is the brain: is there a ‘dark matter’ problem in neuroscience? J. Comp. Physiol. A: Neuroethol. Sens. Neural. Behav. Physiol. 192, 777–784 67 Brecht, M. et al. (2005) Silent neurons in sensorimotor cortices: Implications for cortical plasticity. In Neural Plasticity in Adult Somatic Sensory-Motor Systems (Ebner, F.F., ed.), pp. 1–19, CRC Press 68 Cohen, L. et al. (2013) Time-lapse electrical recordings of single neurons from the mouse neocortex. Proc. Natl. Acad. Sci. U.S.A. 110, 5665–5670 69 O’Keefe, J. and Dostrovsky, J. (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Res. 34, 171–175 70 Kentros, C.G. et al. (2004) Increased attention to spatial context increases both place field stability and spatial memory. Neuron 42, 283–295 71 Dombeck, D.A. et al. (2010) Functional imaging of hippocampal place cells at cellular resolution during virtual navigation. Nat. Neurosci. 13, 1433–1440 72 Ziv, Y. et al. (2013) Long-term dynamics of CA1 hippocampal place codes. Nat. Neurosci. 16, 1–5 73 Molina-Luna, K. et al. (2009) Dopamine in motor cortex is necessary for skill learning and synaptic plasticity. PLoS ONE 4, e7082 74 Bao, S. et al. (2001) Cortical remodelling induced by activity of ventral tegmental dopamine neurons. Nature 412, 79–83 75 Froemke, R.C. et al. (2007) A synaptic memory trace for cortical receptive field plasticity. Nature 450, 425–429 76 Lefort, S. et al. (2009) The excitatory neuronal network of the C2 barrel column in mouse primary somatosensory cortex. Neuron 61, 301–316 77 Turrigiano, G.G. et al. (1998) Activity-dependent scaling of quantal amplitude in neocortical neurons. Nature 391, 892–896 78 Turrigiano, G.G. (2008) The self-tuning neuron: synaptic scaling of excitatory synapses. Cell 135, 422–435 79 Mrsic-Flogel, T.D. et al. (2007) Homeostatic regulation of eye-specific responses in visual cortex during ocular dominance plasticity. Neuron 54, 961–972 80 Goold, C.P. and Nicoll, R.A. (2010) Single-cell optogenetic excitation drives homeostatic synaptic depression. Neuron 68, 512–528 81 Fu, M. et al. (2012) Repetitive motor learning induces coordinated formation of clustered dendritic spines in vivo. Nature 483, 92–95 82 Xu, T. et al. (2009) Rapid formation and selective stabilization of synapses for enduring motor memories. Nature 462, 915–919 83 Otani, S. et al. (2003) Dopaminergic modulation of long-term synaptic plasticity in rat prefrontal neurons. Cereb. Cortex 13, 1251–1256 84 Schultz, W. et al. (1997) A neural substrate of prediction and reward. Science 275, 1593–1599 85 Rossato, J.I. et al. (2009) Dopamine controls persistence of long-term memory storage. Science 325, 1017–1020 86 Lisman, J.E. and Grace, A.A. (2005) The hippocampal-VTA loop: controlling the entry of information into long-term memory. Neuron 46, 703–713 87 Glickfeld, L.L. et al. (2013) Cortico-cortical projections in mouse visual cortex are functionally target specific. Nat. Neurosci. 16, 219–226 88 Xu, N. et al. (2012) Nonlinear dendritic integration of sensory and motor input during an active sensing task. Nature 492, 247–251 89 Chen, J.L. et al. (2012) Improved in vivo detection of single action potentials using yellow cameleon-nano indicators. In 2012 Neuroscience Meeting Planner, Program number 709.14/FFF11, Society for Neuroscience 90 Palmer, A.E. et al. (2006) Ca2+ indicators based on computationally redesigned calmodulin-peptide pairs. Chem. Biol. 13, 521–530 91 Taniguchi, H. et al. (2011) A resource of Cre driver lines for genetic targeting of GABAergic neurons in cerebral cortex. Neuron 71, 995– 1013 92 Madisen, L. et al. (2010) A robust and high-throughput Cre reporting and characterization system for the whole mouse brain. Nat. Neurosci. 13, 133–140 93 Sato, T.R. and Svoboda, K. (2010) The functional properties of barrel cortex neurons projecting to the primary motor cortex. J. Neurosci. 30, 4256–4260 94 Wolfe, J. et al. (2010) Sparse and powerful cortical spikes. Curr. Opin. Neurobiol. 20, 306–312