AndermannpersName, KerlinpersName, ReidpersName (2010date). “Chronic cellular imaging of mouse visual cortex during operant behavior and passive viewing.” Frontiers in Cellular Neurosciencetitle·j.
Object of study — how to record identified neuronsterm in visual cortex of an awaketerm mouseterm while it performs a visual discriminationterm·mth, without the behaviorterm itself corrupting the recording
Research question — why behavioral modulationterm of nearby neuronsterm is heterogeneousterm in magnitude and even in direction, which needs cell identityterm and repeated measurement of the same cells
Contribution — a GO-NOGOterm·mth discriminationterm·mth taskterm that head-fixedterm·mth mice run for hundreds of trialsterm a day over a year, and a three-dimensional imagingterm·mth method whose motionterm correction survives licking
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | behavioral modulation of visual cortical responses — “A poorly understood feature of most studies, however, is that behavioral modulation of responses in nearby neurons is heterogeneous, both in magnitude and even in sign.” | |
| claim-type | method — “Here, we describe a method for three-dimensional cellular imaging of neural activity in the awake mouse visual cortex during active discrimination and passive viewing of visual stimuli.” | |
| method — evidence | — “To record from multiple neurons during operant behavior with single-trial resolution and minimal artifacts, we built a sensitive microscope for two-photon calcium imaging, capable of rapid tracking of neurons in three dimensions.” | |
| object of study — evidence | — “To study how diverse elements within a local cortical microcircuit are engaged to optimize visual processing across changing task demands, we have developed a robust method for three-dimensional cellular imaging of mouse visual cortex during operant visual discrimination behavior.” | |
| research question — evidence | — “A poorly understood feature of most studies, however, is that behavioral modulation of responses in nearby neurons is heterogeneous, both in magnitude and even in sign.” | |
| contribution — evidence | — “Here, we describe a GO–NOGO visual discrimination task in head-fixed mice, suitable for cellular imaging studies of visual perception, perceptual learning, task engagement, and task difficulty. We show that mice will perform this task (or passively view the same stimuli) for several hundred trials a day, and we demonstrate quantitative monitoring of neural and behavioral responses across hours, days and months.” | |
| transfer | from primate behavioural neurophysiology: the head-fixed operant behavioral technique, which currently did not exist for visual discrimination in mice, and the comparison to visual behavior and neurophysiology in carnivores and primates — “The mouse is the pre-eminent mammalian model system for genetic labeling or manipulation of specific cortical cell classes (O'Connor et al., 2009 ). Cellular imaging in mice offers immense potential for understanding cortical function in healthy mice, as well as in mouse models of neurologic disease (e.g., Eichhoff et al., 2008 ; Kuchibhotla et al., 2009 ). While the imaging and behavioral techniques described here can be used in different cortical regions and using different sensory modalities, we focus on visual behaviors that allow direct comparison to the rich literature on visual behavior and neurophysiology in carnivores and primates. Indeed, electrophysiological studies in anesthetized (Dräger, 1975 ; Mangini and Pearlman, 1980 ; Metin et al., 1988 ; Niell and Stryker, 2008 ; Liu et al., 2009 ) and awake (Frenkel et al., 2006 ; Niell and Stryker, 2009) mice suggest that mouse visual cortical neurons demonstrate similar properties to those in carnivores and primates, despite an overall decrease in spatial acuity (Prusky and Douglas, 2004 ; Niell and Stryker, 2008 ). While freely moving mice have been shown to discriminate between oriented gratings (Reuter, 1987 ; Prusky and Douglas, 2004 ; Wong and Brown, 2006 ), no operant behavioral technique currently exists for visual discrimination in head-fixed mice.” | transfers[0] |
| transfer | from signal detection theory: the discriminability measure and its inverse-normal definition — “A useful measure of behavioral discriminability of targets from non-targets is given by d ′ = norminv(HR) − norminv(FA), where norminv() is the inverse of the cumulative normal function (Green and Swets, 1966 ). Values for HR and FA were truncated between 0.01 and 0.99. Higher d ′ values indicate better discriminability (maximum d ′ = 4.65).” | transfers[1] |
| transfer | from mouse genetics: transgenic labeling of an entire inhibitory class, which is what makes the recorded identity more than a waveform — “Excitatory neurons and inhibitory neurons could be classified in vivo in this mouse due to expression of green fluorescent protein in all GABAergic neurons (Figure 2 A, green neurons; Tamamaki et al., 2003 ; Sohya et al., 2007 ; Kerlin et al., 2009 ). An example of orientation-tuned responses to drifting gratings (28°/cycle square wave, 2 Hz) is shown in Figure 2 B. Responses could be detected in single trials in GABAergic and non-GABAergic neurons (Figure 2 C). These data illustrate the value of two-photon calcium imaging in awake, transgenic mice, providing robust responses in genetically identified cell types.” | transfers[2] |
| transfer | from astrocyte imaging: the red calcium-insensitive dye, repurposed here as a motion reference rather than as a cell marker — “A local volume was loaded with a green synthetic calcium indicator (OGB1-AM), and astrocytes were co-loaded with sulforhodamine (SR101; Nimmerjahn et al., 2004 ), a red, calcium-insensitive fluorescent dye that generates a constant average fluorescence level in the absence of brain motion” | transfers[3] |
| mechanism | Licking changes jaw position, which displaces the brain downward relative to the objective and produces the apparent sinking of imaged cells out of the focal plane — “The onset of a bout of licking appeared to generate an apparent sinking of the recorded neuron out of the plane of imaging (Figure 5 ), implying that changes in jaw position may be the cause of the motion artifact” | mechanisms[0] |
| mechanism | Lower behavioural salience of the initial visual stimuli, relative to odors, slows acquisition of the visual task — “These differences could be due to the greater behavioral salience of different odors as compared to initial visual stimuli employed in our task.” | mechanisms[1] |
| mechanism | this cell demonstrated up to 15 μm of sinking Z-displacement as the mouse adjusted his posture, opened his jaw and began to lick, which by monitoring fluorescence in a single plane appears as an artifactual effect of brain motion — “This cell demonstrated up to 15 μm of ‘sinking’ Z-displacement as the mouse adjusted his posture, opened his jaw and began to lick – a displacement that is well-corrected by 3D co-registration methods (Figure 5 B, bottom ; see also Figure S2B in Supplementary Material). By monitoring fluorescence in a single plane (Figure 5 A, orange square), one can observe artifactual effects of brain motion on SR101 signal that would occur using 2D imaging” | mechanisms[2] |
| mechanism | so the artifact is measured in a channel where the signal cannot exist - a calcium-insensitive red dye that should hold constant unless the brain moves — “A local volume was loaded with a green synthetic calcium indicator (OGB1-AM), and astrocytes were co-loaded with sulforhodamine (SR101; Nimmerjahn et al., 2004 ), a red, calcium-insensitive fluorescent dye that generates a constant average fluorescence level in the absence of brain motion” | mechanisms[3] |
| mechanism | brain motion artifacts not related to licking were mostly corrected using standard 2D motion correction techniques, however adequate correction of lick-related motion artifacts required 3D motion correction — “Brain motion artifacts not related to licking were mostly corrected using standard 2D motion correction techniques, (Figure 5 C, middle ), consistent with previous studies (Dombeck et al., 2007 ; Greenberg and Kerr, 2009 ). However, adequate correction of lick-related motion artifacts required 3D motion correction (Figure 5 C, bottom ). Further, 3D motion correction also reduced motion contamination not related to licking (Figure 5 C, histograms), likely caused by body motion, chewing, breathing, and the heart beat” | mechanisms[4] |
| mechanism | and the discriminability measure converts hit and false-alarm rates through the inverse normal, with both truncated so that the maximum is set by the truncation — “A useful measure of behavioral discriminability of targets from non-targets is given by d ′ = norminv(HR) − norminv(FA), where norminv() is the inverse of the cumulative normal function (Green and Swets, 1966 ). Values for HR and FA were truncated between 0.01 and 0.99. Higher d ′ values indicate better discriminability (maximum d ′ = 4.65).” | mechanisms[5] |
| extends | Göbel et al., 2007 (3D two-photon imaging by objective movement) — “To overcome this problem, we modified previous techniques (Göbel et al., 2007 ) to achieve rapid, trapezoidal vertical movement of the microscope objective” | claim_relations[0] |
| replicates | Reuter, 1987 (orientation discrimination thresholds in freely moving mice) — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice (Reuter, 1987 ).” | claim_relations[1] |
| critiques | the inference of cell type from extracellular spike waveform, as limited in scope and reliability — “Although considerable anatomical, neurochemical, and physiological diversity exists in nearby neurons, only a small number of in vivo electrophysiological studies have correlated behavioral modulation of neural responses with information about cell type (Swadlow and Weyand, 1987 ; Mitchell et al., 2007 ; Chen et al., 2008 ). These studies have all inferred cellular identity from extracellular spike waveforms, which has limited scope and reliability.” | claim_relations[2] |
| extends | two-photon imaging in awake head-fixed rodents, to sensory discrimination and passive viewing which had not been imaged — “Two-photon imaging of cellular and neuropil signals has recently been described in awake, head-fixed mice and rats (Dombeck et al., 2007 , 2009 ; Verhagen et al., 2007 ; Greenberg et al., 2008 ; Kuhn et al., 2008 ; Mukamel et al., 2009 ; for review, see Grewe and Helmchen, 2009 ). However, no published work has imaged cellular responses in awake, head-fixed mice during sensory discrimination or during passive viewing of visual stimuli.” | claim_relations[3] |
| builds-on | the trapezoidal objective movement technique for fast volumetric acquisition — “A major obstacle in our early efforts developing two-dimensional calcium imaging during behavior involved systematic, lick response-induced brain motion artifacts. To overcome this problem, we modified previous techniques (Göbel et al., 2007 ) to achieve rapid, trapezoidal vertical movement of the microscope objective (Figure 1 B, Section ‘ Materials and Methods ’), generating complete local volumes at sufficiently fast rates to correct for rapid brain motion (and slow drift) in three dimensions.” | claim_relations[4] |
| confirms | previously reported orientation discrimination thresholds of ten to twenty degrees in freely moving mice — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice (Reuter, 1987 ). Absolute orientation thresholds likely depend on incidence of presentation of each stimulus (Figure 4 B, legend), and task structure (e.g., GO–NOGO vs. 2-AFC).” | claim_relations[5] |
| qualifies | standard two-dimensional motion correction, which is adequate for everything except the motion caused by the response itself — “Brain motion artifacts not related to licking were mostly corrected using standard 2D motion correction techniques, (Figure 5 C, middle ), consistent with previous studies (Dombeck et al., 2007 ; Greenberg and Kerr, 2009 ). However, adequate correction of lick-related motion artifacts required 3D motion correction (Figure 5 C, bottom ). Further, 3D motion correction also reduced motion contamination not related to licking (Figure 5 C, histograms), likely caused by body motion, chewing, breathing, and the heart beat” | claim_relations[6] |
| distinguishes-from | imaging in freely moving animals, which trades stimulus control and trial count for naturalism — “Complementary methods allowing two-photon calcium imaging of visual responses and other calcium activity in freely moving rats and mice have recently been published (Engelbrecht et al., 2008 ; Flusberg et al., 2008 ; Sawinski et al., 2009 ). Imaging in freely moving animals may allow for easier implementation of naturalistic behaviors in the future, albeit with additional limitations in image quality, visual stimulus control, and number of trials per session as compared to head-fixed operant behaviors.” | claim_relations[7] |
| sense | THE artifact IS TIME-LOCKED TO THE BEHAVIOURAL VARIABLE OF INTEREST: lick-response-induced brain motion was the major obstacle, and licking happens at reward, which happens after the target — “A major obstacle in our early efforts developing two-dimensional calcium imaging during behavior involved systematic, lick response-induced brain motion artifacts. To overcome this problem, we modified previous techniques (Göbel et al., 2007 ) to achieve rapid, trapezoidal vertical movement of the microscope objective (Figure 1 B, Section ‘ Materials and Methods ’), generating complete local volumes at sufficiently fast rates to correct for rapid brain motion (and slow drift) in three dimensions.” | senses[0] |
| sense | and it is measured in a channel where the signal CANNOT exist - a calcium-INSENSITIVE dye that generates a constant fluorescence in the absence of brain motion — “A local volume was loaded with a green synthetic calcium indicator (OGB1-AM), and astrocytes were co-loaded with sulforhodamine (SR101; Nimmerjahn et al., 2004 ), a red, calcium-insensitive fluorescent dye that generates a constant average fluorescence level in the absence of brain motion” | senses[1] |
| sense | up to 15 micrometers of sinking as the animal opens its jaw, which a single plane would have recorded as an artifactual fluorescence change — “This cell demonstrated up to 15 μm of ‘sinking’ Z-displacement as the mouse adjusted his posture, opened his jaw and began to lick – a displacement that is well-corrected by 3D co-registration methods (Figure 5 B, bottom ; see also Figure S2B in Supplementary Material). By monitoring fluorescence in a single plane (Figure 5 A, orange square), one can observe artifactual effects of brain motion on SR101 signal that would occur using 2D imaging” | senses[2] |
| sense | THE CORRECTION DECIDES THE FINDING: standard two-dimensional correction removes the motion that is NOT lick-related, and only three-dimensional correction removes the motion that is — “Brain motion artifacts not related to licking were mostly corrected using standard 2D motion correction techniques, (Figure 5 C, middle ), consistent with previous studies (Dombeck et al., 2007 ; Greenberg and Kerr, 2009 ). However, adequate correction of lick-related motion artifacts required 3D motion correction (Figure 5 C, bottom ). Further, 3D motion correction also reduced motion contamination not related to licking (Figure 5 C, histograms), likely caused by body motion, chewing, breathing, and the heart beat” | senses[3] |
| sense | A CLASS ASSIGNED BY A PROXY FEATURE, CRITICISED BY NAME: the few studies that related behavioral modulation to cell type all inferred cellular identity from extracellular spike waveforms, WHICH HAS LIMITED SCOPE AND RELIABILITY — “Although considerable anatomical, neurochemical, and physiological diversity exists in nearby neurons, only a small number of in vivo electrophysiological studies have correlated behavioral modulation of neural responses with information about cell type (Swadlow and Weyand, 1987 ; Mitchell et al., 2007 ; Chen et al., 2008 ). These studies have all inferred cellular identity from extracellular spike waveforms, which has limited scope and reliability.” | senses[4] |
| sense | and the motivating problem is that the modulation of nearby neurons is heterogeneous IN MAGNITUDE AND EVEN IN DIRECTION, so a population average hides opposite effects — “A poorly understood feature of most studies, however, is that behavioral modulation of responses in nearby neurons is heterogeneous, both in magnitude and even in sign.” | senses[5] |
| sense | THE THRESHOLD IS NOT A PROPERTY OF THE ANIMAL ALONE: absolute orientation thresholds likely depend on the incidence of presentation of each stimulus, and on the task structure — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice (Reuter, 1987 ). Absolute orientation thresholds likely depend on incidence of presentation of each stimulus (Figure 4 B, legend), and task structure (e.g., GO–NOGO vs. 2-AFC).” | senses[6] |
| sense | and the ceiling of the discriminability measure is set by truncating the rates at 0.01 and 0.99, giving a maximum of 4.65 that belongs to the arithmetic and not to the mouse — “A useful measure of behavioral discriminability of targets from non-targets is given by d ′ = norminv(HR) − norminv(FA), where norminv() is the inverse of the cumulative normal function (Green and Swets, 1966 ). Values for HR and FA were truncated between 0.01 and 0.99. Higher d ′ values indicate better discriminability (maximum d ′ = 4.65).” | senses[7] |
| sense | THE DEMONSTRATION SESSION IS NOT THE TYPICAL SESSION, AND THE PAPER SAYS SO: the performance in the two imaging figures was below the trained level, in that mouse and even in that session — “While these data clearly illustrate the strength of our 3D imaging techniques for obtaining clean neural traces during active discrimination (including during lick responses), the behavioral performance demonstrated in Figures 6 A and 7 A was below the level achieved during training (Table 1 ), due to an increase in spurious licking compared to other mice (see Figure 3 B), and even compared to other imaging sessions from the same mouse” | senses[8] |
| sense | with about half the mice tested learning the task at all — “About half of the mice tested (13/25) learned to perform the visual discrimination task, but with variable learning rates. Training proceeded as follows: First, mice were rewarded for licking in response to drifting target stimuli, using an inter-stimulus interval that was gradually increased from 2 to 5–8 s. Subsequent training involved gradually increasing the similarity of target and non-target stimuli in 5–15 sessions by presenting non-target stimuli that were (1) orthogonal to target and stationary, (2) orthogonal and of increasingly similar speed to target, (3) of increasingly similar orientation to target gratings and of identical speed.” | senses[9] |
| sense | THE SAMPLED POPULATION DEPENDS ON THE LABEL: the synthetic dye loads every cell in a local volume, while the genetically encoded indicator labels only a subset — “The choice of whether to use synthetic dyes or GECIs depends on the research question at hand. For example, neurons bulk-loaded with OGB1-AM dye are typically brighter than neurons expressing GECIs, and thus can be imaged with less laser power and potentially less cell damage. Further, OGB1-AM loads every cell in a local volume (∼300 × 300 × 300 μm 3 ), thus providing greater yield than GECI experiments where only a subset of neurons are been labeled using viral or electroporation techniques.” | senses[10] |
| sense | and the animal's motivation is a controlled variable with the arithmetic shown - a 25 gram mouse at a 90% hit rate earns about 0.65 cc against a daily minimum of 0.625 — “Because the mean number of active discrimination trials per session was ∼300 (Table 1 ), and ∼40% of trials were target trials, then a 25-g mouse with a 90% hit rate (HR) would receive, on average, ∼0.65 cc of water during a session (minimum total daily water for a mouse of this weight: 0.625 cc).” | senses[11] |
| sense | A behaviorAL CONTROL THAT SHOWS THE ASSOCIATION IS NOT REFLEXIVE: mice stop licking when the spout is withdrawn, and the discrimination does not extinguish when it returns — “We varied task engagement using interleaved blocks of 43 trials in which mice either actively discriminated target and non-target stimuli, or passively viewed these same stimuli (Figure 3 A). During passive viewing, a solenoid removed the lickspout from reach (Figure 1 B), and mice quickly learned to stop licking in response to target stimuli until the following ‘engaged’ block began, suggesting that the learned stimulus association is not purely reflexive (Figure 3 C). Presentation of stimuli during passive viewing, or of novel stimuli during orientation mapping, did not lead to extinction of subsequent discrimination behavior (e.g., Figures 3 B,C and 4 D), allowing for comparison of neural responses obtained during behavior with receptive field mapping obtained during passive viewing.” | senses[12] |
| sense | and the same neurons give the same orientation preference during the behavior and during passive viewing, which is what licenses comparing the two — “The relative response strength of neurons to different stimulus orientations during behavior (Figures 6 A and 6 E, top left ) was consistent with orientation tuning curves obtained during passive viewing” | senses[13] |
| control | a separate group of nine mice run on the discrimination task for several months to over a year — held — rules out an imaging result that holds only for the days after surgery - the separate group establishes that the behaviour itself is stable over the chronic window the method claims — “In a separate group of nine mice, we found that mice could perform the discrimination task routinely for several months to over a year, and typically performed on the order of 300 trials/day (Figure 4 B, Table 1 and Section ‘ Materials and Methods ’; over 1000 trials or 2 h in some sessions).” | controls[0] |
| differs in | standard two-dimensional motion correction, which is adequate for everything except the motion caused by the response itself — THE AXIS IS WHETHER THE MOTION IS CAUSED BY THE ANIMAL'S OWN RESPONSE. 'Brain motion artifacts not related to licking were mostly corrected using standard 2D motion correction techniques ... However, adequate correction of lick-related motion artifacts required 3D MO[TION CORRECTION].' The standard method is adequate for everything except the motion the task itself produces - so the harder case is exactly the one an operant experiment generates, and a methods citation carries only '2D motion correction'. — “Brain motion artifacts not related to licking were mostly corrected using standard 2D motion correction techniques, (Figure 5 C, middle ), consistent with previous studies (Dombeck et al., 2007 ; Greenberg and Kerr, 2009 ). However, adequate correction of lick-related motion artifacts required 3D motion correction (Figure 5 C, bottom ). Further, 3D motion correction also reduced motion contamination not related to licking (Figure 5 C, histograms), likely caused by body motion, chewing, breathing, and the heart beat” | claim_relations[6] |
| differs in | imaging in freely moving animals, which trades stimulus control and trial count for naturalism — THE AXIS IS WHAT THE PREPARATION COSTS. 'Imaging in freely moving animals may allow for easier implementation of naturalistic behaviors in the future, albeit with additional limitations in image quality, visual stimulus control, and number of trials per session as compared to head-fixed operant behaviors.' Three costs named against one benefit, in the authors' own words - the trade-off a methods citation drops. — “Complementary methods allowing two-photon calcium imaging of visual responses and other calcium activity in freely moving rats and mice have recently been published (Engelbrecht et al., 2008 ; Flusberg et al., 2008 ; Sawinski et al., 2009 ). Imaging in freely moving animals may allow for easier implementation of naturalistic behaviors in the future, albeit with additional limitations in image quality, visual stimulus control, and number of trials per session as compared to head-fixed operant behaviors.” — “Imaging in freely moving animals may allow for easier implementation of naturalistic behaviors in the future, albeit with additional limitations in image quality, visual stimulus control, and number of trials per session” | claim_relations[7] |
| audit | GABAergic —operationalised-as→ cell type — Audit 2026-09-28: DOUBTFUL - read against its span and kept, but GABAergic neurons express GFP, which lets cell type be classified; operationalised-as is the wrong type. — “Excitatory neurons and inhibitory neurons could be classified in vivo in this mouse due to expression of green fluorescent protein in all GABAergic neurons (Figure 2 A, green neurons; Tamamaki et al., 2003 ; Sohya et al., 2007 ; Kerlin et al., 2009 ). An example of orientation-tuned responses to drifting gratings (28°/cycle square wave, 2 Hz) is shown in Figure 2 B. Responses could be detected in single trials in GABAergic and non-GABAergic neurons (Figure 2 C). These data illustrate the value of two-photon calcium imaging in awake, transgenic mice, providing robust responses in genetically identified cell types.” | concept_edges[22] |
| note | GABAergic —operationalised-as→ cell type — a whole class labeled genetically, which is the difference between an identified cell and an inferred one — “Excitatory neurons and inhibitory neurons could be classified in vivo in this mouse due to expression of green fluorescent protein in all GABAergic neurons (Figure 2 A, green neurons; Tamamaki et al., 2003 ; Sohya et al., 2007 ; Kerlin et al., 2009 ). An example of orientation-tuned responses to drifting gratings (28°/cycle square wave, 2 Hz) is shown in Figure 2 B. Responses could be detected in single trials in GABAergic and non-GABAergic neurons (Figure 2 C). These data illustrate the value of two-photon calcium imaging in awake, transgenic mice, providing robust responses in genetically identified cell types.” | concept_edges[22] |
| audit | GO–NOGO —contrasted-with→ 2-AFC — Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span names task structure 'e.g., GO-NOGO vs. 2-AFC' as a factor thresholds may depend on; contrasted-with is loose. — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice (Reuter, 1987 ). Absolute orientation thresholds likely depend on incidence of presentation of each stimulus (Figure 4 B, legend), and task structure (e.g., GO–NOGO vs. 2-AFC).” | concept_edges[29] |
| note | 2D —distinct-from→ 3D — THE CORRECTION DECIDES THE FINDING: two-dimensional correction handles the motion that is NOT lick-related, and only three-dimensional handles the motion that is — “Brain motion artifacts not related to licking were mostly corrected using standard 2D motion correction techniques, (Figure 5 C, middle ), consistent with previous studies (Dombeck et al., 2007 ; Greenberg and Kerr, 2009 ). However, adequate correction of lick-related motion artifacts required 3D motion correction (Figure 5 C, bottom ). Further, 3D motion correction also reduced motion contamination not related to licking (Figure 5 C, histograms), likely caused by body motion, chewing, breathing, and the heart beat” | concept_edges[39] |
| outcome (stated by the tagger, no quotation) | control — held | controls[0] |
| for which use | limitation — every behavioural result and the population it describes: ABOUT HALF OF THE MICE TESTED (13/25) LEARNED TO PERFORM THE VISUAL DISCRIMINATION TASK, BUT WITH VARIABLE LEARNING RATES — “About half of the mice tested (13/25) learned to perform the visual discrimination task, but with variable learning rates.” | limitations[0] |
| for which use | limitation — the sensitivity of the behavioural measure: THE GO-NOGO TASK EMPLOYED HERE HAS DISADVANTAGES COMPARED TO TWO-ALTERNATIVE FORCED-CHOICE TASKS, traded for ease of training — “While the GO–NOGO task employed here has disadvantages compared to two-alternative forced-choice (2-AFC) tasks commonly used in primates and freely moving mice, our task is relatively easy for mice to learn.” | limitations[1] |
| for which use | limitation — every calcium measurement reported: CALCIUM INDICATORS ALSO ACT AS CALCIUM BUFFERS THAT, IN SOME CASES, CAN PERTURB NATIVE CALCIUM DYNAMICS - the instrument alters the quantity it reports — “However, calcium indicators also act as calcium buffers that, in some cases, can perturb native calcium dynamics (Holthoff et al., 2002 ; Pologruto et al., 2004 ; but see Schmidt et al., 2003 for an example of negligible perturbation using OGB-1).” | limitations[2] |
| for which use | limitation — any interneuron result specifically: RELIABLE RECONSTRUCTION OF SPIKE TRAINS CAN BE EXTREMELY CHALLENGING, DUE TO SLOWER CALCIUM KINETICS AND HIGHER FIRING RATES — “In the case of interneurons, reliable reconstruction of spike trains can be extremely challenging, due to slower calcium kinetics (Goldberg et al., 2003 ) and higher firing rates (Niell and Stryker, 2008 ).” | limitations[3] |
| for which use | limitation — the same scope as limitations[0] - half the cohort, variable learning — “About half of the mice tested (13/25) learned to perform the visual discrimination task, but with variable learning rates. Training proceeded as follows: First, mice were rewarded for licking in response to drifting target stimuli, using an inter-stimulus interval that was gradually increased from 2 to 5–8 s. Subsequent training involved gradually increasing the similarity of target and non-target stimuli in 5–15 sessions by presenting non-target stimuli that were (1) orthogonal to target and stationary, (2) orthogonal and of increasingly similar speed to target, (3) of increasingly similar orientation to target gratings and of identical speed.” | limitations[4] |
| for which use | limitation — the two imaging figures specifically: the behavioural performance in them was below the level achieved during training, because of the imaging setup — “While these data clearly illustrate the strength of our 3D imaging techniques for obtaining clean neural traces during active discrimination (including during lick responses), the behavioral performance demonstrated in Figures 6 A and 7 A was below the level achieved during training (Table 1 ), due to an increase in spurious licking compared to other mice (see Figure 3 B), and even compared to other imaging sessions from the same mouse” | limitations[5] |
| for which use | limitation — any absolute orientation threshold quoted: it depends on how often each stimulus is presented and on the task structure, so it is not a property of the animal alone — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice (Reuter, 1987 ). Absolute orientation thresholds likely depend on incidence of presentation of each stimulus (Figure 4 B, legend), and task structure (e.g., GO–NOGO vs. 2-AFC).” | limitations[6] |
| for which use | limitation — the strain choice: it is a compromise between ease of handling and speed of learning — “Several factors must be considered when selecting particular mice and mouse strains for head-fixed visual discrimination (Deacon, 2006 ; Wong and Brown, 2006 ). Inbred mice, while demonstrating more uniform behavior, may have retinal deficits or diminished capacity for learning compared to more outbred strains (e.g., Deacon, 2006 ). Thus, when choosing a strain, and when pre-selecting individual mice for training, a compromise must be reached between ease of handling and acclimating to head restraint, and how quickly the animal learns a task” | limitations[7] |
| duplicate of | limitation — limitations[1] — “While the GO–NOGO task employed here has disadvantages compared to two-alternative forced-choice (2-AFC) tasks commonly used in primates and freely moving mice, our task is relatively easy for mice to learn.” | limitations[8] |
| for which use | limitation — the same scope as limitations[1] - the GO-NOGO design is easier for the mice than the two-alternative task used elsewhere, and that ease is paid for in inferential strength — “While the GO–NOGO task employed here has disadvantages compared to two-alternative forced-choice (2-AFC) tasks commonly used in primates and freely moving mice, our task is relatively easy for mice to learn.” | limitations[8] |
| replication | untested | |
| limitation | Only about half the mice trained learned the discrimination, and learning rates varied. — “About half of the mice tested (13/25) learned to perform the visual discrimination task, but with variable learning rates.” | limitations[0] |
| limitation | The GO-NOGO design is weaker than the two-alternative forced-choice tasks used elsewhere, traded off for ease of learning. — “While the GO–NOGO task employed here has disadvantages compared to two-alternative forced-choice (2-AFC) tasks commonly used in primates and freely moving mice, our task is relatively easy for mice to learn.” | limitations[1] |
| limitation | The calcium indicators themselves buffer calcium and can perturb the dynamics being measured. — “However, calcium indicators also act as calcium buffers that, in some cases, can perturb native calcium dynamics (Holthoff et al., 2002 ; Pologruto et al., 2004 ; but see Schmidt et al., 2003 for an example of negligible perturbation using OGB-1).” | limitations[2] |
| limitation | Spike trains cannot be reliably recovered from the calcium signal for interneurons. — “In the case of interneurons, reliable reconstruction of spike trains can be extremely challenging, due to slower calcium kinetics (Goldberg et al., 2003 ) and higher firing rates (Niell and Stryker, 2008 ).” | limitations[3] |
| limitation | about half the mice tested learned the task at all, with variable learning rates — “About half of the mice tested (13/25) learned to perform the visual discrimination task, but with variable learning rates. Training proceeded as follows: First, mice were rewarded for licking in response to drifting target stimuli, using an inter-stimulus interval that was gradually increased from 2 to 5–8 s. Subsequent training involved gradually increasing the similarity of target and non-target stimuli in 5–15 sessions by presenting non-target stimuli that were (1) orthogonal to target and stationary, (2) orthogonal and of increasingly similar speed to target, (3) of increasingly similar orientation to target gratings and of identical speed.” | limitations[4] |
| limitation | the behavioral performance in the two imaging figures was below the level achieved during training, because of spurious licking in that mouse and even in that session — “While these data clearly illustrate the strength of our 3D imaging techniques for obtaining clean neural traces during active discrimination (including during lick responses), the behavioral performance demonstrated in Figures 6 A and 7 A was below the level achieved during training (Table 1 ), due to an increase in spurious licking compared to other mice (see Figure 3 B), and even compared to other imaging sessions from the same mouse” | limitations[5] |
| limitation | absolute orientation thresholds depend on how often each stimulus is presented and on the task structure, so the number is not a property of the animal alone — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice (Reuter, 1987 ). Absolute orientation thresholds likely depend on incidence of presentation of each stimulus (Figure 4 B, legend), and task structure (e.g., GO–NOGO vs. 2-AFC).” | limitations[6] |
| limitation | the strain choice is a compromise between how easily the animal handles and how quickly it learns — “Several factors must be considered when selecting particular mice and mouse strains for head-fixed visual discrimination (Deacon, 2006 ; Wong and Brown, 2006 ). Inbred mice, while demonstrating more uniform behavior, may have retinal deficits or diminished capacity for learning compared to more outbred strains (e.g., Deacon, 2006 ). Thus, when choosing a strain, and when pre-selecting individual mice for training, a compromise must be reached between ease of handling and acclimating to head restraint, and how quickly the animal learns a task” | limitations[7] |
| limitation | and the go-nogo task is easier for the mice to learn than the two-alternative version used in primates, at a cost — “While the GO–NOGO task employed here has disadvantages compared to two-alternative forced-choice (2-AFC) tasks commonly used in primates and freely moving mice, our task is relatively easy for mice to learn.” | limitations[8] |
finding · quantity · variable · definition · key-term — grounded in the full text| finding | Head-fixed mice showed robust discrimination over many hundreds of trials per day once they had acquired the task. — “Head-fixed mice demonstrated robust discrimination for many hundred trials per day after initial task acquisition.” | content_tags.findings[0] |
| finding | About half of the mice attempted acquired the discrimination task, and learning rates varied widely between animals. — “About half of the mice tested (13/25) learned to perform the visual discrimination task, but with variable learning rates.” | content_tags.findings[1] |
| finding | Trained mice maintained the discrimination for months to over a year at roughly 300 trials per day. — “In a separate group of nine mice, we found that mice could perform the discrimination task routinely for several months to over a year, and typically performed on the order of 300 trials/day” | content_tags.findings[2] |
| finding | Orientation discrimination thresholds improved with training to about 10-20 degrees within ten sessions. — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice” | content_tags.findings[3] |
| finding | Standard two-dimensional motion correction was insufficient for lick-related brain motion; only three-dimensional volume co-registration removed those artifacts. — “However, adequate correction of lick-related motion artifacts required 3D motion correction” | content_tags.findings[4] |
| finding | The same genetically labeled neuron could be re-identified and imaged during behaviour across six sessions spanning forty-six days. — “In total, we imaged YC-3.6-labeled neurons in over ten behavioral sessions in this mouse, including one neuron imaged in six different sessions spanning forty-six days” | content_tags.findings[5] |
| finding | Stimulus-evoked calcium responses did not run down over two continuous hours of imaging within a single behavioral session. — “Average responses from this neuron did not attenuate across two consecutive hours of imaging in one behavioral session” | content_tags.findings[6] |
| finding | Mice quickly learned to stop licking to target stimuli until the next engaged block, suggesting the learned stimulus association is not purely reflexive. — “mice quickly learned to stop licking in response to target stimuli until the following ‘engaged’ block began, suggesting that the learned stimulus association is not purely reflexive” | content_tags.findings[7] |
| finding | Horizontal, and even more so vertical, eye movements were rare in mice during passive viewing. — “In general, horizontal eye movements ( Figure S1 in Supplementary Material) and, to an even greater extent, vertical eye movements (not shown) were rare in mice during passive viewing, similar to eye stability during passive viewing in rabbits (Cano et al., 2006 ).” | content_tags.findings[8] |
| finding | THE MOTIVATING PROBLEM: behavioral modulation of nearby neurons is heterogeneous in magnitude AND IN DIRECTION — “A poorly understood feature of most studies, however, is that behavioral modulation of responses in nearby neurons is heterogeneous, both in magnitude and even in sign.” | content_tags.findings[9] |
| finding | with cell class offered as one candidate explanation beside response properties, laminar depth and connectivity — “The diversity of modulatory effects in nearby neurons may be related to differences in each neuron's visual response properties (DeAngelis et al., 1999 ; Martinez-Trujillo and Treue, 2004 ), laminar depth (Mehta et al., 2000 ), connectivity (Bravo and Karten, 1992 ; Brown and Hestrin, 2009 ) or cell type (see below). In particular, cortical inhibitory cell classes may be differentially modulated by behavioral context (Buia and Tiesinga, 2008 ), providing a potential mechanism underlying dynamic regulation of local information flow.” | content_tags.findings[10] |
| finding | A CLASS ASSIGNED BY A PROXY FEATURE, CRITICISED BY NAME: all such studies inferred cellular identity from extracellular spike waveforms, which has limited scope and reliability — “Although considerable anatomical, neurochemical, and physiological diversity exists in nearby neurons, only a small number of in vivo electrophysiological studies have correlated behavioral modulation of neural responses with information about cell type (Swadlow and Weyand, 1987 ; Mitchell et al., 2007 ; Chen et al., 2008 ). These studies have all inferred cellular identity from extracellular spike waveforms, which has limited scope and reliability.” | content_tags.findings[11] |
| finding | and the more informative alternative has extremely low yield and does not go into behaving animals — “Intracellular recordings in vivo , while more informative, have extremely low yield, and are not easily amenable to studies in awake and behaving animals” | content_tags.findings[12] |
| finding | imaging offers tens or hundreds of simultaneous neurons, the same volume across days and months, and labeled anatomical identity — “The ability to visualize tens or hundreds of simultaneously recorded neurons with two-photon calcium imaging provides several advantages over extra-cellular recordings (see Section ‘ Discussion ’). First, this technique offers the possibility of recording from the same volume of neurons across days and months, enabling new directions in the study of sensory perception and perceptual learning. Cellular imaging also facilitates labeling of anatomical identity (cell type, cell connectivity) of recorded neurons, in combination with molecular and genetic techniques.” | content_tags.findings[13] |
| finding | the gap is named precisely - no published work had imaged cellular responses in awake head-fixed mice during discrimination or passive viewing — “Two-photon imaging of cellular and neuropil signals has recently been described in awake, head-fixed mice and rats (Dombeck et al., 2007 , 2009 ; Verhagen et al., 2007 ; Greenberg et al., 2008 ; Kuhn et al., 2008 ; Mukamel et al., 2009 ; for review, see Grewe and Helmchen, 2009 ). However, no published work has imaged cellular responses in awake, head-fixed mice during sensory discrimination or during passive viewing of visual stimuli.” | content_tags.findings[14] |
| finding | freely moving mice had been shown to discriminate gratings, and no head-fixed operant technique existed — “The mouse is the pre-eminent mammalian model system for genetic labeling or manipulation of specific cortical cell classes (O'Connor et al., 2009 ). Cellular imaging in mice offers immense potential for understanding cortical function in healthy mice, as well as in mouse models of neurologic disease (e.g., Eichhoff et al., 2008 ; Kuchibhotla et al., 2009 ). While the imaging and behavioral techniques described here can be used in different cortical regions and using different sensory modalities, we focus on visual behaviors that allow direct comparison to the rich literature on visual behavior and neurophysiology in carnivores and primates. Indeed, electrophysiological studies in anesthetized (Dräger, 1975 ; Mangini and Pearlman, 1980 ; Metin et al., 1988 ; Niell and Stryker, 2008 ; Liu et al., 2009 ) and awake (Frenkel et al., 2006 ; Niell and Stryker, 2009) mice suggest that mouse visual cortical neurons demonstrate similar properties to those in carnivores and primates, despite an overall decrease in spatial acuity (Prusky and Douglas, 2004 ; Niell and Stryker, 2008 ). While freely moving mice have been shown to discriminate between oriented gratings (Reuter, 1987 ; Prusky and Douglas, 2004 ; Wong and Brown, 2006 ), no operant behavioral technique currently exists for visual discrimination in head-fixed mice.” | content_tags.findings[15] |
| finding | AND THE CEILING OF THE DISCRIMINABILITY MEASURE IS SET BY THE TRUNCATION: rates clipped at 0.01 and 0.99, maximum 4.65 — “A useful measure of behavioral discriminability of targets from non-targets is given by d ′ = norminv(HR) − norminv(FA), where norminv() is the inverse of the cumulative normal function (Green and Swets, 1966 ). Values for HR and FA were truncated between 0.01 and 0.99. Higher d ′ values indicate better discriminability (maximum d ′ = 4.65).” | content_tags.findings[16] |
| finding | the threshold is read off a linear fit as the angular distance predicted to give a discriminability of exactly 1 — “To estimate orientation threshold on a given day (Figure 4 C), we performed a linear fit to d ′ values (for all non-targets with d ′ < 2.5) and estimated the angular distance from the target predicted to generate a value of d ′ = 1” | content_tags.findings[17] |
| finding | THE MOTIVATION IS A CONTROLLED VARIABLE WITH THE ARITHMETIC SHOWN: about 0.65 cc earned against a 0.625 cc daily minimum — “Because the mean number of active discrimination trials per session was ∼300 (Table 1 ), and ∼40% of trials were target trials, then a 25-g mouse with a 90% hit rate (HR) would receive, on average, ∼0.65 cc of water during a session (minimum total daily water for a mouse of this weight: 0.625 cc).” | content_tags.findings[18] |
| finding | and only blocks above a hit rate of 0.7 are counted, a selection on the blocks as well as the trials — “Hit rate (HR; fraction of trials with licks following target presentation) and false-alarm rates (FA; fraction of trials with licks following non-targets) for each non-target stimulus type were calculated for each block of 43 trials. The number of active discrimination trials per session (Table 1 and Figure 4 ) only included trials in blocks with HRs exceeding 0.7.” | content_tags.findings[19] |
| finding | neurons are distinguished from astrocytes at the labeling step, by co-injecting a red dye — “In some experiments (Figure 1 A), we performed a bolus loading of the calcium indicator Oregon Green BAPTA-1 AM (OGB1-AM; Invitrogen) with 10% DMSO and Pluronic F-127 (Stosiek et al., 2003 ; Histed et al., 2009 ), 200–300 μm below the cortical surface, under light isoflurane anesthesia. We distinguished neurons from astrocytes by labeling astrocytes via co-injection of the red dye sulforhodamine-101” | content_tags.findings[20] |
| finding | the ratiometric measure is YFP fluorescence divided by CFP fluorescence, converted to fractional change — “For ratiometric imaging (Figure 7 ), fluorescence ratio ( R ) was defined as the ratio of YFP fluorescence (green channel) to CFP fluorescence (blue channel). The mean ( F 0 or R 0 ) was subtracted from each trace, and traces were subsequently high-pass filtered (200 s cutoff, third order Butterworth filter). Traces were then converted to fractional changes: d F / F 0 = [ F ( t ) − F 0 ]/ F 0 , d R / R 0 = [ R ( t ) − R 0 ]/ R 0 .” | content_tags.findings[21] |
| finding | every GABAergic neuron in this mouse expresses GFP, so excitatory and inhibitory neurons could be told apart in vivo — “Excitatory neurons and inhibitory neurons could be classified in vivo in this mouse due to expression of green fluorescent protein in all GABAergic neurons (Figure 2 A, green neurons; Tamamaki et al., 2003 ; Sohya et al., 2007 ; Kerlin et al., 2009 ). An example of orientation-tuned responses to drifting gratings (28°/cycle square wave, 2 Hz) is shown in Figure 2 B. Responses could be detected in single trials in GABAergic and non-GABAergic neurons (Figure 2 C). These data illustrate the value of two-photon calcium imaging in awake, transgenic mice, providing robust responses in genetically identified cell types.” | content_tags.findings[22] |
| finding | small eye motions should not influence cortical responses significantly, given the larger size of cortical receptive fields in mice and the use of sinusoidal gratings of low spatial frequency — “Given the larger size of cortical receptive fields in mice compared to higher mammals (Dräger, 1975 ; Mangini and Pearlman, 1980 ; Metin et al., 1988 ; Niell and Stryker, 2008 ), small eye motions should not influence cortical responses significantly, particularly when using a very large screen and sinusoidal gratings of low spatial frequency. Rare, large eye movements can be tracked and rejected offline” | content_tags.findings[23] |
| finding | ABOUT HALF THE MICE TESTED LEARNED THE TASK AT ALL, with variable learning rates — “About half of the mice tested (13/25) learned to perform the visual discrimination task, but with variable learning rates. Training proceeded as follows: First, mice were rewarded for licking in response to drifting target stimuli, using an inter-stimulus interval that was gradually increased from 2 to 5–8 s. Subsequent training involved gradually increasing the similarity of target and non-target stimuli in 5–15 sessions by presenting non-target stimuli that were (1) orthogonal to target and stationary, (2) orthogonal and of increasingly similar speed to target, (3) of increasingly similar orientation to target gratings and of identical speed.” | content_tags.findings[24] |
| finding | THE THRESHOLD IS NOT A PROPERTY OF THE ANIMAL ALONE: it depends on how often each stimulus appears and on the task structure — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice (Reuter, 1987 ). Absolute orientation thresholds likely depend on incidence of presentation of each stimulus (Figure 4 B, legend), and task structure (e.g., GO–NOGO vs. 2-AFC).” | content_tags.findings[25] |
| finding | nine mice performed for months to over a year at about 300 trials a day — “In a separate group of nine mice, we found that mice could perform the discrimination task routinely for several months to over a year, and typically performed on the order of 300 trials/day (Figure 4 B, Table 1 and Section ‘ Materials and Methods ’; over 1000 trials or 2 h in some sessions). Consistently high levels of discrimination performance were observed for orthogonal as well as for more difficult non-target stimuli (Figure 4 D and Table 1 ). During this time, mice were also gradually habituated to the imaging environment, including the sounds and motions produced during typical microscope operation. These data suggest that head-fixed visual discrimination learning and steady-state performance can be reliably measured across many sessions – a key factor enabling chronic calcium imaging studies of neural correlates of perception and perceptual learning in the same neurons (see below), in both wild-type mice and in mouse models of disease and aging.” | content_tags.findings[26] |
| finding | A CONTROL THAT SHOWS THE ASSOCIATION IS NOT REFLEXIVE: the mice stop licking when the spout is withdrawn and the discrimination does not extinguish — “We varied task engagement using interleaved blocks of 43 trials in which mice either actively discriminated target and non-target stimuli, or passively viewed these same stimuli (Figure 3 A). During passive viewing, a solenoid removed the lickspout from reach (Figure 1 B), and mice quickly learned to stop licking in response to target stimuli until the following ‘engaged’ block began, suggesting that the learned stimulus association is not purely reflexive (Figure 3 C). Presentation of stimuli during passive viewing, or of novel stimuli during orientation mapping, did not lead to extinction of subsequent discrimination behavior (e.g., Figures 3 B,C and 4 D), allowing for comparison of neural responses obtained during behavior with receptive field mapping obtained during passive viewing.” | content_tags.findings[27] |
| finding | THE artifact IS TIME-LOCKED TO THE BEHAVIOURAL VARIABLE OF INTEREST: lick-response-induced brain motion was the major obstacle — “A major obstacle in our early efforts developing two-dimensional calcium imaging during behavior involved systematic, lick response-induced brain motion artifacts. To overcome this problem, we modified previous techniques (Göbel et al., 2007 ) to achieve rapid, trapezoidal vertical movement of the microscope objective (Figure 1 B, Section ‘ Materials and Methods ’), generating complete local volumes at sufficiently fast rates to correct for rapid brain motion (and slow drift) in three dimensions.” | content_tags.findings[28] |
| finding | with an explicit trade of volume size against acquisition rate and therefore against number of neurons — “We found an acceptable trade-off between volume acquisition rate and volume size (and hence number of recorded neurons) by collecting a volume of 14 planes (3.5 μm spacing between planes) at 4 Hz, typically resulting in volumes up to 60 × 60 × 48 μm 3 (containing ∼20 neurons; Figure 5 A).” | content_tags.findings[29] |
| finding | THE artifact IS MEASURED IN A CHANNEL WHERE THE SIGNAL CANNOT EXIST - a calcium-insensitive dye that holds constant unless the brain moves — “A local volume was loaded with a green synthetic calcium indicator (OGB1-AM), and astrocytes were co-loaded with sulforhodamine (SR101; Nimmerjahn et al., 2004 ), a red, calcium-insensitive fluorescent dye that generates a constant average fluorescence level in the absence of brain motion” | content_tags.findings[30] |
| finding | up to 15 micrometers of sinking as the animal opens its jaw, which a single plane would record as a fluorescence change — “This cell demonstrated up to 15 μm of ‘sinking’ Z-displacement as the mouse adjusted his posture, opened his jaw and began to lick – a displacement that is well-corrected by 3D co-registration methods (Figure 5 B, bottom ; see also Figure S2B in Supplementary Material). By monitoring fluorescence in a single plane (Figure 5 A, orange square), one can observe artifactual effects of brain motion on SR101 signal that would occur using 2D imaging” | content_tags.findings[31] |
| finding | THE CORRECTION DECIDES THE FINDING: two-dimensional correction handles the motion that is not lick-related, and only three-dimensional handles the motion that is — “Brain motion artifacts not related to licking were mostly corrected using standard 2D motion correction techniques, (Figure 5 C, middle ), consistent with previous studies (Dombeck et al., 2007 ; Greenberg and Kerr, 2009 ). However, adequate correction of lick-related motion artifacts required 3D motion correction (Figure 5 C, bottom ). Further, 3D motion correction also reduced motion contamination not related to licking (Figure 5 C, histograms), likely caused by body motion, chewing, breathing, and the heart beat” | content_tags.findings[32] |
| finding | with the reference channel showing what a nominally stable signal actually does before correction — “For typical studies imaging from one plane, SR101 fluorescence from the edge of this astrocyte should be stable over time, but instead shows severe motion artifacts ( top panel ). Gray ticks indicate licking, which mostly occurred during engaged blocks but not during passive viewing (black line). Using standard 2D rigid co-registration techniques ( middle panel ), some artifacts are reduced. However, 3D co-registration was necessary for near-complete elimination of artifacts ( bottom panel ), particularly those concurrent with onset of lick responses.” | content_tags.findings[33] |
| finding | THE DEMONSTRATION SESSION IS NOT THE TYPICAL SESSION, AND THE PAPER SAYS SO — “While these data clearly illustrate the strength of our 3D imaging techniques for obtaining clean neural traces during active discrimination (including during lick responses), the behavioral performance demonstrated in Figures 6 A and 7 A was below the level achieved during training (Table 1 ), due to an increase in spurious licking compared to other mice (see Figure 3 B), and even compared to other imaging sessions from the same mouse” | content_tags.findings[34] |
| finding | the same relative orientation preference appears during behavior and during passive viewing — “The relative response strength of neurons to different stimulus orientations during behavior (Figures 6 A and 6 E, top left ) was consistent with orientation tuning curves obtained during passive viewing” | content_tags.findings[35] |
| finding | the same volume was reloaded two days later, dimmer on the third day, with only a subset of cells still usable — “Two days following this initial recording of neural responses during behavior, we reloaded the same volume of neurons with synthetic calcium indicator OGB1-AM (Figure 6 D), and were again able to obtain stable responses during behavior and during subsequent orientation mapping (Figure 6 E, middle row ). The day after this second injection, OGB1-AM expression was dimmer (dye persists in neurons for 8–30 h following injection), but again yielded responses during behavior and consistent orientation tuning from a subset of neurons” | content_tags.findings[36] |
| finding | one neuron was imaged in six sessions spanning forty-six days, refound within minutes using vascular and dural landmarks — “In total, we imaged YC-3.6-labeled neurons in over ten behavioral sessions in this mouse, including one neuron imaged in six different sessions spanning forty-six days (Figure 7 D). The same neurons could be found within minutes of the start of each session, using epifluorescence imaging (Figure 1 B), vascular and dural landmarks, and wide-field two-photon images at the appropriate depth (not shown). These results demonstrate the feasibility of combining well-controlled visual behavior with repeated, simultaneous calcium imaging of the same neurons across days and months.” | content_tags.findings[37] |
| finding | and responses did not attenuate across two consecutive hours of imaging — “Average responses from this neuron did not attenuate across two consecutive hours of imaging in one behavioral session (Figure 7 C, 764 total trials across three orientations), and similar responses were also observed in a previous session in the same cell (not shown).” | content_tags.findings[38] |
| finding | the hope is stated as a hope - cell identity MAY HELP to explain the variability across nearby neurons — “For example, information on cell type (e.g., GABAergic neurons and subtypes, Figure 2 ; O'Connor et al., 2009 ) and potentially inter-areal connectivity (Brown and Hestrin, 2009 ) of recorded neurons can be directly visualized in vivo in transgenic mice. Incorporating information on cell identity may help to explain the variability in context-dependent responses across nearby neurons.” | content_tags.findings[39] |
| finding | inbred strains behave more uniformly and may have retinal deficits or learn worse, so the choice is a compromise — “Several factors must be considered when selecting particular mice and mouse strains for head-fixed visual discrimination (Deacon, 2006 ; Wong and Brown, 2006 ). Inbred mice, while demonstrating more uniform behavior, may have retinal deficits or diminished capacity for learning compared to more outbred strains (e.g., Deacon, 2006 ). Thus, when choosing a strain, and when pre-selecting individual mice for training, a compromise must be reached between ease of handling and acclimating to head restraint, and how quickly the animal learns a task” | content_tags.findings[40] |
| finding | the go-nogo task is easier for mice to learn than the two-alternative version, at a stated cost — “While the GO–NOGO task employed here has disadvantages compared to two-alternative forced-choice (2-AFC) tasks commonly used in primates and freely moving mice, our task is relatively easy for mice to learn.” | content_tags.findings[41] |
| finding | and longer sessions are bought at the price of more false alarms — “Note that the motivation of mice to perform the task was strongly dependent on water scheduling, and increasing session duration must be balanced with the potential for greater false-alarm rates.” | content_tags.findings[42] |
| finding | learning was slower than in a matched olfactory task, attributed to the greater behavioural salience of odors — “Mice demonstrated improved visual thresholds and task performance over the course of many days, which will prove useful in future studies of perceptual learning. The learning rates were slower than for a similar olfactory task in head-fixed mice (Komiyama et al., 2007 ). These differences could be due to the greater behavioral salience of different odors as compared to initial visual stimuli employed in our task.” | content_tags.findings[43] |
| finding | THE SAMPLED POPULATION DEPENDS ON THE LABEL: the synthetic dye loads every cell in a local volume, the genetic one only a subset — “The choice of whether to use synthetic dyes or GECIs depends on the research question at hand. For example, neurons bulk-loaded with OGB1-AM dye are typically brighter than neurons expressing GECIs, and thus can be imaged with less laser power and potentially less cell damage. Further, OGB1-AM loads every cell in a local volume (∼300 × 300 × 300 μm 3 ), thus providing greater yield than GECI experiments where only a subset of neurons are been labeled using viral or electroporation techniques.” | content_tags.findings[44] |
| finding | and imaging in freely moving animals trades image quality, stimulus control and trial count for naturalism — “Complementary methods allowing two-photon calcium imaging of visual responses and other calcium activity in freely moving rats and mice have recently been published (Engelbrecht et al., 2008 ; Flusberg et al., 2008 ; Sawinski et al., 2009 ). Imaging in freely moving animals may allow for easier implementation of naturalistic behaviors in the future, albeit with additional limitations in image quality, visual stimulus control, and number of trials per session as compared to head-fixed operant behaviors.” | content_tags.findings[45] |
| quantity | 13/25 — Proportion of mice that learned the visual discrimination task — “About half of the mice tested (13/25) learned to perform the visual discrimination task, but with variable learning rates.” | content_tags.quantities[0] |
| quantity | Twenty-five — Total mice used in the study across all procedures — “Twenty-five male and female mice aged p30 to p75 at experiment onset were used in this study” | content_tags.quantities[1] |
| quantity | 10 mice — Number of mice in which two-photon calcium imaging was attempted — “Two-photon calcium imaging was attempted in 10 mice.” | content_tags.quantities[2] |
| quantity | >900 — behavioral trials completed before orientation mapping in the example imaging session — “Following cessation of behavioral trials (>900), mapping of orientation tuning in these same cells was performed in 16 orientations” | content_tags.quantities[3] |
| quantity | 167 sessions — Number of stable behavioral sessions from one mouse, spanning more than a year — “This mouse demonstrated stable performance in 167 sessions spanning over one year ( left panel ) and was engaged in the discrimination behavior for several hundred trials a day” | content_tags.quantities[4] |
| quantity | 10–20° — Orientation difference mice could discriminate after about ten training sessions — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice” | content_tags.quantities[5] |
| quantity | 4.65 — the maximum discriminability - a ceiling set by truncating the rates at 0.01 and 0.99, not by the animal — “A useful measure of behavioral discriminability of targets from non-targets is given by d ′ = norminv(HR) − norminv(FA), where norminv() is the inverse of the cumulative normal function (Green and Swets, 1966 ). Values for HR and FA were truncated between 0.01 and 0.99. Higher d ′ values indicate better discriminability (maximum d ′ = 4.65).” | content_tags.quantities[6] |
| quantity | 13/25 — the mice that learned the task at all, which bounds every behavioral number in the paper — “About half of the mice tested (13/25) learned to perform the visual discrimination task, but with variable learning rates. Training proceeded as follows: First, mice were rewarded for licking in response to drifting target stimuli, using an inter-stimulus interval that was gradually increased from 2 to 5–8 s. Subsequent training involved gradually increasing the similarity of target and non-target stimuli in 5–15 sessions by presenting non-target stimuli that were (1) orthogonal to target and stationary, (2) orthogonal and of increasingly similar speed to target, (3) of increasingly similar orientation to target gratings and of identical speed.” | content_tags.quantities[7] |
| quantity | 15 μm — the apparent sinking of a calcium-insensitive cell during licking - the size of the artifact a single imaging plane would have called a response — “This cell demonstrated up to 15 μm of ‘sinking’ Z-displacement as the mouse adjusted his posture, opened his jaw and began to lick – a displacement that is well-corrected by 3D co-registration methods (Figure 5 B, bottom ; see also Figure S2B in Supplementary Material). By monitoring fluorescence in a single plane (Figure 5 A, orange square), one can observe artifactual effects of brain motion on SR101 signal that would occur using 2D imaging” | content_tags.quantities[8] |
| quantity | 0.7 — the hit rate a block must exceed to be counted, so the reported trial counts are already selected — “Hit rate (HR; fraction of trials with licks following target presentation) and false-alarm rates (FA; fraction of trials with licks following non-targets) for each non-target stimulus type were calculated for each block of 43 trials. The number of active discrimination trials per session (Table 1 and Figure 4 ) only included trials in blocks with HRs exceeding 0.7.” | content_tags.quantities[9] |
| quantity | 0.625 cc — the daily water minimum against about 0.65 cc earned in a session - the arithmetic that makes the motivation auditable — “Because the mean number of active discrimination trials per session was ∼300 (Table 1 ), and ∼40% of trials were target trials, then a 25-g mouse with a 90% hit rate (HR) would receive, on average, ∼0.65 cc of water during a session (minimum total daily water for a mouse of this weight: 0.625 cc).” | content_tags.quantities[10] |
| variable | task engagement (independent) — “We varied task engagement using interleaved blocks of 43 trials in which mice either actively discriminated target and non-target stimuli, or passively viewed these same stimuli” | content_tags.variables[0] |
| variable | angular distance (independent) — “Increased angular distance of non-targets from target resulted in better discrimination performance, as expected.” | content_tags.variables[1] |
| variable | behavioral discriminability (dependent) — “(A) Discriminability ( d ′) of target from non-target stimuli improved over days, as shown in four mice (four different colors).” | content_tags.variables[2] |
| variable | time courses of responses (dependent) — “Time courses of responses were obtained by summing signals from all pixels within 3D cell masks.” | content_tags.variables[3] |
| variable | water scheduling (control) — “Note that the motivation of mice to perform the task was strongly dependent on water scheduling, and increasing session duration must be balanced with the potential for greater false-alarm rates.” | content_tags.variables[4] |
| variable | SR101 (control) — “astrocytes were co-loaded with sulforhodamine (SR101; Nimmerjahn et al., 2004 ), a red, calcium-insensitive fluorescent dye that generates a constant average fluorescence level in the absence of brain motion” | content_tags.variables[6] |
| variable | discriminability — “A useful measure of behavioral discriminability of targets from non-targets is given by d ′ = norminv(HR) − norminv(FA), where norminv() is the inverse of the cumulative normal function (Green and Swets, 1966 ). Values for HR and FA were truncated between 0.01 and 0.99. Higher d ′ values indicate better discriminability (maximum d ′ = 4.65).” | content_tags.variables[8] |
| variable | threshold — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice (Reuter, 1987 ). Absolute orientation thresholds likely depend on incidence of presentation of each stimulus (Figure 4 B, legend), and task structure (e.g., GO–NOGO vs. 2-AFC).” | content_tags.variables[9] |
| variable | hit rate — “Hit rate (HR; fraction of trials with licks following target presentation) and false-alarm rates (FA; fraction of trials with licks following non-targets) for each non-target stimulus type were calculated for each block of 43 trials. The number of active discrimination trials per session (Table 1 and Figure 4 ) only included trials in blocks with HRs exceeding 0.7.” | content_tags.variables[10] |
| variable | fluorescence — “For typical studies imaging from one plane, SR101 fluorescence from the edge of this astrocyte should be stable over time, but instead shows severe motion artifacts ( top panel ). Gray ticks indicate licking, which mostly occurred during engaged blocks but not during passive viewing (black line). Using standard 2D rigid co-registration techniques ( middle panel ), some artifacts are reduced. However, 3D co-registration was necessary for near-complete elimination of artifacts ( bottom panel ), particularly those concurrent with onset of lick responses.” | content_tags.variables[11] |
| variable | pupil position (control) — “This provided an offline means of tracking pupil position and rejecting trials with large eye movements.” | content_tags.variables[12] |
| variable | eye movements (control) — “This provided an offline means of tracking pupil position and rejecting trials with large eye movements.” | content_tags.variables[13] |
| definition | visual discrimination task — “Here, we describe a GO–NOGO visual discrimination task in head-fixed mice, suitable for cellular imaging studies of visual perception, perceptual learning, task engagement, and task difficulty.” | content_tags.definitions[0] |
| definition | false-alarm rate — “Hit rate (HR; fraction of trials with licks following target presentation) and false-alarm rates (FA; fraction of trials with licks following non-targets) for each non-target stimulus type were calculated for each block of 43 trials.” | content_tags.definitions[1] |
| definition | target and non-target stimuli — “Target stimuli were upwards drifting gratings, and non-target stimuli were identical gratings drifting in other directions” | content_tags.definitions[2] |
| definition | discriminability — “A useful measure of behavioral discriminability of targets from non-targets is given by d ′ = norminv(HR) − norminv(FA), where norminv() is the inverse of the cumulative normal function (Green and Swets, 1966 ). Values for HR and FA were truncated between 0.01 and 0.99. Higher d ′ values indicate better discriminability (maximum d ′ = 4.65).” | content_tags.definitions[3] |
| definition | threshold — “To estimate orientation threshold on a given day (Figure 4 C), we performed a linear fit to d ′ values (for all non-targets with d ′ < 2.5) and estimated the angular distance from the target predicted to generate a value of d ′ = 1” | content_tags.definitions[4] |
| definition | hit rate — “Hit rate (HR; fraction of trials with licks following target presentation) and false-alarm rates (FA; fraction of trials with licks following non-targets) for each non-target stimulus type were calculated for each block of 43 trials. The number of active discrimination trials per session (Table 1 and Figure 4 ) only included trials in blocks with HRs exceeding 0.7.” | content_tags.definitions[5] |
| definition | SR101 — “A local volume was loaded with a green synthetic calcium indicator (OGB1-AM), and astrocytes were co-loaded with sulforhodamine (SR101; Nimmerjahn et al., 2004 ), a red, calcium-insensitive fluorescent dye that generates a constant average fluorescence level in the absence of brain motion” | content_tags.definitions[6] |
| definition | fluorescence — “For ratiometric imaging (Figure 7 ), fluorescence ratio ( R ) was defined as the ratio of YFP fluorescence (green channel) to CFP fluorescence (blue channel). The mean ( F 0 or R 0 ) was subtracted from each trace, and traces were subsequently high-pass filtered (200 s cutoff, third order Butterworth filter). Traces were then converted to fractional changes: d F / F 0 = [ F ( t ) − F 0 ]/ F 0 , d R / R 0 = [ R ( t ) − R 0 ]/ R 0 .” | content_tags.definitions[7] |
| key-terms | imaging, two-photon, calcium, indicator, OGB1-AM, GECI, SR101, astrocyte, fluorescence, motion, artifact, co-registration, 3D, 2D, volume, plane, lick, behavior, task, discrimination, GO–NOGO, 2-AFC, target, non-target, stimulus, orientation, grating, threshold, discriminability, hit rate, false-alarm, trial, session, engagement, passive viewing, head-fixed, awake, eye, water, learning, perceptual learning, mouse, strain, cell type, GABAergic, inhibitory, identity, waveform, heterogeneous, context, modulation, neuron, yield, salience, chronic |
| d-prime (signal detection discriminability index) d ′ | as written: Discriminability ( d ′) — “(A) Discriminability ( d ′) of target from non-target stimuli improved over days, as shown in four mice (four different colors).” | content_tags.statistics[0] |
| hit rate and false-alarm rate HR, FA | as written: Hit rate (HR; ...) and false-alarm rates (FA; ...) — “Hit rate (HR; fraction of trials with licks following target presentation) and false-alarm rates (FA; fraction of trials with licks following non-targets) for each non-target stimulus type were calculated for each block of 43 trials.” | content_tags.statistics[1] |
| psychometric function | as written: psychometric curves — “Complete psychometric curves across daily sessions are shown for one mouse” | content_tags.statistics[2] |
| cross-correlation | as written: 3D cross-correlation — “These data were analyzed using a rigid, subpixel alignment procedure based on a 3D cross-correlation with an average volume generated from motion-free epochs.” | content_tags.statistics[3] |
| inverse of the cumulative normal function | — “A useful measure of behavioral discriminability of targets from non-targets is given by d ′ = norminv(HR) − norminv(FA), where norminv() is the inverse of the cumulative normal function (Green and Swets, 1966 ). Values for HR and FA were truncated between 0.01 and 0.99. Higher d ′ values indicate better discriminability (maximum d ′ = 4.65).” | content_tags.statistics[4] |
| linear fit | — “To estimate orientation threshold on a given day (Figure 4 C), we performed a linear fit to d ′ values (for all non-targets with d ′ < 2.5) and estimated the angular distance from the target predicted to generate a value of d ′ = 1” | content_tags.statistics[5] |
| psychometric curves | — “Within ten sessions, mice were able to discriminate orientations as little as 10–20° apart, consistent with previous observations in freely moving mice (Reuter, 1987 ). Absolute orientation thresholds likely depend on incidence of presentation of each stimulus (Figure 4 B, legend), and task structure (e.g., GO–NOGO vs. 2-AFC).” | content_tags.statistics[6] |
| claim type | method — a second reading reaching the same value; recorded: method — “Here, we describe a GO–NOGO visual discrimination task in head-fixed mice, suitable for cellular imaging studies of visual perception, perceptual learning, task engagement, and task difficulty. We show that mice will perform this task (or passively view the same stimuli) for several hundred trials a day, and we demonstrate quantitative monitoring of neural and behavioral responses across hours, days and months.” | alternatives.claim_type |
| method | volumes of 14 planes at 4 Hz, rigid three-dimensional co-registration followed by plane-by-plane correction, with a calcium-insensitive dye in a second channel as the motion reference — a second reading DIFFERING from the recorded value; recorded: system-or-tool-building — “We found an acceptable trade-off between volume acquisition rate and volume size (and hence number of recorded neurons) by collecting a volume of 14 planes (3.5 μm spacing between planes) at 4 Hz, typically resulting in volumes up to 60 × 60 × 48 μm 3 (containing ∼20 neurons; Figure 5 A).” | alternatives.method |
| phenomenon | lick-locked brain motion, which in two-dimensional imaging appears as a fluorescence change time-locked to the behaviour — a second reading DIFFERING from the recorded value; recorded: behavioral modulation of visual cortical responses — “A major obstacle in our early efforts developing two-dimensional calcium imaging during behavior involved systematic, lick response-induced brain motion artifacts. To overcome this problem, we modified previous techniques (Göbel et al., 2007 ) to achieve rapid, trapezoidal vertical movement of the microscope objective (Figure 1 B, Section ‘ Materials and Methods ’), generating complete local volumes at sufficiently fast rates to correct for rapid brain motion (and slow drift) in three dimensions.” | alternatives.phenomenon |
Full text — tagged
The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.
Chronicterm Cellular Imagingterm·mth of Mouseterm Visual Cortex During Operant Behaviorterm and Passive Viewingterm·mth Mark L AndermannpersName Mark L AndermannpersName 1 Department of Neurobiology, Harvard Medical School, Boston, MA, USA 1, * , A M KerlinpersName A M KerlinpersName 1 Department of Neurobiology, Harvard Medical School, Boston, MA, USA 1 , R C ReidpersName R C ReidpersName 1 Department of Neurobiology, Harvard Medical School, Boston, MA, USA 1 Department of Neurobiology, Harvard Medical School, Boston, MA, USA Edited by: Hernan Lopez-Schier, Centro de Regulacion Genomica - PRBB, Spain Reviewed by: Jessica Cardin, Yale University, USA; Leopoldo Petreanu, Janelia Farm Research Campus, USA ✉ *Correspondence: Mark L. AndermannpersName, Department of Neurobiology, Harvard Medical School, 220 Longwood Avenue, Room 243, Boston, MA, 02115, USA. e-mail: mark_andermannpersName@hms.harvard.edu Received 2010 Jan 12; Accepted 2010 Feb 18; Prepublished 2010 Jan 25; Collection date 2010.
Copyright © 2010 AndermannpersName, KerlinpersName and ReidpersName.
This is an open-access article subject to an exclusive license agreement between the authors and the Frontiers Research Foundation, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are credited.
PMCID: PMC2854571 PMID: 20407583 Abstract Nearby neuronsterm in mammalian neocortex demonstrate a great diversity of cell typesterm and connectivity patterns. The importance of this diversity for computation is not understood. While extracellular recording studies in visual cortex have provided a particularly rich description of behavioral modulationterm of neural activity, new methods are needed to dissect the contribution of specific circuit elements in guiding visual perception. Here, we describe a method for three-dimensional cellular imagingterm·mth of neural activity in the awaketerm mouseterm visual cortex during active discriminationterm·mth and passive viewingterm·mth of visual stimuli. Head-fixedterm·mth mice demonstrated robust discriminationterm·mth for many hundred trialsterm per day after initial taskterm acquisition. To record from multiple neuronsterm during operant behaviorterm with single-trial resolution and minimal artifactsterm, we built a sensitive microscope for two-photonterm·mth calciumterm imagingterm·mth, capable of rapid tracking of neuronsterm in three dimensions. We demonstrate stable recordings of cellular calciumterm activity during discriminationterm·mth behaviorterm across hours, days, and weeks, using both synthetic and genetically encoded calciumterm indicatorsterm. When combined with molecular and genetic technologies in mice (e.g., cell-type specific transgenic labeling), this approach allows the identification of neuronal classes in vivo . Physiological measurements from distinct classes of neighboring neuronsterm will enrich our understanding of the coordinated roles of diverse elements of cortical microcircuits in guiding sensory perception and perceptual learningterm·dom. Further, our method provides a high-throughput, chronicterm in vivo assay of behavioral influences on cellular activity that is applicable to a wide range of mouseterm models of neurologic disease.
Keywords: awaketerm mouseterm, visual cortex, neuronterm, two-photonterm·mth calciumterm imagingterm·mth, discriminationterm·mth, perception, learningterm, head-fixedterm·mth Introduction Visual cortical responses to identical stimuli in behaving primates demonstrate rapid, task-dependent modulationterm. For instance, responses can be influenced by spatial, temporal and motivational contextterm, arousal, task-difficulty, and selective attention (e.g., Richmond et al., 1983 ; Spitzer and Richmond, 1991 ; Desimone and Duncan, 1995 ; Zipser et al., 1996 ; Parasuraman, 1998 ; Liu and Richmond, 2000 ; O'Connor et al., 2002 ; Boudreau et al., 2006 ; Gilbert and Sigman, 2007 ). Response modulationterm can be observed even at the level of primary visual cortex (V1; McAdams and Maunsell, 1999 ; O'Connor et al., 2002 ; Super, 2002 ; McAdams and ReidpersName, 2005 ; Palmer et al., 2007 ; Chen et al., 2008 ; Herrero et al., 2008 ). A poorly understood feature of most studies, however, is that behavioral modulationterm of responses in nearby neuronsterm is heterogeneousterm, both in magnitude and even in sign.
The diversity of modulatory effects in nearby neuronsterm may be related to differences in each neuronterm's visual response properties (DeAngelis et al., 1999 ; Martinez-Trujillo and Treue, 2004 ), laminar depth (Mehta et al., 2000 ), connectivity (Bravo and Karten, 1992 ; Brown and Hestrin, 2009 ) or cell typeterm (see below). In particular, cortical inhibitoryterm cell classes may be differentially modulated by behavioral contextterm (Buia and Tiesinga, 2008 ), providing a potential mechanism underlying dynamic regulation of local information flow. Although considerable anatomical, neurochemical, and physiological diversity exists in nearby neuronsterm, only a small number of in vivo electrophysiological studies have correlated behavioral modulationterm of neural responses with information about cell typeterm (Swadlow and Weyand, 1987 ; Mitchell et al., 2007 ; Chen et al., 2008 ). These studies have all inferred cellular identityterm from extracellular spike waveformsterm, which has limited scope and reliability. Intracellular recordings in vivo , while more informative, have extremely low yieldterm, and are not easily amenable to studies in awaketerm and behaving animals (but see Margrie et al., 2002 ; Crochet and Petersen, 2006 ; Poulet and Petersen, 2008 ; Harvey et al., 2009 ).
To study how diverse elements within a local cortical microcircuit are engaged to optimize visual processing across changing taskterm demands, we have developed a robust method for three-dimensional cellular imagingterm·mth of mouseterm visual cortex during operant visual discriminationterm·mth behaviorterm. The ability to visualize tens or hundreds of simultaneously recorded neuronsterm with two-photonterm·mth calciumterm imagingterm·mth provides several advantages over extra-cellular recordings (see Section ‘ Discussion ’). First, this technique offers the possibility of recording from the same volumeterm of neuronsterm across days and months, enabling new directions in the study of sensory perception and perceptual learningterm·dom. Cellular imagingterm·mth also facilitates labeling of anatomical identityterm (cell typeterm, cell connectivity) of recorded neuronsterm, in combination with molecular and genetic techniques. Two-photonterm·mth imagingterm·mth of cellular and neuropil signals has recently been described in awaketerm, head-fixedterm·mth mice and rats (Dombeck et al., 2007 , 2009 ; Verhagen et al., 2007 ; Greenberg et al., 2008 ; Kuhn et al., 2008 ; Mukamel et al., 2009 ; for review, see Grewe and Helmchen, 2009 ). However, no published work has imaged cellular responses in awaketerm, head-fixedterm·mth mice during sensory discriminationterm·mth or during passive viewingterm·mth of visual stimuli.
The mouseterm is the pre-eminent mammalian model system for genetic labeling or manipulation of specific cortical cell classes (O'Connor et al., 2009 ). Cellular imagingterm·mth in mice offers immense potential for understanding cortical function in healthy mice, as well as in mouseterm models of neurologic disease (e.g., Eichhoff et al., 2008 ; Kuchibhotla et al., 2009 ). While the imagingterm·mth and behavioral techniques described here can be used in different cortical regions and using different sensory modalities, we focus on visual behaviorsterm that allow direct comparison to the rich literature on visual behaviorterm and neurophysiology in carnivores and primates. Indeed, electrophysiological studies in anesthetized (Dräger, 1975 ; Mangini and Pearlman, 1980 ; Metin et al., 1988 ; Niell and Stryker, 2008 ; Liu et al., 2009 ) and awaketerm (Frenkel et al., 2006 ; Niell and Stryker, 2009) mice suggest that mouseterm visual cortical neuronsterm demonstrate similar properties to those in carnivores and primates, despite an overall decrease in spatial acuity (Prusky and Douglas, 2004 ; Niell and Stryker, 2008 ). While freely moving mice have been shown to discriminate between oriented gratingsterm (Reuter, 1987 ; Prusky and Douglas, 2004 ; Wong and Brown, 2006 ), no operant behavioral technique currently exists for visual discriminationterm·mth in head-fixedterm·mth mice. Visual discrimination tasksterm·def using head fixation would significantly facilitate chronicterm, three-dimensional cellular imagingterm·mth during behaviorterm, in combination with accurate monitoring of eyeterm position.
Here, we describe a GO–NOGOterm·mth visual discrimination taskterm·def in head-fixedterm·mth mice, suitable for cellular imagingterm·mth studies of visual perception, perceptual learningterm·dom, task engagementterm·var, and taskterm difficulty. We show that mice will perform this taskterm (or passively view the same stimuli) for several hundred trialsterm a day, and we demonstrate quantitative monitoring of neural and behavioral responses across hours, days and months. Combining visual behaviorterm and imagingterm·mth techniques with molecular and genetic technologies in mice (for review, see O'Connor et al., 2009 ) will enable new translational approaches to dissecting the cellular mechanisms underlying visual behaviorterm at the level of cortical microcircuits.
Materials and Methods All procedures were conducted in accordance with the ethical guidelines of the National Institutes of Health and were approved by the IACUC at Harvard Medical School. Twenty-five male and female mice aged p30 to p75 at experiment onset were used in this study: 19 wild-type C57BL/6 (Charles River), 3 wild-type Swiss (Charles River), and 3 GAD67-GFP (Tamamaki et al., 2003 ). Two-photonterm·mth calciumterm imagingterm·mth was attempted in 10 mice. Below, we provide detailed behavioral, surgical, and imagingterm·mth procedures for two-photonterm·mth calciumterm imagingterm·mth during visual discriminationterm·mth (see also Figure 1 and Section ‘ Results ’).
Figure 1.
Overview and setup for two photonterm·mth calciumterm imagingterm·mth in awaketerm and behaving mice . (A) Schematic timeline for combining cellular imagingterm·mth with visual behaviorterm. Green and gray paths delineate routes for imagingterm·mth with genetically encoded calciumterm indicatorsterm or bolus-loaded synthetic calciumterm indicatorsterm, respectively. (B) Head-fixedterm·mth mice view an LCD monitor during concurrent two-photonterm·mth calciumterm imagingterm·mth and eyetracking. Optimized fluorescenceterm excitation and light collection, combined with fast 3Dterm imagingterm·mth, allow robust measurements of neural responses on single trialsterm. (C) A titanium headpost secured over visual cortex and an imagingterm·mth well constructed from rubber O-rings provided excellent stability, wide viewing angle, and light shielding (see Section ‘ Materials and Methods ’).
Visual discriminationterm·mth training and behavioral monitoring Visual stimulation Visual stimuli were controlled and behavioral lickterm responses recorded by custom-written Presentation software (Neurobehavioral Systems). Visual stimuli were square- or sine-wave gratingsterm (28°/cycle, 80 or 100% contrast, black and white) drifting at 2 Hz. Stimuli were presented at 60 Hz using a calibrated 19′?LCD monitor (Viewsonic VP930B). For orientationterm mapping, stimuli were either presented in random order, 4 s long and intermittent with 8 s ISI of gray screen at mean luminance (Figure 2 ), or were 9.5 s long (with 9.5 s ISI), presented sequentially in counterclockwise order (Figure 6 and Movie S1 in Supplementary Material).
Figure 2.
Mapping visual responses in identified excitatory and inhibitoryterm neuronsterm in awaketerm mice . (A) Average image of neuronsterm loaded with synthetic calciumterm indicatorterm, OGB1-AMterm, in an awaketerm, transgenic mouseterm expressing GFP in all GABAergicterm neuronsterm (labeled in green; image depth: 145 μm below pia; 45 mW excitation at 800 nm). (B) Orientationterm tuning curve (cell 1; average of eight trialsterm) demonstrating calciumterm responses during presentation of drifting oriented gratingsterm in 16 directions (arrows and gray bars). (C) Examples of single-trial responses to repeated presentations of the same stimulusterm (yellow bar in B ), in one excitatory (gray, cell 1) and one inhibitoryterm (green, cell 2) neuronterm. Thick lines indicate mean responses. (D) Concurrent eyetracking ( left panel ) demonstrated relatively stable pupil positionterm·var (slow drifts <10°, with minor changes in dilation) across hundreds of seconds. Right panel : Time-lapse of cross-section of mouseterm eyeterm (black: pupil). Pupil location (blue line) was estimated using simple offline tracking of pupil edges. Yellow arrows indicate stimulusterm presentation. (E) Pupil time course following alignment to stimulusterm onset [ N = 24 traces; mean pupil location in (4 s, 4 s) was subtracted from each trace; black line: mean trace]. For the stimuli used in this manuscript (28°/cycle, drifting at 2 Hz), stimulus-evoked eye movementsterm·var were not consistently observed, and trialsterm with spontaneous eye movementsterm·var were rare (see also Figures S1 and S2 in Supplementary Material).
Figure 6.
Simultaneous calciumterm responses from multiple cortical neuronsterm during visual behaviorterm . (A) Responses from 3 out of 20 neuronsterm recorded simultaneously using OGB1-AMterm indicatorterm during a 2-h behaviorterm sessionterm (black traces; 130–178 μm deep; 13–16 mW excitation at 800 nm). Colored vertical bars indicate presentation of targetterm (green) and non-targetterm (red, blue) stimuli (1 s stimuli; bar widths are 3 s wide to emphasize stimulusterm selectivity of different neuronsterm). Black line indicates task engagementterm·var/passive viewingterm·mth, gray ticks indicate licking, and blue trace indicates online estimate of angular pupil displacement (downsampled to 50 Hz) after correction for translational motionterm (using corneal reflections). Small nasally directed eyeterm rotations occurred concurrent with reward presentation. Visual responses were not strongly affected by licking or small changes in eyeterm position. (B) Following cessation of behavioral trialsterm (>900), mapping of orientationterm tuning in these same cells was performed in 16 orientationsterm, including the three orientationsterm used during behaviorterm (colored lines). (C) Orientationterm responses in cell 2 for 16 consecutive orientationsterm showed high signal-to-noise ratio in single trialsterm (gray traces) and on average (black trace). (D) Neuronsterm in and surrounding the imaged volumeterm ( upper panel ) expelled the dye over ∼12–36 h (not shown), but were reloaded and imaged 2 days later ( lower panel ). (E) Average responses (mean ± SE) during behaviorterm (cell 2, left panels ) were robust across large numbers of trialsterm and across both these imagingterm·mth sessionsterm, as well as in a third sessionterm (Day 4) imagingterm·mth calciumterm dye loaded a day earlier. Yellow box: stimulusterm presentation. Polar plots of orientationterm tuning mapped following behavioral sessionsterm (normalized to peak response) were also stable across multiple days ( right panels ; different rows indicate different days).
Training procedures Mice were first handled for 3–14 days (>5 min on most days, or until mice began to groom while being held). Water schedulingterm·var then began, in which mice were given a total of 25 cc/kg waterterm per day, during the training sessionterm (see below) or handling (via syringe with soft plastic tip). To achieve head restraint, awaketerm, headposted mice were led into a custom semicircular plastic tube containing support for forepaws and a plastic shield to prevent grasping of the lickspout, and both headpost ends were clamped to 1/2″ posts (Thorlabs) using thumbscrews. The cylinder was held in place by beaker clamps (VWR), and adjusted to minimize stress on neck muscles. An accelerometer glued to the cylinder detected body motionterm ( Figure S2 in Supplementary Material). Waterterm was administered using a syringe or a lickspout at the onset of the restraint period, which was gradually extended from 5 min to >1 h over the course of 1–2 weeks. Several mice could be restrained simultaneously on the same table, further minimizing stress and training time. Once mice sat quietly for extended periods of time, automated training began (for additional details on training procedures, see Section Stable Visual Discriminationterm·mth Behaviorterm in Head-Fixedterm·mth Mice and Figure 3 ).
Figure 3.
GO–NOGOterm·mth operant visual orientationterm discriminationterm·mth taskterm: Basic paradigm and variants . (A) Upper panel : Mice were rewarded with waterterm for licking following upward drifting oriented gratingsterm, and received negative reinforcement for licking to other non-targetterm orientationsterm, including a 500 ms white noise burst, and, early in training, a 2 s time-out and a mild airpuff stimulusterm. Lower panel : Following the end of the 1 s stimulusterm (yellow box), mice had 1 s to lickterm to indicate a response (purple line). Stimuli were presented 7–8 s apart, which encouraged cessation of licking following the end of the previous trialterm. Mice could also be partially deterred from spurious licking by aborting trialsterm with lickterm responses immediately following stimulusterm onset (blue line). Blocks in which the mouseterm performed this basic discriminationterm·mth paradigm could be inter-digitated with passive viewingterm·mth blocks in which the lickspout was removed. (B) Example behavioral sessionterm containing active discriminationterm·mth and passive viewingterm·mth trialsterm. Behavioral responses are shown for active discriminationterm·mth trialsterm only, sorted by stimulusterm type. These data demonstrate correct decisions (green lines) in the majority of trialsterm – Lickterm responses to targetterm stimuli (top), and suppression of responses following difficult non-targetsterm (middle; 11°) and easy non-targetsterm (bottom; 90°). The average performance and discriminability ( d ′)term·stat is shown at the right. Note that minimal licking occurred prior to stimulusterm onset or during the enforced lick-suppression window (blue lines: aborted trialsterm). Behaviorterm did not extinguish following passive viewingterm·mth blocks (gray arrows at right indicate onset of each active discriminationterm·mth block, not including five initial targetterm trialsterm). (C) Same sessionterm, cumulative incidence of targetterm trialsterm containing lickterm responses. Mice quickly learned to cease licking following targetterm stimuli when the lick-spout was removed during passive viewingterm·mth blocks (gray bars).
Rewards during automated training were 5–6 μl drops of waterterm delivered using a computer-controlled solenoid (Minuteman Controls, EV 2-6), via a shaped syringe and plastic spout. Licksterm were detected as breaks in an optical beam (940 nm LED, Edmund), using custom electronics. Mice received auditory feedback following correct and incorrect lickterm responses (2 kHz tone or noise, respectively) presented via computer speakers. Early in training, incorrect licksterm or licksterm early in stimulusterm presentation also triggered presentation of a mild, brief airpuff (100 ms) to the facial skin, using a solenoid (Figure 3 ), as well as additional ‘time-out’ delays (∼2 s) before commencing the next trialterm. In some sessionsterm, we used alternating active discriminationterm·mth and passive viewingterm·mth (no reward) blocks of trialsterm (43 trialsterm/block; see Figure 3 and Section ‘ Variation of Behavioral Contextterm ’, below). We further deterred mice from licking during passive viewingterm·mth blocks by swinging the lickterm spout in and out of reach using a solenoid (see Figure 1 A). Sessionsterm were concluded when mice ceased to be engaged in the taskterm. Because the mean number of active discriminationterm·mth trialsterm per sessionterm was ∼300 (Table 1 ), and ∼40% of trialsterm were targetterm trialsterm, then a 25-g mouseterm with a 90% hit rateterm (HR) would receive, on average, ∼0.65 cc of waterterm during a sessionterm (minimum total daily waterterm for a mouseterm of this weight: 0.625 cc). For sessionsterm with greater incidence of rewarded trialsterm, no additional restriction was used on subsequent days. If less that the minimal amount of waterterm was provided during training, then the mouseterm was supplemented to the minimal amount while in his home cage, >1 h following training.
Table 1.
Mice discriminate well for many trialsterm per sessionterm, and for many sessionsterm .
Mouseterm no. Time since onset (days) No. of sessionsterm with hit rateterm ≥70 for 2+ blocks No. of trialsterm per sessionterm d -Prime for easy non-targetterm, 90° diff.
d -Prime for hard non-targetterm (used in a subset of all sessionsterm) Mean ± SD across sessionsterm of hard non-targetterm (degrees) 1 444 167 285 ± 155 3.3 + 0.9 2.4 ± 0.8 (121) 13 ± 8 2 343 126 257 ± 136 3.6 + 0.7 2.3 ± 1.1 (103) 15 ± 10 3 168 96 270 ± 84 1.8 + 0.8 1.3 ± 0.9 (79) 41 ± 21 4 182 87 280 ± 114 2.4 + 0.9 2.2 ± 0.9 (81) 26 ± 18 5 175 74 205 ± 84 3.0 + 1.1 2.2 ± 0.7 (65) 27 ± 19 6 189 70 301 ± 165 2.4 + 0.9 1.6 ± 0.9 (60) 25 ± 18 7 190 66 329 ± 111 1.8 + 1.4 1.2 ± 0.7 (35) 35 ± 28 8 102 52 387 ± 206 2.7 + 0.9 1.6 ± 0.6 (30) 26 ± 14 9 48 26 373 ± 142 2.1 + 0.3 0.6 ± 0.6 (26) 26 ± 17 Mean ± SD 205 ± 120 84.9 ± 41.5 299 ± 57 2.6 ± 0.6 1.7 ± 0.8 26 ± 9 Values are mean ± SD across trialsterm. Bolded values are mean ± SD of the means across mice .
Eyetracking Eyetracking was performed using an ISCAN eyetracking system (240 Hz) coupled to a macro lens, positioned along the body axis of the mouseterm (Figure 1 A). The eyeterm was illuminated using 940 nm LEDs positioned near the eyeterm and out of the way of the monitor. A 940 nm narrow-band filter (Edmund) was placed in front of the eyetracking camera to block infra-red laser light emanating from the pupil. A second camera collected images of the eyeterm at lower rates (20 Hz) via the ISCAN monitor output. This provided a simple, low-tech solution for detecting large horizontal eyeterm displacements (Figure 2 D), and could be combined with less expensive eyetracking cameras (see also Stahl et al., 2000 ; Sakatani and Isa, 2004 ). We took a collapsed, horizontal cross-section of the eyeterm from the 20 Hz images, located the pupil boundaries in each frame, and took the midpoint as the pupil positionterm·var (Figures 2 D,E). This provided an offline means of tracking pupil positionterm·var and rejecting trialsterm with large eye movementsterm·var. In general, horizontal eye movementsterm·var ( Figure S1 in Supplementary Material) and, to an even greater extent, vertical eye movementsterm·var (not shown) were rare in mice during passive viewingterm·mth, similar to eyeterm stability during passive viewingterm·mth in rabbits (Cano et al., 2006 ).
Online ISCAN estimates of eyeterm rotation using pupil positionterm·var were contaminated by small translational motionsterm (<300 μm) during licking or chewing, which could be removed by online tracking and subtraction of the corneal reflection spot generated by LED illumination ( Figure S2 in Supplementary Material). Pupil displacement estimates using online or offline software were converted into angular displacements by correcting for off-axis camera location and assuming that the center of eyeterm rotation was 1.09 mm behind the pupil (Stahl et al., 2000 ). For synchronization of eyetracking, visual stimulation, behavioral responses, and imagingterm·mth rigs, ISCAN outputs and all other triggers were digitized (2 kHz, National Instruments) and sorted offline (MATLAB, The MathWorks).
Behavioral analyses Targetterm stimuli were upwards drifting gratingsterm, and non-targetterm stimuli were identical gratingsterm drifting in other directions (see Figures 3 and 4 ). Hit rateterm (HR; fraction of trialsterm with licksterm following targetterm presentation) and false-alarm ratesterm·def (FA; fraction of trialsterm with licksterm following non-targetsterm) for each non-targetterm stimulusterm type were calculated for each block of 43 trialsterm. The number of active discriminationterm·mth trialsterm per sessionterm (Table 1 and Figure 4 ) only included trialsterm in blocks with HRs exceeding 0.7. A useful measure of behavioral discriminabilityterm·var of targetsterm from non-targetsterm is given by d ′ = norminv(HR) − norminv(FA), where norminv() is the inverse of the cumulative normal functionterm·stat (Green and Swets, 1966 ). Values for HR and FA were truncated between 0.01 and 0.99. Higher d ′ values indicate better discriminabilityterm (maximum d ′ = 4.65). To estimate orientationterm thresholdterm on a given day (Figure 4 C), we performed a linear fitterm·stat to d ′ values (for all non-targetsterm with d ′ < 2.5) and estimated the angular distanceterm·var from the targetterm predicted to generate a value of d ′ = 1 (see Section ‘ Taskterm Acquisition and Long-term Stability ’).
Figure 4.
Taskterm acquisition and long-term stability . (A) Discriminability ( d ′)term·stat of targetterm from non-targetterm stimuli improved over days, as shown in four mice (four different colors). Dashed lines: d ′ for non-targetsterm oriented 90° from the targetterm. In subsequent sessionsterm, more difficult non-targetterm stimuli including 45° discriminationterm·mth (solid lines) were used. (B) Example of psychometric curvesterm·stat across days for one mouseterm. Increased angular distanceterm·var of non-targetsterm from targetterm resulted in better discriminationterm·mth performance, as expected. Later training days (lighter curves) showed general improvement in performance. Frequency of stimulusterm presentation: (targetterm, difficult non-targetterm,…, easiest non-targetterm) = (35, 7, 7, 31, 14, 6%). (C) Using these daily psychometric curvesterm·stat, orientationsterm thresholdsterm were estimated, for this mouseterm (blue line) and three other mice, as the angular discriminationterm·mth achievable at a performance level of d ′ = 1. Traces in (A,C) were smoothed with a 3-day boxcar filter for presentation. (D) Example of long-term stability in a mouseterm presented with three stimuli: Targetterm (40% of stimuli), difficult non-targetterm (∼15° from targetterm; 45% of stimuli) and easy non-targetterm (90°; 15% of stimuli). This mouseterm demonstrated stable performance in 167 sessionsterm spanning over one year ( left panel ) and was engaged in the discriminationterm·mth behaviorterm for several hundred trialsterm a day ( right panel , number of trialsterm per day, mean ± SD). Later sessionsterm (gray bars) involved alternating discriminationterm·mth blocks and passive viewingterm·mth blocks (the latter blocks, while not included in ordinate estimates, provided a useful control during subsequent imagingterm·mth). For summary statistics of stability and performance across nine mice, see Table 1 . These data suggest that the current behavioral method can be used in studies imagingterm·mth the same neuronsterm across days and months, during taskterm learningterm or disease progression/therapeutic intervention.
Surgery and indicatorterm delivery Expression of genetically encoded calciumterm indicatorterm, YC3.6 Viral expression of the genetically encoded calciumterm indicatorterm (GECIterm) yellow cameleon 3.6 (pAAV-CBA-YC3.6-WPRE, Kuchibhotla et al., 2008 ; original FRET based probe described in Nagai et al., 2004 ) was achieved in the left primary visual cortex, as follows: Mice were anesthetized and the scalp resected using aseptic technique. Approximately 1 μl of YC3.6 was injected over 10 min though a ∼0.5 mm burrhole (2.7 mm lateral, 0.5 mm anterior to lambda), between 200 and 400 μm deep, using a glass pipette (20 μm tip) and a Picospritzer (<5 psi). After leaving the pipette in place for an additional 10 min, the pipette was removed and sutures applied to skin margins.
Headpost and other implants, craniotomy A titanium headpost (grade 2, 0.05′ thick, 0.95 g; Figure 1 C) was implanted using aseptic technique, as follows: Mice were anesthetized with an initial dose of ketamine/xylazine (100 and 5 mg/kg), maintained on isoflurane (1.5% in 100% O 2 ), and positioned in a stereotactic apparatus on a heating blanket. Atropine was administered (0.2 mg/kg, IM) and eyesterm were protected with ointment (Vetropolycin). Following resection of skin and clearing of skull fascia, a skull screw (00-96, Plastics One) was cemented in place placed on the anterior/right bone plate. In some mice, subdural electrocorticogram electrodes and a connector (Plastics One) were positioned over the right visual cortex in order to obtain gross measures of brain state (not shown). The skull was then covered with a thin layer of cyanoacrylate (Vetbond) and allowed to dry, followed by cementing of the headpost in an axis approximately parallel with the skull over the left primary visual cortex. White dental cement powder (Lang) mixed 3:1 with black powder paint (Pearl Arts #101463) was applied for structural support and to prevent stray light from the monitor from entering through the skull. O-rings (#13 and #15; McMaster-Carr) were glued to the headpost concentric with the center of the targeted craniotomy, to form a well for imagingterm·mth (Figure 1 C) and to provide light-tight coupling with a #17 O-ring glued to flexible light shielding material (Thorlabs) that could be wrapped around the imagingterm·mth objective. All wound margins adhered to the perimeter of the headpost and dental cement, and were treated with Vetropolycin prior to recovery. Buprenorphine injections (4 × 0.05 mg/kg) and prophylactic injections of antibiotics (cefazolin 500 mg/kg) were administered, and mice were single housed and allowed to recover with freely available waterterm mixed with sulfatrim antibiotic (1:32) for >5 days before water schedulingterm·var recommenced.
Following habituation to head restraint and training in the visual discrimination taskterm·def (Figure 1 A), a 3-mm diameter craniotomy was performed over primary visual cortex. Procedures were similar to headpost surgery except that dexamethasone (8 mg/kg, IM) was also applied. Care was taken to minimize dural tears or bleeding. Either sterile ACSF (HEPES-based, glucose-free), saline, or 1.5 or 3% agarose (type III-A, Sigma-Aldrich) was applied and a 5-mm glass coverslip (WPI) was sealed over the craniotomy using dental cement. The mouseterm was allowed to recover for at least 1–2 days before imagingterm·mth.
Loading of synthetic calciumterm indicatorterm In some experiments (Figure 1 A), we performed a bolus loading of the calciumterm indicatorterm Oregon Green BAPTA-1 AM (OGB1-AMterm; Invitrogen) with 10% DMSO and Pluronic F-127 (Stosiek et al., 2003 ; Histed et al., 2009 ), 200–300 μm below the cortical surface, under light isoflurane anesthesia. We distinguished neuronsterm from astrocytesterm by labeling astrocytesterm via co-injection of the red dye sulforhodamine-101 (SR101term, Nimmerjahn et al., 2004 ). Passive viewingterm·mth and/or behaviorterm sessionsterm began 1–2 h after recovery from surgery in the home cage, and lasted 2–5 h. Note that no additional systemic drugs (e.g., atropine) were used during this surgery that could influence subsequent behaviorterm, and eyesterm were lightly taped shut to protect them from drying. Re-injection on subsequent days was achieved by removing the coverslip and penetrating the pre-existing agarose during dye injection, which minimized additional dural bleeding.
Two-photonterm·mth calciumterm imagingterm·mth Imagingterm·mth hardware and software The system for two-photonterm·mth imagingterm·mth in behaving mice is shown in Figure 1 B. In vivo calciumterm imagingterm·mth was conducted with a homebuilt two-photonterm·mth microscope. For optimal detection sensitivity, we used a two-inch optical path (2′ lenses and hardware for Thorlabs, 2″ filters and dichroics from Semrock and Chroma), and gallium arsenide photomultiplier tubes (PMTs, 2× increased sensitivity; Hamamatsu #H7422P-40MOD) coupled to electronic shutters (Uniblitz) to prevent PMT damage. The microscope permitted rapid alternation between blue/green and green/red dichroic cubes, in order to isolate signals from CFP or GFP (blue), YFP or OGB1-AMterm (green), or sulforhodamine (red; FF01-457_50, FF01-542_50, and FF01-629_53, respectively; Semrock). We achieved high excitation efficiency using an Olympus 25× (NA 1.05) or Nikon 16× (NA 0.8) objective together with a Mai Tai laser (830 nm for YC3.6, 800 nm or 920 nm for OGB 1-AM) with group delay dispersion compensation (Deep See ‘pre-chirp’ module, Newport). Scanning galvanometers (Cambridge Technology) provided a frame rate up to 64 Hz (32 × 32 pixels/frame, pixel dwell time: 16 μs), which required passive cooling of mirrors and active cooling of mirror drivers (using a chiller, WAtronix).
To achieve complete three-dimensional imagingterm·mth of small volumesterm, we scanned the objective in a trapezoidal pattern using a piezo-scanner (Physik Instrumente, P-721.LLQ; 48 μm displacement, 3.5 μm/planeterm; Figure 1 B) controlled by a PC computer running custom-written Labview software, synchronized to output triggers sent from the two-photonterm·mth acquisition computer. Image acquisition was controlled by a modified version of ScanImage (Pologruto et al., 2003 ) and MATLAB (The MathWorks). To maintain constant fluorescenceterm excitation across multiple depths and to avoid passing laser power during the two ‘flyback’ frames, we often modulated laser power with a Pockels cell (e.g., between 16 and 13 mW in Figure 6 ), which was controlled by the MATLAB program.
Imagingterm·mth acquisition and analysis All image analyses were performed in MATLAB. Imagingterm·mth during behaviorterm typically involved three to four runs, each lasting 25 min (∼200 trialsterm/run, 100000 frames/run). Three-dimensional volumesterm (14 planesterm) were acquired at 4 Hz. These data were analyzed using a rigid, subpixel alignment procedure based on a 3Dterm cross-correlationterm·stat with an average volumeterm generated from motion-free epochs. Alignment was performed on the OGB-1 channel or YFP channel (for YC3.6) and the estimated shifts were then applied to that channel as well as the SR101term or CFP channel, respectively. Volumesterm could also undergo subsequent co-registrationterm·mth that corrected for motionsterm faster than 4 Hz (Figure 5 ), by first performing rigid 2Dterm co-registrationterm·mth on each planeterm in the volumeterm, followed by estimation of shifts of the planeterm in the Z dimension, and interpolation back to an evenly spaced volumeterm.
Figure 5.
Three-dimensional imagingterm·mth during behaviorterm improves recording stability . (A) Average imagingterm·mth volumeterm corresponding to OGB calciumterm indicatorterm ( top left , green fluorescenceterm) and static astrocytic label SR101term ( bottom left , red fluorescenceterm). Individual OGB volumesterm acquired at 4 Hz were co-registered in 3Dterm, and estimated shifts were then applied to SR101term data. (B) SR101term data from a small square region at each imagingterm·mth depth (yellow rectangular solid in A ) surrounding an astrocyteterm was collapsed (in X – Y planeterm) for each 250 ms volumeterm, revealing an apparent ‘sinking’ of the astrocyteterm to deeper planesterm during lickterm/jaw motionterm ( top panel ). Following post-processing and 3Dterm rigid and subsequent plane-by-plane co-registrationterm·mth, the cell motionterm was greatly reduced ( bottom panel ; only a subset of planesterm were used after motionterm correction). (C) For typical studies imagingterm·mth from one planeterm, SR101term fluorescenceterm from the edge of this astrocyteterm should be stable over time, but instead shows severe motionterm artifactsterm ( top panel ). Gray ticks indicate licking, which mostly occurred during engaged blocks but not during passive viewingterm·mth (black line). Using standard 2Dterm rigid co-registrationterm·mth techniques ( middle panel ), some artifactsterm are reduced. However, 3Dterm co-registrationterm·mth was necessary for near-complete elimination of artifactsterm ( bottom panel ), particularly those concurrent with onset of lickterm responses. Histograms on right indicate a narrower bandwidth of d F / F values using 3Dterm correction, likely due to improved robustness to licking and other motionterm artifactsterm (e.g., breathing, heart rate).
Time courses of responsesterm·var were obtained by summing signals from all pixels within 3Dterm cell masks. The cell masks were first derived using an automated algorithm, then corrected by hand to eliminate obvious errors, and finally eroded to minimize contamination by neuropil signals. For ratiometric imagingterm·mth (Figure 7 ), fluorescenceterm ratio ( R ) was defined as the ratio of YFP fluorescenceterm (green channel) to CFP fluorescenceterm (blue channel). The mean ( F 0 or R 0 ) was subtracted from each trace, and traces were subsequently high-pass filtered (200 s cutoff, third order Butterworth filter). Traces were then converted to fractional changes: d F / F 0 = [ F ( t ) − F 0 ]/ F 0 , d R / R 0 = [ R ( t ) − R 0 ]/ R 0 . For average traces (Figures 2 C, 6 D and 7 B,C), the mean of each trace in the 2 s prior to stimulusterm onset was subtracted from all time points.
Figure 7.
Stable recordings during visual behaviorterm using a genetically encoded calciumterm indicatorterm . (A) Strong calciumterm responses (black trace) during visual behaviorterm recorded from a cell with virally induced expression of the ratiometric calciumterm indicatorterm, YC3.6 (132–180 μm deep; 15 mW excitation at 830 nm). The trace was computed as the ratio of YFP to CFP fluorescenceterm (using 4 Hz volumeterm acquisition). All lines and colors are as in Figure 6 A. (B) Top : Color coded stimulusterm directions used. Middle one planeterm in the volumeterm collected during imagingterm·mth in (A) . Bottom : Lower magnification image illustrates sparsity of expression. (C) Average responses (mean ± SE) to targetterm stimuli presented during first and second hour of behaviorterm. Response strength did not decline over two consecutive hours of imagingterm·mth. (D) Another neuronterm in the same mouseterm demonstrated visual responses during behavioral sessionsterm that were spaced over a month apart. Top : Timeline and images of the same neuronterm and its processes across behavioral sessionsterm (and number of behavioral trialsterm per sessionterm). Bottom : Responses to targetterm stimuli during a large number of behavioral trialsterm in four different imagingterm·mth sessionsterm.
Results Overview Our procedures for imagingterm·mth cellular responses during passive viewingterm·mth and visual behaviorterm are summarized in Figure 1 A (see also Section ‘ Materials and Methods ’). The sequence began with daily handling, headpost implant surgery, and habituation of the animal to head restraint and sitting in a plastic half cylinder (Figure 1 B; ease of habituation was a useful indicatorterm of successful taskterm acquisition and stable neural recordings). Next, a cranial window was placed over left visual cortex, and the mouseterm was allowed to recover. At this point, head-fixedterm·mth mice could be trained on visual discrimination tasksterm·def of varying complexity and difficulty (Section ‘ Stable Visual Discriminationterm·mth Behaviorterm in Head-fixedterm·mth Mice ’), allowing combination with three-dimensional cellular imagingterm·mth during visual behaviorterm (Section ‘ Chronicterm Cellular Imagingterm·mth of Visual Responses During Behaviorterm ’). Alternatively, acute mapping of visual responses in non-anesthetized mice could be conducted using a more basic, two-dimensional imagingterm·mth rig (Section ‘ Cellular Imagingterm·mth in Identified Neuronsterm During Awaketerm, Passive Viewingterm·mth ’).
Cellular imagingterm·mth in identified neuronsterm during awaketerm, passive viewingterm·mth Before we attempted to record visual responses during behavioral engagementterm (Section ‘ Chronicterm Cellular Imagingterm·mth of Visual Responses During Behaviorterm ’), we first established simple methods for imagingterm·mth during passive viewingterm·mth. We found that stable neural responses (with minimal brain motionterm or eyeterm motionterm) could be obtained in the majority of trialsterm during passive stimulusterm presentation ( Figure S1 in Supplementary Material), enabling visual mapping studies in un-anesthetized mice using standard two-dimensional imagingterm·mth. Figure 2 provides an example of calciumterm responses of cortical neuronsterm during awaketerm, passive viewingterm·mth, using acute loading of synthetic calciumterm indicatorterm, OGB1-AMterm (cf. Ohki et al., 2005 ; Mrsic-Flögel et al., 2007 ). Excitatory neuronsterm and inhibitoryterm neuronsterm could be classified in vivo in this mouseterm due to expression of green fluorescent protein in all GABAergicterm neuronsterm (Figure 2 A, green neuronsterm; Tamamaki et al., 2003 ; Sohya et al., 2007 ; KerlinpersName et al., 2009 ). An example of orientation-tuned responses to drifting gratingsterm (28°/cycle square wave, 2 Hz) is shown in Figure 2 B. Responses could be detected in single trialsterm in GABAergicterm and non-GABAergic neuronsterm (Figure 2 C). These data illustrate the value of two-photonterm·mth calciumterm imagingterm·mth in awaketerm, transgenic mice, providing robust responses in genetically identified cell typesterm.
Imagingterm·mth in head-fixedterm·mth mice also facilitates monitoring of eye movementsterm·var. In general, drifting gratingsterm in the range of low spatial (0.02–0.1 cyc/deg) and temporal (1–2 Hz) frequencies employed here did not evoke significant eye movementsterm·var. We monitored eyeterm motionterm using an infra-red camera and simple video analyses (20 Hz; e.g., Figure 2 D and Figure S1 in Supplementary Material) or using more sophisticated online tracking software (240 Hz; see Figure S2 in Supplementary Material, and Section ‘ Materials and Methods ’). Figure 2 D demonstrates the relative stability in eyeterm motionterm across several hundred seconds of passive viewingterm·mth from the above experiment (slow drifts <10°, and no or minimal stimulus-evoked eyeterm motionterm, <4°; for additional analyses, see Figure S1 in Supplementary Material). Given the larger size of cortical receptive fields in mice compared to higher mammals (Dräger, 1975 ; Mangini and Pearlman, 1980 ; Metin et al., 1988 ; Niell and Stryker, 2008 ), small eyeterm motionsterm should not influence cortical responses significantly, particularly when using a very large screen and sinusoidal gratingsterm of low spatial frequency. Rare, large eye movementsterm·var can be tracked and rejected offline ( Figure S1 in Supplementary Material).
Stable visual discriminationterm·mth behaviorterm in head-fixedterm·mth mice Behavioral setup Head-fixedterm·mth mice could be trained to discriminate between gratingsterm drifting in different directions, using a ‘GO–NOGOterm·mth’ operant discriminationterm·mth taskterm (see Section ‘ Materials and Methods ’). The taskterm structure is described in Figure 3 A. Water-scheduled mice were trained to lickterm following presentation of an upward drifting ‘targetterm’ gratingterm, and to suppress licking following ‘non-targetterm’ gratingsterm drifting in other directions. Correct detection of a targetterm (‘hit’) was indicated with a pure tone (2 kHz) and rewarded with a drop of waterterm (5–6 μl), while responses to non-targetterm stimuli (‘false-alarmsterm’) were indicated using white noise.
For initial training sessionsterm, we presented only targetterm stimuli (at 0.5 Hz), and delivered a drop of waterterm automatically following each presentation. The inter-stimulus interval was then gradually extended by several seconds across sessionsterm, while maintaining high response rates (>80%). At this point, we switched to an operant discriminationterm·mth taskterm, by (1) making reward presentation contingent on a lickterm response, and (2) by presenting non-targetterm stimuli as well as targetterm stimuli. For several sessionsterm early in training, a mild airpuff stimulusterm and an additional time-out period (∼2 s) were also used to discourage responses to non-targetterm stimuli. Early lickterm responses following stimulusterm presentation, which could diminish accuracy and hasten lick-related brain motionterm (see below) and eyeterm motionterm ( Figure S2 in Supplementary Material), were discouraged during training by aborting trialsterm in which licking occurred in the first several hundred milliseconds (Figures 3 A,B). We maintained 7–8 s between trialsterm to allow for previous lickterm responses to abate (and to permit previous evoked calciumterm signals to fully decay back to baseline). While we did not penalize licking prior to stimulusterm onset, most mice typically did not lickterm for several seconds prior to stimulusterm onset (e.g., Figure 3 B; but see Figure 6 A for an example of more sustained licking between trialsterm).
Taskterm acquisition and long-term stability About half of the mice tested (13/25) learned to perform the visual discrimination taskterm·def, but with variable learningterm rates. Training proceeded as follows: First, mice were rewarded for licking in response to drifting targetterm stimuli, using an inter-stimulus interval that was gradually increased from 2 to 5–8 s. Subsequent training involved gradually increasing the similarity of target and non-target stimuliterm·def in 5–15 sessionsterm by presenting non-targetterm stimuli that were (1) orthogonal to targetterm and stationary, (2) orthogonal and of increasingly similar speed to targetterm, (3) of increasingly similar orientationterm to targetterm gratingsterm and of identical speed.
The latter part of this learningterm curve is shown for four mice, in sessionsterm employing easy non-targetsterm that were orthogonal to targetterm stimuli and of equal speed, followed by sessionsterm in which non-targetsterm were presented at five increasingly difficult orientationsterm (solid lines). We calculated discriminabilityterm of targetterm from non-targetterm stimuli using the ‘ d -prime’ measure ( d ′, see Section ‘ Materials and Methods ’). Supposing a fractional response rate to targetterm stimuli of 0.8, d ′ values 0, 1, 2, and 3 correspond to non-targetterm response rates of 0.8 (chance performance), 0.4, 0.12 and 0.015, respectively. In all mice, d ′ values improved over days using easy (orthogonal) non-targetterm stimuli (Figure 4 A, dashed lines). Discriminationterm·mth rates in sessionsterm involving more difficult non-targetsterm (including 45° non-targetsterm) began at an intermediate baseline performance and improved, suggesting partial transference of initial learningterm (Figure 4 A, solid lines). Complete psychometric curvesterm·stat across daily sessionsterm are shown for one mouseterm (Figure 4 B). Orientationterm discriminationterm·mth thresholdsterm (at the level of d ′ = 1) calculated for each day (see Section ‘ Materials and Methods ’) demonstrated steady improvement (Figure 4 C). Within ten sessionsterm, mice were able to discriminate orientationsterm as little as 10–20° apart, consistent with previous observations in freely moving mice (Reuter, 1987 ). Absolute orientationterm thresholdsterm likely depend on incidence of presentation of each stimulusterm (Figure 4 B, legend), and taskterm structure (e.g., GO–NOGOterm·mth vs. 2-AFCterm·mth).
In a separate group of nine mice, we found that mice could perform the discriminationterm·mth taskterm routinely for several months to over a year, and typically performed on the order of 300 trialsterm/day (Figure 4 B, Table 1 and Section ‘ Materials and Methods ’; over 1000 trialsterm or 2 h in some sessionsterm). Consistently high levels of discriminationterm·mth performance were observed for orthogonal as well as for more difficult non-targetterm stimuli (Figure 4 D and Table 1 ). During this time, mice were also gradually habituated to the imagingterm·mth environment, including the sounds and motionsterm produced during typical microscope operation. These data suggest that head-fixedterm·mth visual discriminationterm·mth learningterm and steady-state performance can be reliably measured across many sessionsterm – a key factor enabling chronicterm calciumterm imagingterm·mth studies of neural correlates of perception and perceptual learningterm·dom in the same neuronsterm (see below), in both wild-type mice and in mouseterm models of disease and aging.
Variation of behavioral contextterm The above taskterm can be readily modified to study context-dependent modulationterm of responses by varying taskterm difficulty (proportional of ‘difficult’ vs. ‘easy’ non-targetterm trialsterm), or task engagementterm·var (presence/absence of reward). We varied task engagementterm·var using interleaved blocks of 43 trialsterm in which mice either actively discriminated target and non-target stimuliterm·def, or passively viewed these same stimuli (Figure 3 A). During passive viewingterm·mth, a solenoid removed the lickspout from reach (Figure 1 B), and mice quickly learned to stop licking in response to targetterm stimuli until the following ‘engaged’ block began, suggesting that the learned stimulusterm association is not purely reflexive (Figure 3 C). Presentation of stimuli during passive viewingterm·mth, or of novel stimuli during orientationterm mapping, did not lead to extinction of subsequent discriminationterm·mth behaviorterm (e.g., Figures 3 B,C and 4 D), allowing for comparison of neural responses obtained during behaviorterm with receptive field mapping obtained during passive viewingterm·mth.
Chronicterm cellular imagingterm·mth of visual responses during behaviorterm Three-dimensional motionterm correction A major obstacle in our early efforts developing two-dimensional calciumterm imagingterm·mth during behaviorterm involved systematic, lickterm response-induced brain motionterm artifactsterm. To overcome this problem, we modified previous techniques (Göbel et al., 2007 ) to achieve rapid, trapezoidal vertical movement of the microscope objective (Figure 1 B, Section ‘ Materials and Methods ’), generating complete local volumesterm at sufficiently fast rates to correct for rapid brain motionterm (and slow drift) in three dimensions. We found an acceptable trade-off between volumeterm acquisition rate and volumeterm size (and hence number of recorded neuronsterm) by collecting a volumeterm of 14 planesterm (3.5 μm spacing between planesterm) at 4 Hz, typically resulting in volumesterm up to 60 × 60 × 48 μm 3 (containing ∼20 neuronsterm; Figure 5 A). Pre-processing of acquired volumesterm involved correcting each volumeterm for rigid translation in 3Dterm. Subsequently, each planeterm (acquired at 64 Hz) could be individually translated in 3Dterm, allowing correction for motionsterm occurring faster than 4 Hz (see Section ‘ Materials and Methods ’).
The improvements in recording stability during behaviorterm gained by incorporating three-dimensional imagingterm·mth are illustrated in Figure 5 . A local volumeterm was loaded with a green synthetic calciumterm indicatorterm (OGB1-AMterm), and astrocytesterm were co-loaded with sulforhodamine (SR101term; Nimmerjahn et al., 2004 ), a red, calcium-insensitive fluorescent dye that generates a constant average fluorescenceterm level in the absence of brain motionterm (Figure 5 A). Figure 5 B ( top panel ) shows a time-lapse image of average SR101term signal from a loaded astrocyteterm at multiple depths (Figure 5 A, yellow cube). This cell demonstrated up to 15 μm of ‘sinking’ Z-displacement as the mouseterm adjusted his posture, opened his jaw and began to lickterm – a displacement that is well-corrected by 3Dterm co-registrationterm·mth methods (Figure 5 B, bottom ; see also Figure S2B in Supplementary Material). By monitoring fluorescenceterm in a single planeterm (Figure 5 A, orange square), one can observe artifactual effects of brain motionterm on SR101term signal that would occur using 2Dterm imagingterm·mth (Figure 5 C, top ). Brain motionterm artifactsterm not related to licking were mostly corrected using standard 2Dterm motionterm correction techniques, (Figure 5 C, middle ), consistent with previous studies (Dombeck et al., 2007 ; Greenberg and Kerr, 2009 ). However, adequate correction of lick-related motionterm artifactsterm required 3Dterm motionterm correction (Figure 5 C, bottom ). Further, 3Dterm motionterm correction also reduced motionterm contamination not related to licking (Figure 5 C, histograms), likely caused by body motionterm, chewing, breathing, and the heart beat (as indicated by the temporal pattern of shift-corrections estimated at 64 Hz, data not shown).
Imagingterm·mth during visual behaviorterm using synthetic dyes Stable neural responses demonstrating high signal-to-noise ratio were achieved during visual behaviorterm, following loading of synthetic calciumterm indicatorterm OGB1-AMterm. Figure 6 A shows the time courses for 3 of 20 simultaneously recorded neuronsterm. The same quality visual responses were obtained during licking (engaged condition) as during quiet, passive viewingterm·mth. While these data clearly illustrate the strength of our 3Dterm imagingterm·mth techniques for obtaining clean neural traces during active discriminationterm·mth (including during lickterm responses), the behavioral performance demonstrated in Figures 6 A and 7 A was below the level achieved during training (Table 1 ), due to an increase in spurious licking compared to other mice (see Figure 3 B), and even compared to other imagingterm·mth sessionsterm from the same mouseterm (e.g., Figure S2 in Supplementary Material).
Following 2 h of behavioral responses (905 trialsterm), we could still perform detailed mapping of orientationterm preference for an additional hour during passive viewingterm·mth (Figures 6 B,C,E). Responses to presentation of drifting gratingterm stimuli (9.5 s duration) were consistent across presentations (Figure 6 C). The relative response strength of neuronsterm to different stimulusterm orientationsterm during behaviorterm (Figures 6 A and 6 E, top left ) was consistent with orientationterm tuning curves obtained during passive viewingterm·mth (Figure 6 E, top right ).
Two days following this initial recording of neural responses during behaviorterm, we reloaded the same volumeterm of neuronsterm with synthetic calciumterm indicatorterm OGB1-AMterm (Figure 6 D), and were again able to obtain stable responses during behaviorterm and during subsequent orientationterm mapping (Figure 6 E, middle row ). The day after this second injection, OGB1-AMterm expression was dimmer (dye persists in neuronsterm for 8–30 h following injection), but again yielded responses during behaviorterm and consistent orientationterm tuning from a subset of neuronsterm (Figure 6 E, bottom row ).
Imagingterm·mth during visual behaviorterm using genetically encoded calciumterm indicatorsterm Although we found that robust imagingterm·mth of neuronsterm during one or several behavioral sessionsterm is possible using synthetic calciumterm dyes, GECIsterm have recently demonstrated increased brightness and signal-to-noise that may soon rival synthetic indicatorsterm (Tamamaki et al., 2003 ; Mank et al., 2008 ; Wallace et al., 2008 ; Tian et al., 2009 ). GECIsterm have several clear advantages for use in behavioral studies (see Section ‘ Discussion ’). To illustrate the use of GECIsterm to record calciumterm responses during behaviorterm, we used a viral vector to produce sparse expression of the ratiometric calciumterm indicatorterm yellow cameleon (pAAV-CBA-YC3.6-WPRE, Nagai et al., 2004 ; Kuchibhotla et al., 2008 ; Figure 7 B) in layer II/III of primary visual cortex of a wild-type mouseterm via burr-hole injection (see Section ‘ Materials and Methods ’). We subsequently performed a craniotomy and cemented a sealed cranial window in place. After several months of training and stable behaviorterm (Table 1 , Mouseterm 7, which was subsequently used for OGB imagingterm·mth, Figure 6 ), we performed 3Dterm neural recordings during behaviorterm. Increased calciumterm levels during visual stimulation lead to increased levels of YFP fluorescenceterm, and decreased CFP fluorescenceterm ( Figure S3 in Supplementary Material). Figure 7 A illustrates ratiometric responses, which had high signal-to-noise even when the animal was licking. In general, measuring neural activity using the ratio of YFP/CFP fluorescenceterm provided increased robustness to motionterm artifactsterm ( Figure S1 in Supplementary Material; cf. Mank et al., 2008 ) as compared with non-ratiometric indicatorsterm, even without 3Dterm co-registrationterm·mth.
Average responses from this neuronterm did not attenuate across two consecutive hours of imagingterm·mth in one behavioral sessionterm (Figure 7 C, 764 total trialsterm across three orientationsterm), and similar responses were also observed in a previous sessionterm in the same cell (not shown). An additional day of recording from this neuronterm during passive viewingterm·mth demonstrated that the neuronterm was tuned for orientationterm (see Movie S1 in Supplementary Material). In total, we imaged YC-3.6-labeled neuronsterm in over ten behavioral sessionsterm in this mouseterm, including one neuronterm imaged in six different sessionsterm spanning forty-six days (Figure 7 D). The same neuronsterm could be found within minutes of the start of each sessionterm, using epifluorescence imagingterm·mth (Figure 1 B), vascular and dural landmarks, and wide-field two-photonterm·mth images at the appropriate depth (not shown). These results demonstrate the feasibility of combining well-controlled visual behaviorterm with repeated, simultaneous calciumterm imagingterm·mth of the same neuronsterm across days and months.
Discussion We have described a method for achieving robust, chronicterm two-photonterm·mth calciumterm responses from layer II/III cortical neuronsterm in head-fixedterm·mth mice performing a visual discrimination taskterm·def. This technique offers many new opportunities for using molecular and genetic techniques to map and manipulate identified classes of neuronsterm during well-controlled operant behaviorterm. Below, we consider the benefits, shortcomings and possible improvements in the choice of (1) species (mice), (2) behavioral paradigm (head-fixedterm·mth, visual GO–NOGOterm·mth), and (3) recording modality (3Dterm two-photonterm·mth calciumterm imagingterm·mth) to study circuit mechanisms underlying sensory perception, perceptual learningterm·dom, and perceptual dysfunction.
Benefits of studying perceptual mechanisms using cellular imagingterm·mth in mice Why use cellular imagingterm·mth in mice to study mechanisms underlying sensory perception? The major advantage lies in the unique accessibility to genetically modified animals. For example, information on cell typeterm (e.g., GABAergicterm neuronsterm and subtypes, Figure 2 ; O'Connor et al., 2009 ) and potentially inter-areal connectivity (Brown and Hestrin, 2009 ) of recorded neuronsterm can be directly visualized in vivo in transgenic mice. Incorporating information on cell identityterm may help to explain the variability in context-dependent responses across nearby neuronsterm. Two-photonterm·mth imagingterm·mth offers ease of access to most neuronsterm in layer II/III across the lissencephalic cortex. Over nine retinotopically organized visual areas exist in mice, allowing comparative studies across multiple cortical areas. The physiological properties of secondary visual areas are largely unexplored (Wang and Burkhalter, 2007 ), and may demonstrate increased multisensory responses (Brady et al., 2009 ) and variations in context-dependent modulationterm (cf. O'Connor et al., 2002 ). As imagingterm·mth techniques continue to improve (cf. Figure 1 B), the small extent (∼0.5–2 mm 2 ) and relatively thin laminae of mouseterm visual cortical areas could conceivably allow near-complete, simultaneous imagingterm·mth of every neuronterm within layer II/III of a given area during behaviorterm. Our methods could also lend a new dimension to many of the genetic mouseterm models of neurologic disease that currently exist (e.g., Eichhoff et al., 2008 ; Kuchibhotla et al., 2008 ), providing a high-throughput, chronicterm assay of perceptual influences on single neuronsterm throughout the course of disease progression or therapeutic treatment.
Several factors must be considered when selecting particular mice and mouseterm strainsterm for head-fixedterm·mth visual discriminationterm·mth (Deacon, 2006 ; Wong and Brown, 2006 ). Inbred mice, while demonstrating more uniform behaviorterm, may have retinal deficits or diminished capacity for learningterm compared to more outbred strainsterm (e.g., Deacon, 2006 ). Thus, when choosing a strainterm, and when pre-selecting individual mice for training, a compromise must be reached between ease of handling and acclimating to head restraint, and how quickly the animal learns a taskterm (Van Dam et al., 2006 ).
Benefits of using GO–NOGOterm·mth visual discriminationterm·mth in head-fixedterm·mth mice The visual discrimination taskterm·def developed here for head-fixedterm·mth mice recapitulates many of the advantages of head-fixedterm·mth behaviorsterm in primates. First, we found that mice will perform hundreds of trialsterm per sessionterm (up to ∼1000 per sessionterm) for several months to a year, allowing the averaging of visual responses across sessionsterm that may be necessary to reveal more subtle behavioral effects. Note that the motivation of mice to perform the taskterm was strongly dependent on water schedulingterm·var, and increasing sessionterm duration must be balanced with the potential for greater false-alarm ratesterm·def. Although we are unaware of any protocol for training a mouseterm to visually fixate, we found that eye movementsterm·var were small relative to the spatial frequencies of the presented stimuli, and any trialsterm containing occasional large eye movementsterm·var could be rejected offline (see Section ‘ Materials and Methods ’). Our use of sine wave stimuli (vs. square-wave) during recordings further decreased the possibility of strong, local luminance transients induced by eye-movements. Thus, our technique provides sufficiently well-controlled recordings of visual responses in awaketerm/behaving animals to enable some comparisons with the wealth of primate visual studies that are most often carried out in non-anesthetized animals. While the GO–NOGOterm·mth taskterm employed here has disadvantages compared to two-alternative forced-choice (2-AFCterm·mth) tasksterm commonly used in primates and freely moving mice, our taskterm is relatively easy for mice to learn. Many of the strategies developed in the course of our work could be applied to the development of 2-AFCterm·mth visual behaviorsterm in head-fixedterm·mth mice using, for example, a track-ball and virtual reality environment (Harvey et al., 2009 ) or multiple lickterm response ports (P. Reinagel et al., personal communication). The use of response indicatorsterm other than licking (e.g., heading motionterm, left/right lever press using forepaws) may also help mitigate challenges associated with lick-induced brain motionterm. Complementary methods allowing two-photonterm·mth calciumterm imagingterm·mth of visual responses and other calciumterm activity in freely moving rats and mice have recently been published (Engelbrecht et al., 2008 ; Flusberg et al., 2008 ; Sawinski et al., 2009 ). Imagingterm·mth in freely moving animals may allow for easier implementation of naturalistic behaviorsterm in the future, albeit with additional limitations in image quality, visual stimulusterm control, and number of trialsterm per sessionterm as compared to head-fixedterm·mth operant behaviorsterm.
Mice demonstrated improved visual thresholdsterm and taskterm performance over the course of many days, which will prove useful in future studies of perceptual learningterm·dom. The learningterm rates were slower than for a similar olfactory taskterm in head-fixedterm·mth mice (Komiyama et al., 2007 ). These differences could be due to the greater behavioral salienceterm of different odors as compared to initial visual stimuli employed in our taskterm. While differences in stimulusterm speed (Section ‘ Taskterm Acquisition and Long-term Stability ’) were useful for attaining initial discriminationterm·mth of stimuli, varying of other stimulusterm parameters such as luminance, or behavioral factors such as running (Niell and Stryker, 2010 ), may further increase visual salienceterm and shorten acquisition time. In addition, the behavioral techniques described in this paper can likely be automated, allowing high-throughput training of large numbers of mice in the home cage (Meier et al., 2009 ). Future experiments could conceivably decrease mouse-to-mouse variability and user-intensive training/surgical procedures by imagingterm·mth a larger number of (locally or commercially acquired) mice that arrive pre-trained and already expressing calciumterm indicatorsterm. Transgenic mice with cell-type specific expression of sensitive GECIsterm are expected to become commercially available soon, and may replace more labor-intensive approaches (virus injection or in utero electroporation).
Head-fixedterm·mth visual studies may offer improved stimulusterm control compared to other head-fixedterm·mth preparations in rodents, such as tactile studies during active whisking (e.g., Harvey et al., 2001 ; O'Connor et al., 2010 ), allowing enhanced differentiation of internally vs. externally mediated changes in neural activity in different behavioral contextsterm (but see Ferezou et al., 2006 ), and more direct comparison with behavioral neurophysiology experiments in higher mammals. Although mice are nocturnal and have less acute vision than carnivores or primates (Prusky and Douglas, 2004 ; Wong and Brown, 2006 ), we and others have demonstrated reasonably fine neural and behavioral discriminationterm·mth of specific stimulusterm features such as orientationterm (e.g., Dräger, 1975 ; Mangini and Pearlman, 1980 ; Niell and Stryker, 2008 ; Liu et al., 2009 ).
Benefits of two-photonterm·mth calciumterm imagingterm·mth using various indicatorsterm We found that robust calciumterm responses from visualized neuronsterm can be recorded across hours and days during visual discriminationterm·mth, using both synthetic calciumterm dyes and GECIsterm. The choice of whether to use synthetic dyes or GECIsterm depends on the research question at hand. For example, neuronsterm bulk-loaded with OGB1-AMterm dye are typically brighter than neuronsterm expressing GECIsterm, and thus can be imaged with less laser power and potentially less cell damage. Further, OGB1-AMterm loads every cell in a local volumeterm (∼300 × 300 × 300 μm 3 ), thus providing greater yieldterm than GECIterm experiments where only a subset of neuronsterm are been labeled using viral or electroporation techniques. Additionally, we found that clear, stable visual responses could be obtained from the dendrites of GECI-expressing neuronsterm in awaketerm mice ( Movie S1 in Supplementary Material). As GECIsterm designed to report activity at individual synapses become available (e.g., Dreosti et al., 2009 ), cellular integration of behaviorally modulated synaptic input could be explored using the head-fixedterm·mth imagingterm·mth approach described here.
Many GECIsterm such as TN-XXL (Mank et al., 2008 ) and YC3.6 (Figure 7 ; Nagai et al., 2004 ) are ratiometric, which is helpful in canceling out residual motionterm artifactsterm. In our experience, however, three-dimensional imagingterm·mth during behaviorterm was still necessary for ruling out motion-related artifactsterm systematically linked to lickterm responses. Non-ratiometric GECIsterm and synthetic dyes may have greater utility in experiments involving operant behaviorterm when co-expressed with calcium-insensitive fluorescent proteins in order to normalize signals during rapid cell motionterm and slow drift (Knopfel et al., 2006 ). The onset of a bout of licking appeared to generate an apparent sinking of the recorded neuronterm out of the planeterm of imagingterm·mth (Figure 5 ), implying that changes in jaw position may be the cause of the motionterm artifactterm (see also Figure S2B in Supplementary Material). Imagingterm·mth at high rates, as well as maintaining a pressurized cranial window, are all important in obtaining clean spontaneous and evoked signals before and during onset of lickterm responses. The 3Dterm volumeterm imagingterm·mth method described in this paper differs from a previous 3Dterm technique that selectively recorded neural activity at pre-determined cellular spots (Göbel et al., 2007 ). The latter technique, while providing higher yieldterm (large number of imaged neuronsterm), would not be effective during operant behaviorterm involving licking, unless it was combined with a predictive model of cell motionterm (Fee, 2000 ; Dombeck et al., 2007 ).
The observed improvement in behavioral orientationterm thresholdsterm across sessionsterm, combined with the ability to image calciumterm indicatorterm signals across sessionsterm in behaving animals, should open the possibility of studying neural correlates of visual task-related plasticity and learningterm. However, calciumterm indicatorsterm also act as calciumterm buffers that, in some cases, can perturb native calciumterm dynamics (Holthoff et al., 2002 ; Pologruto et al., 2004 ; but see Schmidt et al., 2003 for an example of negligible perturbation using OGB-1). These exogenous calciumterm buffers may also enhance or suppress synaptic plasticity (Tonkikh et al., 2006 ). Plasticity experiments would benefit from in situ estimation of the indicatorterm concentration (Stosiek et al., 2003 ) and calciumterm affinity (Pologruto et al., 2004 ; Hendel et al., 2008 ). While these potential changes in native calciumterm dynamics using certain indicatorsterm could also modify neural activity, initial electrophysiological experiments in vivo have shown little effect of bulk loading of calciumterm indicatorsterm on basic cortical response properties (Ohki et al., 2005 ; Sato et al., 2007 ). For example, Ohki and colleagues observed typical orientationterm tuning curves in single-unit recordings from cat visual cortical neuronsterm loaded with OGB-1 AM (Ohki et al., 2005 ).
Finally, while calciumterm signals do increase monotonically with spike rate, dye and calciumterm kinetics are slow compared to spike recordings (∼10 s of ms onset, 100–1000 s of ms decay). Analysis and calibration methods to extract spike rate information from genetically encoded and synthetic calciumterm indicatorterm signals in cortical pyramidal neuronsterm are beginning to come of age (Smetters et al., 1999 ; Kerr et al., 2005 ; Wallace et al., 2008 ; Bonin et al., 2009 ; Greenberg and Kerr, 2009 ; Mukamel et al., 2009 ; Tian et al., 2009 ). In the case of interneurons, reliable reconstruction of spike trains can be extremely challenging, due to slower calciumterm kinetics (Goldberg et al., 2003 ) and higher firing rates (Niell and Stryker, 2008 ). Precise knowledge of spike timing may not be necessary to begin to address basic questions of behavioral modulationterm in visual cortex, such as whether neuronsterm of a given class increase or decrease their evoked or spontaneous activity during different behavioral tasksterm and brain states.
Conflict of Interest Statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Supplementary Material The Supplementary Material for this article can be found online at http://www.frontiersin.org/cellularneuroscience/paper/10.3389/fncel.2010.00003/ Acknowledgments We thank P. Dempsey and W. Danse for assistance with behavioral training, D. Smith, A. Vagodny and M. Henry for assistance with surgical procedures and headpost development, D. Brittain for assistance in developing 3Dterm imagingterm·mth, and V. Bonin, S. Yurgenson, A. Kampff, D. Robson, M. Histed, B. Sabatini, T. LaFratta, and J. LaFratta for additional assistance with microscope construction. T. Komiyama provided critical assistance with behavioral taskterm development. K. Kuchibhotla and B. Bacskai provided advice and samples of YC3.6. We thank M. Cohen, C. Moore and members of the ReidpersName lab for helpful discussions and comments on the manuscript. This work was supported by NIH grant R01 EY018742 and by the Helen Hay Whitney Foundation (M. L. AndermannpersName).