Abstract
We established a low background, Cre-dependent version of the inducible Tet-On system for fast, cell type-specific transgene expression in vivo. Coexpression of a constitutive, Cre-dependent fluorescent marker selectively allowed single-cell analyses before and after inducible, Tet-dependent transgene expression. Here, we used this method for precise, acute manipulation of neuronal activity in the living brain. The goal was to study neuronal network homeostasis at cellular resolution. Single induction of the potassium channel Kir2.1 produced cell type-specific silencing within hours that lasted for at least 3 d. Longitudinal in vivo imaging of spontaneous calcium transients and neuronal morphology demonstrated that prolonged silencing did not alter spine densities or synaptic input strength. Furthermore, selective induction of Kir2.1 in parvalbumin interneurons increased the activity of surrounding neurons in a distance-dependent manner. This high-resolution, inducible interference and interval imaging of individual cells (high I5, HighFive) method thus allows visualizing temporally precise, genetic perturbations of defined cells.
SIGNIFICANCE STATEMENT Gene function is studied by KO or overexpression of a specific gene followed by analyses of phenotypic changes. However, being able to predict and analyze exactly those cells in which genetic manipulation will occur is not possible. We combined two prominent transgene overexpression methods to fluorescently highlight the targeted cells appropriately before cell type-specific transgene induction. By inducing a potassium channel that decreases neuronal firing, we investigated how neuronal networks in the living mouse brain possibly compensate swift changes in cellular activities. Unlike in vitro, known compensatory homeostatic mechanisms, such as changes in synapses, were not observed in vivo. Overall, we demonstrated with our method rapid genetic manipulation and analysis of neuronal activities as well as precision transgene expression.
Introduction
Neuronal homeostasis ensures an exquisite balance of activities across the entire brain while simultaneously allowing function altering perturbations to the networks. Homeostasis plays a governing role during development, learning, and memory, as well as health and disease. However, with hundreds of different types of neurons in the brain (Saunders et al., 2018) each with distinct gene expression profiles, it is not surprising that cell type-specific contributions to maintenance of balanced networks remain mostly unclear. In addition, perturbations to neuronal homeostasis are often compensated for on various time-scales (Pozo and Goda, 2010; Turrigiano, 2011) making it difficult to experimentally dissect and identify fundamental principles. In vitro, studies of homeostasis have substantially helped our understanding of neuronal networks (Turrigiano et al., 1994). An important publication by Burrone et al. (2002) identified key homeostatic mechanisms in vitro by silencing individual neurons with Kir2.1, an inwardly rectifying potassium channel. The perturbed network responded by increasing the input strength to the silenced neuron which interestingly did not alter its synaptic density. For two decades now, these in vitro findings withstood their validation in vivo, mostly because of technical difficulties. It is thus unclear whether similar homeostatic mechanisms occur in the living brain after cell-autonomously silencing individual neurons.
We decided to address this important question. For this, we first had to reinvent inducible transgene expression in vivo because existing conditional transgene methods do not provide fast, cell type-specific expression. Additionally, we needed to be able to characterize transgene-induced downstream changes in surrounding cells neighboring the manipulated cells. Currently, the powerful Cre/lox system is widely used for expression of Cre-dependent (“floxed”) transgenes in specific cell types (Yarmolinsky and Hoess, 2015). Cellular specificity is achieved through well-characterized endogenous or exogenous promoters. The conditional tamoxifen-inducible Cre/lox version (Feil et al., 1996) adds temporal control, although the time resolution is poor as transgene expression is typically induced by injections of tamoxifen over several days. Conversely, the faster Tet-On system was shown to express transgenes within hours after administration (Hasan et al., 2001), but there tend to be issues with elevated background expression and lower inducibility (Zhu et al., 2007). Thus, both the Cre/lox and the Tet systems have strengths and weaknesses. Importantly, there are currently no methods existing that allow prior identification and visualization of those cells in which the respective transgene will be induced. We genetically engineered a system that combines the strengths of both inducible transgene methods by providing fast, Tet-dependent expression which is restricted by the cellular specificity of the Cre/lox system and allows prior identification of cells primed for transgene expression.
By directly expressing Kir2.1 which cell-autonomously hyperpolarizes cells (Paradis et al., 2001), we used longitudinal in vivo two-photon imaging to demonstrate that one can use this approach as an attractive complement to the DREADD system (Armbruster et al., 2007) or optogenetics (Emiliani et al., 2015), especially for long-term manipulation in the range of days. With this approach, we investigated possible homeostatic in vivo network compensations following acute silencing of a few neurons.
Materials and Methods
DNA constructs and plasmids
All plasmids were based on standard plasmid vectors with an rAAV backbone. A polycistronic vector of rtTA, luciferase, and mKO (Kusabira Orange) coupled via self-cleaving 2A peptide bridges and under control of the human synapsin promoter for neuronal expression (plasmid kindly provided by Rolf Sprengel) was used to clone the driver construct by exchanging luciferase with the repressor tTR. The “responder” plasmids were all based on bidirectional TetO7 sequences flanked by a minimal Cytomegalovirus Immediate-Early promoter (Gossen and Bujard, 1992). Viral DNA constructs (1-3) that were sufficiently below the total packaging limit of 5.2 kb had the mRNA stabilizing Woodchuck Hepatitis Virus Post-Transcriptional Regulatory Element inserted ahead of the Bovine Growth Hormone Poly-Adenylation Signal at 3′-end (Wang et al., 2016). For Cre-dependent expression, the Kir2.1 transgene was tagged with HA (hemagglutinin peptide tag) and flanked by DIO sites (double-floxed inverse open reading frame) harboring LoxP and Lox2272 sites (Sohal et al., 2009).
Recombinant adeno-associated viruses (AAVs)
For rAAV production, a helper-free system was used that generates capsid chimeras of AAV serotyped 1 and 2 at a ratio of 1:1. Briefly, 4 × 106 HEK293T cells/cm2 were plated in ten 15 cm Ø cell culture dishes and were grown in DMEM containing 2× MEM NEAA (PAN Biotech), and penicillin-streptomycin (100 U/ml, Invitrogen). Transfections were performed 24 h after plating using the standard calcium phosphate method. DNA of the desired AAV and the two commercial helper plasmids (Plasmidfactory) pDP1rs and pDP2rs coding for the capsid proteins of the AAV1 and AAV2, respectively, were mixed at equimolar ratios to a final concentration of 37.5 µg DNA/plate. The DNA mixture was added to 6.8 ml of dH20 and 1 ml of 2.5 m CaCl2 and vortexed briefly; 12 ml 2× HeBS buffer (280 mm NaCl, 50 mm HEPES, 1.5 Na2HPO4, adjusted to pH 7.05 with NaOH) was added dropwise while vortexing the CaCl2-DNA solution. After 100 s, 2 ml of the transfection mixture was added dropwise to each plate. Eighteen to 24 h later, the transfection efficiency should be at least 25%. The medium was exchanged with 30 ml fresh DMEM and 48 h later, the cells were scraped from dishes and pooled with the supernatants. The suspension was centrifuged at 200 × g and 4°C for 10 min. The supernatant was transferred to separate Falcon tubes and 125 U Benzonase per 50 ml supernatant were added and incubated for 2 h at 37C. The pellet was resuspended in TNT extraction buffer (20 mm Tris-HCl, pH 7.5, 150 mm NaCl, 1% Triton X-100, 10 mm MgCl2) at room temperature for 10 min to lyse the cell membranes; 1000 U of Benzonase was added, and the solution incubated at 37°C for 1 h. Cellular debris was removed by centrifugation (3600 × g at 4°C, 15 min) followed by filtering of the supernatant with a 0.45 µm Millex PVDF membrane filter (Merck Millipore). For subsequent fast protein liquid chromatography purification (ÄKTAprime plus, GE Healthcare), all solutions were degassed by vacuum-driven filtration (Stericup 500 ml Durapore 0.45 µm PVDF, Merck Millipore). Before loading, HiTrap Heparin HP (GE Healthcare) columns were first washed with 5 ml dH20 at 0.5 ml/min and then equilibrated with 5 ml PBS at 0.5 ml/min. The supernatant and the crude lysate were separately filtered using a 0.2 µm syringe Filter (Puradisc FP 30 Cellulose Acetate Syringe Filter, GE Healthcare) and directly loaded onto the column at 1 ml/min one after another. The column was then washed at a rate of 0.5 ml/min with 10 ml PBS and the virus eluted with elution buffer (500 mm NaCl in PBS) at 0.5 ml/min. The eluates containing viral capsids were collected and transferred to an Amicon Ultra-15 filter tube (Merck Millipore). The filter tube was centrifuged at 3000 × g at 4°C until <1 ml solution was left, and the samples in the filter tube were subsequently washed twice with 13 ml PBS. The final run was continued until ∼250 µl was left. This concentrated virus solution was filtered (0.22 µm Millex, Merck Millipore) and stored at 4°C until further use. A simple qualitative assessment of the virus titer was done by transducing primary rat hippocampal cultures with 0.5 or 1 µl of virus and assessing fluorescence up to 7 d later. Viruses that did not produce satisfying fluorescence in cell culture were discarded, and the preparation was repeated. For calcium imaging, a GCaMP6f rAAV (AAV1.Syn.GCaMP6f.WPRE.SV40) under the synapsin promoter was purchased from UPenn Vector Core (University of Pennsylvania), diluted 1:10, and mixed with the driver/responder viruses 1:1:1. The Synapsin-Cre AAV was coinjected in dilution of ∼1:500-1:2000 to achieve sparse transduction of neurons in vivo.
Neuronal cell culture and in vitro assays
Neuronal cell cultures
Primary rat (Sprague Dawley, Janvier) hippocampal neurons from P18.5 embryos were produced by a standard protocol (Banker and Goslin, 1998). The cells were plated in 24 well plates (Falcon Clear Flat Bottom TC-Treated Multiwell, Corning) at ∼30,000 cells per well using 12 mm diameter poly-L-lysine (Sigma-Aldrich)-coated coverslips.
Immunofluorescence staining
Cultured neurons were transduced with rAAVs, induced with 10 μm doxycycline, and were subsequently fixated at defined time points by incubation in 1 ml 4% PFA (Sigma-Aldrich) in PBS for 5 min. Immunofluorescence staining was performed by a standard protocol using an anti-HA rabbit primary antibody (C29F4, Cell Signaling Technology, diluted 1:1600) and a secondary antibody goat anti-rabbit AlexaFluor-633 (Invitrogen, 1:1000 dilution). Analysis and quantification of epifluorescence images were achieved using ImageJ Fiji (Schindelin et al., 2012).
In vitro calcium imaging
To study the effect of Kir2.1 or NaChBac on the activity of neurons in vitro, poly-L-lysine-coated cell culture dishes with a lid (µ-Dish 35 mm, #81151, Ibidi, Martinsried) were used. Rat hippocampal cultures were transduced between DIV7 and DIV10 with 1 µl each of the Driver AAV, one of the Responder AAVs, and a 1:10 dilution of the commercial GCaMP6f AAV. After DIV15, cultures were continuously induced with a final concentration of 10 μm doxycycline and were continuously imaged for 5 min each hour for 5 h. The imaging was performed on a Leica SP8 confocal microscope with a live cell imaging chamber at 37°C. 5% CO2 was superfused onto the medium with a cannula through the lid of dishes. For analysis, cell somata were manually selected by morphology independent of firing, calcium traces extracted, and the peaks quantified with a custom MATLAB script.
Animals
All experiments were conducted in accordance with the German animal welfare guidelines and were approved by the Regierungspräsidium Karlsruhe (G-54/12; G-130/17; T-34/21). Mice were kept at a 12/12 h dark/light cycle synchronized with the local day-night cycle. Water and food were available ad libitum, except for the imaging sessions. Mice were generally housed in ventilated racks with up to 3 animals per cage, but were separated after surgery to minimize the risk of injury. Transgenic mice were either a homozygous Ai14 (B6.Cg-Gt(ROSA)26Sortm14(CAG-tdTomato)Hze/J) mouse line with Cre-dependent tdTomato expression (JAX stock #007914) (Madisen et al., 2010). Or, for cell type-specific manipulation of parvalbumin (PV) interneurons, a double homozygous line derived from the Ai14 line and a PV-Cre (B6.129P2-Pvalbtm1(cre)Arbr/J) line (JAX stock #017320) (Hippenmeyer et al., 2005).
Stereotactic injection, chronic cranial window implantation, and doxycycline administration
Surgery was performed on 8- to 12-week-old mice. Injection of viruses and the implantation of a chronic cranial window were performed following craniectomy. The animals were anesthetized by an intraperitoneal injection of a mixture of 40 µl fentanyl (1 mg/ml, Janssen-Cilag), 160 µl midazolam (5 mg/ml, Hameln Pharma Plus), and 60 µl medetomidin (1 mg/ml, Sedan Alvetra-Werfft) at a dosage of 3.1 µl/g body weight. Bepanthen eye lotion (Bayer) was applied to both eyes to prevent dryness. To antagonize anesthesia at the end of the surgery, animals received a subcutaneous injection of a mixture of 120 µl naloxone (0.4 mg/ml, Inresa), 800 µl flumazenil (0.1 mg/ml, Fresenius Kabi), and 60 µl antipamezole (5 mg/ml, Pfizer) at a dosage of 6.2 µl/g body weight. The body temperature of the animals was maintained at 37°C-38°C during the whole surgical procedure by a feedback-controlled heating pad (FHC). After the surgery was finished, the animals received a subcutaneous injection of carprofen (Rimadyl, Pfizer) dosed 10 µg/g body weight as an analgesic. The analgesic treatment was continued for at least 2 consecutive days or as long as the mice showed behavioral signs of pain (Langford et al., 2010). For the surgical procedure, the scalps of the mice were shaved and subsequently mounted in a stereotactic EM70G manipulator (Kopf Instruments). Before severing the skin at ∼9 mm diameter around bregma, ∼100 µl of 1% xylocaine solution (Astra Zeneca) was applied subcutaneously, and the skin locally removed. After cleaning the skull surface, a 6 mm diameter coverslip (VWR) was used to mark a circular area with a pencil on the skull. This template was used to thin the skull at ∼6.5 mm diameter using a hand-held dental drill (Osada EXL-40). The excised circular bone was removed by carefully lifting it up with blunt forceps. Potential bleeding was stopped with moistened coagulant foam (Equimedical) that contains fibrinogen. Subsequently, the dura was removed to improve long-term imaging results. Using a custom-made stereotactic injection device, up to ∼1.2 µl of viral solution per brain hemisphere was injected at cortical depths of 400-500 µm with a glass micropipette (Blaubrand IntraMARK 5 µl) pulled to ∼10-20 µm inner diameter at the tips using a horizontal puller (Sutter Instruments). Usually, 2 or 3 sites per hemisphere in S1 cortex were injected to increase the transduced brain area. The viruses were injected at a speed of ∼3 µl/h with manually controlled pressure applied using a 10 ml syringe. After positioning the injection micropipette at the desired depth, the tissue was allowed to tighten around the micropipette by waiting for ∼2 min before starting the injection. Meanwhile, the brain surface was continuously kept moist with sterile PBS at all times to avoid drying out. When the injections were completed, the glass coverslip with a silicone filled access hole (Roome and Kuhn, 2014) (∼0.9 mm diameter, custom-made) was sterilized in 70% ethanol and chronically implanted. The access hole was placed above the ventricle (from bregma: y = –0.7 mm, x = ±1.5 mm) and the coverslip glued to the skull with dental acrylic cement (mixture of glue: Cyano [Hager & Werken, Duisburg] and powder: Paladur [Heraeus, Hanau]), while avoiding contact between cement and brain tissue. Additionally, the wounded skin around the surgery area was also sealed with dental cement. For animal mounting at the microscope setup, a round 3D-printed (Ultimaker 2) plastic crown was glued onto the skull. Following surgery, animals were treated with dexamethasone (100 µl s.c. injection 1 mg/ml, Bela-Pharm) for 2 consecutive days to minimize bleeding and inflammation below the cranial window (surgery protocol adapted from Holtmaat et al., 2009). Animals were allowed to recover from the surgical procedure, and the inflammation reaction was allowed to cease for 3-10 weeks before experiments were conducted. For intracerebroventricular injection of doxycycline, animals were anesthetized with 1% isoflurane and mounted onto the stereotactic setup. The micropipette was leveled at glass surface and advanced slowly through the silicone access hole to 2.2 mm depth aiming for the ventricle (from bregma: x = ±1.5 mm/y = –0.7 mm/z = –2.2 mm); 1.5 µl 8 mm doxycycline hyclate (Sigma-Aldrich, sterile-filtered) in PBS were injected slowly (0.15 µl/min). To achieve maximal expression levels by a single induction protocol, 2.5 mg of 9TB doxycycline (Echelon) in 500 µl PBS was intraperitoneally injected additionally to the intracerebroventricular injections at the same time.
In vivo two-photon imaging
Mice were initially anesthetized with 2% isoflurane (Baxter) and mounted onto the imaging stage. Before imaging, the cranial window was thoroughly cleaned with water and the volume above the cranial window was filled with distilled water. The site and expression of virus-mediated fluorescence were verified by epifluorescence microscopy coupled to a TriM Scope II multiphoton microscope (LaVision) using the imaging software Imspector of the same company. Imaging was performed with a 16× (NA = 0.8) or 25× (NA = 1.1) water immersion objective (both Nikon) using the appropriate filter sets (Venus or GCaMP6f: 535/70 nm and tdTomato: 650/100; Chroma). Fluorescence emission was detected with low-noise high-sensitivity photomultiplier tubes (H7422-40-LV 5 M; Hamamatsu). Two-photon imaging was performed in cortical M1/S1 regions at depths between 80 and 130 µm, and focal planes were chosen where many bright fluorescent layer 2/3 neurons were located. The duration of each in vivo imaging session was ∼9 min corresponding to 8000 images acquired at 15.3 Hz. During imaging sessions, isoflurane levels were continuously adjusted to ∼0.8%-1.5% to yield respiratory rates between 110 and 130 per minute. Respiratory rates were monitored with a piezoelectric transducer (3 cm diameter, V/AC 1.3 ± 0.5 kHz, FT-31T-1.3A1-472) (Zehendner et al., 2013) and an amplifier from HEKA using a custom-written MATLAB script. In this mildly anesthetized state, the animals exhibited substantial spontaneous cortical activity but no voluntary movement. For the respective panels in Figures 7, 9, and 11, the baseline recordings before doxycycline injection were averaged and grouped as time point 0 h.
Semiautomatic pipeline: image postprocessing and spike extraction
Mild anesthesia allowed analyzing relative changes of spontaneous activity, and the anesthesia depth was continuously adjusted to a breathing rate of ∼120 breaths/min (Ellwardt et al., 2018) to ensure comparable activities across the longitudinal experiments. Because both the spontaneous activity and the breathing rates covary with the depth of anesthesia (see Fig. 3A), the breathing rate was used as a proxy for the spontaneous activity. The breathing rate could be acutely monitored with a piezo element placed under the mouse thorax, and the levels of isoflurane were adjusted manually if necessary. This ensured constant breathing rates during and across imaging sessions (see Fig. 3B). Otherwise, we found that the variability of spontaneous activities was much more pronounced than the experimentally induced changes (data not shown). In addition, great care was taken to record the same cells in the same imaging plane despite returning the animals to their home cages in between time points.
Maximum intensity projections (MIP) of all 10-12 imaging planes (i.e., time points) from each animal were first aligned and registered so that all the cells in the planes could be superimposed (see Fig. 4A,B). Remarkably, for some experiments, a near perfect superimposition was achieved (see Fig. 4C). By concatenating the corresponding movies, the activity of individual neurons could be viewed and ROIs around the somata were manually determined for all recordable cells. To exclude “contaminating” neuropil signal from our analyses, a second, doughnut-shaped ROI surrounding each cell was digitally created and subtracted from the somata ROI (see Fig. 5A,B). Finally, shifting baselines of the concatenated GCaMP6f calcium transients were corrected with a customized MATLAB curve fitting toolbox (The MathWorks) (see Fig. 5C). The MLspike algorithm was used to extract and quantify action potential spiking from calcium transients (Deneux et al., 2016).
Motion correction and movie registration
Nonrigid correction for respiratory or mechanical motion drifts between individual imaging sessions was accomplished using the ImageJ Fiji distribution and the plugin Moco (Dubbs et al., 2016). For the analysis of consecutive time points, the respective imaging planes had to be registered onto each other. A time point before doxycycline induction was chosen as the template image to which all the other time points were registered. To achieve a high-fidelity registration of calcium imaging movies, we created a Calcium Movie Registration (CaMoReg) toolbox based on MATLAB 2016b (The MathWorks). CaMoReg was developed by using the MATLAB-App called Feature-Based Image Registration 1.0.0.1 by Brett Shoelson. CaMoReg works semiautomatically as it requires the user to choose the best registration among six different nonrigid image registration algorithm results. This App utilizes all of MATLAB intrinsic Computer Vision System Toolbox parameters on an interactive user interface and displays the immediate results.
Calcium trace extraction
For the extraction of calcium traces from regions that were typically densely packed with GCaMP6f+ neurons, the contaminant neuropil signal was removed. For this, cell somata were manually selected in ImageJ using the MIPs of all registered time points. This ROI set was loaded for every experiment into a custom-written MATLAB based algorithm that processes each single ROI separately. Each ROI was expanded by 4 pixels around the initial soma-ROI, but not to overlap with neighboring ROIs. This created a donut-shaped, second neuropil-ROI around the initial soma-ROI. The neuropil-ROI represented “contaminant” signal from dendrites and axons in the vicinity of the soma-ROI. For each frame, the average fluorescence value of the neuropil-ROI was subtracted from the average fluorescence value of the soma-ROI. After processing every ROI for every time point of the experiment, the results of the neuropil corrected calcium traces were saved for each ROI in a matrix for every imaging frame.
Fluctuating or shifting baselines of calcium signals were corrected using a MATLAB 2016b script based on the intrinsic Curve Fitting Toolbox 3.5.4 with either one of four constrained curve fitting models, from rigid to flexible. The grade of flexibility was determined by the degree of the polynomial fit that was used for every model. A user interface was added to the script to display the result of every curve fitting model and allow the user to confirm or choose the best fit manually. The whole process was reiteratively performed on all single-cell calcium traces to achieve to best possible baselines for each calcium trace.
Spike estimation
With the established image processing pipeline, calcium traces could be extracted from single cells and analyzed over multiple time points. To analyze the activity of neuronal cell populations and inferring their spiking patterns, the MLspike algorithm (Deneux et al., 2016) was used. The algorithm was first calibrated with a published in vivo dataset (Chen et al., 2013) that was very similar to our recordings. In total, 16 different parameter combinations were tested to approximate spiking from calcium traces. The combination of the MLspike intrinsic parameters Hill 2.80, C(0) 0.65, and Drift 3.00 produced only a minor spike average underestimation of −1.85% compared with patch-clamp determined spiking numbers.
Distance and activity correlations
The lengths of all distances between neurons that only expressed GCaMP6f and those neurons that were genetically silenced (Cre+, tdTomato+) were summed up and divided by the number of Cre+ cells. This value indicated the relative average distance each GCaMP6+ cell had to the silenced cells that surrounded it. Calculations were performed in MATLAB; cell positions were extracted from ImageJ ROIs using the MATLAB extension ReadImageJROI (Muir and Kampa, 2014). The spontaneous activity before and after induction of Kir2.1 was averaged, and a ratio calculated. To avoid divisions by zero, the values were increased by 1, and all ratios >1 were plotted as follows:
The average distance of each cell to all silenced cells was correlated to the activity change after induction.
Statistical analyses
Statistical analyses, including tests for normality, were done using Prism software. Numbers are given as mean ± SD or ± SEM as indicated.
Results
To be able to study homeostatic mechanisms at cellular resolution in vivo, we needed to develop an inducible approach for fast and cell type-specific expression in the living brain. For this, we combined the Cre/lox and the Tet system so that transgene expression required both doxycycline induction and Cre recombinase. Both systems have been combined differently before (Rodda and McMahon, 2006); however, it was essential for the experimental success of our project that the same transgene concomitantly depended on the Cre/lox as well as the Tet system.
The general principle is depicted in Figure 1A. Quintuple (5×) genetic engineering of brain tissue with transgenic mice and/or AAV provided rapid manipulation via the floxed, two-component Tet system (2×), cell type specificity mediated by Cre recombinase (1×), prior labeling of cells primed for Tet-dependent expression (1×, here: transgenic floxed tdTomato), and optical analyses of the induced genetic interference (1×, here: viral GCaMP6f). Cre recombinase could either be provided through transgenic mice (e.g., PV-Cre mice) or through injection of a virus (e.g., Synapsin-Cre). The exciting strength of the Cre virus approach is that the concentration of virus can be titrated according to the experimental needs. If only very few cells or even a single cell are intended to be manipulated, the Cre virus could simply be diluted considerably to transduce the approximate cell number. Conversely, for interference with many cells, a higher titer may be used. Thus, for all different Cre virus dilutions, the same high titer of the two Tet-On viruses can be used to achieve efficient transgene expression, which in turn also promotes high cotransduction of cells. Before doxycycline administration, extensive baseline recordings (activity, morphology, etc.) of the same tdTomato+ cells and their surrounding cells were performed for a few days by means of in vivo optical imaging. Following Tet-dependent transgene induction, the same cells and cellular parameters were repeatedly imaged in a longitudinal experimental design ranging from hours to days and potentially weeks. The unique approach of extensive imaging before, during, and after transgene induction of specifically tailored fluorescent readouts, we term “optical interrogation.”
Establishment of a Cre-dependent Tet system. A, Scheme and timeline for Cre- and Tet-dependent manipulation of cells in vitro and in vivo. Coexpression of a Cre-dependent fluorescent marker (tdTomato) marks those cells that are primed for fast, Tet-dependent expression of the desired transgene. B, The optimized driver AAV containing rtTA and tTR was used with one of three different responder AAVs. C, In hippocampal cell culture, neurons were transduced with AAVs [1] and [4]. For cells labeled with “+Cre,” viral Synapsin-Cre [5] was coadministered. Expression of blue fluorescent mtagBFP required doxycycline for induction while immunodetection of the HA-tag required doxycycline and Cre recombinase. Scale bar, 20 µm. D, Overexpressed Kir2.1 was detected in the soma and dendrites, including spines (Howe et al., 2008). E, A 4 h pulse of 10 μm doxycycline produced long-lasting Kir2.1 protein presence in vitro. Neuronal cultures were transduced with the standard driver AAV [1] and an AAV containing a Tet-dependent HA-tagged Kir2.1 construct [4]. Cultures were fixed, immunostained for the HA-tag at the indicated times, and the immunofluorescence quantified using ImageJ Fiji. At least 60 cells from two different cultures were quantified per time point (Wilcoxon signed-rank test, p < 0.0001). Error bars indicate SD.
For the two-component Tet-On system, we produced a “driver” virus with the reverse tetracycline-dependent transcriptional activator (rtTA) [1] (Fig. 1B) and “responder” viruses containing the TetO7 promoter with the respective transgene [2-4]. We flanked the Kir2.1 transgene of one responder virus [4] with Cre recombinase recognition sequences while the mtagBFP transgene on the other side of the bidirectional TetO7 promoter was not modified. In dissociated hippocampal cultures, the dependency of the Kir2.1 transgene on doxycycline and Cre recombinase could be demonstrated with the respective AAVs [1,4,5]. Doxycycline administration alone induced blue mtagBFP fluorescence, while the presence of both, Cre recombinase and doxycycline, was necessary for Kir2.1 expression (Fig. 1C).
For genetic manipulation of neuronal activity, we chose the inward rectifying potassium channel Kir2.1 (Burrone et al., 2002; Auffenberg et al., 2016) (neuronal hyperpolarization, eventually silencing the neuron) and the bacterially derived sodium channel NaChBac (Lin et al., 2010) (neuronal depolarization, eventually activating the neuron). Overexpression of Kir2.1 in rat neuronal cultures correctly targeted the exogenous Kir2.1 channel to the soma and dendritic compartments (Fig. 1D). In these cultures, HA-tagged Kir2.1 could be detected by immunofluorescence up to at least 96 h after a 4 h pulse of doxycycline (Fig. 1E). This suggested that a single induction leads to the expression of Kir2.1 protein for several days and that silencing via the Tet system was long-lasting.
To mimic the in vivo situation, analysis of Tet-dependent manipulation of neuronal activity in vitro was performed by recording calcium transients via imaging of GCaMP6f fluorescence. Doxycycline application and induction of transgenes led to a rapid and significant decrease (Kir2.1) or increase (NaChBac) of neuronal activity within hours (Fig. 2A). Our early attempts to establish an efficient method for inducible transgene expression based on the Tet-On system were often flawed by considerable background expression in the absence of doxycycline (data not shown). This “leak” expression from the TetO7 promoter is likely because of the presence of a minimal CMV promoter in the construct. This element cannot be removed because it is necessary for transcription. Thus, we decided to increase tightness via the driver virus by introducing an additional tetracycline-dependent repressor (tTR) (Deuschle et al., 1995). In the absence of doxycycline, the repressor binds to the TetO7 promoter, thereby blocking unspecific “leak” expression, but dissociates from DNA on binding to tetracyclines. In cell lines, this approach has been shown to reduce leak expression at low doxycycline concentrations (Freundlieb et al., 1999). In dissociated hippocampal neurons, we found that the addition of tTR decreased the effective concentration (EC50) threefold (3.3 μm with tTR, 9.3 μm without tTR) (Fig. 2B). While we expected a higher EC50 value with tTR and not without, this finding did not impact the effectiveness of our method. Moreover, compared with the more gradual increase detected without tTR, its presence produced a step-like dose–response curve, which should also improve tightness of the system.
Genetic manipulation of neurons in vitro and in vivo. A, Tet-dependent induction of Kir2.1 [AAVs 1, 2, 6] or NaChBac [AAVs 1, 3, 6] significantly decreased (p = 0.0011) or increased (p = 0.0035) calcium signals recorded with GCaMP6f, respectively (≥4 hippocampal cultures and >400 neurons per condition; statistical test: linear regression fit). Error bars indicate SD. B, Quantification of Venus fluorescence and curve-fitting computed two distinct EC50 values (on average, 33 neurons quantified per condition) for transduced hippocampal neurons with tTR (blue) or without tTR (red). C, In vivo leak expression in the absence of doxycycline was reduced by the presence of the repressor tTR. Different virus combinations were injected into either hemisphere of a mouse brain. Left, Green Venus leak expression was apparent in the left hemisphere without tTR. Right, The presence of tTR substantially reduced Venus fluorescence. Stippled lines roughly outline the injection areas. There is considerable green background fluorescence of brain tissue and skull. Middle, Blue channel only display of the left panel image highlights minor blue Venus fluorescence on the left, but not the right hemisphere. Right, Leak expression was because of the Tet promoter and not residual rtTA activity because the responder virus alone produced Venus fluorescence on the left. D, Top, Direct injection of doxycycline through a small hole in the chronic window. The hole could be resealed with silicone and thus allowed immediate and continued access to the brain. Imaging was possible ipsilateral or contralateral to the injection site. Bottom, Significantly increased ipsilateral (green) Venus in vivo expression compared with intraperitoneal injection (blue) [AAVs 1, 2] (Kolmogorov–Smirnov test, p = 0.026). Error bars indicate SD. Expression contralateral (red) to the doxycycline injection produced transgene levels similar to intraperitoneal injection (ipsilateral: n = 5 mice; contralateral: n = 3 mice; intraperitoneal: n = 6 mice). E, Comparison of different administration routes of doxycycline in vivo. After injection of the standard driver and the Tet-Venus responder viruses [AAVs 1, 2], doxycycline was administered intraperitoneally or by direct intracerebroventricular injection, either ipsilateral or contralateral to the imaging site. Shown are images of 1 mouse of each doxycycline administration route visualized by two-photon in vivo imaging. In each animal, the same cortical region was imaged over time, from before doxycycline administration up to 72 h thereafter. Scale bar, 50 µm. For the displayed examples, Venus fluorescence increased from baseline to 72 h for intracerebroventricular (ipsilateral) injections 11.1-fold, 2.5-fold for intracerebroventricular (contralateral) injections, and 2.8-fold following intraperitoneal injections of doxycycline.
Monitoring and manual adjustment of breathing rates. A, Sample recordings of calcium transients of a single cell (top) and the corresponding breathing rate (bottom) of a mildly anesthetized mouse. Increased spontaneous calcium transients were detected at higher breathing rates/lower anesthesia. B, Continuous manual adjustment of the isoflurane levels produced stable breathing rates across longitudinal experiments.
Registration of calcium movies from different time points of longitudinal imaging experiments. A, Scheme of the registration process. The individual imaging planes of the raw movies were all registered to the initial time point (left), and the result was visualized by an MIP across all time points (right). BI-BVI, A custom MATLAB toolbox provided six different registration results of affine transformation matrices from which the user had to choose the preferred registration manually. BI, BV, Good registration between the two pseudo-colored (green and purple) MIPs from two different time points. C, An MIP of registered imaging planes from 11 different time points produced a near complete overlap.
Workflow for calcium movie processing. A, Scheme for removal of neuropil signal that typically “contaminates” extraction of calcium transients. B, After subtraction of the neuropil signal from the ROI signal, the corrected trace is an adequate representation of the somatic calcium transients. C, A concatenation of calcium traces (extracted from movies) of the same cell at different time points. The uncorrected shifting baselines would impair spike estimation. Correction was achieved with a customized MATLAB 2016b intrinsic Curve Fitting Toolbox 3.5.4.
In the living brain, the “leak” expression could also be successfully blocked by the presence of tTR. Within the same brain, a side-by-side comparison between a virus pair with tTR (AAVs 1,2) and a pair without tTR (AAV driver with luciferase instead of tTR + AAV 2) evidently showed that Venus expression was not apparent on the right hemisphere with the standard driver containing tTR (Fig. 2C, left). To clearly differentiate green Venus signal from the pervasive green background fluorescence of the tissue, the blue channel of the same RGB image is shown with the minor blue Venus fluorescence readily visible only in the left hemisphere (Fig. 2C, middle). To address the possibility that the leak expression of the TetO7 promoter stems from residual rtTA activity in the absence of doxycycline, we injected one hemisphere only with the Venus responder virus and no driver. Similar to the previous experiment with the tTR-less driver, abundant Venus expression was observed (Fig. 2C). Again, the control hemisphere with AAVs [1,2] exhibited negligible Venus fluorescence. We concluded that tTR was a critical component for the Tet system in vivo, essentially limiting leak expression from the TetO7 promoter.
To achieve rapid and robust onset of doxycycline-induced transgene expression in vivo, we first compared two different routes of doxycycline administration (i.e., intraperitoneal injection vs direct injection into the brain). For direct injections, we implanted chronic window glass coverslips with an acentric, resealable hole (Roome and Kuhn, 2014) of ∼0.5 mm (Fig. 2D, top). This allowed repeated access to the brain so that doxycycline could be directly injected at any time after the surgery and window implantation. We also assessed transgene expression in hemispheres ipsilateral versus contralateral to the injection site. In longitudinal two-photon microscopy experiments, for each mouse, the levels of Venus fluorescence were repeatedly monitored in the same imaging plane. For all in vivo experiments, imaging typically occurred at a depth of ∼150-250 µm below the brain surface in layer 2/3 of M1 or S1 depending on the injection site. Quantification of Tet-dependent Venus fluorescence in vivo revealed that different routes of doxycycline administration produced different levels of transgene induction (Fig. 2D, bottom). We found that transgene expression ipsilateral to the injection (green bars) was substantially higher than in the contralateral hemisphere (red bars) or after intraperitoneal injection (blue bars) (Fig. 2D,E). The Venus fluorescence before doxycycline treatment was negligible. More importantly, because our approach allows analysis of the same cells before and after transgene induction, those very few cells with “leak” expression could potentially be excluded in the post hoc image processing, if necessary. Thus, for all subsequent GCaMP6f-based (Chen et al., 2013) calcium imaging experiments, doxycycline-induced silencing of neurons was assessed ipsilateral to the injection. Overall, we consider the levels of transgene expression that can be achieved with our approach in vivo after a single induction of doxycycline more than adequate for typical experimental needs.
To genetically manipulate and quantify neuronal activity in vivo in longitudinal imaging experiments, we revisited the same cortical areas and imaging depths within each mouse to record from the same neurons over time, using mildly anesthetized animals. Recording 9 min videos of neuronal GCaMP6f fluorescence in the same cortical image plane at each time point allowed to repeatedly monitor spontaneous neuronal activity on a cell-by-cell basis. It is important to note that it was almost impossible to achieve comparable levels of baseline network activities from one time point to the next with fixed levels of anesthesia. Since animals were returned to their home cages in between time points, for each imaging session we monitored and adjusted the breathing rate as a proxy for the depth of anesthesia and neuronal activity (Fig. 3A) (Zehendner et al., 2013). The assumption was that similar breathing rates would reflect similar depths of anesthesia, which in turn should cause comparable baseline network activities during the different time points. This adjustment was necessary because in pilot experiments we initially used the same fixed level of anesthesia at each time point, which in turn produced highly variable levels of baseline activity across the longitudinal experiments. We assume this was a consequence of different levels of arousal before anesthesia and different experimental time points during the circadian rhythm. We could achieve stable, reproducible breathing rates at all time points (Fig. 3B) by acutely adjusting the levels of anesthesia, which in turn substantially reduced the variability of baseline activity and thus allowed genetic manipulation and detection of neuronal activity.
We established a semiautomatic pipeline using MLSpike (Deneux et al., 2016) to be able to extract and quantify these neuronal activities in longitudinal experiments. The details of this pipeline are described in Materials and Methods. Most importantly, the images had to be processed through multiple steps to produce a value of activity for each neuron at each time point. Images were first semiautomatically registered and aligned (Fig. 4A) by manually choosing the best match from different outputs of a custom MATLAB toolbox (Fig. 4B). This allowed superimposing images to precisely define ROIs across all time points for further quantification. An MIP of images of the same cortical region from 11 different time points highlighted the precision with which the images could be aligned (Fig. 4C). Somatic ROIs were manually delineated followed by subtraction of “contaminating” GCaMP6f signal coming from the surrounding neuropil (Fig. 5A,B). Finally, using a customized MATLAB Curve Fitting algorithm, shifting baselines were “flattened” and concatenated for automated spike estimation (Fig. 5C). Example raw traces of a few cells from experiments described below are shown at baseline and 25 h (Fig. 6A,B) where each trace corresponds to the activity of one neuronal soma during a 9 min recording (left panels: Kir2.1 induction; right panels: controls). The MLSpike spike extraction output from such a raw trace of calcium transients is depicted in Figure 6C. The computed number of spikes for each transient is indicated by the red numbers. Together, we found that our semiautomatic pipeline yielded reproducible and accurate estimations of neuronal activity from two-photon recordings. In combination with the adjusted anesthesia levels, our approach was thus suitable to detect and characterize genetically induced changes in activity of individual neurons in vivo.
Raw in vivo calcium traces before (baseline) and after doxycycline induction (25 h time point). A, Kir2.1 silencing in a sparse population of cortical neurons [correspond to data in Fig. 7]. Left, Kir2.1 silencing. GCaMP6f fluorescence traces of six cells recorded during a 9 min video at baseline and of the same six cells at 25 h after doxycycline-mediated induction of Kir2.1 expression. At 25 h, the overall activity was substantially reduced compared with baseline. Right, Six control neurons without Kir2.1. At 25 h, no obvious changes in activity compared with baseline. B, Kir2.1 silencing in PV neurons [correspond to data in Fig. 9]. Left, Kir2.1 silencing. GCaMP6f fluorescence traces of seven cells surrounding PV neurons recorded during a 9 min video at baseline and of the same seven cells at 25 h. At 25 h, the overall activity was substantially increased compared with baseline. Right, Eight control neurons surrounding PV neurons that were not silenced. At 25 h, no obvious changes in activity compared with baseline. C, Representative spike extraction output from a raw trace using MLSpike. Numbers indicate extracted spikes for corresponding GCaMP6f transients.
Rapid and long-lasting in vivo genetic silencing by induction of Kir2.1 in a sparse population of cortical neurons. A, Genetic engineering of brain tissue to allow Cre- and Tet-dependent transgene expression in vivo. Single in vivo two-photon imaging plane shows that yellow Cre+ neurons (red tdTomato + green GCaMP6f) can be clearly distinguished from green Cre– neurons (GCaMP6f only). Scale bar, 100 µm. B, In vivo two-photon imaging of GCaMP6f: spontaneous cortical activity in Cre+ neurons (left) significantly decreased after Tet-dependent induction of Kir2.1 compared with baseline (0 h), while control Cre– neurons on the same hemisphere without Kir2.1 did not change (right). Displayed is the mean spiking of all cells per mouse (6 mice, ∼30 neurons for each mouse; linear regression analysis: Cre+ [0-9 h] p = 0.0066, Cre– [0-9 h] not significant, [0-80 h] not significant). Error bars indicate SEM. C, Representative example of longitudinal imaging of tdTomato fluorescence to assess morphologic changes of manipulated neurons over time. Scale bar, 5 µm. D, Blinded, manual quantification of spines on dendritic stretches of neurons that were silenced with Kir2.1 revealed no changes in spine numbers after prolonged silencing of more than 3 d (4 mice, 24 dendritic stretches, statistical test: nonparametric Wilcoxon matched-pairs signed rank test, p = 0.4774). E, Each before-after pair represents one dendritic stretch before doxycycline injection and 80 h after Kir2.1 induction. Dendritic stretches from each mouse were grouped together. No statistical difference was observed for either mouse (statistical test: nonparametric Wilcoxon matched-pairs signed rank test; from left to right: p = 0.7813, p = 0.6250, p = 0.250, p > 0.999).
For the in vivo experiments, the population of neurons in which Kir2.1 eventually would be induced by doxycycline was defined by the cotransduction with a dilute Synapsin-Cre virus [5]. A quadruple cocktail of viruses (driver [1], floxed Kir2.1 responder [4], dilute Synapsin-Cre [5], Synapsin-GCaMP6f [6]) was injected into M1/S1 cortical regions of transgenic Ai14 mice containing a floxed tdTomato reporter (Madisen et al., 2010) (Fig. 7A). The contralateral hemisphere only received the GCaMP6f virus but not the two Tet-On viruses and served as control. Sparse, neuronal Cre expression (Cre+) could be visualized by tdTomato fluorescence, which in turn identified those cells that were also primed for silencing via Kir2.1 and the Tet system. Because the Cre and GCaMP6f virus were both driven by the Synapsin promoter, the tdTomato signal typically colocalized with the green GCaMP6f fluorescence (Fig. 7A). Because of the higher titer of the GCaMP6f virus, many additional neurons only exhibited GCaMP6f fluorescence but were negative for Cre recombinase (Cre–); the activity of these surrounding control neurons was assessed in parallel. Thus, each time point consisted of at least several dozen cells in the imaging plane whose spontaneous GCaMP6f activity could be recorded during a 9 min movie. A single dose of doxycycline induced rapid Kir2.1-dependent silencing in tdTomato-positive (Cre+) neurons within hours, while surrounding tdTomato-negative neurons remained unaffected (Fig. 7B). Thus, these surrounding neurons did not adjust their input strength in vivo, unlike during the aforementioned in vitro experiments (Burrone et al., 2002). Similar to the in vitro detection of Kir2.1 for up to 96 h, reduced spontaneous activity persisted in vivo for at least 80 h.
Burrone et al. (2002) also demonstrated in neuronal cultures that Kir2.1-mediated silencing after synapse formation did not lead to homeostatic changes in spine number. We used dendritic tdTomato fluorescence for longitudinal analysis of neuronal morphology in silenced neurons (Fig. 7C). Notably, the same Cre+ neurons and their dendritic spines could be repeatedly reidentified in consecutive imaging sessions which permitted quantification of spine density changes. No significant spine changes could be detected 80 h after prolonged reduction of neuronal activity (Fig. 7D). This overall trend of spine stability was also seen when analyzing individual dendritic stretches from 4 different mice as there were only small changes observed (Fig. 7E). Our data thus extend recent findings that showed synapse formation during development in the absence of neuronal activity (Sando et al., 2017) to synapse maintenance despite acute, prolonged silencing in adult mice.
Overall, we find that most analyzed Cre+ neurons had decreased spontaneous activity while a minor fraction exhibited an increase (Fig. 8A,B, left). This is in stark contrast to ipsilateral control Cre– neurons for which about the same number of neurons showed an increase or decrease of spontaneous activity, respectively (Fig. 8A,B, right). The relative surplus of cells that exhibited a Tet-dependent activity decrease on the treatment hemisphere compared with the spontaneous activity changes seen on the control hemisphere indicated that ∼40% of the Cre+ neurons must have been transduced by all four viruses. There are two important additional points to consider here. First, a temporally correlated, doxycycline-induced decrease in activity demonstrated that all four viruses must have transduced a neuron; otherwise, we would not have observed this change in activity. Second, the detected decrease of neuronal activity in Cre+ cells highly correlated in time with the doxycycline injection, while the activity of Cre– cells did not change significantly. This strongly suggested that Cre+ cells did not simply deteriorate because of the doxycycline administration since neighboring Cre– cells remained unaffected by the injection. Also, all viruses were transduced at least 3 weeks before imaging; and after this time, almost all GCaMP6f+ cells appeared healthy, which in turn were those chosen for experimental analyses.
Kir2.1-dependent changes of activity in individual neurons in vivo. A, Left, The absolute number of spikes was quantified in the same Cre+ neurons before (predox) and after (postdox) induction of Kir2.1. The number of spikes decreased for most neurons. Right, Control. The number of neurons with increasing or decreasing number of spikes was about the same in Cre– neurons. B, Same neurons as in A, but showing the relative change for each individual cell. Left, Red line indicates the overall average decrease. Error bars indicate SD. Right, The red line being close to zero suggests stable spontaneous activities across the imaged neuronal population.
Cell type-specific silencing of inhibitory PV neurons increased neuronal activity of surrounding cells. A, Genetic engineering of brain tissue to allow Tet-dependent silencing of PV neurons. Single in vivo imaging plane shows that yellow PV neurons (red tdTomato + GCaMP6f) can be clearly distinguished from the surrounding green Cre– neurons (GCaMP6f only). B, Silencing of PV neurons significantly increased neuronal activity of surrounding cells (left). Neurons in control hemispheres of the same mice but without Kir2.1 did not show any significant change in activity (right). Displayed is the mean spiking of all cells (6 of 7 mice [because of technical reasons, the experimental hemisphere of one mouse could not be recorded], ∼50-100 neurons for each mouse; linear regression analysis: Cre+ [0-80 h] p = 0.0073, Cre– [0-80 h] not significant) Error bars indicate SEM.
The use of a diluted Synapsin-Cre virus afforded great flexibility in the number of transduced Cre+ cells with our method. Alternatively, the abundance of transgenic Cre mice offers great flexibility for cell type-specific manipulation with our method. Taking advantage of such mice, we wanted to demonstrate cell type specificity by silencing inhibitory PV neurons in PV-Cre mice in vivo. The prediction was that, in this set of experiments, the surrounding non-PV cells change their activity because of the importance of PV cells in regulating neuronal activity and homeostasis. A triple virus cocktail ([1],[4],[6]) was injected into double-transgenic PV-Cre, Ai14 mice. Red tdTomato fluorescence clearly differentiated PV neurons from non-PV cells (Fig. 9A). Silencing inhibitory PV-cells led to rapid and significant increases in spontaneous activity of surrounding non-PV neurons (Fig. 9B). The control hemisphere was only injected with GCaMP6f and the corresponding changes of the calcium transients over time were not significant. Overall, the effect of silencing PV neurons persisted in non-PV cells for at least 80 h as their activity remained high. Because the observed changes in surrounding neurons temporally correlated with the Kir2.1 induction, were of increasing nature as expected from silencing inhibitory neurons, and were based on a well-characterized PV-Cre mouse line, we conclude that PV neurons were specifically and acutely silenced in vivo. We next wondered whether there was also a spatial correlation between the induced manipulation of Cre+ neurons and the observed changes in the surrounding Cre– cells. We analyzed the changes in spontaneous activity in non-PV neurons as a function of average distance to silenced PV neurons. Within the imaging plane (Fig. 10A), tdTomato+ and GCaMP6f+ cells were digitally identified, catalogued, and the computed activity change of GCaMP6f+ cells assigned. The degree of activity change was highlighted as a heat map of the imaging plane with PV cells shown in black (Fig. 10B). We found that non-PV cells, which had a smaller average distance to all detected PV neurons, were significantly more affected (Fig. 10C). The observed maximum “functional reach” of PV cells of ∼500 µm (intersection with x axis) is in good agreement with previously published morphologic characteristics of PV neurons, suggesting a “morphologic reach” of 760 ± 130 µm (Halasy et al., 1996). For comparison, no distance dependency was seen between PV cells and the surrounding neurons in control hemispheres that were only injected with GCaMP6f, but without driver and responder (Fig. 10D). We also repeated this type of analysis for the Synapsin-Cre injected Ai14 mice. While the surrounding Cre– neurons as a population did not display any change in activity (Fig. 7B), possibly small local changes of nearby Cre– neurons could have been missed. Within the imaging plane (Fig. 10E), a heat map of activity changes for Cre+ (black) and Cre– neurons was generated (Fig. 10F). However, no distance-dependent changes could be observed for either neuron population (Fig. 10G,H). Together, the data suggest that induced changes in activity in one neuronal population can have secondary effects on surrounding neurons and that these secondary activity changes may be a function of distance and connectivity.
Silencing a neuronal subpopulation in vivo and its distance-dependent effect on the activity of surrounding neurons. A, In vivo image of a cortical region showing red/yellow PV cells and green fluorescent surrounding neurons (GCamP6f). B, Heat map representation of the same region as in A. The changes in average spiking relative to the initial activity over all recordings were color-coded. Black represents the positions of PV cells, not in red. Scale bar, 100 µm. C, Each dot represents one neuron displaying its average distance to all detected nearby PV neurons and its change in neuronal firing after doxycycline. Surrounding neurons with a shorter average distance to PV cells displayed significantly higher increases in activity compared with those further away (linear regression analysis: p < 0.001). D, Control neurons on the other hemisphere with PV cells which did not express Kir2.1 showed no distance-dependent changes in neuronal activity (linear regression analysis: not significant). E, In vivo image of a cortical region showing red/yellow Cre+ cells and green fluorescent surrounding neurons (GCamP6f). Injection of a dilute Synapsin-Cre AAV into Ai14 mice defined the Cre+/tdTomato+ neuronal subpopulation which was silenced with Kir2.1. F, Heat map representation of the same region as in E. The changes in average spiking relative to the initial activity over all recordings were color-coded. The positions of Cre+ cells are displayed in black. Scale bar, 100 µm. G, Each dot represents one neuron displaying its average distance to all detected nearby Cre+ cells and its change in neuronal firing. The average distance of manipulated neurons to GCamP6f-only neurons had no significant effect on the observed changes in activity. H, Control hemispheres in the same animal without Kir2.1 silencing. No distance-dependent change in activity was detected.
Cell type-specific silencing of inhibitory PV neurons. A, In 1 animal, induction of Kir2.1 in PV neurons induced an ∼10-fold increase in spontaneous activity in the surrounding non-PV cells over the course of the experiment. Error bars indicate SEM. Following a rapid increase in activity of the surrounding neurons, their activity tended to decrease temporarily (hours 30-55) and then tended to rebound again to levels before the decrease (hour 25 vs hour 80). B, Average of the 3 animals in which the intermittent decrease trend was observed. Error bars indicate SEM.
Discussion
Our method provides high-resolution, inducible interference and interval imaging of individual cells (high I5), which is why we designate it the HighFive approach. The HighFive approach proved to be versatile, robust, and is readily implemented in longitudinal in vivo imaging paradigms. Prior marking of cells with a Cre-dependent, fluorescent reporter accurately predicts when and where Tet-dependent transgene expression occurs, thereby allowing precisely tailored in vivo optical interrogation and manipulation of defined cell types. This is currently not possible with any other conditional expression method and opens new possibilities for cellular and network-level neuroscience research. With the HighFive approach, transgene expression can exclusively be correlated in time and space to potentially subtle, cellular phenotypic changes, within the same cell, but also within the context of the cellular network in vivo. Thus, our method offers exciting new possibilities to study cell biology in vivo at the level of single cells. We showed that the HighFive method worked well in vitro in rat neuronal cultures and in mice in vivo. This demonstrated the versatility of the approach as it could be applied in different species.
By combining the Tet with the Cre/lox system, our method offers both: rapid transgene expression conferred by the Tet system and exquisite cellular specificity defined by the Cre/lox system. We are not aware of any publication that demonstrated similarly fast expression times for the tamoxifen-inducible Cre/lox system. The HighFive method takes advantage of the cellular specificity provided by the many available Cre mouse lines, particularly since both standard Cre lines and inducible CreERT2 lines can be used as long as Cre recombinase is active before doxycycline induction. We also improved the Tet-On system for in vivo work by introducing the tTR repressor for less background expression and by direct brain injections for higher inducibility compared with intraperitoneal injections.
The second major methodological advancement is the development of a new approach for conditional manipulation of neuronal activity. We used the cellular precision of the HighFive method to successfully interfere with neuronal activity in defined neuronal populations. The high temporal resolution of genetic transgene expression allowed acute manipulation of neuronal networks with the experimental intention of being able to study network homeostasis in vivo. Here, we used standard two-photon imaging of GCaMP6f fluorescence to visualize and extract neuronal activity from calcium transients. We note that calcium transients are not fully understood in respect to the underlying neuronal activity; therefore, the activity values we obtained represent only an approximation (Siegle et al., 2021).
We silenced a subpopulation of neurons as defined by the Synapsin-Cre expression but did not observe any obvious homeostatic compensatory mechanisms. The silenced neurons did not exhibit significant morphologic changes, while the activity of the surrounding neurons also appeared unaffected. After the bulk of synapse formation had occurred, Kir2.1 expression in vitro also did not alter spine density. However, Kir2.1 expression in cultured neurons first reduced their spiking frequencies which eventually returned to control levels after 3 d (Burrone et al., 2002). To accommodate this finding, the authors proposed a homeostatic increase in synaptic input strength that we could not detect with the analysis method available to us (i.e., somatic GCaMP6f transients). Not surprisingly, different homeostatic mechanisms may thus operate in vitro versus in vivo. Conversely, we find that silencing PV cells did increase the activity of surrounding neurons in vivo. Most likely this increase is simply a reflection of less inhibition by the PV neurons, but theoretically, a possible minor homeostatic effect cannot be excluded. Indeed, we would like to speculate that the temporary reduction seen after the first rapid rise of activity (individual mouse, hours 30-55, Fig. 11A) could be a consequence of the network attempting to compensate the activity increase only to eventually succumb to the strong impact of the cumulative Kir2.1 expression. We did observe a similar decrease in activity in 2 other mice, albeit not as pronounced (average of 3 mice, Fig. 11B). Future work will have to investigate whether this observation is reproducible and will have to aim at identifying possible underlying mechanisms.
Regarding the effectiveness of the HighFive method, we noted the difference in baseline activity at t = 0 h between the two imaged Cre+ and Cre– cell populations (Fig. 7B). Cre+ neurons displayed less baseline activity which could possibly be related to “leak” expression of Kir2.1 before doxycycline induction. However, in experiments silencing PV neurons (Fig. 9B), the surrounding neurons to PV neurons with Kir2.1 also exhibited a lower baseline activity than neurons surrounding control PV neurons without Kir2.1. It is our interpretation that, if there truly was a “leak” of Kir2.1 before doxycycline induction, the surrounding neurons to PV neurons with Kir2.1 should have a higher, not lower, baseline activity than neurons surrounding PV neurons without Kir2.1. We therefore believe that Kir2.1 “leakage” was not the reason for the different baseline activities, but we currently do not have a good alternative explanation for the observed trends.
In general, being able to silence or hyperactivate neuronal firing is central to many research questions. The genetic manipulation of neuronal activity via Kir2.1 or NaChBac represents a desirable complement to the DREADD system (Armbruster et al., 2007) whose active ligand was recently found to be the psychoactive compound clozapine (Gomez et al., 2017). Nevertheless, we do not see our method as competition to optogenetics or DREADDs but rather as an important, complementary approach which is perhaps better suited for behavioral experiments. More importantly, unlike the other two approaches, the HighFive method foremost allows transgene manipulation in vivo at cellular resolution.
Footnotes
This work was supported in part by Friedrich-Ebert-Stiftung doctoral fellowship to F.T.; Frontiers grant to S.B.C.; and the SFB 1134 to S.B.C. We thank Gabriele Krämer, Michaela Kaiser, and Claudia Kocksch for technical assistance; Albrecht Stroh and Amit Agarwal for valuable input to the manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to Sidney B. Cambridge at cambridge{at}med.uni-frankfurt.de
