Abstract
Genetically encoded biosensors can be used to track signaling events in living cells by measuring changes in fluorescence emitted by one or more fluorescent proteins. Here, we describe the use of genetically encoded biosensors based on Förster resonance energy transfer (FRET), combined with high-content microscopy, to image dynamic signaling events simultaneously in thousands of neurons in response to drug treatments. We first applied this approach to examine intercellular variation in signaling responses among cultured striatal neurons stimulated with multiple drugs. Using high-content FRET imaging and immunofluorescence, we identified neuronal subpopulations with unique responses to pharmacological manipulation and used nuclear morphology to identify medium spiny neurons within these heterogeneous striatal cultures. Focusing on protein kinase A (PKA) and extracellular signal-regulated kinase 1/2 (ERK1/2) signaling in the cytoplasm and nucleus, we noted pronounced intercellular differences among putative medium spiny neurons, in both the magnitude and kinetics of signaling responses to drug application. Importantly, a conventional “bulk” analysis that pooled all cells in culture yielded a different rank order of drug potency than that revealed by single-cell analysis. Using a single-cell analytical approach, we dissected the relative contributions of PKA and ERK1/2 signaling in striatal neurons and unexpectedly identified a novel role for ERK1/2 in promoting nuclear activation of PKA in striatal neurons. This finding adds a new dimension of signaling crosstalk between PKA and ERK1/2 with relevance to dopamine D1 receptor signaling in striatal neurons. In conclusion, high-content single-cell imaging can complement and extend traditional population-level analyses and provides a novel vantage point from which to study cellular signaling.
SIGNIFICANCE STATEMENT High-content imaging revealed substantial intercellular variation in the magnitude and pattern of intracellular signaling events driven by receptor stimulation. Since individual neurons within the same population can respond differently to a given agonist, interpreting measures of intracellular signaling derived from the averaged response of entire neuronal populations may not always reflect what happened at the single-cell level. This study uses this approach to identify a new form of cross-talk between PKA and ERK1/2 signaling in the nucleus of striatal neurons.
Introduction
Neurons in the dorsal striatum regulate goal-directed locomotion, behavioral action selection, and specific forms of learning. Medium spiny GABAergic projection neurons (MSNs) account for 95% of the neurons found in the rodent striatum (Ivkovic and Ehrlich, 1999; Schambra et al., 1994). These neurons integrate dopaminergic and glutamatergic inputs to the striatum and send inhibitory projections to several downstream nuclei. Based on both anatomic and molecular features, MSNs are broadly divided into two subpopulations: “direct-pathway” MSNs, which express the Gαs/olf -coupled D1 dopamine receptor (D1R), and “indirect-pathway” MSNs, which express the Gαi/o-coupled D2 receptor (Thibault et al., 2013; Gerfen et al., 1990). Recent studies using single-cell transcriptomics have demonstrated that even within these primary striatal subpopulations, variable expression of distinct molecular features can further define distinct subclasses (Gokce et al., 2016; Märtin et al., 2019; Saunders et al., 2018; Stanley et al., 2020). However, the functional consequences of this transcriptional diversity remain unexplored. In the present study, we were interested in understanding diversity in signaling responses to drugs among neurons belonging to a single striatal population, i.e., MSNs. More specifically, we sought to answer two main questions: firstly, how much do individual MSNs differ in their signaling responses to pharmacological manipulation, and secondly, how does this variability inform our understanding of the way drugs act on these cells?
In addition to dopamine receptors, both of the predominant MSN subtypes also express NMDA, N-methyl-d-aspartic acid (NMDA)–sensitive glutamate receptors (NMDARs) (Flores-Hernández et al., 2002; Mao et al., 2004). In MSNs that form the direct pathway, the D1R and NMDAR cooperatively regulate synaptic plasticity and gene expression through activation of protein kinase A (PKA) and extracellular signal-regulated kinase (ERK) 1/2 protein kinases (Paul et al., 2000; Saxena et al., 1999; Valjent et al., 2005). Activation of these intracellular signaling pathways by D1R and NMDAR plays an important role, for example, in the effects of cocaine, which increases both dopamine and glutamate in the striatum (Fasano et al., 2009; Pascoli et al., 2011). However, little is known about intercellular heterogeneity in PKA or ERK1/2 responses. Interestingly, in a recent study using single-cell transcriptomics, only a small proportion of molecularly defined direct-pathway MSNs were found to respond to cocaine in vivo (Savell et al., 2020), even though all direct-pathway MSNs express both D1R and NMDAR. It is unknown how much of this in vivo variation can be attributed to circuit factors, such as heterogeneity in dopaminergic or glutamatergic inputs to the striatum, versus intrinsic properties of striatal MSNs. We therefore sought initially to determine whether the signaling responses of cultured primary striatal neurons show intercellular heterogeneity, by measuring D1R- and NMDAR-regulated PKA and ERK1/2 signaling in single cells at a large scale.
Genetically encoded biosensors based on fluorescent or bioluminescent proteins allow biochemical processes to be tracked in living cells with imaging technologies that detect changes in fluorescence or luminescence emitted by one or more tagged proteins. Many such sensors are based on the principle of fluorescence (or Förster) resonance energy transfer (FRET) and allow endogenous signaling events to be monitored without overexpression of signaling molecules (Miyawaki, 2011). Several studies have used such biosensors to probe endogenous signaling dynamics in striatal neurons, employing a variety of imaging modalities including wide-field and confocal microscopy (Castro et al., 2013; Yapo et al., 2017; Yapo et al., 2018; Mariani et al., 2019; Muntean et al., 2018). Evidence for intercellular response variability can be seen in many of these studies, but with analysis typically restricted to a few dozen cells, intercellular differences were not explored in detail. Here, we performed FRET biosensor imaging using high-content microscopy to rapidly image hundreds or thousands of cells over time and combined this with immunofluorescence to characterize multiple dimensions of signaling pathway activation.
Using this approach in primary striatal neurons, we explore intercellular variability in PKA and ERK1/2 responses to pharmacological stimulation of D1Rs or NMDARs and to direct activation of adenylyl cyclase. Using biosensors localized to either the cytosol or nucleus, we measure compartment-specific signaling in single cells segmented by morphology, to identify cell types that exhibit distinct response profiles. Focusing on striatal MSNs we also identify novel interactions between PKA and ERK1/2 signaling that occur in a cell compartment–specific manner.
Materials and Methods
Drugs and Reagents
Unless otherwise noted, products were purchased from Sigma-Aldrich. SKF 81297 hydrobromide (Toronto Research Chemicals), SCH 772984 (Selleckem), forskolin, phorbol-12-myristate-13-acetate (VWR), and U0126 stocks were prepared at 10 μM in DMSO and stored at −80°C. NMDA was prepared fresh in Hanks’ balanced salt solution (HBSS) (Wisent). The drug concentrations used in experiments are reported upon the first usage of each drug and remain unchanged for subsequent experiments.
Animals
Sprague-Dawley dams with postnatal day 1 pups were purchased from Charles River, Saint-Constant Québec, Canada. Animals were maintained on a 12/12-hour light/dark cycle with free access to food and water. Rats expressing Cre recombinase under the control of the Drd1 promoter (LE-Tg(Drd1a-iCre)3Ottc, Rat Genome Database 10412325) were obtained from the Rat Resource and Research Center (Strain #767, donors Brandon Harvey and Jim Pickel, National Institute on Drug Abuse/National Institute of Mental Health). All procedures were approved by the McGill University Animal Care Committee, in accordance with Canadian Council on Animal Care guidelines.
Isolation and Culture of Primary Striatal Neurons
Primary striatal neurons were prepared from mixed male and female postnatal day 1 pups as previously described (Jones-Tabah et al., 2020). One day in advance, 96-well optical bottom imaging plates (Nunc) were coated overnight with 0.1 mg/ml poly-d-lysine dissolved in PBS (Sigma). Prior to dissection, plates were washed three times with sterile water and allowed to dry, and all required solutions were warmed to 37°C. Pups were decapitated and brains were rapidly removed and placed in ice cold HBSS without calcium and magnesium (Wisent). Striata were dissected, and then digested on a rotator at 37°C for 18 minutes with papain diluted to a final concentration of 20 units/ml in a neuronal medium (Hibernate A Minus Calcium, BrainBits). Next, to halt digestion, tissue was transferred for 2 minutes to HBSS containing 10% fetal bovine serum, 12 mM MgSO4, and 10 units/ml DNase1 (Roche). Striata were pelleted by centrifugation at 300g for 5 minutes and supernatant was removed. Tissue was then triturated in HBSS with 12 mM MgSO4 and 10 units/ml DNase1 using a fire-polished Pasteur pipette. Once a cell suspension was produced (approximately 20 up and down triturations), it was passed through a 40 μm mesh filter (Fisher) to remove undigested tissue and then centrifuged on an OptiPrep gradient as previously described (Brewer and Torricelli, 2007) to remove cell debris. Purified neurons were then counted and diluted in neurobasal-A medium (NBA) with 1× final concentration of B27 supplement (Gibco), 1% GlutaMAX (Gibco), and 1% penicillin/streptomycin (henceforth referred to as complete NBA) supplemented with 10% fetal bovine serum. On a precoated 96-well plate (see above), 50,000 cells were plated in 75 µl volume per well. Sixteen hours after plating, cells were washed with HBSS without calcium and magnesium, and medium was changed for complete NBA, containing 5 µM cytosine-d-arabinoside to inhibit glial cell proliferation. Cultures were maintained in complete NBA, and medium was refreshed by exchanging 30% of the volume with fresh medium every 3 days.
Virus Production and Transduction of Primary Neurons
The FRET-based protein kinase biosensors AKAR3-EV and EKAR-EV were generously provided by Dr. Michiyuki Matsuda (Komatsu et al., 2011) and expressed with either nuclear export signal (NES) or nuclear localization signal (NLS) peptide sequences. FRET biosensors were expressed using a neuron-specific adeno-associated virus plasmid, pAAV-SynTetOFF (Sohn et al., 2017), kindly provided by Dr. Hiroyuki Hioki. All experiments were performed with adeno-associated virus (AAV) serotype 1 produced by the Neurophotonics Platform Viral Vector Core at Laval University, Québec. Primary striatal neurons were transduced by adding AAV directly into the culture medium three days after cell plating, using a multiplicity of infection of 5000 viral genomes per cell. Neurons were then maintained as described above for 7 days prior to imaging.
High-Content FRET Imaging in Primary Neurons
Live-cell imaging was performed at 37°C and with 3% CO2 using an Opera Phenix high-content confocal microscopy system (Perkin Elmer). One hour prior to imaging, medium was replaced with 90 μl of HBSS with calcium, magnesium, and sodium bicarbonate (Wisent). For antagonist experiments, antagonist or vehicle was also added at the appropriate concentration. Plates were transferred to the Opera Phenix and allowed to acclimatize in the live-cell chamber for 10 minutes before baseline image acquisition. Images were acquired using a 40× water-immersion objective using a 425 nm laser for excitation of cyan fluorescent protein (CFP). Emissions were detected with filters at 435–515 nm (CFP) and 500–550 nm [yellow fluorescent protein (YFP)]. For time-course experiments, approximately 20 fields were imaged per well, with fields being evenly spaced across the well. Baseline images were acquired (time = 0) and then vehicle or drug solution was added directly to each well at a volume of 10 μl to achieve the required final concentration in the well. Images were then acquired at the indicated intervals. To accommodate for imaging delay between sequentially imaged wells of a plate, drugs were added with staggered timing to ensure that images of each well were acquired after the specified time had elapsed. After each imaging session, cells were fixed and could be processed for immunofluorescence, as described next.
Cell Fixation and Immunofluorescence
Cells were fixed for 10 minutes in 2% paraformaldehyde prepared in PBS. Fixed cells were then washed twice with PBS and permeabilized for 10 minutes using 0.3% Triton X-100 in PBS. Blocking was performed for 3 hours at 4°C with 5% bovine serum albumin (BSA) in PBS. Next, cells were incubated overnight at 4°C with primary antibodies diluted in PBS with 5% BSA, then washed twice with PBS, and subsequently incubated for 3 hours at room temperature with secondary antibodies diluted in PBS with 5% BSA. After two additional washes with PBS, cells were incubated with Hoechst dye (Invitrogen) diluted 1:10,000 in PBS. The following primary antibodies and dilutions were used: anti–dopamine-and-cAMP–regulated phosphoprotein of 32 kDa (DARPP-32) (1:1000, catalog #2302, Cell Signaling Technology), anti-cFos (1:2000, catalog #sc-52, Lot #C1010, Santa Cruz), anti–phospho-H3-Ser10 (1:1000, catalog #5176, Abcam), anti–protein kinase A catalytic subunit (1:1000, catalog #610980, BD Biosciences). Secondary antibodies were Alexa 488 anti-mouse (1:1000; A21236) and Alexa 647 anti-rabbit (1:1000; A21245), both from Invitrogen. Fluorescence imaging of fixed cells was performed in an Opera Phenix high-content confocal microscope at 40X magnification.
Image Processing and Fluorescence Quantification
All image analysis was performed using Columbus analysis software (Perkin Elmer) using the following generalized workflow. For FRET imaging, transduced cells were automatically identified using “Method B” run on the combined CFP and YFP fluorescence intensity. Specific thresholds for object seize, brightness and contrast were optimized for each sensor and then kept constant for all experiments. The identified objects were then further filtered based on morphology and fluorescence intensity to exclude dead cells, noncell objects and clusters of overlapping cells. Specific parameters were determined by visual inspection and were adjusted for each sensor. After filtering the object population, CFP and YFP intensities at each pixel were used to calculate the FRET ratio (YFP/CFP), which was then averaged to create a single FRET ratio for each object at each time point. Values for fluorescence intensities, FRET ratio, object area, object roundness and spatial coordinates within the image were calculated and exported as text files for further analysis. The same pipeline was applied for quantitative immunofluorescence experiments, except that nuclei were first detected using the Hoechst stain so as to create a mask to define a region in which fluorescence signals from the Alexa-488 and Alexa-647 channels were then quantified.
Single-Cell Analysis
For time-course FRET imaging, single cells were tracked across time points using an analysis script written in R (code and sample data are available on the Hébert laboratory GitHub https://github.com/HebertLab/Single-Cell-FRET, and the complete data set is available upon request). This script was applied to the output of the image analysis. Objects were matched between time points using the Cartesian coordinates of each object in the image, with a threshold set for maximum acceptable displacement between subsequent images. To ensure correct matching of objects at each time point, the size and fluorescence intensity of each object was cross-checked, and objects which exhibited a change in size or brightness of greater than 30% were excluded from analysis. Only objects that could be positively matched across all six time points in the experiment were included for analysis. ΔFRET values were then calculated for each object relative to the object's baseline FRET and converted into percent changes in FRET (%ΔF/F). Here, the denominator (F) was calculated as the average baseline FRET, not of the particular cell in question, but rather of all cells sampled in the same well. This approach avoided extremely high %ΔF/F values that tended to occur with cells having a very low baseline FRET value. The number of cells analyzed in each experiment is indicated in the figure legends.
Values corresponding to single cells from independent experiments were then pooled for data visualization in heat maps. Grouping of FRET responses by either magnitude or kinetics was performed by time-series clustering provided by the TSclust package, in R using the “pam” and “shape” functions for clustering by response magnitude and kinetics, respectively (Montero and Vilar, 2014). All figures were generated as R markdown files. R scripts for analysis and data visualization are available on the Hébert laboratory GitHub (https://github.com/HebertLab/Single-Cell-FRET).
Response Clustering Analysis
For FRET data sets, response types (e.g., high, medium, low, none) were assigned based on the time-series clustering described in the previous section. Mean response profiles for each of the defined clusters from Figs. 4 and 5 can be seen in Supplemental Fig. 2. Mean response profiles for each cluster in Figs. 6 and 7 are shown within the respective figures. For immunofluorescence analysis, z-scores were calculated for each cell using the fluorescent intensities in the vehicle treated condition as reference population. Responses were then defined as follows: z < 1, no response, 1 < z < 2, low response, 2 < z < 3, medium response, z > 3, high response.
Statistical Analysis
All statistical testing was performed in R using the rstatix package or in GraphPad Prism 9. Data in Fig. 1 and Fig. 6 were analyzed using Bonferroni corrected t tests, with each treatment being compared with DMSO. In Fig. 3 and Fig. 7, data were analyzed first by two-way ANOVA, using treatment and cell type as factors (Fig. 3) or pretreatment and response cluster as factors (Fig. 7). ANOVA was followed by multiple comparisons performed using Bonferroni corrected t tests, with significant results indicated. All statistical comparisons were performed using the mean and variability of biologic replicates (as opposed to individual cells), where a biologic replicate was defined by cells isolated from a distinct group of rats i.e., cells isolated from a partial or entire litter of pups. Curve fitting was performed in GraphPad Prism 9 using the Logistic equation for sigmoidal curves, and the Rise-and-fall time-response equation from the Pharmechanics package for rise to steady state curves (Hoare et al., 2020).
Results
In the experiment presented in Fig. 1, we expressed substrate-based reporters of PKA [A-kinase activity reporter (AKAR)] and ERK1/2 [ERK kinase activity reporter (EKAR)] activity with either nuclear export (NES) or nuclear localization (NLS) sequences in primary striatal neurons and then performed live-cell high-content imaging (Fig. 1, A and D, respectively). Neuronal cultures were treated with either vehicle, SKF 81297 (a selective agonist of D1-like receptors), NMDA, a combination of these two drugs, or forskolin (a direct activator of adenylyl cyclase). Images were acquired before treatment, and again 10 minutes after stimulation and the FRET changes of individual cells (shown as violin plots) and biologic replicates (points) are shown (Fig. 1, B and C; Fig. 1, E and F). Broadly, we observed activation of cytosolic and nuclear PKA and ERK1/2 by all drug conditions, with the exception that NMDA activated nuclear, but not cytosolic PKA.
Focusing on nuclear-localized (NLS) biosensors, we next explored the use of morphologic criteria in our analysis to explore the intercellular variability of nuclear FRET responses. We first compared the size and roundness of transduced nuclei to the basal FRET ratio of each cell (calculated as above); basal FRET ratios were found to be variable, ranging from 0.5 to 1.25, but evenly distributed across cells with varying size and roundness (Fig. 2, A and D). Published evidence suggests that MSNs can be distinguished from other striatal cell types based on a distinctive nuclear morphology (Matamales et al., 2009). Accordingly, we hypothesized that cells with 50–100 μm2 round nuclei (roundness score > 0.9) (Fig. 2, A and D, purple squares) would be enriched for MSNs. The cutoffs for defining these cells, which we refer to as “medium-round,” were based on the previous finding that striatal MSNs in vivo were found to have highly round nuclei, with cross-sectional area of 75–95 μm2 (Matamales et al., 2009). We also assigned the remaining cells to three additional categories based on their nuclear morphology (Fig. 2, A and D, colored boxes). We next compared the FRET response of each cell plotted against the CFP intensity (a surrogate measure for the expression level of the biosensor) and observed that although FRET responses appeared independent of expression level, cells with this “medium round” morphology tended to exhibit higher biosensor expression and larger FRET responses compared with the other categories of cell (Fig. 2, B and E). Within these “medium round” cells there appeared to be no overall association between the expression level of the biosensor and the response magnitude to stimulation (Fig. 2, C and F). However, for the EKAR biosensor specifically, we did observe a qualitative reduction in the number of high responses among cells with the highest biosensor expression levels (Fig. 2F). Although no corrective action was taken in our analysis, it does raise the possibility that when expression levels are too high, biosensor responses may be blunted due to presence of unphosphorylated “spare biosensors.” Overall, we concluded that the observed differences in biosensor expression level do not account for differences in response magnitude.
To further explore our hypothesis that the “medium round” cell group was enriched in striatal MSNs we performed immunofluorescence to measure the striatal cell marker DARPP-32 (Ivkovic and Ehrlich, 1999) and used Hoechst dye to stain nuclei and determine nuclear morphology (Fig. 3A). We plotted relative DARPP-32 expression levels on a coordinate plot of nuclear size and roundness, and found that as predicted, cells with high levels of DARPP-32 expression were concentrated in the “medium round” cell grouping (Fig. 3B). Specifically, out of 34,429 cells imaged, 12,815 had high DARPP-32 expression, (defined by a z-score > 0), and of those, 12,296 (or 95.9%) were found in the “medium round” grouping. To confirm the enrichment of D1R-expressing MSNs in the “medium round” cell group, we cultured striatal neurons from rats expressing Cre-recombinase under the control of the Drd1 promoter and transduced these with a Cre-dependent mCherry-expressing AAV (Supplemental Fig. 1A). When we again quantified nuclear morphology, we found that mCherry-positive neurons were concentrated in the “medium round” grouping (Supplemental Fig. 1B).
Reanalysis of the FRET data presented in Fig. 1 by segregating cells into these nuclear morphology–defined groups showed that, for both nuclear PKA (AKAR-NLS) and nuclear ERK1/2 (EKAR-NLS), the response to drug treatments significantly differed between cell types (two-way ANOVA, Treatment × Cell Type interaction, P < 0.0001 and P < 0.001, for AKAR-NLS and EKAR-NLS, respectively) (Fig. 3, C and D). Specifically, “medium round” cells were more responsive to all drugs with respect to both PKA and ERK1/2 signaling, with several individual comparisons being statistically significant. Finally, as an independent method of verifying these differences in drug response, we performed cFos immunofluorescence on striatal cultures after 3-hour drug exposure (Fig. 3E). Here we again found that cFos induction significantly differed between cell types (two-way ANOVA, Treatment × Cell Type interaction P < 0.0001) and that relative to all other cell groups, “medium round” cells were significantly more responsive in regard to cFos induction by SKF 81297, NMDA and forskolin.
Using this morphologic approach to focus our analysis on the MSN-enriched cells defined by “medium round” nuclear morphology, we next investigated the intercellular variability in nuclear PKA and ERK1/2 signaling. Single-cell analysis allowed us to visualize the responses of individual cells, imaged at 10-minute intervals for 50 minutes after stimulation (Fig. 4, A and B). To group cell with similar responses, we performed time-series clustering (see Materials and Methods) to assign each cell to a specific response type based on the magnitude of FRET response. Clustering was performed independent of biologic replicate or treatment condition. Categories were subsequently labeled as nonresponsive cells, low-, medium-, and high-responding cells, based on the response profile of the cluster (Supplemental Fig. 2). We then analyzed drug responses both in terms of the average percentage of cells in each replicate displaying a given response type (Fig. 4, C and D) and by the average response magnitudes, either when all cells were included, or when cells in the nonresponsive cluster were excluded (Fig. 4, E and F). This analysis revealed that although the D1R agonist SKF 81297 initially appeared to be the weakest activator of nuclear PKA and ERK1/2, this impression was driven by a lower percentage of responding cells, not by a lower average magnitude of response per cell (Fig. 4, E and F). Importantly, the rank order of drug effects differed, depending on whether analysis was restricted to drug-responsive cells or if cells that were unresponsive to drug were included (Fig. 4, E and F).
We also performed a similar single-cell analysis of PKA and ERK1/2 signaling in the cytosol, using biosensors fused to a nuclear export sequence (NES) (Fig. 5, A–F). Whereas in the case of NLS-tagged biosensors, the nuclear morphology was used to filter cells and enrich for MSNs, in the analysis of NES-tagged biosensors, this segmentation was not possible, and so the total population of imaged cells was analyzed. Cytosolic signaling responses largely paralleled those in the nucleus, with three notable differences. First, whereas NMDA activated nuclear PKA (Fig. 4A), it had a minimal effect on cytosolic PKA (Fig. 5A). Second, when only responding cells were considered, SKF 81297 was as efficacious as forskolin in activating nuclear PKA (Fig. 4, E and F) but less efficacious than forskolin in activating cytosolic PKA (Fig. 5E). Lastly, whereas nuclear responses were almost exclusively stimulatory (i.e., increased FRET) and responses to vehicle were minimal, in the cytosol we observed more pronounced inhibitory responses to ligand stimulation and larger vehicle responses, both in the positive and negative direction.
The rapid and sustained activation of nuclear PKA by D1R agonists in rat striatal neurons which we observed (Fig. 4, A and E) has been previously described in mouse brain slices (Yapo et al., 2018). In cultured striatal neurons, nuclear ERK1/2 was also activated to a similar, if not greater, degree by stimulation of either the D1R or adenylyl cyclase (Fig. 4, B and F). Although the strong and rapid activation of nuclear PKA would suggest a potentially important functional role in this structure, the specific nuclear targets of PKA in striatal neurons remain largely uncharacterized. To investigate the relative contributions of PKA and ERK1/2 to the regulation of specific nuclear signaling events, we used two inhibitors of ERK1/2-dependent signaling: U0126, an inhibitor of the upstream protein kinase mitogen-activated protein kinase (MAPK) and ERK kinase, and SCH772984, an inhibitor of ERK1/2. Pretreatment with either inhibitor abolished nuclear ERK1/2 signaling in response to SKF 81297 or forskolin (Fig. 6A). Similar results were obtained for cFos expression, except that the inhibitors did not fully inhibit the response to forskolin (Fig. 6B). At the single-cell level, reduction in cFos immunofluorescence were evident both in terms of maximum levels of induction, and the percentage of responding cells, but with forskolin treatment, some cFos induction was again still evident (Supplemental Fig. 3, A and B). We next examined phosphorylation of histone H3 on serine 10 (pH3-S10), a transcription-activating event that has been associated with D1R-dependent ERK1/2 signaling in striatal neurons in vivo (Bertran-Gonzalez et al., 2008; Brami-Cherrier et al., 2005). Here, pH3-S10 was largely unaffected by either inhibitor, with a partial but statistically nonsignificant reduction observed only for SCH772984 (Fig. 6C; Supplemental Fig. 3, C and D). This suggests a role for PKA-dependent, and ERK1/2-independent signaling in the deposition of pH3-S10 in primary striatal neurons.
Although to our knowledge neither U0126 nor SCH772984 is known to have off-target effects on PKA, we nonetheless measured PKA-dependent signaling to confirm that it was preserved in the presence of these inhibitors. Unexpectedly, we observed that the ERK1/2 inhibitor reduced nuclear, but not cytosolic PKA activation after a 10-minute treatment with SKF 81297 or forskolin (Fig. 6, D and E). This decrease was transient, and levels of nuclear PKA activation similar to that observed in vehicle pretreated cells were still achieved in the presence of either inhibitor 20–50 minutes after stimulation (Supplemental Fig. 4). To investigate the altered kinetics of nuclear PKA activation, single-cell responses were subjected to kinetic clustering analysis (see Materials and Methods), to categorize cells according to the time-course of their responses to drug. This analysis revealed four predominant response types (Fig. 6F). Both inhibitors tended to increase the percentage of “slow” responding cells at the expense of “fast” responding cells, but these changes did not reach statistical significance (Fig. 6G).
We next performed high-content imaging on a shorter time scale (1-minute intervals from 1 to 9 minutes) to more precisely characterize the timing of early cytosolic and nuclear signaling events. Stimulation with either SKF 81297 or forskolin rapidly activated cytosolic PKA, which reached maximum amplitude within 60 seconds, whereas the onset of nuclear PKA activity was slower, peaking 6–7 minutes after stimulation with either SKF 81297 or forskolin (Fig. 7, A and B). The time-course of nuclear PKA activation in primary striatal neurons appeared sigmoidal, similar to previous findings in mouse brain sections (Yapo et al., 2018). Neither U0126 or SCH772984 detectably altered cytosolic PKA activity when measured on this shorter time scale (Fig. 7, C and D) but significant differences were found in the extent of nuclear PKA activation (two-way ANOVA, Pretreatment × Time interaction, P < 0.0001 for both SKF 81297 and forskolin) (Fig. 7, E and F) with either inhibitor appearing to reduce nuclear PKA. Nuclear ERK1/2 activation followed similar kinetics to that of nuclear PKA, reaching its maximum 5–6 minutes after SKF 81297 or forskolin treatment, and as expected, was abolished by pretreatment with the two inhibitors of ERK1/2-dependent signaling, U0126 or SCH772984 (Fig. 7, G and H). Response-magnitude clustering (Fig. 7I) revealed that within this 10-minute time frame, U0126 and SCH772984 did not affect the percentage of nonresponsive or low-responsive cells, but significantly reduced the number of high-responding cells to either SKF 81297 or forskolin (Fig. 7, J and K). In the case of forskolin, this was matched by a significant increase in cells with medium response magnitude (Fig. 7K). Taken together, these findings suggest that ERK1/2 plays a facilitatory role in the activation of nuclear PKA and that so called “high-responders” at the level of nuclear PKA exhibit the greatest sensitivity to inhibition of the ERK1/2 pathway.
A potential mechanism that could explain the reductions in nuclear, but not cytosolic, PKA signaling would be ERK1/2-dependent facilitation of PKA nuclear translocation. To measure nuclear translocation of PKA, we used immunofluorescent detection of the catalytic subunit of PKA (PKAC) and used Hoechst dye to demarcate the nuclear compartment. Prior to stimulation, PKAC was largely absent from the nucleus, but its levels were increased upon stimulation with SKF 81297 or forskolin (Fig. 8, A and B). This nuclear localization of PKAC was apparent only in a subset of cells (approximately 25% and 45% of putative MSNs for SKF 81297 and forskolin, respectively) (Fig. 8C; Supplemental Fig. 5). Pretreatment with U0126 or SCH772984 did not alter the translocation of PKAC to the nucleus (Fig. 8, D and E; Supplemental Fig. 5, A and B) indicating that ERK1/2 inhibition does not attenuate nuclear PKA signaling by reducing nuclear accumulation of PKAC.
The finding that U0126 and SCH772984 did not affect the movement of PKAC to the nucleus would instead suggest that ERK1/2 promotes the relative activity PKA without altering the quantity of PKA in the nucleus. To test this hypothesis, we sought to determine whether co-activation of ERK1/2 signaling through an independent pathway could enhance nuclear PKA activation stimulated by forskolin. In the striatum in vivo, independent activation of PKA and ERK1/2 signaling could potentially occur via co-activation of D1R and NMDARs. However, since we had previously observed that NMDA can independently activate nuclear PKA signaling (Fig. 4) we opted to use phorbol 12-myristate 13-acetate (PMA) to activate ERK1/2 signaling via protein kinase C (PKC) (Castagna et al., 1982). As expected, PMA or forskolin activated nuclear ERK1/2 (Fig. 9A; Supplemental Fig. 6A). Compared with treatment with either drug alone, co-treatment with forskolin and PMA resulted in a faster onset but produced similar maximum levels of ERK1/2 activity in the nucleus, and SCH772984 abolished the signal (Fig. 9A; Supplemental Fig. 6A). PMA activated nuclear PKA to a similar degree as forskolin, both in terms of mean response and percentage of responding cells (Fig. 9B; Supplemental Fig. 6B). This effect was initially unexpected given the pharmacology of PMA, but activation of PKA through a PKC-dependent mechanism has previously been described to occur in neurons (Chen et al., 2017). We further found that nuclear PKA activation by PMA was abolished by SCH772984, suggesting that this effect was ERK1/2-dependent and not directly mediated by PKC. Compared with treatment with either drug alone, the combination of PMA and forskolin increased the rate of nuclear PKA activation, and similar to our previous observations, this was attenuated by SCH772984 (Fig. 9B; Supplemental Fig. 6B). PMA also activated cytosolic PKA (Fig. 9C; Supplemental Fig. 6C), but this cytosolic response differed from the nuclear response in three respects: it occurred more promptly, it was smaller, and it was not clearly inhibited by SCH772984 (compare Fig. 9B with Fig. 9C). Taken together, findings further argue for nuclear-specific cross-talk whereby ERK1/2 activation can promote PKA signaling in striatal neurons.
Discussion
Cells are often categorized by anatomic location, morphologic features, functional properties, or the expression of specific molecular markers. Cell type designations are inherently limited by the availability of technologies used to isolate, sort or label cells. As technologies have become available for single-cell profiling at the genomic, proteomic, and functional levels, previously inaccessible heterogeneity has been revealed within cell populations that were once treated, out of necessity, as homogenous (Cadwell et al., 2016; Fuzik et al., 2016). This “homogeneity assumption” has also affected the study of cellular signaling and molecular pharmacology. For example, drug effects are often expressed in terms of a single averaged drug response determined from a population of interest, but biologically important differences may exist in the way drugs engage individual cells within such a population. Importantly, some generalization allows us to conceptualize complex systems such as signaling pathways and the way drugs act on them. However, this simplification also arises from limitations in the available technology. Recently, new techniques including high-content imaging have allowed researchers to investigate the existence and importance of intercellular heterogeneity in signaling pathway engagement (Handly, Yao, and Wollman, 2016; Singh et al., 2010; Yao, Pilko, and Wollman, 2016; Bao et al., 2010; Tsou et al., 2011; Chavez-Abiega et al., 2021). Here, we illustrate a simple approach in which the combination of biosensors and high-content microscopy allowed us to capture the signaling dynamics of thousands of primary cells in culture.
The single-cell approach reported here offers at least one major advantage and limitation. The main advantage is throughput. The ability to image thousands of cells decreases experimenter bias and increases replicability. These large data sets also allow us to consider the proportion of responding cells when investigating drug effects. As an example, seemingly identical results could be produced by a few strongly responding cells or many weakly responding cells. A case in point is the activation of nuclear PKA by D1R versus NMDAR agonists. In the aggregate response, the two drugs appeared to produce similar effects (Fig. 1H); the single-cell results disambiguated this observation, revealing that although NMDA activates nuclear PKA in a larger proportion of cells, the D1R agonist produces larger responses. The main limitation of our single-cell approach is the lack of temporal control of drug release within the high-content imaging system. For this reason, this approach would not be useful for studying cellular responses to transient receptor stimulation.
This approach applied to striatal neurons yielded three main sets of findings. First, we uncovered substantial intercellular variability in the magnitude and kinetics of signaling pathway activation in response to various pharmacological stimuli. Second, as well as confirming a previous report that D1R activation in striatal neurons rapidly activates both nuclear and cytosolic PKA (Yapo et al., 2018), we showed that ERK1/2 is also rapidly and persistently activated in the nucleus of primary striatal neurons; moreover, nuclear PKA and ERK1/2 activation were induced not only by D1R agonist but also by NMDAR agonists, cAMP signaling, and by an activator of PKC. These findings indicate that many distinct pathways converge on these nuclear signals. Finally, we describe a novel mode of cross-talk between PKA and ERK1/2, whereby ERK1/2 signaling regulates the activity of PKA specifically in the nucleus.
The molecular features that give rise to the signaling heterogeneity observed in our experiments remain to be determined, but could potentially arise from differences in receptor density, or the regulation of intracellular signaling cascades. Since our striatal cultures were unlikely to comprise pure populations of striatal MSNs, we used nuclear morphology measurements to refine the population of analyzed cells to include only those cells most likely to be MSNs. Within this MSN-enriched population, the D1R agonist SKF 81297 elicited nuclear PKA and ERK1/2 responses in ∼30% and ∼40% of cells, respectively. If one assumes that most “medium round” cells were in fact MSNs, with around 50% of those MSNs expressing D1Rs as reported in vivo (Schambra et al., 1994; Thibault et al., 2013), then our results suggest that most of the cultured D1R-expressing MSNs responded to the D1 agonist, albeit with widely differing response magnitudes. Even forskolin, which directly activates adenylyl cyclase and is commonly used as a positive control in studies of G protein–coupled receptor–mediated cAMP signaling, produced a range of effect sizes, with approximately 10% of neurons unresponsive to forskolin within the timeframe measured. Importantly, the response magnitude in individual cells was unrelated to the expression level of the biosensor.
Compared with other neurons, striatal MSNs reportedly exhibit larger fluxes in cAMP and PKA signaling in both the cytosol (Castro et al., 2013; Yapo et al., 2017) and nucleus (Yapo et al., 2018). The mechanisms of G protein–coupled receptor–mediated protein kinase signaling to the nucleus have not been fully elucidated, and as argued below, are likely cell-type specific. Although the activated PKA catalytic subunit can enter the nucleus by passive diffusion from the cytoplasm (Harootunian et al., 1993), the inactive PKA holoenzyme has also been observed in the nucleus in neurons (Ilouz et al., 2017). In a previous report, D1R-dependent nuclear PKA activation in cortical neurons followed slow linear kinetics, consistent with passive diffusion from the cytosol, whereas in striatal neurons, activation of nuclear PKA was rapid, yet with a consistent delay relative to activation of cytosolic PKA (Yapo et al., 2018). Our results largely confirm these previous findings. The ∼5-minute delay in nuclear relative to cytosolic PKA activation would be inconsistent with free diffusion of cytosolic cAMP to activate PKA preexisting in the nucleus. It was previously suggested that activated PKA could facilitate its own entry into the nucleus (Yapo et al., 2018). Our findings suggest that, independent of any effect on the nuclear translocation of PKA, ERK1/2 activation also contributes to enhancing nuclear PKA signaling in striatal neurons.
ERK1/2 has been described as a coincidence detector, synergistically responding to combined D1R and NMDAR signaling (Valjent et al., 2005). We found that ERK1/2 signaling was stimulated by individual application of the D1R or NMDAR agonists, with effects occurring in both the cytosol and nucleus. However, in contrast to previous work, the D1R and NMDAR agonists in combination produced a subadditive effect. These findings may be best explained by two factors: first, our primary striatal cultures were effectively dopamine-free, given that the striatal dopaminergic innervation is entirely extrinsic, and second, in vivo striatal dopamine depletion enhances coupling between D1R and ERK1/2 (Darmopil et al., 2009; Gerfen et al., 2002; Pavón et al., 2006; Santini et al., 2009; Westin et al., 2007). The lack of dopaminergic innervation could thus feasibly explain the potent ERK1/2 activation by D1R seen in our primary cultures. Furthermore, the activation of cytosolic and nuclear ERK1/2 in nearly all forskolin-stimulated striatal neurons suggest that in these cells, the cAMP cascade is strongly coupled to MAPK activation.
Our data also describe two novel features of nuclear signal transduction in striatal neurons. Firstly, our findings argue against previous studies that have indicated ERK1/2 activity is required for the induction of histone 3 phosphorylation (pH3-S10) in striatal MSNs (Bertran-Gonzalez et al., 2008; Brami-Cherrier et al., 2005). Thus, in our primary striatal cultures, stimulation of D1R or AC produced strong pH3-S10 induction even under conditions where no ERK1/2 signaling was detected.
We also observed that ERK1/2 signaling contributes to the activation of nuclear PKA. Notably, in striatal neurons stimulated with D1R agonist or forskolin, nuclear PKA and ERK1/2 were activated with comparable kinetics, and when ERK1/2 signaling was blocked, the rate of nuclear PKA activation was significantly diminished. Moreover, we found that independently promoting ERK1/2 signaling through the PKC pathway could also activate nuclear PKA. This novel mode of cross-talk could contribute to the previously described synergistic effects of co-incident D1R and NMDA signaling in vivo, particularly with respect to the regulation of nuclear functions such as transcription and the epigenetic modifications that regulate it. Although the underlying mechanism remains to be determined, our findings argue that ERK1/2 does not regulate the translocation of PKA to the nucleus, but rather the balance of PKA activity in the nucleus. Given that AKAR is a substrate-based reporter, such an effect could occur through facilitation of PKA activation, or inhibition of phosphatases that normally antagonize PKA-dependent phosphorylation; the specific ERK1/2 substrates mediating this effect remain to be investigated. It is well established that D1R-dependent signaling can promote ERK1/2 activation through multiple routes within striatal MSNs, including via cAMP/PKA/DARPP-32–dependent mechanisms (Paul et al., 2000; Saxena et al., 1999; Valjent et al., 2005) and the numerous points of interaction between these signaling cascades complicates the functional delineation of individual signaling pathways.
Here, we have used high-content microscopy and FRET biosensors to characterize PKA and ERK1/2 signaling in the cytoplasm and nucleus of cultured striatal neurons, and to identify intercellular differences, in both the magnitude and kinetics of signaling responses to drug application. This approach has yielded new observations about the signaling of PKA and ERK1/2 in striatal neurons, and more generally can provide a novel vantage point from which to capture the intercellular nuances of cellular signaling dynamics.
Acknowledgments
The authors thank Dr. Michiyuki Matsuda (Kyoto University) for providing them with biosensor constructs and Dr. Hiroyuki Hioki (Kyoto University) for providing AAV plasmids. T.E.H. is the holder of the Canadian Pacific Chair in Biotechnology. The authors thank the McGill Pharmacology and Therapeutics Imaging and Molecular Biology platform, as well as the McGill Advanced BioImaging facility for training and assistance with microscopy and image analysis. The authors thank Hanan Mohammad for help in colony management for the Drd1-Cre transgenic rats and Emma Paulus for assistance in maintaining neuronal cultures. Lastly, the authors thank members of the Hébert, Clarke, and Tanny laboratories for feedback and guidance throughout the development of the project and for critical reading of the manuscript.
Authorship Contributions
Participated in research design: Jones-Tabah, Tanny, Clarke, Hébert.
Conducted experiments: Jones-Tabah, Martin.
Contributed new reagents or analytic tools: Martin.
Performed data analysis: Jones-Tabah, Tanny, Martin, Clarke, Hébert.
Wrote or contributed to the writing of the manuscript: Jones-Tabah, Tanny, Clarke, Hébert.
Footnotes
- Received March 27, 2021.
- Accepted September 7, 2021.
This work was supported by grants from the Weston Brain Institute [RR181011] and Canadian Institutes of Health Research [PJT-256524]. J.J.-T. was supported by doctoral studentships from the Canadian Institutes of Health Research and the McGill Healthy Brains for Healthy Lives initiative. R.M. was supported by studentships from the McGill-CIHR Drug Development Training Program and the McGill Faculty of Medicine.
A version of this article has been submitted to the preprint server bioRxiv (https://www.biorxiv.org/).
↵This article has supplemental material available at molpharm.aspetjournals.org.
Abbreviations
- AAV
- adeno-associated virus
- AKAR
- A-kinase activity reporter (FRET biosensor)
- BSA
- bovine serum albumin
- CFP
- cyan fluorescent protein
- DARPP-32
- dopamine-and-cAMP–regulated phosphoprotein of 32 kDa
- D1R
- dopamine receptor subtype D1
- EKAR
- ERK kinase activity reporter (FRET biosensor)
- ERK
- extracellular signal-regulated kinase
- FRET
- Förster resonance energy transfer
- HBSS
- Hanks’ balanced salt solution
- MAPK
- mitogen-activated protein kinase
- MSN
- medium spiny GABAergic projection neuron
- NBA
- neurobasal-A medium
- NES
- nuclear export signal
- NLS
- nuclear localization signal
- NMDA
- N-methyl-d-aspartic acid
- NMDAR
- NMDA receptor
- pH3-S10
- phosphorylation of histone H3 on serine 10
- PKA
- protein kinase A
- PKAC
- catalytic subunit of PKA
- PKC
- protein kinase C
- PMA
- phorbol 12-myristate 13-acetate
- YFP
- yellow fluorescent protein
- Copyright © 2021 by The American Society for Pharmacology and Experimental Therapeutics