Labeling of active neural circuits in vivo with designed calcium integrators

Brain function relies on patterns of synaptic input and action potential firing, which are accompanied by transient changes in free intracellular calcium (Ca²⁺) concentration ([Ca²⁺]) (1, 2). Genetically encoded calcium indicators (GECIs) are useful for monitoring the activity of populations of neurons and synapses in behaving organisms (3–6). However, the transient nature of GECI responses following [Ca²⁺] rises requires continuous monitoring during behavior using sophisticated imaging equipment with limited fields of view, and often physical restraint to prevent brain movement. Alternatively, expression of immediate early genes (IEGs) such as Arc and cFos can be measured following free behavior (7) within several hours, but is only weakly correlated with neural electrical activity (8, 9) and is constrained neither by genetic cell type nor by a precise temporal window. The creation of molecular tools to allow genetic targeting of direct reporters of neural activity that can be rendered permanent for post hoc analysis across entire brains, and that only mark neurons active during short, user-defined behavioral epochs, would be transformative (10, 11).

The GCaMP calcium indicator reports changes in free [Ca²⁺] [rise and decay times <1 s (12)] through mechanisms that modulate the environment of the 488-nm–absorbing, fluorescent 4-(p-hydroxybenzylidene)-5-imidazolinone chromophore of circularly permuted green fluorescent protein (cpGFP) (13). The photoconvertible fluorescent protein (FP) EosFP is a bright green FP that irreversibly converts to a bright red fluorescent species {backbone cleavage produces 2-[(1E)-2-(5-imidazolyl)ethenyl]-4-(p-hydroxybenzylidene)-5-imidazolinone} upon illumination with violet light (14). Previously, circular permutation of a photoconvertible FP and attachment to the Ca²⁺-binding protein calmodulin (CaM) and its associated M13 peptide (M13) gave rise to a photoconvertible GECI that produced GCaMP-like rises in fluorescent intensity both in the unconverted, green state and in the converted, red state (15). We reasoned that related constructs would undergo Ca²⁺-dependent allosteric modulation of the chromophore photoconvertibility. Such a protein species would convert from green to red only in the simultaneous presence of high [Ca²⁺] and user-supplied violet light (Fig. 1A). By combining library screening and structure-guided mutagenesis (16) we produced such a protein, which we named CaMPARI (calcium-modulated photoactivatable ratiometric integrator).

CaMPARI has the primary structure depicted in Fig. 1B and fig. S1, and includes a nuclear export signal (NES) to exclude it from the nucleus when expressed in eukaryotic cells. CaMPARI exhibits 60 and 40% of the green and red fluorescence brightness of mEos2, respectively, and photoconverts to the red form 21 times as fast in the presence of calcium (Fig. 1C and table S1). Photoconversion (PC) of the calcium-bound and calcium-free states of CaMPARI occurs three times and one-seventh as fast, respectively, as that of the parent EosFP variant (fig. S2). No effect of calcium on PC of the parent EosFP variant was observed (fig. S2).

Measuring the PC rate while titrating calcium gave an apparent dissociation constant (K_d) of 128 ± 6 nM (Fig. 1D). We modified the affinity of CaMPARI for calcium via mutagenesis at the CaM/M13 interface (fig. S3), thus increasing dynamic range in cells with a wide range of baseline and peak calcium concentrations. In addition to a calcium-dependent green-to-red PC rate, the fluorescence of both green and red CaMPARI states is also calcium-dependent, decreasing by a factor of 9 and 23, respectively, upon calcium binding (Fig. 1E, fig. S4, and table S1).

After initial characterization in cultured HeLa cells (fig. S5), we expressed CaMPARI in primary cultured rat hippocampal neurons. Whereas a 2-s pulse of PC light (1.5 W/cm²) resulted in only minimal CaMPARI PC in unstimulated neurons, the same light exposure induced significant PC in field-stimulated neurons (Fig. 1, F to H), with a roughly linear correspondence between the frequency of field stimulation and the extent of CaMPARI PC (Fig. 1I). Stimulation of CaMPARI-expressing neurons without exposure to 405-nm light led to transient decreases of its green fluorescence (fig. S6), but with no conversion to the red form. We additionally depolarized neurons either by increasing extracellular potassium concentration or via pharmacogenetic activation of exogenously expressed P2X channels with adenosine 5′-triphosphate (ATP), both in concert with application of PC light. Cells were fixed with 4% paraformaldehyde (PFA) and colabeled with traditional antibody-staining techniques (Fig. 1, J and K). The green and red fluorescence of CaMPARI each decreased ~50% upon PFA fixation (fig. S7). Retention of the CaMPARI PC signal through fixation, permeabilization, and multicolor antibody labeling enables integrated Ca²⁺ concentrations within individual cells to be directly correlated to independent proteomic metrics typically used for post hoc estimation of cell-type identification and examination of a wide range of biochemical events related to plasticity and other key signaling cascades. We coimmunolabeled fixed neurons for P2X expression level (fig. S8, an indication of the magnitude of the exogenous pharmacogenetic manipulation) and phosphorylated nuclear cyclic adenosine 3′,5′-monophosphate (cAMP)–responsive element-binding protein (pCREB), which increased with elevated intracellular calcium (Fig. 1L). CaMPARI signal after fixation was similar to that of conventional immunofluorescence and in situ hybridization markers, making it possible to image a direct measure of Ca²⁺ in parallel with proteomic markers that can only be accessed after fixation.

We next used CaMPARI to mark active neurons in vivo, during sensory stimuli or behavior, in four preparations of three model organisms: larval zebrafish and Drosophila melanogaster to demonstrate whole-brain and genetically targeted neural ensemble mapping during free behavior; mouse primary visual cortex (V1) for large-scale post hoc confirmation of known distributions of direction-selective cells in a mammalian brain; and adult Drosophila to highlight “labeled lines” of synaptically connected ensembles following naturalistic or optogenetic stimulation of sensory pathways. We chose calcium affinity variants (CaMPARI in mouse and adult Drosophila; V398D, with about one-seventh the affinity for Ca²⁺, in larval and adult Drosophila; W391F+V398L, with about one-half the affinity for Ca²⁺, in larval zebrafish; fig. S3) based on performance of different GCaMP variants in these preparations, as well as literature estimates of intracellular calcium.

To demonstrate the utility of CaMPARI for whole-brain neural circuit marking in larval zebrafish (Danio rerio) during free behavior, we generated stable transgenic zebrafish expressing CaMPARI in all neurons from the elavl3 promoter (16) (Fig. 2A). After exposure of larvae (4 to 5 days after fertilization) to 10 s of PC light (405-nm light-emitting diode array, 400 mW/cm², Fig. 2B), confocal stacks were acquired to generate a cellular-resolution snapshot of the calcium state of all neurons in the zebrafish brain (Fig. 2C and movie S1). Exposure of larvae anesthetized with tricaine methanesulfonate (MS-222) to PC light resulted in a lack of PC throughout the brain (Fig. 2C). Freely swimming fish with no additional stimulation showed a large amount of PC in the forebrain and habenula, but very little in the optic tectum. Additionally, patterns of hindbrain activity consistent with motor output for swimming (17, 18) were often observed. Treatment of larvae with proconvulsive compounds such as 4-aminopyridine (4-AP), exposure to noxious heat or cold stimuli (Fig. 2C), or exposure to turbulent water (fig. S9) during the PC light pulse resulted in qualitatively different patterns of CaMPARI PC throughout the brain, consistent with permanent marking of the subset of neurons activated by these stimuli. CaMPARI signals following each stimulus were consistent between fish (figs. S9 and S10) and showed clear cellular resolution (Fig. 2C, bottom panels).

Next, we tracked activity in a freely moving organism with CaMPARI expression isolated to a genetically specified subset of neurons. We expressed CaMPARI in peripheral sensory neurons (PSNs) of Drosophila melanogaster larvae using the P0163-Gal4 line (19). In third-instar larvae, CaMPARI fluorescence was clearly visible in PSN axonal projections in the ventral nerve cord (VNC) and could be segmented into proprioceptive, chordotonal, and nociceptive axon terminals (Fig. 2E). To highlight PSN responses to vibrational stimuli (20), we administered a single 5-s pulse of PC light to freely crawling larvae on an agar plate during presentation of a 1000-Hz tone delivered from a speaker below the plate (Fig. 2F). The vibration stimulus produced a twofold increase in green-to-red PC only within the axonal projections of chordotonal neurons, not the axons of proprioceptive or nociceptive neurons (Fig. 2G). The extent of CaMPARI PC within chordotonal axons increased in a dose-dependent manner with increasing amplitude of the vibration stimulus, with more sensitive responses in lateral compared with ventromedial axon terminals (Fig. 2H).

Within layer 2/3 of mouse V1, neurons responsive to different directions and orientations of moving bar gratings are interspersed throughout the tissue (21). CaMPARI was delivered to layer 2/3 pyramidal cells with an adeno-associated virus 1 (AAV1)-synapsin1 virus. Expression was bright and excluded from the nucleus (Fig. 3B). First, in vivo calcium imaging was performed through a cranial window by imaging two-photon excited CaMPARI fluorescence as moving gratings were presented to the contralateral eye (16) (Fig. 3A). This allowed generation of orientation-tuning maps of segmented cell bodies. After identification of cells responsive to specific orientations (Fig. 3C), the visual stimulus (a single direction— “northwest,” preferred by cell 1 but not cell 2 in Fig. 3B) was replayed, and PC light was delivered through the cranial window in 500-ms pulses. After 20 PC pulses spaced 12 s apart, noticeable red fluorescence was present in a subset of cells (cell 1, but not cell 2, in Fig. 3B and movie S2). Neurons responsive to the “northwest” moving grating direction (like cell 1 in Fig. 3) exhibited a red/green fluorescence ratio significantly higher than cells not responsive to that direction (like cell 2 in Fig. 3, fig. S11). We relocated the in vivo two-photon fields of view within fixed, sectioned tissue using a confocal microscope (Fig. 3, B and E, and fig. S12). The relative red/green ratio of cells was maintained after fixation (fig. S13), and we were additionally able to image the CaMPARI signal over a much larger volume (Fig. 3F) than would have been easily accessible through the cranial window in vivo. When perfusion and fixation occurred 24 hours after PC, marked cells could still be easily distinguished in fixed sections with confocal microscopy (fig. S14).

Genetic tools in Drosophila do not include methods for trans-synaptic circuit tracing and whole-brain functional mapping. Although photoactivatable GFP (22) and optogenetics in combination with calcium imaging (23) have been employed to detect potential neuronal connections, these methods require knowledge of presynaptic neurons and operate on a limited field of view. We set out to test whether CaMPARI could provide a more comprehensive solution. We affixed intact flies with pan-neuronal expression of low-affinity CaMPARI (V398D) to a physiology holder via the head cuticle and delivered PC light through a water-dipping objective while stimulating the animals with a panel of odors (Fig. 4A and fig. S15): (i) 3-octanol (3-Oct), which activates multiple glomeruli in the antennal lobe [AL; see (16) for a list of anatomical abbreviations used] (24); (ii) geosmin (Geo), an odorant that specifically activates the DA2 glomerulus (25); and (iii) phenylacetic acid (PAA), which strongly activates ionotropic receptor neurons projecting to the VL2a glomerulus (26). CaMPARI mapping confirmed the known neural representations of these odors in the antennal lobe (Fig. 4C, figs. S16 and S17, and movies S3 to S8). We also found additional, previously undescribed representations of some odors (Fig. 4, D and E). In the previous Geo study (25), a specific driver, GH146-Gal4, was used to express GCaMP3 in a limited subset of cells, facilitating segmentation of live-cell imaging data at the cost of ensemble undersampling. We similarly saw only the DA2 glomerulus activated when we drove CaMPARI expression with GH146-Gal4 (fig. S18), but additionally observed the DC1, VA1d, VA1v, and VM7 glomeruli (Fig. 4, D and E) with pan-neuronal CaMPARI. In the case of PAA, the original study (26) assessed odor responses using electrophysiology from specific Ir84a receptor neurons projecting to a single glomerulus (VL2a), rather than sampling the entire AL. When pan-neuronal CaMPARI was used, VL2a was the most salient responder, but weaker PC of VA1v and VM7 was also seen. Increasing the stimulation and PC time increased the overall red/green ratio without altering the relative amount of signal in different glomeruli (Fig. 4F).

Olfactory receptor neuron (ORN) responses likely contribute substantially to the activation patterns we observe in the AL with the broad R57C10-Gal4 driver. Although local neurons (LNs) likely contribute to AL labeling as well, we found evidence of projection neuron (PN) activation (fig. S19). The activation of these secondary neurons was clearer when we used the PN-specific GH146-Gal4 (figs. S18, S20, and S21). With GH146-Gal4, we also found photoconverted PN axons projecting to the mushroom body (MB) calyx (CA) and the lateral horn (LH) (figs. S20 and S21 and movie S6), confirming that CaMPARI enables trans-synaptic circuit mapping. The labeled axons in LH arborize primarily ventrally (fig. S21), consistent with previous reports (26). Thus, pan-neuronal expression enables large-scale unbiased ensemble mapping across the brain. Coregistration of active cells to standardized atlases or the use of specific drivers to identify activated neurons (27) could be a powerful strategy for mapping pathways involved in specific behaviors.

We reasoned that sensory-driven circuit mapping with CaMPARI could also be complemented by the use of optogenetic stimulation to activate circuits downstream of an arbitrary cellular point of entry. We expressed the red light–sensitive channelrhodopsin CsChrimson (28) in Ir84a receptor neurons (fig. S22) using a LexA driver while expressing CaMPARI pan-neuronally with a Gal4 driver. We then mapped active circuits throughout the brain during optogenetic stimulation of this single-receptor neuron type (Fig. 4G). In addition to the glomeruli in the antennal lobe, where axons of ORNs and dendrites of PNs synapse (Fig. 4H), we observed axons of PNs projecting to the posterior areas of the brain (Fig. 4, I to L). One axon commissure, the medial antennal lobe tract (mALT), targets both CA and LH (Fig. 4, I and K), whereas another commissure, the medial-lateral antennal lobe tract (mlALT), targets LH directly (Fig. 4, I to L). As with PAA odor-driven activation, photoconverted PN axons mostly target the ventral part of LH. In addition, we detected activity in the MB vertical lobe (VL, Fig. 4, J and L) and VLP (Fig. 4, I and K), which are putative postsynaptic partners of PNs (Fig. 4J and fig. S23). More specifically, the α′ and β′ MB lobes were more photoconverted than α and β (fig. S24, E, F, G, and I compared to D), consistent with previous reports (29, 30). Using the higher-affinity CaMPARI, we attempted to mark circuit components further downstream (Fig. 4, K and L, and fig. S25). Although we found similar patterns of active PNs and VLP neurons as before (Fig. 4K), we also found other putative downstream circuit components such as LH neurons (Fig. 4L, cyan arrows), mushroom body output neurons (Fig. 4, J and L, and fig. S24), and descending neuron projections to the ventral nerve cord (fig. S25). Activation of deeper layers of the circuit suggests that CsChrimson may be able to drive the firing rates (and thus Ca²⁺ levels) of sensory neurons and downstream neurons more potently than natural stimuli.

In summary, CaMPARI was validated in four preparations in three organisms. Recording neural circuit activity free from restraint creates opportunities for studying complex behaviors such as social interaction, courtship, and so forth. In the case of direction-selective L2/3 pyramidal cells in mouse V1, CaMPARI labeling is consistent with results obtained from imaging of GECIs (12) and small-molecule dyes (31), as well as electrical recordings (21). In freely moving Drosophila larvae, vibration-evoked signals were restricted to chordotonal, but not proprioceptive or nociceptive, sensory neurons; the results show a functional difference in sensitivity of two distinct subtypes of chordotonal axons to vibration. In adult Drosophila brains, PC during both naturalistic and optogenetic stimulation of olfactory sensory pathways labeled activity in the expected OSN/PN circuit and, further, labeled PN projections to the LH and MB, and potential quaternary descending neurons in the VNC. In freely swimming zebrafish, activation patterns are consistent with hindbrain premotor and spinal cord motor output, as well as sensory modality-specific responses, which will be followed up experimentally. In all in vivo test cases, CaMPARI validates known results and offers extensions not otherwise possible: free movement, whole-brain activity mapping, a permanent signal compatible with fixation, and follow-up experiments such as targeted electrical recordings, multichannel immunohistochemistry (allowing cell type identification), and isolation and genetic profiling of labeled cells. These properties facilitate “forward functional neuromics,” in which active neural populations are labeled during behavior, drug treatment, sensory stimulation, and so forth in an unbiased fashion, with subsequent validation and characterization of labeled cells.

CaMPARI represents a new class of genetically encoded indicator that enables light-gated integration of Ca²⁺ flux in vivo with a time resolution of seconds to hours, specified by the user. Converted signal is persistent, amenable to multitrial accumulation, and compatible with fixation for post hoc whole-brain (or targeted-area) activity reconstruction. CaMPARI signal is ratiometric, correcting for expression-level differences. The protein engineering principles applied here could be extended to allow permanent marking of cell states or analytes other than Ca²⁺ by fusion of circularly permuted Eos variants to various ligand-binding or sensor domains, similar to previous GFP-based fluorescence intensity sensors (32–34).