Main

Neural circuits underlying sleep1, interoception2, tactile sensory processing3, sexual behavior4, digestion5 and parenting6 have been studied by identifying critical cellular components using activity-dependent reporters and activating those cells optogenetically or through sensory stimulation. However, fully defining the anatomy and function of these circuits requires the identification of cells upstream and downstream of starter cells. This task is difficult due to the heterogeneity of cell types present in relatively small volumes of neural tissue7. Thus, classical tracing methods using proteins such as wheat germ agglutinin and horse radish peroxidase8, which are taken up by most neurons, cannot be used to define most neuronal circuits precisely.

A genetically modified form of rabies virus was developed that specifically infects neurons expressing the Avian sarcoma leucosis virus receptor, TVA, enabling tracing from specific cell types9. Neural circuits carry information from presynaptic to postsynaptic cells, or anterogradely, but modified rabies virus infects neurons that are presynaptic to starter neurons, traveling retrogradely10. In addition, it is engineered to jump a single synapse, allowing for precise mapping11,12. Although modified rabies virus has been broadly used for tracing circuits retrogradely13, there is currently no comparable method for tracing circuits from genetically determined cells anterogradely. In particular, although two viruses have recently been used for anterograde tracing from nonspecific cells, an adeno-associated virus (AAV1) and a genetically modified yellow fever virus (YFV), neither can mediate tracing from genetically determined cells14,15,16.

Here, we introduce ATLAS (anterograde transsynaptic label based on antibody-like sensors), a rationally designed protein that mediates anterograde transsynaptic labeling from genetically determined cells. ATLAS can be expressed specifically in Cre-expressing neurons in transgenic animals and label neurons to which they project. Labeling is strictly anterograde, monosynaptic and activity-dependent in vivo. Furthermore, we provide direct evidence that transneuronal labeling mediated by ATLAS is exclusively transsynaptic. Finally, we show that components of ATLAS can be independently replaced, demonstrating that it is a versatile platform that can be upgraded or modified to perform specific tasks.

Results

Generating AF, a binder to the N terminus of GluA1

We used messenger RNA (mRNA) display, an in vitro selection procedure17, to identify binders to amino acids 1–394 of Rat GluA1 from a library of mRNA–protein fusions encoding the 10FNIII domain of human FN1, with 17 random residues engineered into the BC and FG loops18,19. We expressed the selected fibronectin clones in COS7 cells with GluA1 fused to a hemagglutinin tag (GluA1-HA) and Stargazin-V5. Nonpermeabilized labeling of MYC showed that four fibronectin (Fn) clones labeled the surface of COS7 cells colabeled for HA (Extended Data Fig. 1). Following expression in dissociated cultures of cortical neurons, clone Fn9.3 gave the most robust labeling (Extended Data Fig. 1). We refer to this clone as AMPA.FingR (AF; FingR stands for fibronectin intrabody generated with mRNA display18).

We generated AF-HA protein and found that it bound specifically to a band colabeled with a GluA1 antibody on a western blot of neuronal lysate (Fig. 1a,b). To characterize AF further, we coexpressed a secreted version of it, ss-AF-YFP (ss refers to the GluA1 signal sequence), in dissociated cortical neurons with PSD-95.FingR-tagRFP, which labels postsynaptic sites18. We found that AF labeling partially overlapped with PSD-95.FingR-tagRFP, suggesting a perisynaptic localization (Fig. 1c and Extended Data Fig. 1). AF also colocalized with endogenous clathrin, consistent with localization to endocytic zones (Fig. 1d). We then tested whether AF-HA protein is endocytosed by adding it to the medium of cortical neurons in culture. We found that virtually all neurons displayed HA labeling (37 of 38, n = 5 independent cultures, Fig. 1e) within 1 h. Furthermore, the addition of 10 mg ml−1 chlorpromazine, a clathrin inhibitor, significantly reduced HA labeling (4.2 ± 0.8 versus 2.1 ± 0.5, P = 0.012, n = 5 cultures, Mann–Whitney U-test, Fig. 1f), confirming that internalization was mediated by endocytosis. All values are expressed as mean ± s.e.m.

Fig. 1: Mechanisms of release and uptake by ATLAS.
figure 1

a, Western blot of lysate from cortical cultures. IB, immunoblot. b, DNA constructs for testing of ATLAS. Arrows denote promoters. c, Live cultured neuron expressing ss-AF-YFP (green) and PSD-95.FingR-tagRFP (magenta). Scale bar, 5 μm. d, Cultured neurons expressing ss-AF-YFP fixed and colabeled for endogenous clathrin. Scale bar, 5 μm. e, Neurons exposed to AF-HA protein (green) for 1 h and colabeled for endogenous MAP2 (magenta). Scale bar 25 μm. f, Same as e, but with the addition of 10 mg ml−1 chlorpromazine to the culture medium. Scale bar, 25 μm. g, Release of SEP from presynaptic sites in cultured cortical neurons expressing VAMP2-At-BACEcs-SEP. PSD-95.FingR-tagRFP (magenta), SEP (green). Scale bar, 1 μm. h, Same as g, but with VAMP2-At-BACEcs-AF-SEP. Scale bar, 1 μm. i, τ of VAMP2-At-BACEcs-AF-SEP is significantly increased versus VAMP2-At-BACEcs-SEP (n = 5, unpaired two-tailed t-test with Welch’s correction). j, Adding 100 mM KCl to the culture medium causes a significant increase in the frequency of release events of VAMP2-At-BACEcs-SEP (n = 5, paired two-tailed t-test). k,l, Graphs of the differential (DF) of SEP fluorescence for VAMP2-At-BACEcs-AF-SEP with control medium (k) versus with medium containing 100 mM KCl (l). Each spike denotes a single release event and does not reflect the duration of visible SEP fluorescence. m, Paired patch clamp recording of CA1 neurons in hippocampal slices (np). EC, entorhinal cortex. n, Biolistic transfection of CA1 pyramidal neurons with ss-AF-YFP did not significantly affect AMPAR-eEPSC amplitude (n = 10 pairs, P = 0.31, paired, two-tailed t-test). Scatterplots show eEPSC amplitudes for pairs of neighboring untransfected and transfected cells (open circles) with corresponding mean ± s.e.m. (filled circles). Insets show representative current traces from control and transfected neurons with stimulation artifacts subtracted. o, Same as n for NMDAR-eEPSC amplitude (n = 8 pairs, P = 0.70, paired, two-tailed t-test). p, Right, bar graph shows mean ± s.e.m. amplitude ratio of second AMPAR-eEPSC versus first (n = 6, P = 0.19, paired, two-tailed t-test). Left, representative current traces from control (gray) and transfected (orange) without stimulation artifacts. Scale bar, 50 ms. NS, not significant.

Source data

Since AF binds to GluA1 receptors and is endocytosed postsynaptically, we reasoned that we could make a transsynaptic tracer if we targeted AF to synaptic vesicles by fusing it to the luminal domain of the presynaptic protein, synaptobrevin 2 (VAMP2) (ref. 20). We inserted a BACE1 (β-secretase) cleavage site (BACEcs)21 between AF and VAMP2 (to give VAMP2-BACEcs-AF), enabling endogenous BACE to cleave AF from VAMP2, leaving it free to exit the lumen of the synaptic vesicle on exocytosis. In addition, an ALFA-tag (At) was added to label VAMP2 (ref. 22). To test our strategy, we substituted super ecliptic pHluorin (SEP)20 for AF and coexpressed the resulting plasmid (VAMP2-At-BACEcs-SEP, Fig. 1b) with PSD-95.FingR-tagRFP in cortical neurons in culture. The release from acidic synaptic vesicles into the neutral pH of the synaptic cleft is expected to cause a sudden increase in the fluorescence of SEP, allowing it to be visualized20. Following overnight expression of VAMP2-At-BACEcs-SEP in neuronal cultures, we imaged SEP using widefield epifluorescence microscopy. We saw flashes of fluorescence lasting approximately 2–8 s (images taken at 1 Hz) appearing at points opposite postsynaptic sites (Fig. 1g). This result is consistent with VAMP2 delivering SEP to synaptic vesicles, BACE cleaving off SEP, and SEP being released into the synaptic cleft. We also fused AF to SEP and expressed the resulting protein (VAMP2-At-BACEcs-AF-SEP, Fig. 1b) in culture and observed flashes that were brighter and decayed with a longer time constant (6 ± 2 versus 405 ± 60 s, n = 5 independent cultures, Fig. 1h,i), a significant difference (P < 0.0001, t-test). This result is consistent with AF binding to the postsynaptic membrane following release and remaining until it is endocytosed. To further test whether AF-SEP is released from synaptic vesicles, we examined whether depolarization increases the number of release events. We imaged the same cultures before and after adding 100 mM KCl and counted the number of bright flashes corresponding to release events. We found a roughly 15-fold increase in release events with KCl (50 ± 6 versus 3 ± 1 for control, P = 0.02, t-test, n = 5 independent cultures; Fig. 1j–l). Together, these results are consistent with AF-SEP being released from synaptic vesicles, moving across the synaptic cleft, binding to the adjacent postsynaptic neuron perisynaptically and then being endocytosed.

Given that AF likely binds to the extracellular N-terminal domain of GluA1, we tested whether the expression of ss-AF-YFP affects synaptic transmission to hippocampal CA1 neurons. We found that in cells in hippocampal slices transfected with ss-AF-YFP using biolistics, the amplitudes of evoked excitatory postsynaptic currents (eEPSCs) of AMPA (n = 10, P = 0.31, t-test), NMDA receptors (n = 8, P = 0.70, t-test) and paired-pulse facilitation (n = 6, P = 0.19, t-test) were not significantly different from control cells, consistent with the expression of AF not affecting synaptic transmission (Fig. 1m–p). All pairs of recordings were made from distinct cultured slices and, thus, each is a biological replicate.

Transsynaptic tracing in vitro and in vivo

To generate a protein capable of mediating transsynaptic tracing, we replaced SEP with Cre in VAMP2-At-BACEcs-AF-SEP to give VAMP2-At-BACEcs-AF-Cre (ATLASCre, Fig. 1b). Following expression of ATLASCre, AF-Cre should be released into the synaptic cleft and then bind to AMPA receptors at the postsynaptic membrane. Subsequently, it needs to be endocytosed and carried to the nucleus to provide genetic access to the postsynaptic cell (Fig. 2a). To facilitate this final step, we added the DNA binding domain from CCR5 (refs. 18,23) to the ATLAS protein, as it contains a robust nuclear localization signal. To test our strategy, we transfected ATLASCre into dissociated cultures infected with AAV encoding a floxed green fluorescent protein (GFP) reporter (AAV8-DIO-GFP). We observed cells with GFP and Cre that were not labeled with the starter cell marker, At, consistent with the transneuronal transport of Cre and recombination of the floxed allele, implying transport of Cre to the nucleus (Fig. 2b–e and Extended Data Fig. 2). Furthermore, At-labeled presynaptic structures were found to be overlapping or immediately adjacent to dendrites belonging to each of the GFP-labeled neurons (Extended Data Fig. 2). Thus, our results are consistent with ATLASCre mediating transsynaptic tracing and providing genetic access to postsynaptic cells.

Fig. 2: Schematic of ATLAS transsynaptic transport and testing in vitro.
figure 2

a, ATLASCre protein (VAMP2-At-BACEcs-AF-Cre) is expressed in a presynaptic starter cell and (1) targeted to synaptic vesicles by a VAMP2 domain. (2) A BACE cleavage site within the ATLAS protein is cleaved before release. (3) A peptide domain consisting of AF fused to Cre is released into the synaptic cleft and (4) subsequently binds to the extracellular domain of postsynaptic GluA1. (5) It is internalized through endocytosis and transported to the endosome. (6) AF-Cre moves to the nucleus, where Cre can catalyze the expression of a floxed reporter gene. b, Cortical neuron in culture expressing ATLASCre protein (arrow) following transfection, stained with anti-ALFA-tag (At) nanobody. Four other neurons shown (arrowheads) do not express At. c, Cre staining of the same neurons as in b including a neuron expressing ATLASCre (arrow), and four not expressing ATLASCre (arrowheads). d, GFP labeling of postsynaptic cells (arrowheads, green) following infection of AAV8-DIO-GFP. e, Combined labeling of At, GFP and Cre showing presynaptic (arrow) and postsynaptic neurons (arrowheads). Scale bars, 100 μm.

To test our strategy in vivo, we injected AAV8-ATLAScre into the medial prefrontal cortex (mPFC) and AAV8-DIO-mCherry into the striatum (Str) of an adult mouse (Fig. 3a). After 1 week, we found At staining in mPFC and abundant mCherry staining in Str (Fig. 3b,c). In contrast, there was negligible mCherry staining in a control experiment without AAV8-ATLAScre (Fig. 3d). Because the projection from mPFC to Str is strictly anterograde24, this result is consistent with AAV8-ATLAScre mediating the transsynaptic transport of Cre into the nucleus of the postsynaptic cell, causing mCherry labeling.

Fig. 3: ATLAS mediates transsynaptic tracing in vivo.
figure 3

a, Schematic depicting injections of AAV8-ATLASCre into the mPFC (cyan) and AAV8-DIO-mCherry into the Str (red) to trace the unidirectional projection from the mPFC to Str. b, ALFA-tag (At) staining of ATLASCre in presynaptic cells in mPFC in a wild-type (WT) mouse. c, mCherry staining of postsynaptic cells in Str in the same brain as b. d, Control brain injected only with AAV8-DIO-mCherry in Str. e, Schematic depicting injections of AAV8-DIO-ATLASFLP and Lenti-CamkII-Cre into the primary visual cortex (V1, cyan) and AAV8-fDIO-mCherry into the SC (red) to trace the unidirectional projection from V1 to SC. f, At staining of ATLASFLP in presynaptic cells in V1 in a wild-type mouse. g, mCherry staining of postsynaptic cells in SC in the same brain as f. h, Control brain injected only with AAV8-fDIO-mCherry in SC. i, Schematic depicting injections of AAV8-DIO-ATLASFLP into the dLGN (cyan) and AAV8-fDIO-mCherry into V1 (red) to trace the projection from dLGN to V1 in a VIPR2-Cre mouse. j, At staining of ATLASFLP in presynaptic cells in dLGN. k, mCherry staining of postsynaptic cells in the V1 in the same brain as j. l, Control brain injected with only AAV8-fDIO-mCherry in V1. Scale bars, 100 μm.

Tracing from genetically determined cells

To mediate genetically determined tracing, we generated AAV8-DIO-ATLASFLP (Fig. 1b), where expression of ATLAS is Cre-dependent and the payload is FLP (Flip recombinase). To test this construct, we traced the projection from the primary visual cortex (V1) to the superior colliculus (SC), which is strictly anterograde25. We coinjected Lenti-CamkII-Cre (a lentivirus expressing Cre specifically in excitatory neurons) and AAV8-DIO-ATLASFLP into V1 and AAV8-fDIO-mCherry into SC of wild-type mice. In addition to At staining in V1, we saw abundant staining for mCherry in SC (391 ± 37 cells, n = 7 samples from three mice, Fig. 3e–h and Supplementary Table 1) and a small number of faint cells on the control (35 ± 6, n = 6 samples from three mice), which was subtracted from the previous count. All cell counts are expressed as the number of cells per 0.1 mm unless otherwise specified to allow for comparisons. An advantage of this strategy is that lentivirus has no intrinsic retrograde or anterograde transneuronal transport that could cloud the interpretation of the results, unlike virtually every pseudotype of AAV15.

To test ATLAS tracing in transgenic Cre mouse lines, we examined the projection from the dorsal lateral geniculate nucleus (dLGN) to V1 in a VIPR2-Cre line expressing Cre in neurons in the dLGN26. We injected AAV8-DIO-ATLASFLP into dLGN and AAV8-fDIO-mCherry into V1. Staining of At was found specifically in the dLGN, and mCherry staining in V1 (267 ± 54 cells, n = 5 samples from three mice) was dense in layer IV and less dense in layers II/III and V (Fig. 3i–l), consistent with known anatomy27. These results are consistent with ATLAS tracing anterogradely from Cre-expressing neurons in wild-type and transgenic mice.

Increasing efficiency of transsynaptic labeling with BACE

To measure the efficiency of ATLAS tracing, we injected AAV8-ATLASCre in mPFC and AAV8-DIO-mCherry in Str and counted the number of labeled cells as before after 1 week of incubation. Initially, we observed 144 ± 17 labeled cells (n = 6 samples from three mice). One possible inefficiency in our system is that there may be insufficient endogenous BACE in presynaptic cells to cleave all the AF-Cre from VAMP2. To test this hypothesis, we coexpressed exogenous BACE with ATLASCre and measured the resulting efficiency. When we coinjected AAV8-BACE-HA and AAV8-ATLASCre into mPFC in equal amounts, it did not significantly affect the number of postsynaptic cells labeled with mCherry (180 ± 20 cells, n = 6 samples from three mice, P = 0.38, Tukey’s multiple comparisons, Extended Data Fig. 3). However, when we injected AAV8-BACE-HA and AAV8-ATLASCre in a 5:1 ratio (viral particles), we saw a significant increase in the number of cells in Str labeled with mCherry, compared to both AAV8-ATLAS + BACE 1:1 (P = 0.0027) and AAV-ATLAS alone (286 ± 17, P = 0.0002), n = 6 samples from three mice; Tukey’s multiple comparisons test, Extended Data Fig. 3 and Supplementary Table 1). With this ratio of BACE to ATLAS, we saw approximately 10% of postsynaptic neurons labeled (Extended Data Fig. 3). Thus, when supplemented with exogenous BACE, ATLAS labels cells in the Str at a similar density to YFV14.

Labeling is transsynaptic

Given its design, it is likely that transneuronal labeling mediated by ATLAS occurs exclusively through synapses. To test this hypothesis in vivo, we asked whether transneuronal labeling could happen from inhibitory neurons. Since the uptake of ATLAS is mediated by GluA1 receptors, which are not found at inhibitory synapses, if we expressed ATLAS in inhibitory neurons, any transneuronal labeling would have to be nonsynaptic (Fig. 4a). Accordingly, we injected AAV8-DIO-ATLASFLP and AAV8-fDIO-mCherry into V1 of an SST-Cre mouse, where somatostatin-positive inhibitory interneurons express Cre28. When we examined labeling in the cortex after 2 weeks, we saw 483 cells labeled with mCherry and, of those, only 20 cells were not colabeled with At (Fig. 4b–d and Supplementary Table 1, n = 3 mice). This labeling rate was less than the background rate measured over an equivalent number of sections for control experiments where only AAV8-fDIO-mCherry was injected into V1 (100 cells, n = 3 mice), consistent with no transneuronal labeling above background. This result leads us to conclude that ATLAS labels postsynaptic neurons exclusively through the transsynaptic transfer of proteins.

Fig. 4: ATLAS does not mediate nonsynaptic transneuronal labeling and transsynaptic labeling is specific.
figure 4

a, Schematic of possible nonsynaptic pathways for transneuronal labeling from an SST+ neuron. b, SST-Cre mouse with AAV8-DIO-ATLASFLP and AAV-fDIO-mCherry coinjected into V1. Labeling of inhibitory presynaptic neurons with At. This experiment was performed in two mice and bilaterally in one of the mice, for a total of three biological replicates. c, mCherry-labeled neurons from the same brain as in a. d, Merged mCherry and At labeling shows that all neurons labeled with mCherry are also labeled with At. e, Schematic of configuration for whole-cell patch clamp recording from random pairs of mCherry- or tdTomato-labeled and unlabeled striatal spiny projection neurons following ATLAS-mediated tracing from V1 cortex. EPSCs were generated as a result of optical activation of inputs from the V1 cortex. f, Percentage of cells of different types with nonzero amplitude EPSCs following exposure to blue light: 100% of mCherry-labeled neurons, 70% of paired unlabeled neurons, ~40% of random, unlabeled neurons. g, Comparison of average EPSC trace from all optically evoked EPSCs from mCherry+ and paired mCherry striatal spiny projection neurons. h, EPSC amplitudes of mCherry+ and mCherry neurons (n = 10). In ten out of ten pairs, the amplitude of the EPSC recorded from mCherry-labeled neuron is larger than that of the EPSC recorded from the corresponding unlabeled neuron (Wilcoxon signed-rank test). Box plot, the central mark indicates the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers. Scale bar, 100 μm.

Source data

ATLAS tracing does not block synaptic transmission

To test whether ATLAS disrupts synaptic transmission, we coinfected AAV8-ATLASsnCre (Fig. 1b), AAV8-BACE-HA and AAV8-DIO-Channelrhodopsin 2 (ChR2)-GFP into the mPFC and AAV8-DIO-mCherry in the Str of wild-type mice. Since we found that ATLAS labels ~10% of target neurons in this circuit (Extended Data Fig. 3), we reasoned that it should be possible to find both labeled and nonlabeled cells that are synaptically connected to ChR2-expressing neurons to compare whether ATLAS labeling affects the cell’s physiological properties. Four weeks after the virus injection, we cut coronal slices of Str approximately 350 μm thick and incubated them in artificial cerebral spinal fluid for in vitro slice recording. Whole-cell patch clamp recording was performed to measure 470-nm optically evoked EPSC (eEPSC) from striatal cells (Extended Data Fig. 4). To evaluate the potential effect of ATLAS tracing on synaptic transmission, we compared the striatal cells with and without mCherry expression. Robust evoked EPSCs were recorded in mCherry-expressing cells with an onset latency of <5 ms following the light flash in eight out of eight cells (from three mice), indicating that the presynaptic ChR2-expressing cells were synaptically connected with mCherry-labeled cells (Extended Data Fig. 4). Furthermore, when labeled postsynaptic cells were recorded from, the amplitude of ChR2-evoked EPSCs exhibited a nonsignificant increase compared with those of unlabeled cells (Extended Data Fig. 4, n = 8 labeled cells, n = 7 unlabeled cells from three mice for both, P = 0.28, t-test). Thus, ATLAS did not block synaptic transmission. In addition, there were no significant changes in onset latency (P = 0.22, t-test) and time constant (P = 0.78, t-test, Extended Data Fig. 4). This result is consistent with transsynaptic labeling with ATLAS not causing detectable toxicity. Moreover, we verified that evoked currents are monosynaptic through the sequential application of Tetrodotoxin, which blocks action potentials and polysynaptic transmission, and 4-aminopyridine, which blocks K+ channels and facilitates direct activation of presynaptic terminals.

Electrophysiology confirms specificity of ATLAS tracing

To test for the specificity of ATLAS tracing, we measured optogenetically evoked currents in the circuit from the primary visual cortex to the Str29. We coinfected AAV8-ATLASsnCre (Fig. 1b) and AAV8-BACE-HA in V1, and AAV8-DIO-mCherry/tdTomato in the dorsomedial Str (DMS). Two weeks later, we injected AAV8-DIO-ChR2-GFP in the same area in V1 as the first injection to manipulate ATLAS-labeled presynaptic inputs while minimizing expression competition between ATLAS and ChR2. Histology 2 weeks after the second injection revealed efficient expression of At, HA, ChR2 in V1 and Cre-dependent mCherry/tdTomato in DMS, consistent with the projection pattern of DMS-projecting V1 neurons (Extended Data Fig. 4). To compare V1 innervation between ATLAS-labeled versus unlabeled postsynaptic neurons, we measured optically evoked EPSCs (oEPSCs) following a 2-ms flash of 473-nm blue light in acute brain slices prepared from mice injected with virus 4 weeks earlier (Fig. 4e). We performed sequential, paired whole-cell recordings where we measured oEPSCs of an ATLAS-labeled (mCherry/tdTomato+) and an unlabeled (mCherry/tdTomato) neuron in the same field of view (100 × 100 μm; Fig. 4e–h). Ten pairs of cells (from four mice) were recorded, in which ten out of ten labeled cells carried a current, compared with seven out of ten unlabeled cells (Fig. 4f). To assess the baseline connection probability of the V1-DMS circuit, we randomly patched onto DMS neurons in striatal areas where we saw prominent fiber labeling with GFP. Of the 17 random cells we patched (n = 3 mice), 41% of cells were connected (7 out of 17, Fig. 4f). oEPSCs recorded from labeled cells were significantly larger in amplitude and total charge compared with those of paired unlabeled cells (amplitude −134 ± 38 versus −66 ± 21 pA, P = 0.0020, Wilcoxon signed-rank test, total charge −0.121 ± 0.03 versus −0.038 ± 0.01 pC, P = 0.0020, Wilcoxon signed-rank test, n = 10 mice, plotted as absolute values in Fig. 4g,h). Furthermore, in ten out of ten pairs, the current recorded from the labeled cell had a larger amplitude than that from the unlabeled cell (Fig. 4h). Together, these results indicate that when ATLASCre and a Cre-dependent reporter are simultaneously expressed in different brain regions, cells exhibiting Cre-dependent labeling are likely to receive synaptic inputs from cells expressing ATLASsnCre. Furthermore, they demonstrate that cells with stronger synaptic connections are more likely to be labeled by ATLAS.

ATLAS is exclusively anterograde and monosynaptic

A limitation of many virally based anterograde transneuronal labeling methods is that they also mediate labeling in the retrograde direction14,15. To determine whether ATLAS mediates retrograde labeling in vivo, we injected AAV8-DIO-ATLASFLP and Lenti-hsyn-Cre into Str and AAV8-fDIO-mCherry into mPFC (Fig. 5a). Because the mPFC to Str projection is unidirectional and lentivirus does not spread retrogradely, any labeling in mPFC would be due to retrograde transneuronal labeling mediated by ATLAS. However, despite abundant At labeling in Str, indicating that ATLAS was expressed at a high level, we found no labeling above background in mPFC (20 ± 4 cells for experiment, n = 6 samples from three mice, versus 45 ± 14 for control, n = 6 samples from three mice, P = 0.11, t-test, Fig. 5b–e and Supplementary Table 1). These data are consistent with ATLAS not mediating retrograde labeling.

Fig. 5: Transsynaptic labeling by ATLAS is exclusively anterograde and monosynaptic.
figure 5

a, Schematic of the projection from mPFC to Str showing injections of AAV8-DIO-ATLASFLP and Lenti-hsyn-Cre (Str, cyan), and AAV8-fDIO-mCherry (mPFC, red). b, At labeling of cells in Str. c, mCherry labeling of cells in mPFC in the same brain as b. d, mCherry labeling of cells in mPFC in a control mouse brain with an injection of AAV8-fDIO-mCherry into the mPFC only. e, Number of labeled postsynaptic cells following Str-mPFC labeling is not significantly different from control mPFC (t-test, n = 6, Str-mPFC; n = 6, control). f, Schematic of the disynaptic pathway from V1 to SC to PBG showing injection of ATLASCre (V1) and AAV8-DIO-mCherry (SC and PBG) for fh. g, At labeling of presynaptic cells in V1. h, mCherry labeling of postsynaptic cells in the superficial gray layer of the SC (SCsg) from the same brain as g. i, mCherry labeling in PBG from the same brain as f and g. j, Schematic of projection from SCsg to PBG showing injections of ATLASCre (SCsg) and AAV8-DIO-mCherry (PBG). k, At labeling of presynaptic cells in SCsg. l, mCherry labeling of postsynaptic cells in PBG from the same brain as k. m, mCherry labeling in PBG following injection of AAV8-DIO-mCherry into PBG only in a control mouse brain. n, Number of labeled postsynaptic cells following SC-PBG labeling is significantly more than following V1-PBG labeling (one-way ANOVA). Scale bars, 100 μm.

Source data

To test whether ATLAS can mediate transsynaptic labeling that is monosynaptic, a critical property for transsynaptic tracing results to be interpretable, we tried tracing the polysynaptic pathway from V1 to SC to the parabigeminal nucleus (PBG)14. We injected AAV8-ATLASCre into V1 and AAV8-DIO-mCherry into SC and PBG (Fig. 5f). We found AT labeling in V1 and abundant transsynaptic mCherry labeling in SC (Fig. 5g,h) but no staining significantly above background in PBG (3 ± 1 cells for experiment, n = 8 samples from three mice, versus 19 ± 3 for control, n = 6 samples from three mice, P = 0.35, one-way analysis of variance (ANOVA) (Tukey’s test, Fig. 5i,n). Furthermore, when we injected AAV8-ATLASCre into SC, we saw abundant labeling in PBG (181 ± 30 cells, n = 3 samples from three mice) that was significantly above both background and labeling from V1 to PBG (P < 0.0001 for both, one-way ANOVA, Tukey’s multiple comparisons, Fig. 5j–n), demonstrating that ATLAS is capable of tracing the SC to PBG pathway. Thus, transsynaptic labeling mediated by ATLAS is exclusively anterograde and monosynaptic.

Activity dependence of ATLAS

Because ATLAS mediates transsynaptic labeling by releasing AF-recombinase from synaptic vesicles, transsynaptic labeling is likely activity-dependent. To test for activity dependence of transsynaptic labeling in vivo, we injected AAV8-ATLASsnCre + AAV8-hHM3Dq-mCherry, an excitatory Dreadd30, into mPFC and AAV8-DIO-GFP into Str (Extended Data Fig. 5) in two mice. We then injected one mouse daily with clozapine N-oxide (CNO) intraperitoneally but not the other. After 1 week, we perfused the mice and performed immunocytochemisty. We found abundant c-FOS labeling in the mPFC of the CNO+ mouse, which overlapped with At labeling (Extended Data Fig. 5 and Supplementary Table 1). In contrast, there was negligible c-FOS labeling in the mPFC of the CNO mouse (Extended Data Fig. 5 and Supplementary Table 1). In addition, GFP labeling appeared to increase in terms of total fluorescence and number of labeled cells in the CNO+ mouse (Extended Data Fig. 5). We then pooled these data with those from similar experiments using AAV8-ATLASCre and found that CNO+ mice had significantly higher postsynaptic GFP labeling in the Str versus CNO- mice both in terms of the number of cells labeled (88 ± 12 versus 52 ± 7, n = 5 mice, P = 0.0357, t-test, Extended Data Fig. 5) and total fluorescence (38 ± 4 versus 19 ± 2 a.u., P = 0.0025, t-test, Extended Data Fig. 5). In contrast, the number of presynaptic At stained cells was not significantly different between CNO+ and CNO mice (1,200 ± 100 versus 1,200 ± 100, P = 0.7492, t-test, Extended Data Fig. 5). Thus, our results indicate that the efficiency of transsynaptic labeling was significantly higher in CNO+ mice versus CNO mice, consistent with ATLAS mediating activity-dependent transsynaptic labeling.

An alternate synaptic vesicle targeting domain

ATLAS is composed of multiple independent components, each of which can, in principle, be modified or upgraded to improve efficiency or alter function. We took advantage of this property to optimize the VAMP2 component of ATLAS, which is responsible for presynaptic targeting. Although VAMP2 is widely used as a presynaptic marker, it can produce toxicity when expressed at a high level. Indeed, we saw evidence of toxicity when ATLAS was expressed in the mPFC for 2 and 4 weeks (Extended Data Fig. 6). Accordingly, we exchanged the cytoplasmic domain of VAMP2 for a nanobody against Synaptotagmin 1 (SYTnb)31 to give SYTnb-VAMP2Δ-At-BACEcs-AF-Cre (ATLASsnCre, Figs. 1b and 6a), where VAMP2Δ consists of the lumenal and transmembrane domains of VAMP2. When we expressed ATLASsnCre in the mPFC for 2 or 4 weeks, we saw no evidence of toxicity (Extended Data Fig. 6). To test ATLASsnCre, we transfected it into dissociated cultures of cortical neurons while infecting them with AAV8-DIO-mCherry. We saw abundant cells labeled with mCherry but not with At, indicating they were labeled transsynaptically (Fig. 6b). A closeup image of such a postsynaptic cell revealed At labeling opposite its dendritic spines, consistent with presynaptic labeling (Fig. 6b, inset).

Fig. 6: Transsynaptic tracing with ATLASsn.
figure 6

a, Schematic of ATLASsn. (1) Synaptotagmin nanobody (SYTnb) fused to the transmembrane domain of VAMP2 (VAMP2Δ) targets ATLAS to synaptic vesicles. (2) AF-Cre is cleaved from VAMP2Δ. (3) The synaptic vesicle fuses with the plasma membrane at the presynaptic terminal. (4) AF-Cre is released into the synaptic cleft. (5) SYTnb-VAMP2Δ-At is recycled back into the terminal, labeling the presynaptic site. b, ATLAS tracing in dissociated cortical culture transfected with AAV8-ATLASsnCre (labeled with At, cyan) and AAV8-BACE-HA and infected with AAV8-DIO-mCherry. Scale bar, 10 μm. c, mPFC of the mouse injected with AAV8-ATLASsnCre (cyan) and incubated for 2 weeks. This experiment was performed in three mice. d, Str of the same mouse as in c, injected with AAV8-DIO-mCherry (red). e, Schematic showing projections from the M1 to Str and thalamus (Thal). This experiment was performed in three mice. f, M1 of a mouse coinjected with AAV8-BACE-GFP (green) and AAV8-ATLASsnCre. g, Str of the same mouse as in f injected with AAV8-DIO-mCherry showing mCherry-labeled neurons (red). This experiment was performed in three mice. h, Thalamus of the same mouse as in f and g injected with AAV8-DIO-mCherry showing mCherry-labeled neurons (red). i, Schematic showing pathway from the ventral CA1 to the nucleus accumbens (ACB) and then to the LHA in the rat. This experiment was performed in three rats. j, Ventral CA1 region of a rat injected with AAV8-ATLASsnCre showing cell labeled with At (cyan). k, ACB of the same rat as in j injected with AAV8-DIO-tdTomato, showing tdTomato-labeled neurons (red): aco refers to the anterior commissure. l, LHA of the same rat as in j and k showing nerve terminals labeled with tdTomato (red). Scale bars 100 μm in c,d,f,g,h, 500 μm in j; 200 μm in k,l.

To test AAV8-ATLASsnCre in vivo, we injected it and AAV8-BACE-HA into the mPFC and AAV8-DIO-mCherry into the Str. After 2 weeks of incubation, we saw expression of At in the mPFC and abundant mCherry labeling in Str (Fig. 6c,d and Extended Data Fig. 7). The dorsal region of postsynaptic mCherry labeling colocalized almost perfectly with At labeling representing presynaptic termini (Fig. 6d and Extended Data Fig. 7), consistent with ATLASsn mediating transsynaptic labeling. To further test ATLASsn in vivo, we coinfected AAV8-ATLASsnCre and AAV8-BACE-GFP in M1 cortex and AAV8-DIO-mCherry in the Str and thalamus (Thal, Fig. 6e). After 2 weeks of incubation, staining of GFP and mCherry revealed labeled presynaptic cells in M1 (Fig. 6f) and postsynaptic cells in Str (Fig. 6g) and Thal (Fig. 6h). Overlap of GFP and mCherry labeling in Str and Thal (Extended Data Fig. 7) is consistent with transsynaptic labeling. Furthermore, minimal background was observed in the Str when AAV8-DIO-mCherry alone was injected (Extended Data Fig. 7).

Labeling of cortical to subcortical circuits

To further test ATLASsn, we used it to trace three additional pathways: from the auditory cortex (A1) to the inferior colliculus, the mPFC to the claustrum and the entorhinal cortex to the hippocampus, respectively. In each case, we injected AAV8-ATLASsnCre and AAV8-BACE-HA into the presynaptic area and AAV8-DIO-GFP into the corresponding postsynaptic area. After 2 weeks, we stained presynaptic cells for At and postsynaptic cells for GFP. In each experiment, we found cells with At and GFP labeling, consistent with efficient transsynaptic labeling (inferior colliculus, 173 ± 13 cells, n = 8 samples from three mice; claustrum, 71 ± 14 cells, n = 7 samples from three mice; hippocampus, 132 ± 30 cells, n = 4 samples from two mice, Extended Data Fig. 8 and Supplementary Table 1). Thus, ATLAS is capable of tracing multiple different pathways.

In addition, to understand the time course over which transsynaptic labeling happens, we coexpressed AAV8-ATLASsnCre and AAV8-BACE-HA in the mPFC and AAV8-DIO-GFP in the Str and waited 4, 7, 14 and 21 days (n = 8 samples from four mice in each case) before perfusing animals and staining for presynaptic and postsynaptic labels. After 4 days, we found faint labeling of presynaptic cells with At, although labeled cells were relatively abundant (534 ± 114 cells, Extended Data Fig. 9). In contrast, GFP-labeled postsynaptic cells were very sparse at this time point (2 ± 1 cells), and total GFP labeling was minimal (3 ± 1 a.u.). After 7 days, the number of presynaptic cells labeled with At increased (1,150 ± 300, Extended Data Fig. 9), but not significantly (P = 0.3279, one-way ANOVA). However, both the number of labeled postsynaptic cells (52 ± 7 cells) and the total GFP label (35 ± 3 a.u.) were significantly higher (P < 0.0001, ANOVA; Extended Data Fig. 9). Thus, postsynaptically labeled cells become visible between 4 and 7 days postinjection. After 14 and 21 days, the number of labeled presynaptic cells (1,270 ± 260 and 1,070 ± 270 cells) was not significantly different from 7 days (P = 0.9874, P = 9943, ANOVA; Extended Data Fig. 9). The number of postsynaptic cells labeled with GFP at 14 days (63 ± 5 cells) was not significantly different from the number at 7 days (P = 0.4514, ANOVA; Extended Data Fig. 9). However, the cell number at 21 days (86 ± 5 cells) was significantly higher than at 14 days (P = 0.0116, ANOVA). The labeling intensity increased significantly at 14 and 21 days versus 7 days (65 ± 2 and 82 ± 6 a.u., P < 0.0001, ANOVA; Extended Data Fig. 9). Thus, ATLAS transsynaptic labeling increases notably in the first 7 days and at a slower rate thereafter in terms of cell number, but increases in a relatively linear manner in total intensity. In contrast, labeling of presynaptic cells is constant after the first week.

ATLAS mediates transsynaptic labeling in the rat

It is highly desirable for transsynaptic labeling techniques to work in systems other than the mouse. Accordingly, we tested whether ATLAS would work in the rat by injecting AAV8-ATLASsnCre and AAV8-BACE-HA into the CA1 region of the ventral hippocampus and AAV8-DIO-tdTomato into the nucleus accumbens (ACB, Fig. 6i), to which the CA1 neurons project. We observed At staining in CA1 (Fig. 6j) and abundant tdTomato-labeled cells in ACB (115 ± 11, n = 3 rats, Fig. 6k), consistent with transsynaptic labeling. Finally, tdTomato labeling of axons within the lateral hypothalamic area (LHA) (Fig. 6l), which receives axonal projections from ACB32, confirms the postsynaptic labeling of neurons within ACB.

Discussion

ATLAS is a rationally designed protein that mediates transsynaptic tracing from genetically determined neurons via a precisely defined mechanism. In contrast, viruses such as rabies, YFV, AAVs and the protein used in the nonviral method Trans-Seq have modes of egress from starter cells and entry into recipient cells that remain mysterious10,14,15,33. In addition, AAV1, YFV, vesicular stomatitis virus (VSV) and herpes simplex virus-1 (HSV1) undergo a measurable component of retrograde transport in addition to anterograde and have not been shown to mediate tracing that is exclusively transsynaptic14,15,16,34,35.

Because ATLAS protein does not have an intrinsic amplification mechanism like rabies and similar viruses9, tracing using ATLAS requires the expression of a recombinase-dependent reporter in the postsynaptic cell. While this could be interpreted as a drawback, it is also a benefit because amplification that occurs in rabies or VSV infection is associated with toxicity36,37. In the case of tracing using AAV1, which also requires amplification, a transgenic mouse with a floxed reporter transgene can be used instead of postsynaptic injections15,16. Efficient and consistent transsynaptic labeling occurred in some but not all cases when we expressed ATLASCre in the ai75 floxed reporter mice38. Thus, broad regions containing postsynaptic target areas must be identified a priori and injected with a floxed reporter, which is not required for many virus-based transneuronal labeling techniques. However, this should not be a difficult problem to surmount as regions postsynaptic to specific cell types are easily identified by exogenously expressing a presynaptic marker such as VAMP2 (ref. 39).

Our finding that 100% of postsynaptic striatal cells labeled following expression of ATLAS in V1 were synaptically connected to cells expressing ChR2-GFP, which was coinfected with ATLAS, indicates that postsynaptic cells with Cre-dependent labeling are likely to be synaptically connected with presynaptic starter cells in which ATLASCre is expressed. Also, 70% of neurons not labeled with ATLAS but immediately adjacent to labeled cells were found to be connected to presynaptic cells expressing ChR2-GFP. This result could be due to many factors, including the fact that transsynaptic labeling with ATLAS is not 100% efficient. In addition, Cre-dependent expression of ChR2-GFP could occur in neurons in V1 that are connected with starter cells through local connections, leading to more widespread expression of ChR2-GFP than of ATLAS. In contrast to paired unlabeled neurons, ~40% of random unlabeled cells chosen near labeled presynaptic terminals from V1 were found to be connected to those terminals. These data are consistent with postsynaptic neurons labeled with ATLAS being connected to labeled presynaptic neurons at a rate considerably higher than chance.

Because ATLAS is a protein, it can be evolved to change its properties and improve performance similar to GCaMP40 or GRAB sensors41. For instance, ATLAS, as described in this paper, works only for anterograde tracing from excitatory neurons. However, AF can, in principle, be replaced by a recombinant binder to a neurotransmitter receptor, such as a GABA (γ-aminobutyric acid) or acetylcholine receptor, or even to a receptor for a neuromodulator, such as dopamine or serotonin. The resulting modified ATLAS could be used for tracing virtually any neuronal circuit mediated by the corresponding neurotransmitter–receptor combination. Also, because of its activity dependence, ATLAS could be used to mark circuits that respond to specific stimuli, especially if ATLAS tracing could be gated to occur during a particular time interval. Finally, it may be possible to modify ATLAS to label circuit changes during events such as memory formation42 or specific developmental processes such as synaptic pruning43.

Methods

AF production

Amino acids 1–394 of GluA1 grown in Sf9 (Spodoptera frugiperda) cells was used as a target for an mRNA display selection, carried out as described in ref. 44. FingRs amplified from the ninth round of selection were fused to a Myc tag and coexpressed in Cos7 cells (ATCC, CRL-1651) with ss-GluA1-HA (ss, signal sequence of GluA1) and Stargazin-V5. Cos7 cells were stained live with anti-MYC antibody, then fixed, permeabilized and stained for HA and V5. Fibronectins labeling the surface of cells expressing GluA1 and Stargazin were then expressed in dissociated cultures. Fibronectin 9.3-labeled endogenous GluA1 and was designated AF.

Plasmid constructs

The ss-AF-YFP consists of DNA encoding amino acids 1–18 of GluA1 inserted upstream of AF and yellow fluorescent protein (YFP). For ATLAS constructs, VAMP2 complementary DNA (cDNA) was inserted upstream of ALFA-tag (SRLEEELRRRLTE) followed by a BACEcs (β-site Amyloid Precursor Protein-cleaving enzyme 1) cut site (EISEVNLDAEFR). VAMP2-BACEcs-SEP has SEP downstream of BACEcs. VAMP2-BACEcs-AF-SEP has AF and SEP downstream of BACEcs. ATLASCre has AF and Cre cDNA downstream of BACEcs. DIO-ATLASFlp has a Myc tag (EQKLISEEDL) followed by FLP1 cDNA downstream of BACEcs and nested LoxP sites on either side of the coding sequence. GFP-HA-BACE1 plasmid was purchased (Addgene, no. 165032), and GFP was removed to make BACE-HA. The above sequences were ligated into a pAAV plasmid containing the human synapsin (hsyn) promoter and a woodchuck hepatitis virus posttranscriptional regulatory element (WPRE). Lenti-hsyn-Cre was purchased from Addgene (no. 86641), and the hSYN1 promoter was replaced with the CamKIIα promoter to generate Lenti-CamKII-Cre.

Dissociated cultures

Pups from timed pregnant Sprague Dawley rats (E19) were used for cultured cortical neurons, prepared as in ref. 45. All procedures were approved by the Institutional Animal Care and Use Committee of the University of Southern California.

Protein purification

Human embryonic kidney 293T cells were transfected with the plasmid pCAG-ss-AF-3XHA-TEVcs-HALOTag and the secreted protein AF-HA-TEVcs-HALOTag was collected and incubated with HaloLink resin (Promega, cat. no. G1914) overnight at 4 °C. and then with tobacco etch virus (TEV) protease (New England Biolabs, cat. no. P8112S) overnight at 4 °C. Supernatant was collected and incubated with NEBExpress Ni Resin (New England Biolabs, cat. no. S1428S) overnight at 4 °C to remove the TEV protease. Purified protein was dialyzed into 20 mM HEPES, 250 mM NaCl pH 7.4 and concentrated using the Amicon Ultra centrifugal filters (cat. no. UFC501096).

Endocytosis

Neuronal cultures were incubated with 1 mM purified AF-HA protein for 1 h at 37 °C. Next, cultures were washed three times with 1× Hanks Balanced Salt Solution containing 10 mM HEPES, fixed with 4% paraformaldehyde and stained for anti-HA (Cell Signaling, cat. no. 3724S) and anti-Map2 (Abcam, cat. no. ab5392).

Presynaptic release

Plasmids encoding VAMP2-BACEcs-SEP (1 μg) or VAMP2-BACEcs-AF-SEP (1 μg) with BACE-HA (2 μg) were transfected into 15–18 day in vitro (DIV) cultured neurons. Two days before transfection, cultures were infected with AAV8-Cre and AAV8-DIO-PSD-95.FingR-TdTomato to label postsynaptic sites. Neurons were imaged 3 to 5 days later with or without 100 mM KCl. Time-lapse imaging was performed for either 100 or 200 frames at an interval of 1,000 ms and an exposure of 500 ms. Time constants were measured using curve fitting functions in GraphPad Prism. Videos of SEP release events were generated using ImageJ by taking the difference of adjacent image frames to eliminate background and SEP fluorescence beyond the initial frame after release.

Viruses

The following viruses were purchased from Addgene: AAV8-hSyn-DIO-mCherry no. 50459, AAV8-EF1a-fDIO-mCherry no. 114471, AAV8-hsyn-DIO-eGFP no. 50457 and AAV8.hSyn.hM3D.mCherry no. 50474. All other viruses were generated in the Arnold laboratory using standard methods.

Transsynaptic tracing in culture

Neuronal cultures (14 DIV) were transfected with plasmids for ATLAS (1 μg) and BACE-HA (2 μg) using the calcium phosphate method. Transfection was followed immediately by infection with AAV8-hsyn-DIO-GFP and incubated for 7 days. Neurons were fixed and probed as described.

Immunocytochemistry for dissociated cultures

Neurons were stained using: anti-GFP, 1:10,000, Aves GFP-1020; anti-MYC, 1:1,000, Novus Biologicals NB600-334; anti-Map2, 1:1,000, Abcam ab5392; anti-clathrin, 1:1,000, GeneTex GTX80218; anti-mCherry, 1:1,000, Rockland M11217; anti-RFP, 1:1,000, Invitrogen MA5-15257; anti-GluA1, 1:1,000, EMD Millipore ABN241; anti-PSD-95, 1:3,000, NeuroMab 75-348; anti-HA, 1:1,000, Cell Signaling 3724S; anti-Cre, 1:1,000, EMD Millipore MAB3120; Alexa Fluor 488 goat anti-chicken, 1:1,000, Invitrogen A-11006; Alexa Fluor 594 goat anti-rabbit, 1:1,000, Invitrogen A-11012; Alexa Fluor 594 goat anti-mouse, 1:1,000, Invitrogen A-11005; Alexa Fluor 647 goat anti-mouse, 1:1,000, Invitrogen A-21235; Alexa Fluor 647 goat anti-rabbit, 1:1,000, Invitrogen A-21244; JF549 halotag ligand, 1:1,000, Janelia Farm; JF647 halotag ligand, 1:1,000, Janelia Farm; FluoTag-X2 anti-alfa 647, 1:1,000, NanoTag biotechnologies N1502 AF647. Neurons were mounted using Dapi-fluoromount-G (Electron Microscopy Services, 17984-24).

Electrophysiology (testing of AF)

All experiments were performed in accordance with National Institutes of Health (NIH) Guidelines for the Care and Use of Laboratory Animals, and all procedures were approved by the Institutional Animal Care and Use Committee of the University of Southern California. As previously described, 400-μm-thick rat organotypic hippocampal slice cultures were prepared from male and female P6 to P8 Sprague Dawley rats46. Sparse biolistic transfections were performed on DIV1 as previously described47,48. Next, 50 μg of mixed plasmid DNA encoding ss-AF-YFP was coated on 1-μm diameter gold particles. Construct expression was confirmed by YFP epifluorescence. Dual whole-cell recordings of CA1 pyramidal neurons were performed on DIV7–8 through simultaneous recordings from a transfected neuron and a neighboring, nontransfected control neuron. Synaptic responses were evoked by stimulating with a monopolar glass electrode filled with artificial cerebrospinal fluid (ACSF) in the stratum radiatum. AMPAR-evoked EPSCS (-eEPSCs) were measured at −70 mV. NMDAR-eEPSCs were measured at +40 mV and were temporally isolated by measuring amplitudes 150 ms after the stimulus. For paired-pulse facilitation, paired pulses at 40-ms interpulse interval were delivered and the ratio of peak amplitudes was analyzed to generate barplots depicting mean ± s.e.m. Data analysis was performed using Igor Pro v.9.

Animal preparation and stereotaxic surgery (mouse)

All procedures in Figs. 3, 4b–d, 5 and 6c,d and Extended Data Figs. 3, 5, 6, 7a,b, 8 and 9 were performed in accordance with NIH Guidelines for the Care and Use of Laboratory Animals and were approved by the Animal Care and Use Committee at the University of Southern California. Male and female wild-type mice (C57BL/6J, Jackson Laboratories, strain no. 000664), Vipr2-Cre (Cre expression in dLGN, Jackson Laboratories, strain no. 031332) and Sst-IRES-Cre (Cre expression in somatostatin interneurons, Jackson Laboratories strain no. 013044) mice aged 2–6 months were used. For stereotaxic viral injection, mice were anesthetized in an induction chamber containing 4% isoflurane mixed with oxygen and transferred to a stereotaxic frame with a heating pad. Then, 1.5% isoflurane was delivered throughout the procedure through a nose cone at a rate of 0.5 l min−1. Injection coordinates based on ref. 49 were used to mark the location where a small hole (0.5 mm diameter) was drilled. Viruses were delivered through pulled glass micropipettes using pressure injection via a micropump. Total injection volumes ranged from 50 to 800 nl, at 20–60 nl min−1. Following injection, the micropipette was left in place for 15 min to minimize virus diffusion into the pipette track. Animals were administered Ketoprofen (5 mg kg−1) at the beginning of the surgical procedure and again every 24 h for 2 d following surgery.

Injections of viruses (mouse)

For experiments in Figs. 3, 4b–d, 5, 6c,d and Extended Data Figs. 3, 5, 6, 7a,b, 8 and 9, anterograde tracing experiments using ATLAScre, 400–600 nl of either ATLASCre or ATLASsnCre virus was injected into presynaptic cells in the PFC, Primary auditory cortex (A1) or entorhinal cortex. AAV8-DIO reporter virus (400–800 nl), diluted to minimize the background signal, was injected into postsynaptic cells in the Str, ipsilateral claustrum, bilateral inferior colliculus or ipsilateral dentate gyrus. The animals were euthanized 2 weeks following the injections.

In experiments with DIO-ATLASFlp, transgenic VIPR2-Cre mice were injected with AAV8-DIO-ATLASFlp (600 nl, 1.0 × 1013) at the ipsilateral LGN. A second injection of AAV8-EF1a-fDIO-mCherry (600 nl, 5.0 × 1012) was administered at the primary visual cortex (V1). The animals were euthanized after 2 weeks. To test AAV-DIO-ATLASFlp, C57BL/6J mice were injected with a viral mixture of AAV8-DIO-ATLASFlp and Lenti-hsyn-Cre at V1 and AAV8-EF1a-fDIO-mCherry (600 nl, 5.0 × 12) in the ipsilateral SC and euthanized 2 weeks later. Transgenic SST-Cre mice were injected with a mixture of AAV8-DIO-ATLASFlp and AAV8-ef1a-fDIO-mCherry (at 1:1 ratio based on viral particles) at V1 and euthanized 1 week later.

For testing the potential retrograde spread of ATLAS, AAV8-DIO-ATLASFlp and Lenti-hsyn-Cre were injected into Str of C57BL/6J mice, and AAV8-EF1a-fDIO-mCherry (800 μl, 5.0 × 1012) was injected into mPFC. As the negative control, a group of C57BL/6J mice was injected with AAV8-ef1a-fDIO-mCherry (800 μl, 5.0 × 1012) in mPFC. All animals were then euthanized 2 weeks following the injections.

To test for monosynaptic tracing, C57BL/6J mice were injected with AAV8-ATLASCre (600 nl, 1.0 × 1013) in V1 and AAV8-hsyn-DIO-mCherry (400 nl, 1.0 × 1013) in the PBG. As the positive control, C57BL/6J mice were injected with AAV8-ATLASCre (600 nl, 1.0 × 1013) in SC and AAV8-hsyn-DIO-mCherry (400 nl, 1.0 × 1013) in PBG. As a negative control, wild-type mice were injected with AAV8-hsyn-DIO-mCherry (400 nl, 1.0 × 1013) at PBG. Animals were euthanized 2 weeks following injections.

To test supplementing ATLAS with BACE, mixtures with the following ratios of AAV8-ATLASCre and AAV8-hsyn-BACE-HA: 1:1 ATLAS: BACE, 1:5 ATLAS: BACE, or no BACE based on number of viral particles. These were injected into the C57BL/6J mice in mPFC. AAV8-hsyn-DIO-GFP (800 nl, 1.0 × 1013) in ipsilateral Str to visualize postsynaptic cells. Animals were euthanized 2 weeks following injections. To examine the effect of incubation time on efficiency of ATLAS, C57BL/6J mice were injected with a viral mixture of AAV8-ATLASsnCre and AAV8-hsyn-BACE-HA (600 nl, 1:3 based on viral particles) in mPFC and AAV8-hsyn-DIO-GFP (800 nl, 1.0 × 1013) in ipsilateral Str. Animals were euthanized 4, 7, 14 or 21 days following injections.

For testing the effect of neuronal activity on ATLAS tracing, C57BL/6J mice were injected with AAV8-ATLASsnCre or AAV8-ATLASCre, AAV8-hsyn-BACE and AAV8-hHM3Dq-mCherry (600 nl, 1:3:1 based on viral particles) in mPFC and AAV8-hsyn-DIO-GFP (800 nl, 1.0 × 1013) in ipsilateral Str. CNO (200 μl, 1.1 mM) was administered to experiment animals through intraperitoneal injection daily following viral injections. Animals were euthanized 7 days after viral injections. A final dose of CNO was administered 45 min before the perfusion.

To determine whether ATLAS shows toxicity compared to ATLASsn, one group of C57BL/6J mice were injected with 600 nl of AAV8-ATLASCre and AAV8-hsyn-BACE (1:3 based on viral particles), and the other group of mice were injected with 600 nl of viral mixture of AAV8-ATLASsnCre and AAV8-hsyn-BACE (at a 1:3 ratio based on viral particles) at mPFC. AAV8-hsyn-DIO-GFP (800 nl) was injected into the ipsilateral Str of both groups. Animals were euthanized 14 and 28 days following injections.

In all experiments listed above, presynaptic and postsynaptic injections were performed during the same surgery. Bregma and Lambda coordinates are in millimeters as follows: anteroposterior is the distance from Lambda, mediolateral is the distance from midline and dorsoventral is the distance from Lambda except for V1, A1 and SC, which use the distance from the surface of the cortex. mPFC (6.3, 0.3, 1.7); Str (5.1, 1.8, 3.4); dLGN (2.2, 2.2, 2.9); primary visual cortex (V1) (0.8, 2.4, 0.6); primary auditory cortex (A1) (1.5, 4.2, 0.6); SC (0.6, 0.8, 1.2); inferior colliculus (−0.7, 0.8, 0.6); PBG (0.1, 1.8, 3.3); claustrum (5.4, 2.5, 4.0); entorhinal cortex (0.0, 3.6, 3.6) and dentate gyrus (0.6, 2.8, 2.5).

Histology (mouse)

Mouse brains were perfused with 40 ml of phosphate-buffered saline (PBS), then 40 ml of 4% paraformaldehyde in PBS (Figs. 3, 4b–d, 5, 6c,d and Extended Data Figs. 3, 5, 6, 7a,b, 8 and 9). The brains were fixed overnight in 4% paraformaldehyde at 4oC and dehydrated in 30% sucrose solution for at least 48 h. Brains were embedded in tissue freezing medium at −45 °C and sectioned to a thickness of 25–35 µm using a cryostat. For immunohistochemistry, antibodies used were: rabbit anti-RFP, Rockland (cat. no. 600-401-379), 1:1,000; chicken anti-GFP antibody, AVES (gfp-1020, cat. no. NC9510598), 1:1,000; Fluotag-X2 anti-ALFA, NanoTag (cat. no. N1502-ALFA647-L), 1:1,000; mouse anti-NeuN antibody, EMD Millipore (cat. no. MAB377), 1:1,000, c-Fos (cat. no. 9F6) Rabbit mAb, Cell Signaling Technology (cat. no. 2250T), 1:1,000; Alexa Flour 488-conjugated goat anti-chicken (A-11039) or anti-mouse IgG (cat. no. A-28175), Thermo Fisher, 1:1,000; Alexa Flour 594-conjugated anti-rabbit IgG, Thermo Fisher (cat. no. A-11012), 1:1,000; Alexa Flour 647-conjugated goat anti-mouse IgG, Thermo Fisher, 1:1,000 (cat. no. A-21235).

Quantification of transsynaptic labeling

Labeled presynaptic cells (At) and postsynaptic cells (GFP or mCherry) were counted within an identical 1-mm2 region of 25-μm-thick stained sections of target regions. All sections examined were roughly equal distance from the site of AAV injection. We quantified cells in two representative sections, each 25 µm apart. Then, the data from the selected sections was pooled and multiplied by two to estimate the number of labeled cells within 100 µm. To account for the background expression from spontaneous recombination by the DIO reporter construct, we subtracted the cell counts from control experiments from those in corresponding ATLAS experiments. To quantify activity-dependent transsynaptic labeling, total GFP fluorescence minus background was measured identically in Str in sections at equivalent distances from injection sites using ImageJ. For the SST experiment, At-labeled cells and mCherry-labeled cells were counted within an identical 1 mm2 region of 25-μm-thick stained sections. The number of mCherry+ cells over the mCherry and At colabeled cells was then calculated.

Stereotaxic surgery and histology (M1 to Str and Thal)

All experiments in Fig. 6e–h were performed in accordance with NIH Guidelines for the Care and Use of Laboratory Animals, and all procedures were approved by the Institutional Animal Care and Use Committee of NIH. The ATLASsnCre viral vector AAV8-SYTnb-VAMP2Δ-At-BACEcs-AF-Cre and AAV8-BACE-GFP (1:1, 200 nl) were injected stereotaxically into the motor cortex of wild-type mice at (1.0, 2.0, 1.8) (Fig. 6e–h). They also received an injection of a Cre-dependent reporter vector: AAV8-Ef1a-FLEX-tdTomato (Addgene, 28306) into the Str (0.5, 2.0, 4.0) and thalamus (−1.55, 1.5, 3.75). Following a survival time of 2 weeks, animals were perfused and 50-μm sections were cut with a freezing microtome. The following antibodies were used for immunocytochemistry: rabbit anti-RFP (cat. no. 600-401-379, Rockland Immunochemicals), chicken anti-GFP (cat. no. A10262, Invitrogen), goat anti-chicken 488 (cat. no. A32931, Invitrogen) and goat anti-rabbit 555 (cat. no. A32732, Invitrogen). Sections were counterstained with Neurotrace Blue 435 (cat. no. N21479, Invitrogen) and coverslipped with Aqua-polymount (cat. no. 18606, Polysciences). Slides with 12–15 coronal brain sections were imaged on a Zeiss EpiFluorescence Microscope equipped with Xylis broad spectrum LED illumination (Excelitas Technologies), a Hammamatsu Orca Flash4.0 camera and Ludl stage drive controlled by Neurolucida 360 Imaging software (MBF Bioscience). Images were registered using NeuroInfo v.2022.1.1.

Animal preparation and stereotaxic surgery (rat)

All experiments were performed in accordance with NIH Guidelines for the Care and Use of Laboratory Animals, and all procedures were approved by the Institutional Animal Care and Use Committee of the University of Southern California. Male Sprague Dawley rats (Envigo) weighing 300–400 g were injected with viruses delivered using a micro-infusion pump (Harvard Apparatus) connected to a 33-gauge microsyringe injector attached to a PE20 catheter and Hamilton syringe. Flow rate was 5 μl min−1. Injectors were left in place for 3 min postinjection. Viruses were incubated for 21 days after injection. AAV8-ATLASsnCre was unilaterally injected into the CA1v at: −4.9 mm anteroposterior, ±4.8 mm mediolateral, −7.8 mm dorsoventral (zeroing at skull site). AAV-pCAG-FLEX-tdTomato-WPRE (Addgene; cat. no. 51503) was unilaterally injected into the ACB: +1.2 mm anteroposterior, ±1.0 mm mediolateral, −6.75 mm dorsoventral (zeroing at dura). Injection volume was 100 nl per site. For control experiments animals were injected with AAV-pCAG-FLEX-tdTomato-WPRE unilaterally in the ACB, but no injection was made in the ventral CA1. Photomicrographs were acquired using a Nikon 80i (Nikon DS-QI1, 1,280 × 1,024 resolution, 1.45 megapixel) microscope under epifluorescence controlled by NIS-Elements v.C6.10.01.

Histology (rat)

Following perfusion and fixation, Brains were sectioned into 30-μm sections on a microtome cooled with dry ice, collected in an antifreeze solution and stored in a −20 °C freezer. The following antibodies were used for immunocytochemistry: Rabbit anti-RFP (1:2,000, Rockland Inc.; cat. no. 600-401-379; Clonality: Polyclonal) and FluoTag-X2 anti-ALFA (1:1,000, NanoTag Biotechnologies; cat. no. N1502).

Quantification of activity-dependent transsynaptic labeling

For experiments to quantify transsynaptic labeling with and without BACE, mCherry-expressing cells were counted within an identical 1-mm2 region over four consecutive 25-μm stained sections of Str. For experiments to quantify activity-dependent transsynaptic labeling, total GFP fluorescence minus background was measured identically in Str sections using ImageJ v.1.49b, Slidebook v.8 and Zen Microscopy v.3.11.

Electrophysiology (PFC to Str)

For experiments in Extended Data Fig. 4, wild-type C57BL/6J mice coinfected with AAV8-ATLASsnFLP, AAV8-BACE-HA and AAV8-DIO-ChR2-GFP into the mPFC, and DIO-mCherry in Str were used for slice recording. Details of the stereotaxic surgery are the same as in Figs. 3a–d and 6c,d. Four weeks following injections, animals were decapitated following anesthesia, and the brain was rapidly removed and immersed in an ice-cold dissection buffer (composition 60 mM NaCl, 3 mM KCl, 1.25 mM NaH2PO4, 25 mM NaHCO3, 115 mM sucrose, 10 mM glucose, 7 mM MgCl2, 0.5 mM CaCl2; saturated with 95% O2 and 5% CO2; pH 7.4). Coronal slices of 350 µm thickness were sectioned by a vibrating microtome (Leica VT1000s) and recovered for 30 min in a submersion chamber filled with warmed (35 °C) ACSF (composition 119 mM NaCl, 26.2 mM NaHCO3, 11 mM glucose, 2.5 mM KCl, 2 mM CaCl2, 2 mM MgCl2, and 1.2 NaH2PO4, 2 mM sodium pyruvate, 0.5 mM VC). Str neurons surrounded by GFP+ fibers were visualized under a fluorescence microscope (Olympus BX51 WI). Patch pipettes (~4–5 MΩ resistance) filled with a cesium-based internal solution (composition 125 mM cesium gluconate, 5 mM TEA-Cl, 2 mM NaCl, 2 mM CsCl, 10 mM HEPES, 10 mM EGTA, 4 mM ATP, 0.3 mM GTP and 10 mM phosphocreatine; pH 7.25; 290 mOsm) were used for whole-cell recordings. Signals were recorded under voltage-clamp mode at a holding voltage of −70 mV for excitatory currents, filtered at 2 kHz and sampled at 10 kHz. Tetrodotoxin (1 μM) and 4-aminopyridine (1 mM) were added to the external solution for recording monosynaptic responses to blue light stimulation (5-ms pulse, 3-mW power, 5–10 trials).

Surgery (V1 to Str)

Experiments in Fig. 4e–h were performed in accordance with NIH Guidelines for the Care and Use of Laboratory Animals, and all procedures were approved by the Institutional Animal Care and Use Committee of Harvard Medical School. Injections were performed through a pulled glass at a rate of 30 nl min−1. After injections, pipettes were slowly withdrawn (less than 100 µm s−1) at least 5 min after the end of the infusion. Viruses were injected in two rounds of surgeries to minimize expression competition between ATLAS viruses (that is ATLASsnCre, BACE-HA) and Cre-dependent ChR2. Wild-type animals undergo the first round of surgery between 4 and 8 weeks of age (Fig. 4e–h). Before the start of surgery, presynaptic virus cocktail was prepared by mixing AAV8-hSyn-ATLASsnCre and AAV8-hSyn-BACE-HA in 3:1 ratio (titer is 2 × 1013 gc ml−1 for both viruses). During surgery, presynaptic virus cocktail was injected at three separate sites in V1. In addition, AAV8-hSyn-DIO-mCherry or AAV9-CAG-DIO-tdTomato was injected into two separate sites in Str. Then 14 days after the first surgery, AAV8-hSyn-DIO-ChR2-GFP was injected at the same three V1 sites as the first surgery.

Coordinates of V1 injection sites were as follows (coordinates are relative to bregma and dorsoventral is relative to the brain surface, below). Here, 300 nl of presynaptic virus cocktail (AAV8-ATLASsnCre and AAV8-BACE-HA) was injected per site for the first surgery. Then 200 nl of AAV8-hSyn-DIO-ChR2 was injected per site for the second surgery.

  • V1 site 1: −2.75 mm anteroposterior, ±2.5 mm mediolateral, −0.65 mm dorsoventral

  • V1 site 2: −3.80 mm anteroposterior, ±2.9 mm mediolateral, −0.65 mm dorsoventral

  • V1 site 3: −3.80 mm anteroposterior, ±2.0 mm mediolateral, −0.65 mm dorsoventral

Next, 500 nl of postsynaptic virus (Cre-dependent mCherry or tdTomato) was injected at Str site 1 (anterior) while 400 nl of virus was injected at Str site 2 (posterior).

  • Str site 1: 0.70 mm anteroposterior, ±1.60 mm mediolateral, −2.22 mm dorsoventral

  • Str site 2: −0.58 mm anteroposterior, ±2.75 mm mediolateral, −2.75 mm dorsoventral

Electrophysiology (V1 to Str)

For experiments in Fig. 4e–h, brain slices were obtained from 2- to 3-month-old mice (both male and female) using standard techniques. The experiments closely followed the procedures outlined in previous studies50,51. Mice were anesthetized using isoflurane inhalation and subsequently subjected to transcardial perfusion with ice-cold ACSF composed of the following (in mM): 125 NaCl, 2.5 KCl, 25 NaHCO3, 2 CaCl2, 1 MgCl2, 1.25 NaH2PO4 and 11 glucose (final osmolarity of 300–305 mOsm kg−1). The brain was removed from the skull, placed in ice-cold ACSF and 250–300-μm coronal brain slices were cut and initially placed in a holding chamber for 10 min at 34 °C in a choline-based solution composed of (in mM): 110 choline chloride, 25 NaHCO3, 2.5 KCl, 7 MgCl2, 0.5 CaCl2, 1.25 NaH2PO4, 25 glucose, 11.6 ascorbic acid and 3.1 pyruvic acid. Following choline incubation, slices were transferred to a second chamber with ACSF at 34 °C for a minimum of 30 min. The chamber was subsequently shifted to room temperature for the duration of the experiment. All recordings were made within 4–5 h of slicing. The temperature was maintained at 32 °C and carbogen-bubbled ACSF was perfused at a rate of 2–3 ml min−1 for all experiments.

Monosynaptically connected striatal neurons were identified by their mCherry-positive or tdTomato-positive somata. Fluorescence-negative neurons with GABAergic interneuron physiological properties (membrane tau decay <0.8 ms for both fast-spiking and persistent low-threshold spiking subtypes; input resistance >500 MΩ in persistent low-threshold spiking subtype) were excluded from the analysis.

Striatal neurons were patched in whole-cell voltage-clamp configurations using borosilicate glass electrodes (3–5 MΩ) filled with cesium-based (voltage clamp) internal solution to measure EPSCs. The solution contained (in mM): 135 CsMeSO3, 10 HEPES, 1 EGTA, 3.3 QX-314 (Cl− salt), 4 Mg-ATP, 0.3 Na-GTP, 8 Na2-phosphocreatine, with pH adjusted to 7.3 using CsOH, resulting in an osmolarity of 295 mOsm kg−1. All recorded neurons exhibited electrophysiological characteristics of spiny projection neurons. All synaptic currents were recorded with a cesium-based internal and monitored at a holding potential of −70 mV. Fluorescent positive and negative neurons were patched as randomly ordered sequential pairs, strictly within one field of view of each other. ChR2-mediated synaptic currents from V1 were optically evoked using 2-ms pulses of 473-nm light at light powers of 2 mW controlled through an acousto-optic modulator.

To assess the baseline connection probability of V1-DMS circuit viruses were injected in V1 at identical locations, titer and volume and at identical times as described above (see Surgery section for Fig. 4e–h). After 4 weeks, brain slices were obtained and patched in whole-cell voltage-clamp configurations as described above. We first found the field of view with maximal ChR2 fiber expression in DMS and randomly patched cells within it. ChR2-mediated synaptic currents from V1 were optically evoked using as above.

Quality control criteria were a stable baseline membrane resistance (variation in Rm < 10%) and a stable baseline resting potential with median standard deviation of ≤0.5 mV across recordings. Optically evoked EPSCs were analyzed post hoc using MATLAB v.2024a.

Statistics and reproducibility

Experiments in Fig. 1c were performed in three independent cultures; in Fig. 1d, two cultures; in Fig. 1e–g, five cultures; in Fig. 1h, four cultures; in Fig. 1n, ten slices; in Fig. 1o, eight slices; in Fig. 1p, six slices; in Fig. 2b–e, in three independent cultures. Experiments in Figs. 3l and 4b–d were performed in one mouse bilaterally and another mouse unilaterally. Experiments in Fig. 6b were performed in three cultures, in Fig. 6c–h, in three mice and in Fig. 6j–l, in three rats.

Experiments in Extended Data Figs. 1a–h and 2a–g were performed in two cultures; in Extended Data Fig. 6, in two mice; in Extended Data Fig. 7, in three mice. All values are expressed as mean ± s.e.m.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.