All-viral tracing of monosynaptic inputs to single birthdate-defined neurons in the intact brain

Summary Neuronal firing patterns are the result of inputs converging onto single cells. Identifying these inputs, anatomically and functionally, is essential to understand how neurons integrate information. Single-cell electroporation of helper genes and subsequent local injection of recombinant rabies viruses enable precise mapping of inputs to individual cells in superficial layers of the intact cortex. However, access to neurons in deeper structures requires more invasive procedures, including removal of overlying tissue. We developed a method that, through a combination of virus injections, allows us to target 4 or fewer hippocampal cells 48% of the time and a single cell 16% of the time in wild-type mice without use of electroporation or tissue aspiration. We identify local and distant monosynaptic inputs that can be functionally characterized in vivo. By expanding the toolbox for monosynaptic circuit tracing, this method will help further our understanding of neuronal integration at the level of single cells.

In brief Jacobsen et al. develop a method that enables inputs to individual cells in deep brain areas to be identified without first removing overlaying tissue. Cells can be targeted based on when they are born during development, and their inputs can be characterized with electrophysiology or imaging techniques in vivo.

INTRODUCTION
The firing properties of individual neurons are the result of a convergence of multiple inputs. Genetically engineered rabies viruses enable identification of monosynaptic inputs to populations of cells (Wickersham et al., 2007;Reardon et al., 2016;Ciabatti et al., 2017;Chatterjee et al., 2018) and have had a significant impact on our ability to dissect neural circuits. However, to understand the input/output transformation in neurons, we need to map the inputs to individual cells, which is not easy to achieve in vivo. Currently, it is possible to target a single ''starter cell'' in superficial cortical layers in vivo using single-cell electroporation (Marshel et al., 2010;Wertz et al., 2015;Rompani et al., 2017;Rossi et al., 2020;Geiller et al., 2021), patch clamping (Rancz et al., 2011;Vé lez-Fort et al., 2014), or stamping (Schubert et al., 2017) techniques. However, targeting cells in deeper structures with these methods requires invasive procedures, including removal of overlaying tissue (Rompani et al., 2017) that may be part of the circuit under investigation.
We set out to develop a method that allows us to target individual neurons for monosynaptic rabies tracing in deep brain areas without the need for tissue aspiration. To demonstrate our method, we chose the hippocampal trisynaptic circuit as this is a network that has received considerable attention because of its importance for learning, memory, and navigation. As a result, the different hippocampal subfields and firing properties of the principal cell types within it, as well as the gross connectivity of the circuit, are well understood. Monosynaptic rabies tracing has been used previously to identify inputs to large populations of hippocampal neurons (Nakashiba et al., 2012;Deshpande et al., 2013;Sun et al., 2014Sun et al., , 2017Tao et al., 2021) and, very recently, also inputs to individual CA1 pyramidal cells (Geiller et al., 2021). However, the constellation of inputs that single cells in deeper hippocampal subfields receive remains unknown. This is in part because rabies-mediated input mapping of single hippocampal cells with existing methods requires removal of substantial amounts of overlaying tissue, including parts of CA1 to reach the dentate gyrus and CA3, which disrupts the trisynaptic circuit. To target single cells in these subfields and identify their monosynaptic inputs in the intact hippocampus, we first performed an in utero injection of a virus carrying Cre recombinase to achieve sparse expression of Cre in the adult brain.
MOTIVATION Identifying the inputs to a neuron is essential to understand its output. Although this is possible with current methods in many neocortical areas, reaching cells in deeper brain regions in vivo requires removal of tissue that may be part of the circuit under investigation. We developed a virus-based method that enables individual cells in subneocortical areas of wild-type mice to be targeted and their monosynaptic inputs to be identified without removing overlaying tissue.
Subsequently, by locally injecting a custom helper virus that reaches only a single neuron in this sparsely Cre-labelled population, we can, in a single cell, express the receptor and G-protein necessary for rabies infection and spread, respectively. We show that our approach enables identification of local and distant inputs to these individual starter cells in the hippocampal-entorhinal circuit. Our in utero approach has the added benefit of enabling cells to be targeted based on their birthdate during development, making it possible to investigate inputs to cells with specific developmental origins. We demonstrate how this method can be used to determine the functional properties of monosynaptic inputs to such cell populations with in vivo electrophysiology combined with optogenetic tagging or calcium imaging. Specifically, we provide proof of concept for determining the spatial tuning properties of cells in the medial entorhinal cortex (MEC) that project directly to birthdate-defined cell populations in the hippocampus.
By enabling individual cells in subneocortical brain structures to be targeted for monosynaptic input mapping in a minimally invasive manner, this method extends rabies virus-based circuit mapping tools to deeper regions. This, in turn, will expand our understanding of afferent connectivity and the functional integration of inputs that occurs in single cells in these circuits.

RESULTS
A virus-based method for targeting a single cell and identifying its inputs We first designed and produced the viruses necessary to perform rabies-mediated tracing based on the Challenge Virus Standard N2c variant (CVS-N2c) strain because this strain spreads to more input cells and is less cytotoxic than the Street-Alabama-Dufferin B19 variant (SAD-B19) strain (Reardon et al., 2016). We produced three different EnvA-pseudotyped, G-protein-deleted CVS-N2c recombinant rabies viruses expressing a gene of interest (RABV-GOI) each; specifically, tdTomato (RABV-tdTomato), channelrhodopsin 2 (RABV-ChR2-YFP), or GCaMP6f (RABV-GCaMP6f), to enable anatomical circuit mapping, optogenetic tagging, and calcium imaging of input cells, respectively (Figure 1A). These pseudotyped RABVs require the TVA receptor to enter cells and a G-protein to spread to input cells ( Figure S1A). We incorporated TVA and the G-protein from the CVS-N2c strain into a single Cre-dependent adeno-associated virus (helper virus; AAV-hS-FLEX-TVA-HA-N2cG, Figure 1A), preventing unspecific and undetected starter cell labeling, which can occur when these components are introduced separately using multiple helper viruses; for example, by resulting in secondary jumping from cells that express G but not TVA. We selected an hemagglutinin (HA) tag as a surrogate marker instead of a fluorescent one to facilitate efficient packaging of viral particles. The transgenes were separated by 2A self-cleaving peptide sequences (de Felipe et al., 2006;Tang et al., 2009), providing 2A peptide epitopes for immunodetection. These strategies enabled us to perform immunohistological identification of HA or 2A ''helper tags'' ( Figure 1B) in cells infected with our single helper virus and ensured compatibility with RABVs expressing different fluorescent payloads.
Next we used these viruses to label monosynaptic inputs to cells in the hippocampus by first expressing Cre in wild-type mice. To target a single cell, we performed in utero injections of an AAV carrying Cre under the control of the Ca 2+ /calmodulin-dependent protein kinase II (CaMKII) promoter (AAV-CKII-Cre) diluted in PBS on embryonic day 13 (E13; Figure 1B; Table S1), when cells in the hippocampal subfields are born (Angevine, 1965;Deguchi et al., 2011), and the lateral ventricles are still clearly visible ( Figure 1B). Using ultrasound to guide the pipette, we targeted one of the lateral ventricles and injected approximately 300 nL into every embryo. Because of the rapid turnover of cerebrospinal fluid and, hence, the restricted temporal exposure of cells lining the ventricles to the virus (Donato et al., 2017), only a few cells express Cre in the adult hippocampus of these animals ( Figure S1B). Next we injected approximately 50 nL of the helper virus carrying Cre-dependent TVA and G under control of the human synapsin promoter (AAV-hS-FLEX-TVA-HA-N2cG) unilaterally into the adult hippocampus, targeting the CA3 and dentate gyrus ( Figure 1B). At least 2 weeks later, a separate injection of the pseudotyped RABV carrying tdTomato (RABV-tdTomato) resulted in infection of TVA-expressing cells (starter cells) and, with time, G-protein-mediated spread of the RABV carrying tdTomato to monosynaptically connected input cells ( Figure 1B). To determine the number of starter cells targeted, we perfused all animals within 1 month of the rabies injection (when we waited longer, the infected cells started to look unhealthy; Figure 7A). We next sectioned the ipsilateral hemisphere, keeping all sections and taking care to maintain their original order. Antibodies were then used to label cells expressing a helper tag ( Figure 1B) as well as those expressing tdTomato (to ensure that rabies-infected cells were not missed because of weak tdTomato expression). All tdTomato-expressing cells in the dorsal hippocampus and entorhinal cortex were scanned at 403 with a confocal microscope to check for co-expression with the helper tag. Cells labeled with both the helper tag and tdTomato were identified as starter cells, whereas cells only expressing tdTomato were classified as input cells ( Figure 1B). No starter cells were found in the MEC. Because of the difficulties associated with antibody detection of helper tags (Discussion), we split the animals into two groups (STAR Methods): those in which starter cells could be clearly detected (''conclusive'' animals) and those in which it was ambiguous whether starter cells were present (''inconclusive'' animals).
With this approach, we could unambiguously target four or fewer starter cells in 48% of animals (15 of 31) (Figure 2A). In this group, a single starter cell was labeled 33% of the time (5 of 15) ( Figure 2A), resulting in an overall success rate of 16% (5 of 31) for targeting a single cell in the hippocampus. As expected, longer expression times of the RABV increased the number of rabies-positive cells ( Figure 2B), the anatomical area over which they were found in the dorsal hippocampus ( Figure S2A), as well as the number of input cells labeled in the more distant MEC (Figure S2B). Hence, we can identify local hippocampal and distant MEC inputs to single cells in the hippocampus (Figures 3 and 4). The identity of the inputs reflected the identity of the starter cell (Figures 3 and 4), which is an important indication that the RABV is indeed transsynaptic. For example, the constellation of inputs a single pyramidal neuron and a single interneuron received was very different despite them being located in neighboring regions (distal CA3 and CA2) in the hippocampus ( Figure 3).
However, it is important to consider potential spurious rabies expression that can occur in the absence of Cre around the 2 Cell Reports Methods 2, 100221, May 23, 2022 Article ll OPEN ACCESS injection site (Watabe-Uchida et al., 2012;Miyamichi et al., 2013). This can be due to the presence of unpseudotyped RABVs, which is a possibility given the currently used viral production processes (Watabe-Uchida et al., 2012), and/or leaky, Cre-independent expression of TVA (Miyamichi et al., 2013). The former does not seem to be an issue here ( Figure S1A), and to minimize the effect of the latter, we used the split-ATG strategy (Fischer et al., 2019) when designing the helper virus. We also checked the extent of non-specific labeling by repeating the method described above in animals not injected with Cre. Although some spurious rabies expression was observed in the hippocampus, the number and anatomical , and EnvA-pseudotyped DG-CVS-N2c rabies virus (RABV-GOI, right). EnvA, envelope protein from the subgroup A avian sarcoma and leukosis virus; ITR, inverted terminal repeat; hS, human synapsin promoter; ATG, start codon; blue overlapping triangles, FLEX cassette consisting of loxP and lox2272 sites; N2cG, CVS-N2c rabies glycoprotein; 2A, self-cleaving peptide element; HA, hemagglutinin; TVA, start codon-removed avian-specific receptor; W, woodchuck hepatitis virus post-transcriptional regulatory element; pA, bovine growth hormone polyadenylation signal; CKII, CaM kinase II promoter; N, P, M, and L, genes coding the necessary viral proteins for the RABV; GOI, gene of interest. (B) Timeline of the experimental protocol, which starts when a male and female mouse are introduced for breeding at E0 (embryonic day 0) for up to 24 h. E13: a diluted AAV carrying Cre is injected into one of the lateral ventricles (Lvs) of a mouse embryo under ultrasound guidance, resulting in Cre expression (gray cells in the schematic) in a sparse number of cells in the adult hippocampus. Post-natal day 90 (P90)+: in combination with sparse Cre expression, a small-volume injection of a single rabies helper virus (green) into the hippocampus leads to expression of TVA and G in a single hippocampal cell. A large-volume injection of a G-protein-deleted, EnvA-pseudotyped RABV (magenta) in the same area weeks later (R2 weeks) results in infection of the single TVA-and G-expressing cell (green and magenta cells in the schematic) and spread of the RABV to monosynaptically connected input cells (magenta-only cells) locally in the hippocampus and in distant regions, such as the MEC. P118+: starter and input cells can be distinguished with antibody staining against HA or 2A epitope proteins introduced by the helper virus (i.e., providing a helper tag). Cells that show colocalization of the helper tag and the GOI introduced by the RABV (such as tdTomato) are identified as starter cells, whereas cells only expressing the GOI are identified as input cells. Images are single z-plane confocal images.
Cell Reports Methods 2, 100221, May 23, 2022 3 Article ll OPEN ACCESS spread of rabies-positive cells was significantly lower than in conclusive animals ( Figures 2B and S2A), and spread to the MEC did not occur ( Figure S2B), not even when a larger volume (2.5 times more) and higher titer (10 11 infectious particles/ml) of the helper virus were used ( Figure S1C). These control animals were also inconclusive, suggesting that minimal expression of Cre-independent TVA led to rabies infection of a subset of cells, but not enough to detect them as starter cells (Discussion) or enough G to permit distant spread. These results highlight the importance of including survival-timematched controls to estimate local, Cre-independent background expression, especially in cases where spread to distant input areas is not observed or when those areas are not the main regions of interest.
Targeting single starter cells in the hippocampus highlights granule-granule cell connectivity Keeping these limitations in mind, a striking finding from the animals in which we achieved single starter cell labeling of hip- After 20 days of rabies expression, one granule cell was found to receive input from another granule cell located several hundred micrometers away along the longitudinal axis of the hippocampus ( Figure 4A). After 26 days (in a different animal), the virus had spread to a total of 35 input cells in the hippocampus, 80% of which were other granule cells ( Figure 4B). Even when accounting for background expression, granule cells were still the dominant input type (subtracting the highest number of rabiespositive cells observed in the control animals from only the number of granule cell inputs observed here: (28 observed granule cell inputs -18 background cells)/(35 total observed hippocampal input cells -18 background cells) = 58.8% granule cell inputs). Although the starter cells themselves were located in the middle of the granule cell layer, indicating that they are mature granule cells, a substantial number of their inputs was found in the subgranular zone (1 of 1 granule cell inputs after 20 days, 11 of 28 after 26 days), which suggests that these are immature granule cells. Although this type of granulegranule cell connection was unexpected, weak connections from adult-born to mature granule cells have been described (Drew et al., 2016), and connections going in the opposite direction have been identified previously with rabies tracing (Vivar et al., 2012). Additional studies are necessary to confirm our preliminary findings, but, in combination with the mentioned literature, they suggest that in the case of hippocampal granule cells, RABVs might preferentially target synapses shared with other granule cells.
The functional properties of input cells to birthdatedefined cell populations can be determined with optogenetics in vivo Apart from enabling sparse Cre expression in adult animals, our in utero approach also provides unique opportunities for developmental studies. By varying the time point at which the virus carrying Cre is injected, different populations of birthdatematched cells can be targeted in wild-type animals (Donato et al., 2017). To determine the functional properties of inputs to such cell populations, our protocol can be adapted to (1) increase the number of starter cells and (2) include techniques for measuring neural activity.
The number of starter cells can be adjusted by varying the amount of the Cre-carrying virus injected in utero and the amount of helper virus injected in the adult brain. By increasing both, we were able to target populations of hippocampal cells ( Figure 5A) born on different embryonic days for RABV-mediated tracing (E12, Figures 5B and S1A; E14, Figure 6A). Inputs to these birthdate-defined populations can also be compared with the inputs a general population of hippocampal cells receives by leaving out the in utero injection and introducing the Cre-carrying virus along with the helper virus in the adult brain instead ( Figure S3A). Both approaches led to broad labeling of these populations' input cells in the superficial MEC (Figures 5B, 6A, S1A, and S3A), which we next sought to characterize functionally. To do so, we first tested the feasibility of using in vivo electrophysiology combined with optogenetic tagging. We injected a RABV carrying ChR2 (RABV-ChR2-YFP) in the dorsal hippocampus and implanted tetrodes with an optic fiber attached into the ipsilateral MEC ( Figure 5A), a major input region containing cell types that have been well characterized with in vivo electrophysiology . In these experiments, input cells could be detected in the MEC from approximately 2 weeks after the rabies injection. We started each session by recording singleunit activity in the MEC while the mice explored an open field environment. We found several functional cell types, including grid, head direction, and border cells ( Figures 5C and S4), indicating that the network as a whole was not affected by the widespread rabies expression in the entorhinal-hippocampal network. Next we placed the mice in a separate holding box and shone a 473-nm laser through the implanted optic fiber to determine whether any of these cells responded to laser stimulation; i.e., whether they express ChR2 introduced by the RABV and, therefore, project directly to the hippocampus. Cells that responded robustly to laser stimulation with a 2-to 4-ms latency (demonstrating direct activation as opposed to later,   Figure 5C). This finding confirms a previous study (Zhang et al., 2013) and indicates that rabies expression does not preclude functional identification of input cells. The firing properties of an individual input cell were also found to remain stable for at least 3 days of continuous rabies expression ( Figure S3), and responsive cells with clear spatial firing patterns could be identified for at least 19 days after the rabies injection ( Figure 5C). These results show that it is possible to functionally identify and characterize distant inputs to developmentally defined cell populations through rabies-mediated optogenetic tagging in vivo.
In vivo two-photon calcium imaging enables functional detection of distant input cells to birthdate-defined cell populations Although optogenetics is a powerful way to identify and characterize cells that are monosynaptically connected to birthdatedefined cell populations in the hippocampus, this approach is limited by the number of cells that can be recorded and stimulated at the same time. This is especially true when the labeled input cells are anatomically sparse. This becomes a pertinent issue when using RABVs because it usually takes days or weeks to position tetrodes, and long-term expression of RABVs will eventually lead to deterioration of cell health, which limits the time window for locating cells. We therefore tested the possibility of using calcium imaging to see whether a RABV carrying GCaMP (RABV-GCaMP6f) injected into the hippocampus could be used to identify several input cells simultaneously in the MEC. We employed a similar approach to that described for the optogenetic experiments, except we implanted a prism between the MEC and cerebellum to obtain optical access to the superficial layers of the MEC ( Figure 6A). This enabled us to perform twophoton imaging in awake, head-fixed mice and identify several cells projecting to a population of birthdate-matched cells in the hippocampus ( Figure 6B).
Finally, to extend the time window for imaging experiments and ensure that the calcium signal itself is not altered by overexpression of rabies proteins in the cell, we considered an alternative approach. By first introducing the calcium indicator with a separate AAV in the MEC of Cre-and helper virus-injected animals, dense population labeling can be achieved, and imaging can be performed for as long as required. Then a RABV carrying a red fluorescent tag can be injected into the hippocampus to identify which of the imaged cells project there before the brain is preserved. This also enables activity profiles of targeted input cells to be compared with neighboring cells ( Figure 7A). For these reasons, this approach may be especially useful when attempting to identify inputs to a small number of starter cells to help ensure their identification before their health deteriorates because of RABV overexpression. When following our protocol for targeting a single hippocampal starter cell, we injected an AAV carrying GCaMP6m (AAV-Syn-GCaMP6m) into the MEC and implanted a gradient-index (GRIN) lens and prism doublet during the same surgery as the helper virus injection ( Figure 7A). By keeping the craniotomy used for the helper virus injection free from cement and temporarily closing it with a silicone sealant, we could reopen it several weeks later to introduce the RABV carrying tdTomato (RABV-tdTomato). When the RABV reached the MEC, we were able to identify a cell expressing both the calcium indicator and tdTomato in vivo with a two-photon miniscope ( Figure 7B; Zong et al., 2017Obenhaus et al., 2022), demonstrating the feasibility of using this approach to characterize the spatial firing properties of input cells in freely moving mice.

DISCUSSION
We developed a virus-based method that allows individual neurons to be targeted for monosynaptic rabies tracing in deep brain structures without the need for tissue aspiration. By applying this method to the hippocampus, we identify local and distant inputs to single cells. We also demonstrate how this method can be used to determine the functional properties of monosynaptic inputs to birthdate-defined cell populations with in vivo electrophysiology combined with optogenetic tagging, or two-photon calcium imaging. This method achieves a success rate similar to that of existing approaches (few studies report the overall success rate, but the 12% [successful transsynaptic spread in 14 of 119 electroporated cells] attained by Marshel et al. (2010) seems representative). However, it also offers several additional and unique advantages. First, it allows rabies-mediated circuit mapping to be extended to single cells in subneocortical regions without the need for removal of overlaying tissue. Second, the in utero viral injection approach enables cells born at specific times to be targeted and, hence, identification of inputs to cells with different developmental origins to be compared (in utero electroporation can also be used here, although this requires an additional electrical pulse delivery step). Third, the use of viruses allows specific cell types, dividing cells, or clones to be targeted by utilizing cell-type-specific gene-regulatory elements (Nair et al., 2020) or different virus types (such as retroviruses) during embryonic development (with the Cre-carrying virus) or in adult animals (with the TVA-and G-carrying helper virus). Fourth, by simply varying the concentration and/or amount of virus used, any number of cells can be targeted for input mapping, from single neurons to populations of cells. Fifth, it can be used in wildtype animals, circumventing the need to generate and breed transgenic lines. Sixth, in theory, this approach can be used in any species in which these viruses express and in utero/early postnatal injections are possible.
However, although modified RABVs are currently the only tools available for large-scale monosynaptic input tracing in vivo, they also have disadvantages that need to be considered. Specifically, though the CVS-N2c strain used here is less toxic than Article ll the SAD-B19 strain (Reardon et al., 2016), it does eventually lead to death of the infected cells, a constraint it shares with all RABVbased approaches regardless of how cells are targeted. Although the virus needs time to spread to distant input regions (e.g., to the MEC from the hippocampus; Figure S2B), waiting too long will result in cytotoxicity of the initially infected cells (Figure 7A) and, hence, an inability to determine the number of starter cells. From our experience, it appears that the rate at which this occurs differs based on the concentration and amount of virus that the cells are exposed to (Lavin et al., 2020), their cell type, and the circuit in which targeted cells are embedded. However, with recent bioengineering efforts, the limitation of rabiesmediated cytotoxicity can now be overcome and long-term functional studies enabled, though additional viruses and/or transgenic animals are required to achieve this (Ciabatti et al., 2017(Ciabatti et al., , 2020Chatterjee et al., 2018). Another limitation, also inde-pendent of the targeting approach, is the lack of knowledge regarding how RABVs spread to input cells. The number of input cells that project to a single cell remains subject to estimates but is likely to be substantially higher than what has been reported in single-cell rabies tracing studies (typically around 100 input cells and up to 846 in the visual cortex; Wertz et al., 2015). Although time is a major determinant of how many input cells are labeled, the relatively low number typically seen raises the question of whether the virus exhibits a bias to cross synapses of certain types, strengths, or activity profiles. This remains an open question because a detailed ''ground truth'' connectivity map to which rabies tracing results can be compared does not yet exist. The question of whether transmission of RABVs is strictly synapse specific is also under debate. It is therefore advisable to use RABV tracing for comparative and explorative studies rather than aiming for complete quantification of all monosynaptic The inset is a single z-plane confocal image highlighting two example cells (arrows) near the tetrode track. (C) The functional identity of cells was determined in an open field with a rectangular cue card on one wall while the mouse foraged for cookie crumbs. Afterward, the mouse was placed in a separate holding box for laser stimulation trials to identify cells that express ChR2, which means they provide monosynaptic input to the hippocampus. Corresponding path and laser stimulation plots from three example cells that show robust and reliable responses to laser stimulation (blue vertical line) are shown. These include a grid cell (left, 19 days after rabies injection), a high-firing-rate cell (center, 17 days after rabies injection), and a low-firingrate cell (right, 17 days after rabies injection). Gray lines indicate the path of the mouse in the open field, and black dots indicate locations (path plots) or timing relative to the laser stimulation (laser stimulation plots) at which units from an individual cell were recorded. Right-centered values above path plots show the average firing rate for the open field session. T, tetrode number; C, cell ID; L, latency; F, fidelity. See also Figures S3 and S4 Rogers and Beier [2021] for a comprehensive review of these limitations). This point becomes particularly pertinent when using Cre-dependent viruses because leaky expression of TVA can result in spurious rabies expression in cells that (combined with the staining issues described below) might be wrongly identified as input cells ( Figure 2B; Miyamichi et al., 2013). Importantly, distant input cells should also be checked for the presence of helper tags because some AAV serotypes infect cells via axons (see e.g., Aschauer et al., 2013). Background expression because of the potential presence of unpseudotyped RABVs, on the other hand, shows promise of being eliminated with the use of new cell lines (Sumser et al., 2021).
The success rate of our method is chiefly determined by the ability to achieve satisfactory labeling of starter cells. In roughly half of the animals (Figure 2A), we could not determine how many starter cells were present because the antibody staining of helper tags was ambiguous. The reason for this could be (1) death of starter cells or (2) staining issues (e.g., limited Cre availability resulting in few recombination events and/or inadequate amounts of helper proteins for our detection method). The latter explanation seems more likely for several reasons. First, we typically did not see cell damage until more than 30 days after the rabies injection when attempting to target single cells ( Figure 7A). Second, we limited the amount of Cre and helper virus used to keep the number of starter cells low, which could result in reduced transgene expression from the helper virus (low expression levels of G could also explain the relatively low number of input cells observed). Third, animals that went through the same protocol but were not injected with the Cre virus ( Figures 2B and S2) or not injected with a RABV (n = 4 animals; data not shown), also showed inconclusive staining of helper tags. Fourth, antibody staining for helper proteins is clearer and easier to detect in animals injected with Cell Reports Methods 2, 100221, May 23, 2022 9 Article ll OPEN ACCESS larger amounts of Cre and helper virus (Figures 5, 6 and S3). Fifth, the spread of the RABV over time followed a similar trend for the inconclusive and conclusive cases ( Figures 2B and S2), indicating that the overall numbers of starter cells were similar in both groups (a notable exception is the inconclusive cases, where a low number of local rabies-positive cells is present despite long survival times; these likely represent cases of background expression because they seem to overlap with the Creindependent controls; Figure 2B). Sixth, based on the relatively low number of input cells observed, it seems unlikely that we are underestimating the number of starter cells. If anything, based on the low number of input cells we see, the number of starter cells should be even lower than what we report. Finally, antibody staining to detect starter cells is known to be challenging because highly efficient interactions between the EnvApseudotyped RABV and its receptor TVA enable minimal expression of the latter for infection (Federspiel et al., 1994;Seidler et al., 2008). As a result, direct detection of the receptor or asso-ciated protein expression for starter cell identification can be difficult to achieve (Hafner et al., 2019;Lavin et al., 2020;Faulkner et al., 2021). Development of more sensitive and/or novel antibodies (for example, against TVA or G specifically) will therefore likely improve the success rate of the present method and rabies tracing studies in general.
Rabies-mediated tracing from single cells is a powerful technique to map monosynaptic inputs, which is necessary to understand how neurons integrate information. With the present in-utero-based virus injection method, rabies tracing can be achieved from individual cells located far beneath the brain's surface. It thus extends detailed circuit mapping and functional characterization of local and distant inputs to deeper brain structures and to cells with defined developmental origins.
Limitations of the study Although our method can target single cells in subneocortical regions for input mapping, it relies on RABV-mediated tracing and Figure 7. A dual labeling strategy facilitating long-term two-photon calcium imaging of distant input cells to birthdatedefined cell populations in vivo (A) Schematic of the experimental design and resulting histology 34 days after rabies injection. Top left: a population of birthdate-matched (E13) starter cells in the dorsal hippocampus is targeted to express TVA and G using the method described in Figure 1 (i.e., the number of starter cells is smaller than in Figure 6). In addition, an AAV virus carrying GCaMP6m is injected into the MEC (brain region marked in light gray) for ubiquitous expression of the calcium indicator. A RABV carrying tdTomato injected into the hippocampus and a GRIN lens with a prism attached implanted by the MEC can then be used to identify cells in the MEC that provide monosynaptic input to the hippocampus. This approach enables cells providing input to the targeted starter cells, and neighboring cells to be compared before (avoiding the issue of rabies-mediated toxicity) and after rabies has been introduced. Bottom left: confocal maximum-intensity projection of the dorsal hippocampus, showing cells expressing tdTomato. The image inset shows a magnification of the region indicated by the white box, showing dendritic blebbing 34 days after rabies injection. Right: confocal maximum-intensity projection showing general GCaMP6m expression in the MEC and hippocampal input cells expressing tdTomato. Image insets are single z-plane confocal images showing colocalization of both markers in one example cell (i.e., an input cell also expressing GCaMP6m, highlighted with an arrow and yellow dotted circle). The white dotted line shows the impression made by the GRIN lens and prism on the MEC. (B) In vivo imaging example. Left: imaging is performed in head-fixed mice with a two-photon miniscope. Center: maximum-intensity projection and mean projection (over time) of the GCaMP6m and tdTomato channels, respectively, acquired 29 days after the rabies injection. Both cells express GCaMP6m, but only one expresses tdTomato, meaning it provides monosynaptic input to the starter cell(s) targeted in the hippocampus. Right: DF/F GCaMP6m traces of the two cells highlighted with corresponding colors on the left. in utero ventricular injections. The latter limits the cells that can be targeted to those that line the ventricles at the time of injection. Although that does include cortical cells types, existing methods to target these cells, such as single-cell electroporation (Marshel et al., 2010;Wertz et al., 2015;Rossi et al., 2020), patch clamping (Rancz et al., 2011;Vé lez-Fort et al., 2014) or stamping (Schubert et al., 2017) may be preferable. This is because reaching single cells in more readily accessible cortical areas involves minimal tissue damage and does not require embryonic injections. Our in utero protocol does, however, also enable our method to be used for developmental studies.
Rabies-mediated tracing itself carries several limitations (Discussion), and here unambiguous detection of starter cells is key. This is currently limited by spuriously leaky viral vectors and low antibody sensitivity, which can lead to Cre-independent expression of TVA at levels undetectable by current antibodies and result in ''invisible'' starter cells. Although these cells do not enable spread of the RABV to distant input regions, development of tighter viral vectors and more sensitive and/or novel antibodies against helper proteins will be important to improve rabies tracing studies in general and the present method in particular.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

ACKNOWLEDGMENTS
We thank S. Ball, A.M. Amundsgå rd, K. Haugen, K. Jenssen, E. Krå kvik, I. Ulsaker-Janke, and H. Waade for technical assistance; M.P. Witter for help with identifying cell types; and C. Kentros for discussions regarding the experimental protocol. We are also grateful to Tom M. Jessell and his lab at Columbia University for gifting us the essential reagents and cell lines to produce CVS-N2c rabies strain viruses when the project started in 2016. We are also thankful to Weijian Zong for providing us with a two-photon miniscope for recording GCaMP fluorescent activity in rabies virus-injected mice. In addition, we want to thank Scidraw.io for making high-quality drawings available, including contributions by Ann Kennedy of a mouse profile (

INCLUSION AND DIVERSITY
We worked to ensure sex balance in the selection of non-human subjects. One or more of the authors of this paper self-identifies as a member of the LGBTQ+ community.

RESOURCE AVAILABILITY
Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Edvard I Moser (edvard.moser@ntnu.no).

Materials availability
The helper and rabies viruses used in this paper are available from the Viral Vector Core at the Kavli Institute for Systems Neuroscience, Norwegian University of Science and Technology (NTNU) in Trondheim, Norway.
Data and code availability d Electrophysiology data used in this manuscript have been deposited online (The DANDI Archive: https://doi.org/10.48324/ dandi.000230/0.220506.1516). d This paper does not report original code. d Any additional information required to reanalyse the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS
Experiments were performed in accordance with the Norwegian Animal Welfare Act and the European Convention for the Protection of Vertebrate Animals used for Experimental and Other Scientific Purposes (permit numbers 6021, 7163 and 18013). C57BL/6JBomTac (Cat#B6JBOM; RRID: IMSR_TAC:b6jbom, Taconic) mice were housed in temperature-and humidity-controlled cages with sex-matched siblings when possible with food and water provided ad libitum. A reversed 12h light/12h darkness schedule was maintained throughout. Both sexes were used, which did not seem to influence the results (Tables S1 and S2), at embryonic and/ or adult developmental stages.

In utero injections
Two females and one male were put in the same cage for 24 h. Embryonic day 1 (E1) was defined as the day on which the male was removed. Females that were suspected to be pregnant were anaesthetised with isoflurane (IsoFlo, Zoetis, 5% isoflurane vaporised in medical air delivered at 0.8-1 L/min for induction, then 1.5%-3%) and the fur covering the abdomen was removed with hair removal cream for sensitive skin (Veet, Canada). If pregnancy was confirmed with ultrasound (Vevo 1100 System with MS-550S probe, Fujifilm Visualsonics) the mouse was prepared for surgery by injecting an analgesic subcutaneously (Metacam, Boehringer Ingelheim, 5 mg/ kg) and placing her on a heated surgery stage (37 C). Midline incisions were made in the skin and peritoneum, and warm (37-39 C) saline (NaCl 9 mg/ml, B. Braun Medical) was used throughout to keep the abdominal cavity warm and wet. Embryos were carefully transferred either one by one or in pairs to a decontaminated chamber placed on top of the female's abdomen and covered with warm ultrasound gel (37-39 C, Aquasonic 100, Parker). One lateral ventricle in all embryos in both uterine horns were injected with AAV-CKII-Cre (AAV1.CamKII0.4.Cre.SV40, Cat#AV-1-PV2396, titre: 2.71e13 GC/ml, UPenn Vector Core, University of Pennsylvania, USA), diluted when applicable (see Tables S1 and S2) in sterile DPBS (1x Dulbecco's Phosphate Buffered Saline, Gibco, ThermoFisher), with a pressure-pump (Visualsonics) under ultrasound guidance. Custom made glass micropipettes (outer tip opening: 200 mm, inner tip opening: 50 mm, Origio) were used to inject the virus in six pulses of 50.6 nL each. After the pipette had been retracted, the embryos were wiped clean with sterile gauze and gently returned to the abdominal cavity. Once injections were complete, the peritoneum and skin were sutured (Supramid DS 13, Resorba Medical, Germany) separately. The entire procedure, from first incision to last stitch, was limited to maximum 90 min to minimise stress to the animals. Recovery took place in a heated chamber (33 C) until mobility and alertness was restored.

Adult injections
Once in utero injected embryos had reached adulthood (minimum 90 days after birth), they were prepared for the second virus injection. After being anaesthetised with isoflurane (IsoFlo, Zoetis, 5% isoflurane vaporised in medical air delivered at 0.8-1 L/min) two analgesics were provided subcutaneously (Metacam, Boehringer Ingelheim, 5 mg/kg or Rimadyl, Pfizer, 5 mg/kg, and Temgesic, Indivior, 0.05-0.1 mg/kg) and one local anesthetic underneath the skin covering the skull (Marcain, Aspen, 1-3 mg/kg). The fur on the top of the head was then shaved and the mouse placed in a stereotaxic frame, the isoflurane level adjusted to 1%-2% and eye ointment applied to prevent dry eyes (Simplex, Actavis). Next the skull was exposed and a small hole drilled (with Cat#1RF HP 330 104 001 001 005 drill bit from Hagen and Meisinger, Germany) at 1.55 mm lateral from the midline and 1.58 mm posterior from bregma over the right hemisphere. The same coordinates were used to inject the helper virus (AAV-hS-FLEX-TVA-HA-N2cG, made in house) into the dorsal hippocampus at 1.55 mm depth from the dura with a pressure-pump (Nanoject, Visualsonics) and pipettes (Cat#504949, WPI) pulled and cut to produce an extended tip with 20-50 mm outer diameter. The required volume (see Tables S1 and S2) was injected at a rate of 23 nL/s and the pipette left for 10 min before retracting. The exposed brain was kept wet with saline (NaCl 9 mg/ml, B. Braun Medical) throughout. The craniotomy was then covered with UV curable cement (Venus Diamond Flow, Kulzer) and the skin sutured (Supramid DS 13, Resorba Medical, Germany). The third virus injection was made at least two weeks later using the same procedure (after removing the cement-cover with forceps and reopening the craniotomy with a drill if necessary) except 180-560 nL of the modified rabies virus (EnvA-pseudotyped DG-CVS-N2c RABV-GOI, see ''Rabies virus (RABV-GOI)'') was injected across 1.7-1.40 mm below the dura over at least 10 min. Recovery took place in a heated chamber (33 C) until mobility and alertness was restored. A second subcutaneous injection of the analgesic Metacam (Boehringer Ingelheim, 5 mg/kg) or Rimadyl (Pfizer, 5 mg/kg) was provided under isoflurane anesthesia once per day for 1-2 days after surgery.
Animals not following the 'single cell experiment' protocol underwent the same procedure but modified to suit the particular experiment as indicated in the text, figure legends and/or Table S2 (e.g. only one injection and/or different viruses, including AAV1-CAG-FLEX-tdTomato (titre: 5.88e12 GC/ml, Cat#AV-1-ALL864, UPenn Vector Core, University of Pennsylvania, USA) in Figure S1B).

Optrode construction
Four 17 mm polyimide-coated platinum/iridium (90/10%) wires (California Fine Wire, CA) were twisted to create one tetrode. Four tetrodes were wired to a 16 channel microdrive (Axona Ltd., Herts, UK) via a connected 22G metal cannula encased in an 18G outer cannula (with one end cut and polished smooth to accommodate the brain surface curvature during implantation) and cut to length with sharp scissors. A 100 mm in diameter optic fibre (Cat#MFC_100/125-0.37_17 mm_ZF1.25-C45, Doric Lenses) was e3 Cell Reports Methods 2, 100221, May 23, 2022 Article ll glued to the anterior side of the tetrodes with the tip approximately 200 mm above the tetrode tips. Electrode impedance was reduced to between 183 and 220 kOhm by plating the electrode tips with platinum at 1 kHz and À0.17 mA using a NanoZ device (Neuralynx, Bozeman, USA).

Optrode implantation
One E12 in utero AAV-CKII-Cre (no dilution) injected mouse and one that had the Cre virus introduced with the AAV-hS-FLEX-TVA-HA-N2cG virus (mixed 1:1, total volume 303.6 nL) at the adult stage were used. Virus injections were performed as described under ''Adult injections'' with the rabies virus carrying ChR2 and YFP (RABV-ChR2-YFP). An optrode was also implanted in the MEC in the ipsilateral (right) hemisphere in the same surgery as the rabies injection as follows: after exposing and scratching the skull with a needle to enhance skull-dental cement adherence, one ground screw with a short wire soldered on (which was then soldered to the optrode at the end/after the surgery) was attached to the occipital bone and one extra screw to the parietal bone, both in the left hemisphere. Tissue adhesive (Histoacryl, B. Braun Surgical) was applied to secure the screws and to attach the skin edges to the lateral edges of the skull. A circular craniotomy was made over and anterior to the transverse sinus in the right hemisphere. An area of the dura slightly bigger than the optrode to be implanted was removed anterior to the sinus. The optrode was implanted 3.5 mm lateral from the midline, 0.4 mm anterior to the sinus and at a 6 degree angle (pointing posteriorly) down to 0.9-1 mm depth below the dura. The outer cannula was gently lowered to touch the dura surrounding the implant. The implant was secured to the skull with UV curable dental cement (Venus Diamond Flow, Kulzer), and dental cement (Paladur, Kulzer) mixed with carbon (CAS number 7440-44-0, Sigma-Aldrich) to reduce transparency and light reflections. 0.5 ml saline (NaCl 9 mg/ml, B. Braun Medical) warmed to body temperature was administered 1-2 times during the procedure to maintain hydration. Recovery took place in a heated chamber (33 C) until mobility and alertness was restored. A second subcutaneous injection of the analgesic Temgesic (Indivior, 0.05-0.1 mg/ kg) was provided approximately 8 h after the first and Metacam (Boehringer Ingelheim, 5 mg/kg) or Rimadyl (Pfizer, 5 mg/kg) was readministered once per day for at least two days after the surgery, all under isoflurane anaesthesia.
In vivo electrophysiology Animals implanted with an optrode were recorded up to two times per day. Before each recording session the microdrive was connected to a digital acquisition system (Axona) to record spikes, and two LEDs were attached to the drive to enable tracking of position and head direction with a tracking system (Axona). Signals from the tetrodes were amplified 5 000-10 000 times and a 0.3-7 kHz bandpass filter was applied. Individual spikes were stored at 48 kHz (8 bits/sample and 50 samples per waveform) with a 32-bit time stamp (96 kHz clock rate). LED position data were recorded at 50 Hz. Each session consisted of two parts. First, the mouse explored an open field environment (1 3 1 3 0.5 m black box with a white, rectangular cue card fixed to one of its walls) for approximately 30 min. Cookie crumbs were thrown into the box throughout until (s)he was satisfied and stopped eating. Second, the mouse was placed in a plexiglass holding box (20 3 25 3 15 cm, lined with a cotton towel) outside and above the open field arena. In this second part, a 473 nm laser (Cobolt 06-MLD, output power max: 100 mW, H€ ubner Photonics) was connected to the implanted optic fibre via a patch cable (Doric Lenses, Canada) after the power had been adjusted to deliver 7.5-15.14 mW at the end of the patch cable. TTL pulses sent from an Arduino Uno Rev3 (Arduino, Italy) controlled the on/off state of the laser to deliver 1 ms light pulses at 10 Hz continuously. The same TTL was fed in parallel into a spare digital input channel of the Axona acquisition system for precise synchronisation of laser stimuli and electrophysiological data. At the end of each recording session the tetrodes were turned down by 25-50 mm (except when trying to record from the same cell across multiple sessions, Figure S3). To encourage exploration in the open field environment, animals were put on a food restriction schedule outside of the recording sessions. Their weight was monitored daily and on average was maintained at 96% of their baseline weight.

Prism implantation with RABV-GCaMP6f
To visualise distant inputs located in the MEC in vivo, the protocol illustrated in Figure 1 was modified as follows: during the in utero injection (as described under ''In utero injections''), the AAV-CKII-Cre virus was not diluted and when the AAV-hS-FLEX-TVA-HA-N2cG virus was injected (as described under ''Adult injections'') a larger volume was used (Table S2). This resulted in a large population of starter cells in the hippocampus. In the same surgery as the rabies virus injection (as described under ''Adult injections'' except here RABV-GCaMP6f was used, see ''Rabies virus (RABV-GOI)''), the skull was exposed, the rabies virus injected, and tissue adhesive (Histoacryl, B. Braun Surgical) was applied to attach the skin edges to the lateral edges of the skull. Next, the skull was scratched with a needle to enhance skull-dental cement adherence and a large, circular craniotomy over and behind the transverse sinus in the right hemisphere was made. The dura connecting the sinus to the cerebellum was carefully cut along the posterior side of the sinus before a 2 mm square microprism with reflective enhanced aluminium coating on the hypotenuse (Tower Optical, USA) glued (Norland Optical Adhesive 71, Norland Products) to a 4 mm in diameter circular coverslip (64-0724, Warner Instruments) was inserted immediately posterior to the sinus by hand. No brain tissue was removed. A thin, metal cannula attached to the stereotaxic frame was used to hold the prism in place after insertion while the coverslip was secured to the skull with UV curable cement (Venus Diamond Flow, Kulzer). Dental cement (Paladur, Kulzer) mixed with carbon (CAS number 7440-44-0, Sigma-Aldrich; to reduce transparency and light reflections) was used to attach a custom-made head bar and to cover the rest of the skull. 0.9 ml saline (NaCl 9 mg/ml, B. Braun Medical) warmed to body temperature was administered subcutaneously at the end of the procedure for hydration. Recovery took place in a heated chamber (33 C) until mobility and alertness was