Integrating physiological and transcriptomic analyses at the single-neuron level

.


Introduction
Over the past half-century, a large number of neurophysiological studies have revealed how the cooperative spike patterns of diverse neuronal ensembles control brain functions such as learning, behavior, and sensory processing in living animals.These insights have progressed further owing to recent advancements in in vivo large-scale recording techniques using genetically encoded molecular sensors (e.g., the GCaMP series) (Nakai et al., 2001;Zhang et al., 2023) and recording devices (e.g., multichannel electrode probes and wide-field optical microscopes) (Terada et al., 2018;Steinmetz et al., 2021).Together, these neurophysiological studies have demonstrated that spike properties relevant to behavior vary prominently among individual cells.
In addition, numerous studies from molecular biology have elucidated the gene expression and molecular characteristics of diverse neuronal types in the brain (Graves et al., 2012;Thome et al., 2014;Cembrowski et al., 2016;Habib et al., 2016;Saunders et al., 2018;Cembrowski and Spruston, 2019).Furthermore, advancements in single-cell transcriptome analyses have revealed the comprehensive gene expression profiles of individual neurons, as exemplified by the BRAIN Initiative Cell Census Network projects (BICCN, 2021).Furthermore, spatial transcriptome analysis is beginning to be widely adopted as a powerful tool to map the spatial distribution of gene expression patterns within specific brain tissue regions (Rodriques et al., 2019;Stickels et al., 2021).
Previous physiological and molecular biological studies have progressed independently over this half-century.Therefore, there are few studies that have integrated these independent insights, specifically addressing how behavior-relevant neuronal spike patterns are supported by complex molecular mechanisms.The next major research challenge is achieving integration of these insights to clarify how the spike patterns of individual neurons are associated with their genetic and molecular characteristics.To address this issue, experimental techniques allowing the physical collected of neurons (or their cytoplasmic contents) for which physiological activity has been measured are required.In this review, we systematically summarize the results of several pioneering studies that have addressed these methodological requirements by combining physiological recordings and transcriptomic analyses, while avoid recapitulating the general descriptions already present in numerous transcriptome analyses studies.

Electrophysiological recordings and gene expression analysis
The first experimental approach, which aims to link physiological and molecular characteristics, utilizes electrophysiological recordings with high temporal resolution (tens of kilohertz) and high signal-tonoise ratios of spike unit signals (often greater than 10).As each electrophysiological recording using a recording pipette can generally target only one neuron, the throughput of cell sampling for genetic analyses is limited.Here, we describe two major methods: patch-clamp recordings and juxta-cellular recordings.While each method has various advantages and disadvantages, as detailed in the subsequent paragraphs, the adoption of either method heavily relies on the background and environment of the laboratory as these recordings under in vivo conditions require considerable expertise.

Patch-clamp recording
Initial ideas to link the physiological characteristics of single neurons with their gene expression involved performing patch-clamp recordings and PCR assays on in vitro cell samples (Lambolez et al., 1992;Sucher and Deitcher, 1995) (Fig. 1A).The methodological procedures involved obtaining a whole-cell patch-clamp configuration from a neuron using a glass pipette, and subsequently sampling the cytoplasmic content of the neuron by applying negative pressure to the pipette.Cell samples were then subjected to PCR to quantify gene expression (Chiang, 1998).Such combinations of patch-clamp recordings and PCR analyses facilitate a deeper understanding of the electrophysiological properties associated with gene expression patterns encoding diverse receptors (Lambolez et al., 1992(Lambolez et al., , 1996;;Grigorenko and Yeh, 1994;Poth et al., 1997) and ion channels (Gurantz et al., 1996;Plant et al., 1998;Koizumi et al., 2004;Theriault and Chahine, 2014) within a single neuron.These methods have been further refined using quantitative PCR techniques targeting multiple genes (Veys et al., 2012;Tricoire et al., 2019), contributing to the detailed classification of inhibitory neuronal types (Perrenoud et al., 2022), the identification of genes related to long-term potentiation (Friend et al., 2019) and epileptic conditions (Kirchheim et al., 2013).
In the 2010s, RNA sequencing (RNA-seq) techniques using nextgeneration sequencing became widespread in the field of molecular biology, allowing comprehensive analysis of the expression profiles of thousands of genes.Following this development, RNA-seq technologies have been applied to neuronal samples collected by patch-clamp recordings, a process termed patch-seq (Cadwell et al., 2016;Chen et al., 2016;Fuzik et al., 2016), to reveal the relationships between diverse electrophysiological properties and gene expression profiles in patch-clamped neurons (Kalmbach et al., 2021;Paraskevopoulou et al., 2021;Mouradian et al., 2022;van den Hurk et al., 2022;Chartrand et al., 2023;Lee et al., 2023).In principle, patch-seq could be applied to living animals using in vivo patch-clamp recordings (Cadwell et al., 2016;Yagishita et al., 2020;Lipovsek et al., 2021;Nishimura et al., 2021;Noguchi et al., 2021) (Fig. 1B).The most challenging aspect of integrating in vivo patch-clamp recordings with RNA-seq is to maintain stable whole-cell recording conditions for several minutes, as required for collecting an adequate amount of cytoplasm for RNA-seq.

Juxtacellular recording
We recently developed another electrophysiology-based experimental technique, an in vivo juxtacellular recording, to identify the gene expression profiles of single neurons, whose spike patterns were recorded from a head-fixed animal (Yagishita et al., 2024) (Fig. 1C).A juxtacellular recording technique cannot capture subthreshold membrane potential changes but can record spike activity of a single neuron and further label the recorded neuron with fluorescent dyes through electroporation (Pinault, 1996;Oyama et al., 2013;Dempsey et al., 2015).In this extracellular recording technique, the cytoplasmic contents of a recorded neuron cannot be directly sampled from a recording pipette as the pipette is located outside the neuron.Therefore, the labeled neuron needs to be identified on acute brain slices after sacrificing the animals and the labeled neurons are collected using a glass pipette, similar to patch-clamp recordings, for RNA-seq.The advantages of our method are as follows; (1) glass pipettes for juxtacellular recording can more easily access deeper brain regions, compared with brain regions that are accessible by optical imaging and patch-clamp recordings, (2) juxtacellular recordings are applicable to freely moving animals (Herfst et al., 2012;Tang et al., 2014), which may further uncover how the spike patterns of behavior-relevant neurons (i.e.neuronal correlates) are associated with their comprehensive gene expression profiles, and (3) depending on recording conditions, juxtacellular recordings can capture extracellular local field potential oscillations, as well as spike unit Fig. 1.Sampling of electrophysiologically-recorded neurons.(A, B) A patch-clamp recording is obtained from a neuron using a glass pipette, and the RNA content of the neuron is collected from an ex vivo slice (A) or from a living animal (B).(C) A juxtacellular recording is obtained from a neuron using a glass pipette and the neuron is labeled with a fluorescent dye in a living animal.Acute slices are subsequently prepared from the brain and the juxtacellularly labeled neuron is collected using a glass pipette.
H. Yagishita and T. Sasaki signals, enabling the analysis of how spike trains are entrained by the corresponding oscillatory activity of neuronal ensembles (e.g., theta oscillations).

Optical imaging and gene expression analysis
The second approach utilizes optical imaging.Recent advancements in imaging tools, such as genetically encoded sensors and large-scale optical microscopes, have enabled the simultaneous visualization of the physiological activity of tens to thousands of neurons.Along with these advancements, new techniques have been developed for the transcriptomic analysis of neurons whose activity patterns have been imaged.

Calcium imaging and single cell sampling
In vivo calcium imaging using two-photon microscopy and fiber photometry is a useful method to monitor the activity of distinct neuronal ensembles in the living brain.To examine the gene expression profiles of neurons imaged in vivo, a direct approach involving the physical collection of a neuron by suctioning within an imaging field in vivo can be applied (Fig. 2A).The cell samples are then analyzed by single-cell RNA-seq, as described in the electrophysiology section.This technique has been applied to the mouse visual cortex to identify several genes that differentiate between responsive and non-responsive neurons to visual stimuli in vivo (Liu et al., 2020).
Another approach is to perform single-cell sampling on ex vivo slices prepared from the brains of living animals that have been monitored by calcium imaging.This method requires sophisticated techniques, including slicing the brain tissue at an angle that perfectly matches the plane as the imaging field and registering identical neurons in the slices with those imaged in vivo.Phototagging is a useful method to mitigate this technical challenge; in this method, a subset of neurons imaged in vivo is labeled with a fluorescent molecule distinct from that used for calcium imaging.For example, after two-photon GCaMP6s calcium imaging of neurons at a wavelength of 950 nm, the imaged neurons are labeled with histone-bound photoactivatable GFP (H2BpaGFP) by applying a pulsed laser at a wavelength of 750 nm (Pfeffer and Beltramo, 2017) (Fig. 2B).Phototagged neurons can be easily visualized on ex vivo acute brain slices, allowing the cell content to be harvested using a patch-clamp pipette for subsequent single-cell RNA-seq.

Voltage imaging and single cell sampling
Although calcium imaging from neuronal populations can monitor suprathreshold spike patterns, it is generally unsuitable for the measurement of subthreshold changes in membrane potentials representing synaptic inputs.Genetically-encoded voltage indicators with higher signal-to-noise ratios have been developed for optical imaging of subthreshold dynamics in neurons (i.e., voltage imaging) (Mollinedo-Gajate et al., 2021).Recently, a new method that combines voltage imaging, single-cell sampling, and RNA-seq, termed voltage-seq, has been proposed (Csillag et al., 2023).Similar to calcium imaging, hundreds of neurons are initially imaged with voltage indicators on ex vivo slices, and the neurons of interest identified on-site are subsequently harvested using a patch-clamp pipette for RNA-seq.Using this method, the gene expression patterns of periaqueductal gray neurons can be identified, which are related to their subthreshold dynamics in response to the activation of upstream ventromedial hypothalamic terminals.While the imaging in this study is performed on ex vivo slice preparations, voltage imaging is now feasible in living mice (Kannan et al., 2018;Villette et al., 2019).In the future, it may be possible to identify the transcriptome profiles of neurons subjected to in vivo voltage imaging, similar to in vivo calcium imaging.

Calcium imaging and spatial transcriptomics
Spatial transcriptomics with fluorescent in situ hybridization (FISH) is a powerful method for mapping gene expression patterns at the singleneuron level by preserving the histological positions of neurons in brain sections (Jung and Kim, 2023).Recently, several pioneering studies have succeeded in performing post hoc spatial transcriptomic analyses of neuronal ensembles subjected to in vivo calcium imaging (Fig. 2C) (Xu et al., 2020;Bugeon et al., 2022;Condylis et al., 2022), enabling extensive comparison of the functional and molecular characteristics of particular neuronal ensembles in living animals.This combination requires highly advanced techniques that match the spatial distribution of neuronal ensembles in brain sections to those imaged in vivo.These alignment and registration procedures can be performed either manually (Xu et al., 2020) or automatically (Bugeon et al., 2022;Condylis et al., 2022), with assistance by several key landmarks in the imaging field, such as the distinctive structures of blood vessels and ventricles, fluorescent beads injected into the tissue, and sparsely labeled cells.
The first study to apply this cutting-edge technique was conducted by Xu et al. (2020), who demonstrated that the expression level of Npy1r, a gene encoding neuropeptide Y receptor type 1 in neurons in the H. Yagishita and T. Sasaki paraventricular nucleus of the hypothalamus, was associated with unique calcium activity patterns in vivo.Using similar methodological approaches, Condylis et al. (2022) revealed that layer 2/3 excitatory neurons expressing Baz1a in the primary sensory cortex process information related to tactile feature selectivity observed in vivo.Bugeon et al. (2022) further revealed that 35 subtypes of interneurons identified by combinatorial padlock-probe-amplified (coppa) FISH in the primary visual cortex exhibit distinct calcium activity patterns depending on different brain states in vivo.As the combination of in vivo calcium imaging and spatial transcriptomics continues to be an innovative technique, its success is expected to yield additional novel insights into various brain regions.
One limitation of spatial transcriptomics is its limited ability to analyze only tens of targeted genes.However, this technical issue may be resolved by recent rapid advancements that have enabled the analysis of a growing number of target genes.In particular, new sequencingbased platforms such as Visium (10X genomics) and GeoMx (nano-String) can reveal the expression of thousands of genes (Wang et al., 2023) from a single tissue sample.Through these innovative platforms, spatial transcriptome data may be directly correlated with in vivo calcium imaging data at the single neuron level.Another issue with spatial transcriptomics is the cost.Although this challenge may not be immediately addressed, we anticipate the further development of more cost-effective methods in the future.

Other approaches using bulk sorting of fluorescently labeled neurons
In addition to the methods for directly recording neuronal activity described in the previous sections, there are useful techniques for cell labeling with specific molecular markers by utilizing genetic tools.Compared to the recording methods described above, these cell labeling techniques offer the advantage of sampling larger numbers of neurons simultaneously and are relatively easier to handle since they do not require advanced specialized skills and recording equipments (e.g.complex surgery, recording, and cell identification).These genetic methods include viral vectors and transgenic animals that can specifically label active neuronal populations with fluorescent proteins such as green fluorescent protein (GFP) or its variants.In addition, several genetic cell-labeling methods have been developed to specifically express fluorescent proteins in certain neuronal populations, only during periods of activation.After collecting labeled neurons from bulk brain tissue by fluorescence-activated cell sorting (FACS), they can be analyzed by single-cell RNA-seq.In these approaches, it is crucial to confirm that neurons tagged by molecular markers truly exhibit expected neuronal physiological activity, such as by creating promoters based on genes identified by RNA-seq (O'Toole et al., 2023).

Labeling of neurons with photoconvertible fluorescent proteins
Calcium modulated photoactivatable ratiometric integrator (CaM-PARI) is an engineered fluorescent protein that switches from green to red fluorescence when calcium elevation occurs during exposure to ultraviolet (405 nm) light (Fig. 3A).This photoresponsive property of CaMPARI allows active neurons to be tagged, which increases the spikeinduced intracellular calcium concentrations within defined time windows, such as specific behavioral states.By utilizing the advantage of the photoconversion of CaMPARI2, O'Toole et al. (2023) succeeded in the FACS collection of primary visual cortical layer 2/3 neurons that were responsive to specific visual stimuli during in vivo photoconversion of CaMPARI2.Subsequent single-cell RNA-seq analysis of the collected neurons revealed that neurons expressing abundant Adamts2 and Rrad were specifically responsive to negative and positive prediction errors, respectively.
In recent years, novel photoconvertible proteins have been developed such as Cal-Light (Lee et al., 2017;Hyun et al., 2022) and FLARE (Wang et al., 2017;Jung et al., 2023), which are designed to induce gene expressions when intracellular calcium is elevated by neuronal activity during light exposure.The photoconversion of these molecules requires longer (tens of minutes) light exposure than that of CaMPARI (less than one minute).These systems are expected to not only label cells but also express functional molecules (e.g., ChR2 and NpHR) in active neuronal ensembles, enabling the investigation of causal relationships between these ensembles and behavioral phenotypes.

Labeling of neurons with promoters for immediate early genes
Immediate early genes (IEGs), such as c-fos and Arc, are rapidly and transiently expressed in response to neuronal activity or plasticity.In neuroscience, promoters of IEG expression (i.e., c-fos promoter and Arc promoter) have been widely used as tools to express functional molecules, such as GFP and Cre.By utilizing these promoters in conjunction with the Cre-loxP and Tet systems, neurons that become active during specific time windows in response to various conditions such as sensory stimuli and learning can be selectively labeled with fluorescent proteins in the living brain (Fig. 3B).The labeled cell samples can subsequently be collected for single-cell RNA-seq using FACS.These techniques have primarily been applied to hippocampal neurons, and have provided novel insights into the transcriptomic profiles of neurons involved in memory engrams (Chen et al., 2020;Marco et al., 2020;Shpokayte et al., 2022;Sun et al., 2024).

Conclusions and perspectives
For over half a century, evidence has been accumulating regarding how various spike patterns of neuronal ensembles are related to specific behaviors.However, little is known about detailed molecular mechanisms that support such behavior-relevant neuronal activity.As reviewed in this paper, the integration of in vivo spike recordings and gene expression analyses has opened new avenues to address longstanding questions regarding how neurons exhibiting various spike patterns in the living brain are related to their molecular profiles.Currently, the primary issue in these studies is technical difficulty.For example, in slice preparations, the identification of neurons that have been physiologically recorded in vivo is technically challenging.Another issue is its cost.It is essential to devise efficient and cost-effective research plans by comparing them with the abundant atlas datasets that are openly accessible.Because the success of these studies strongly depends on the skill of experimenters and considerable costs, replicating these findings is challenging for other researchers.To improve research reproducibility, it is important to validate collected datasets against comprehensive atlas and make them accessible to others, possibly through open-access repositories.
The goal of most neuroscientific studies is to identify gene expression markers representing spike patterns, and to elucidate how such gene expression causally leads to the formation of spike patterns.Testing this causal hypothesis necessitates the manipulation of specific gene expression, emphasizing the importance of speculating on the most appropriate target genes in advance.Such ideas depend on a thorough examination of existing insights and the intuition of the researchers.In conclusion, while the field of neuroscience is becoming increasingly complex, recent advancements in the new techniques introduced in this review open the possibility of obtaining genuine insights that bridge molecules to cells, circuits, and behavior.In the future, the accumulation of ideas and datasets from researchers across diverse areas is expected to advance our understanding of the field of neuroscience.

Fig. 2 .
Fig. 2. Sampling of neurons imaged in vivo.(A) An imaged neuron is collected from a living animal.(B) An imaged neuron is labeled with H2BpaGFP by a pulsed laser and the labeled neuron is collected after slicing.(C) Imaged neuronal populations are identified and subjected to a spatial transcriptomic analysis on a slice.

Fig. 3 .
Fig. 3. Sampling of fluorescently labeled neurons by FACS.(A) Active neurons are labeled with red CaMPARI by exposure to ultraviolet light in vivo.The labeled neurons are then analyzed by FACS.(B) Neurons are labeled with fluorescent proteins expressed under the IEG-promoter in vivo.The labeled neurons are analyzed by FACS.