INTRODUCTION
Single-cell sequencing was first reported in 2009, following which this technology has rapidly developed and has been adopted across various research areas, including the nervous system (Mu et al. 2019; Tang et al. 2009). The nervous system is one of the most essential biological systems in higher animals, composed of large numbers of cell types of different physiological functions. Cells are the fundamental structural units of metazoans. Thus, clarifying different cell types, their molecular signatures, and the relationships between them are necessary steps to elucidate the complexity of the nervous system.
The use of single-cell sequencing technology to investigate cell types in the nervous system was first reported in 2014, in which 301 cells from the human cortex germinal zone were sequenced. These cells could be designated as different progenitors and neurons according to their expression of classical marker genes (Pollen et al. 2014). According to single-cell RNA-seq data, FGF, Wnt, and Notch signal pathways were confirmed to play important roles in neurogenesis (Basak et al. 2012; Bengoa-Vergniory and Kypta 2015; Engler et al. 2018; Hatakeyama et al. 2004; Krol et al. 2011; Sahara and O'Leary 2009). These findings demonstrated the utility of the single-cell technique as a tool for investigating neural development. However, this original study drew conclusions with a very limited cell population. With further technical development, researchers can accomplish sequencing from low-throughput to as many as tens of thousands of single cells for one sample (Cardona-Alberich et al. 2021; Dulken et al. 2019; Hrvatin et al. 2018; Hu et al. 2017; Mu et al. 2019; Tasic et al. 2018; Vanlandewijck et al. 2018). One recent project even obtained single-cell data from over two million cells (Cao et al. 2019). These studies provide detailed information on the nervous system at the single-cell level, dissecting various cell types and clarifying the complex construction of the brain.
So far, the majority of single-cell sequencing studies have focused on the transcriptome, as the transcriptome reflects both partial genomic sequence information and partial protein translation information via its role in the central dogma, produced by transcription using genomic DNA as a template and a template itself in translating the proteome. In fact, proteins are the executors of cell function, and mRNA levels can roughly infer the types and levels of proteins. On the other hand, protein production is not merely determined by the abundance of the corresponding mRNA. Protein quantity is also affected by other factors including RNA modifications and regulators such as non-coding RNA. Therefore, the transcriptome only reveals partial information on the state of the cell. Further, gene expression is affected by many epigenetic factors including genomic mutation, DNA methylation, histone modification, chromatin accessibility, and three-dimensional (3D) conformation. As the transcriptome is controlled by such a complex process, it is very unstable between different cells, even within a defined cell type.
Previous studies argue that patterns of DNA methylation and chromatin accessibility should be more accurate markers of cell type, whereas the transcriptome reflects the physiological state of cells (Kozlenkov et al. 2016; Luo et al. 2019; Mo et al. 2015; Zhu et al. 2020). Therefore, many different methods have been applied to detect different omics beyond the transcriptome at a single-cell resolution. For example, using bisulfite treatment to convert dCTP to dUTP, the DNA methylomes of more than 6,000 single nuclei from human and mouse frontal cortices were sequenced. There were 16 and 21 neuronal subtypes identified in the mouse and human cerebral cortex, respectively. The human brain was specifically identified with a subtype of cortical layer six excitatory neurons and an inhibitory neuron subtype (Luo et al. 2017). An additional study using high-throughput single-cell ATAC-seq (Assay for Targeting Accessible-Chromatin with high-throughput sequencing) showed that chromatin accessibility patterns could also distinguish neural subtypes in the human cerebral cortex and cerebellum, and the authors identified a pool of important cis-regulators for specific cell types (Lake et al. 2018).
Cellular heterogeneity is determined in multiple dimensions beyond the transcriptome. A single omics profile can help us to understand the physiological state of cells in one aspect, but bias also exists when measuring only one modality, preventing us from comprehensively resolving the heterogeneity of cells. Different omics as approaches can provide complementary information; the half-life of mRNA is short while covalent modifications of DNA, and some histone modifications, can persist for months to years (Armand et al. 2021; Chen et al. 2008; Colomé-Tatché and Theis 2018; Maze et al. 2015). As such, epigenetic factors are complementary to the transcriptome and can be used beyond gene expression levels to more accurately define cell types (Maze et al. 2015). In addition, different omics approaches reveal coordinated responses to signals from endogenous development and exogenous stimuli. Therefore, to better understand cellular heterogeneity and functions, single-cell multi-omics sequencing is gaining popularity, and is widely applied to investigate biological questions. In this review, we focus on the application of single-cell multi-omics sequencing in the nervous system. Existing reviews also summarize single-cell single-omics sequencing and its application to other research fields (Ma et al. 2020a; Mu et al. 2019; Zhu et al. 2019).
Currently there are several methods capable of measuring at least two omics layers in a single cell, most of which simultaneously measure the transcriptome and an additional omics layer, such as genomic mutation, DNA methylation, chromatin accessibility, etc. (Table 1 ). Below, we highlight seven categories of single-cell multi-omics approaches to discerning the structure of the nervous system. These methods are ordered according to the frequency of application to the nervous system and the timeline of method development.
1 The reported multi-omics technologies |
| Method | Multi-omics | References |
| RNA: transcriptome; 5mC: DNA methylome; ChrA: chromatin accessibility; 3DChr: chromatin 3D structure; Spa: cell spatial information; HisM: histone modification; Pro: proteome; gDNA: genome mutations | ||
| scMT-seq | 5mC and RNA | Hu et al. 2016 |
| scM&T-seq | 5mC and RNA | Angermueller et al. 2016 |
| scGEM | 5mC and RNA | Cheow et al. 2016 |
| snmCT | 5mC and RNA | Luo et al. 2018 |
| scTrio-seq | RNA, 5mC, and gDNA | Bian et al. 2018; Hou et al. 2016 |
| scNOMe-seq | 5mC and ChrA | Guo et al. 2017b; Pott 2017 |
| scCOOL-seq | 5mC, gDNA and ChrA | Guo et al. 2017a |
| iscCOOL-seq | 5mC, gDNA, and ChrA | Gu et al. 2019 |
| snmC2T | 5mC, RNA and ChrA | Luo et al. 2019 |
| scNOMeRe-seq | 5mC, RNA and ChrA | Wang et al. 2021 |
| scNMT-seq | 5mC, RNA and ChrA | Clark et al. 2018 |
| sciCAR-seq | RNA and ChrA | Cao et al. 2018 |
| scCAT-seq | RNA and ChrA | Liu et al. 2019 |
| ATAC-RNA-seq | RNA and ChrA | Reyes et al. 2019 |
| Paired-seq | RNA and ChrA | Zhu et al. 2019 |
| SNARE-seq | RNA and ChrA | Chen et al. 2019 |
| SHARE-seq | RNA and ChrA | Ma et al. 2020b |
| 10X Single Cell Multiome ATAC+ Gene Expression | RNA and ChrA | https://www.10xgenomics.com/products/single-cell-multiome-atac-plus-gene-expression |
| Methyl-HiC | 5mC and 3DChr | Li et al. 2019 |
| snm3C-seq | 5mC and 3DChr | Lee et al. 2019 |
| Paired-Tag | RNA and HisM | Zhu et al. 2021 |
| CoTECH | RNA and HisM | Xiong et al. 2021 |
| Slide-seq | RNA and Spa | Rodriques et al. 2019 |
| Spatial transcriptomics | RNA and Spa | Stahl et al. 2016 |
| HDST | RNA and Spa | Vickovic et al. 2019 |
| MERFISH | RNA and Spa | Moffitt and Zhuang 2016 |
| STAR-map | RNA and Spa | Wang et al. 2018 |
| seqFISH+ | RNA and Spa | Eng et al. 2019 |
| CITE-seq | RNA and Pro | Stoeckius et al. 2017 |
| REAP-seq | RNA and Pro | Peterson et al. 2017 |
| PLAYR | RNA and Pro | Frei et al. 2016 |
| PEA/STA | RNA and Pro | Genshaft et al. 2016 |
| SPARC | RNA and Pro | Reimegard et al. 2021 |
TRANSCRIPTOME AND DNA METHYLOME
Thus far, the transcriptome is the most informative dimension in single-cell sequencing. Regulation of the transcriptome by the DNA methylome has long been established (Hohn et al. 1996; Jones 2012; Schubeler 2015). Therefore, sequencing the transcriptome and the DNA methylome at a single-cell level can enhance cell type resolution and reveal gene expression regulatory networks (Armand et al. 2021). In this context, different methods had been developed such as scMT-seq, scM&T-seq, scGEM, snmCT and scTrio-seq (Angermuelleret al. 2016; Bian et al. 2018; Cheow et al. 2016; Hou et al. 2016; Hu et al. 2016; Luo et al. 2018). The sensory neurons from adult mouse dorsal root ganglia were analyzed by scMT-seq (Hu et al. 2016). To obtain multi-omics data within the same cell, researchers manually separated the cytoplasm and nucleus to prepare libraries for the single-cell transcriptome and DNA methylome respectively (Fig. 1 A). The direct correlation between mRNA expression and DNA methylation was confirmed in this study. For instance, the expression level of genes containing CpG island promoters showed a positive correlation with the methylation level of the corresponding gene body (Hu et al. 2016). In a similar way, mouse embryonic stem cells and hepatocellular carcinomas samples were analyzed and the researchers discovered regulatory pairs of methylated distal regulatory elements and downstream genes (Angermueller et al. 2016; Hou et al. 2016). For example, the expression of Esrrb negatively correlated with the DNA methylation level of several super enhancers (Angermueller et al. 2016; Papp and Plath 2012).
1 The schematic diagram of single-cell multi-omics sequencing methods detecting transcriptome, DNA methylation and chromatin accessibility. The schematic diagram of scM&T (A), snmCT and snmC2T (B). A In scM&T, genome and transcriptome of the same cell are originally obtained by separating the cytoplasm from the nuclei. Then single cell DNA methylome and transcriptome sequencing are carried out independently.B In snmCT, there is no step to physically separate transcriptome from the genome. Instead, the DNA and cDNA are amplified in the same reaction and distinguished according to the methylation of cytosine in the fragments. All reads from cDNA maintain cytosine while most C in genome fragments are converted into thymine. The snmC2T method further induces exogenous methylation (GC) to measure the chromatin accessibility level |
However, these technologies require physical separation of the cell nuclei and cytoplasm, leading to limited throughput and preventing wide application in the nervous system across complex cell types. High-throughput strategies have been used in other research fields. For example, scGEM was used in cancer studies (Cheow et al. 2016), and snmCT-seq was evaluated using H1 and HEK293 cells (Luo et al. 2018). These methods are suitable for nervous system exploration, especially considering snmCT-seq was further improved to also obtain chromatin accessibility information in the same cells, named snmC2T-seq (Luo et al. 2018, 2019). Based on analysis from snmC2T-seq, we infer that single-cell multi-omics with the transcriptome and DNA methylome could significantly enhance resolution for cell classification. The snmC2T-seq technology is discussed in a later section.
TRANSCRIPTOME AND CHROMATIN ACCESSIBILITY
To initiate transcription, genomic DNA must be accessible for transcription factor binding. Chromatin accessibility is highly associated with gene expression regulation. In 2018, Lake et al. harvested different regions of the human cerebral cortex and cerebellum, and sequenced the single-cell transcriptome and chromatin accessibility separately. Their integrative analysis across both kinds of single-cell datasets identified a series of genes in which promoter/enhancer accessibility was highly correlated with expression level. Therefore, some key fate decision factors were identified with higher confidence (Lake et al. 2018). In the same way, the transcription factor OLIG2 was found to play a vital role in stem cell-derived oligodendrogenesis during injured spinal cord repair (Llorens-Bobadilla et al. 2020). However, separate sequencing on two sample sets could not reveal direct interactions between chromatin accessibility and gene expression.
Different methods have been developed to achieve simultaneous detection of chromatin accessibility and gene expression in the same cells with high throughput in recent years, including sciCAR-seq, scCAT-seq, ATAC-RNA-seq, SNARE-seq, Paired-seq, and SHARE-seq (Table 1 ) (Cao et al. 2018; Chen et al. 2019; Liu et al. 2019; Ma et al. 2020b; Reyes et al. 2019; Zhu et al. 2019). The company 10X genomics has developed the commercial Next GEM Single Cell Multiome ATAC + Gene Expression kit in 2020. All methods depend on the Tn5 transposase to obtain accessible genome fragments first, following which the nuclear transcriptome and genome fragments are synchronously pre-amplified. SNARE-seq and 10X Next GEM Single Cell Multiome ATAC + Gene Expression kit are both based on the microdroplet platform while Paired-seq uses the strategy of splitting and pooling to obtain abundant single-cell data (Fig. 2 ). In 10X Next GEM Single Cell Multiome ATAC + Gene Expression kit, there are two types of primers, each specific for capturing mRNA and transposed DNA, respectively. In contrast, SNARE-seq uses the same primer to capture both mRNA and transposed DNA by incorporating a splint primer (Fig. 2 ). Next, the pre-amplified mixture of cDNA and transposed DNA fragments is subdivided into two parts and prepared in different libraries using specific primers (Fig. 2 ). In Paired-seq, the authors used different restriction enzymes targeting the pre-designed sites in each amplicon to digest either the cDNA or the transposed DNA fragments in a sequencing library. SHARE-seq shared a similar strategy with Paired-seq, and they used biotin-labeled oligos to separate the cDNA from genomic DNA (Fig. 2 ) (Ma et al. 2020b ).
2 Schematic illustration of different methods simultaneously detecting transcriptome and chromatin accessibility. The four reported methods for high throughput sequencing of transcriptome and chromatin accessibility. SNARE-seq and 10X ATAC + Gene Expression kit are based on microfluidic systems, while Paired-seq and SHARE-seq use the pool-split barcoding strategy. They each adopts specific way for cDNA and genome enrichment. SNARE-seq uses a sprint primer to add cell barcode and UMI for the tagmentated DNA fragments; the gel beads of 10X ATAC + Gene Expression kit contain two different primers for RNA and tagmentated DNA capture, and these two methods enriched each modality by amplifying with different primers; Paired-seq and SHARE-seq use the same PCR primers for both cDNA and DNA fragments, while the former induces different endonuclease site to enrich each modality and the later uses biotin to separate cDNA and DNA fragments before amplification. CB: cell barcode; RT: reverse transcription |
Combining the chromatin accessibility dataset with the transcriptome dataset, finer distinctions can be made to elucidate cellular heterogeneity. For example, intermediate progenitor cells were designated as a single cluster by transcriptome, while the same population of cells was divided into three subclusters by SNARE-seq (Chen et al. 2019). Meanwhile, the cell-type boundaries were relatively unclear and some minor cell types were not detected when only referred by the single-cell ATAC-seq dataset (Chen et al. 2019). With single-cell pair-wise datasets, the direct correlation of gene expression with the accessibility of cis-regulatory elements was confirmed for cell type specific genes (Chen et al. 2019; Trevino et al. 2021; Zhu et al. 2019). By identifying potential regulatory programs during neurogenesis and pathogenesis, we can evaluate the risks of neural disorder diseases at much earlier stages. Trevino et al. used the 10X Next GEM Single Cell Multiome ATAC + Gene Expression kit to analyze the human fetal cerebral cortex, and they found that the chromatin states of lineage-defining transcription factors are already accessible before the progenitor cells enter the cell cycle (Trevino et al. 2021). The researchers also infer the regulatory impact of autism spectrum disorder (ASD) relevant non-coding genetic variants in different cell types based on the chromatin accessibility patterns, including identification of loss of function of NFIA during the pathogenesis of ASD (Iossifov et al. 2014). Specifically, one G to A mutation in the enhancer, located within the intron of NFIA, may disrupt accessibility and silence NFIA in some excitatory neurons (Trevino et al. 2021). In addition to marker genes, the ATAC-seq peaks can also act as biomarkers for specific cell clusters. In SHARE-seq, the authors also identified temporal sequential activation of genome regions and gene expression in mouse skin development. Therefore, the chromatin accessibility of enhancers has the potential to foreshadow gene expression and predict lineage choice prior to the expression of key genes (Ma et al. 2020b).
TRANSCRIPTOME, DNA METHYLOME AND CHROMATIN ACCESSIBILITY
As mentioned above, both chromatin accessibility and DNA methylation are related to gene expression. Moreover, they can be analyzed at the single cell level by using different methods such as scNOMe-seq, scCOOL-seq and iscCOOL-seq (Gu et al. 2019; Guo et al. 2017a, b; Pott 2017). In order to understand cell state and the regulatory network more comprehensively, triple-omics sequencing including the transcriptome, DNA methylome, and chromatin accessibility within the same cell was developed. The reported methods include scNMT-seq, snmC2T-seq, and scNOMeRe-seq (Clark et al. 2018; Luo et al. 2019; Wang et al. 2021). All of these technologies adopted the same strategy by inducing exogenous GpC methylation to measure the open chromatin sites, while endogenous CpG methylation represents the DNA methylome. In scNMT-seq and scNOMeRe-seq, genomic DNA and mRNA in a single cell were physically separated. As such, these methods are not practicable for high throughput analysis, nor have these methods been applied to nervous system research yet. We instead focus on snmC2T-seq as an example, which is compatible with high-throughput operations such as split-pooling and the drop-seq platform.
In snmC2T-seq, the GpC methyltransferase M.CviPI was used to add a methyl group to accessible GpC sites. Furthermore, the researchers utilized 5’-methyl-dCTP instead of dCTP during cDNA synthesis to distinguish from genomic DNA after bisulfite conversion (Fig. 1 B) (Luo et al. 2018, 2019). Using snmC2T-seq to combine the transcriptome with the DNA methylome and chromatin accessibility enabled investigation of fine-grained cell subtypes in the human brain with higher sensitivity. For instance, analysis relying on only gene expression pinpointed the genetic risk of schizophrenia across broad cortical neuronal cell populations, such as the neocortical somatosensory pyramidal cells and cortical interneurons (Skene et al. 2018). Triple-omics analysis accurately targeted intratelencephalic neuron types, including L1-3 CUX2, L4-5 FOXP2 and L5-6 PDZEN4 and interneuron subtype MGE CALB1 (Luo et al. 2019). This indicates that multi-omics has the ability to pinpoint more accurate cell populations that are responsible for the pathogenesis of psychiatric disorders.
Vestigial enhancers are usually hypomethylated elements that exhibit activity during embryonic development but lack activity in adult tissues. They can be good biomarkers for the genetic risk of neuropsychiatric traits, as the low methylation level is maintained constitutively (Hon et al. 2013). To find the cell types possibly responsible for the pathogenetic process, the researchers used a partitioned heritability analysis to investigate those vestigial enhancers (Luo et al. 2019). Combining the DNA methylome and chromatin accessibility with the genetic risk locus and genes identified by a genome-wide association study (GWAS), some vestigial enhancers in or around the genetic risk locus of certain psychiatric diseases were enriched in specific cells. For example, vestigial enhancers that had been reported to be responsible for ASD genetic risk also showed strong enrichment in L4 PLCH1 and L6 TLE4 cell populations, indicating that they may be the cell populations essential in ASD pathogenesis (Grove et al. 2019; Luo et al. 2019).
By using scNMT-seq, the negative correlation between promoter DNA methylation and chromatin accessibility in single mouse embryonic stem cells was confirmed (Clark et al. 2018; Thurman et al. 2012). According to correlation analyses of different omics layers, although DNA methylation showed dynamic correlations with gene expression, it was more widely correlated with gene expression than chromatin accessibility (Clark et al. 2018). Beyond the two regulators, other epigenetic layers also play essential roles in gene expression regulation, such as histone modification which was reported to be essential for the function of vestigial enhancers (Hon et al. 2013).
TRANSCRIPTOME AND HISTONE MODIFICATIONS
Histone modifications varied greatly in their cellular specificity and relationship with cell-type-specific gene expression (Stadhouders et al. 2019). Investigating the direct relationships between various histone modifications and gene expression is important for parsing the regulatory network. The most studied types of histone modification affecting cell states include lysine methylation and acetylation in H3. The same modification in different lysine sites can cause inverse effects on gene expression. H3K4me3 is associated with active transcription, while H3K9me3 and H3K27me3 play an important role in chromatin heterogenization (Bannister and Kouzarides 2011; Clayton et al. 1993; Pogo et al. 1966). H3K4me3 was mainly distributed at the promoter region while H3K27me3 was enriched at the promoter region and poised enhancer regulated genes that played important roles in the process of differentiation in mouse and human embryonic stem cells (Banaszynski et al. 2013; Barski et al. 2007; Rada-Iglesias et al. 2011). H3K27ac generally localizes in the promoter and enhancer region, and is considered an active marker (Creyghton et al. 2010).
Single-cell histone modification sequencing was first reported in 2015, using microfluidics-based chromatin immunoprecipitation and sequencing (ChIP-seq), named Drop-ChIP (Rotem et al. 2015). On average 1000 unique reads of H3K4me3 or H3K4me2 were detected per cell. Distinct from previous ChIP-seq methods which fragmented the genome by ultrasonication or micrococcal nuclease, researchers adopted the nuclease or the Tn5 transposase fused to the protein A/G which enabled direct library amplification after antibody enrichment. These ChIP-seq methods include CUT&RUN, CUT&TAG (Kaya-Okuret al. 2019; Skene and Henikoff 2017), and were further optimized for single-cell sequencing (Grosselin et al. 2019; Wang et al. 2019). In these single-cell methods, on average 12,000 unique reads could be obtained in each single cell (Wang et al. 2019). Using these methods, it was difficult to unveil the direct regulatory relationship between histone modification and gene expression.
Later, Paired-Tag (parallel analysis of individual cells for RNA expression and DNA from targeted tagmentation by sequencing) was developed by extending from Paired-seq and CUT&TAG to profile one histone modification and the transcriptome in single cells (Kaya-Okuret al. 2019; Zhu et al. 2019, 2021). This method was applied to reveal the regulatory system, including H3K4me1, H3K4me3, H3K27ac, H3K27me3 and H3K9me3 in the adult mouse frontal cortex and hippocampus. Using combined analysis of histone modifications and the transcriptome, the authors found that many genes may be suppressed by similar mechanisms. For instance, H3K9me3 was mainly enriched in genes related to sensory pathways, especially olfactory receptor genes and vomeronasal receptor genes. H3K27me3 showed distributions on silenced genes relating to embryonic organ development. In addition to gene promoters, histone modifications in the distal cis-regulatory elements were also overrepresented. The candidate cis-regulatory elements resided in intergenic regions preferentially enriched with H3K9me3, while H3K4me1 and H3K27ac tended to occur in genic regions. Furthermore, a series of cis-regulatory element-gene pairs were inferred. An additional high throughput method named Co-TECH was reported recently (Xiong et al. 2021). In that work, the researchers integrated single-cell H3K27ac and transcriptome data to explore the regulatory mechanism in hematopoietic stem cell genesis (Xiong et al. 2021).
Here we note that simultaneously measuring two or more histone marks at the single-cell level is still unattainable. There are many kinds of histone modifications in a genome, but few of them have been widely investigated, especially in single-cell studies. The antibody-reliant approaches for detecting histone modification slow research progress compared to other, more easily detected epigenomic layers, such as chromatin accessibility and DNA methylation. Moreover, the direct interaction between histone modification and other epigenetic factors remains unstudied, as currently there is no way to detect them simultaneously.
DNA METHYLOME AND 3D GENOME
3D genome structure refers to the higher-order conformation of chromosomes which brings widely separated functional elements into close spatial proximity to regulate gene expression (Lieberman-Aiden et al. 2009; Rowley and Corces 2016). Parsing the folding state of chromosomes can provide deeper insights into the complex relationships between chromatin structure and the functional state of a cell (Rowley and Corces 2016). For instance, topological genome reorganization to express pluripotent stem cell marker genes can reprogram B lymphocytes into pluripotent stem cells (Stadhouders et al. 2018). Dynamic 3D chromatin conformation is also highly related to gene expression regulation during neurogenesis and in neuropsychiatric disorders (Lee et al. 2019; Rowley and Corces 2016).
The sn-m3C-seq method was initially developed to capture 3D genome structure information and DNA methylation in the same individual cell (Lee et al. 2019). Previous studies applied this technique to study neurons in the human cerebral cortex and found there was a chromatin loop surrounding the cell-type-specific marker genes. This included the excitatory neuron marker gene SATB2, which exhibited a 1.15 Mb loop between its promoter, and the LINC01923 locus, and this was only found in excitatory neurons other than interneurons. Together with the DNA methylome dataset, the authors found that DNA methylation was significantly correlated with chromatin 3D structure. For example, the hypomethylation of CTCF binding domains was suitable for CTCF binding and promoted chromatin interactions to form higher-order architecture (Lee et al. 2019). Compared to other omics layers, 3D chromatin structure maps play a limited role as indicators for cell-type atlases (Regev et al. 2017).
SPATIAL TRANSCRIPTOMICS
Metazoans contain numerous cell types which spatially interact with each other in complex ways to fulfill physiological roles. However, in most single-cell studies, tissues were dissociated into single-cell or single-nucleus suspensions for amplification and sequencing, resulting in the loss of spatial information (Stahl et al. 2016; Wang et al. 2009). The spatial architecture of the brain is fundamentally related to its function. The brain’s gross morphology is comprised of different cells in specific locations, and the morphology of each individual neuron, including soma position, dendritic architecture, and axonal projections, determines its roles in functional circuitry (Lein et al. 2017). The spatial transcriptome can be used to investigate brain development, neuronal connectivity, and brain disease (Lein et al. 2017). On the other hand, the cerebral cortex is a perfect model for the evaluation of spatial transcriptome detection methods, because of its highly organized structure.
To recover spatial information, different strategies have been reported. One of the classical strategies is to transfer cellular mRNA to a solid surface containing spatial barcode primers (Rodriques et al. 2019; Stahl et al. 2016; Stickels et al. 2021; Vickovic et al. 2019). According to the barcode information, we can reconstruct gene expression patterns on the tissue surface. Researchers utilized this method to investigate cellular distributions in the mouse olfactory bulb (Stahl et al. 2016), which is consistent with the results of high throughput in situ hybridization. Incorporating spatial transcriptomics in the mouse spinal cord over the course of amyotrophic lateral sclerosis, Maniatis et al. distinguished regional differences for microglia and astrocytes (Maniatis et al. 2019). In another study of single-cell spatial transcriptome analysis of the cerebellum (Rodriques et al. 2019), the researchers validated the parasagittal bands of gene expression in the Purkinje layer (Barmack et al. 2000; Gravel and Hawkes 1990). They confirmed banded expressed genes such as Aldoc and Plcb4 (Bailey et al. 2014; Brochu et al. 1990; Demilly et al. 2011). Cell types defined by transcriptome data were further divided into subpopulations according to their spatial characters (Rodriques et al. 2019). Recently, Slide-seq2 was used to analyze the transcriptome of subcellular compartments, where the soma and proximal neuropil of the mouse hippocampus CA1 cells were investigated and more than 200 dendritically enriched genes were identified (Stickels et al. 2021).
The above spatial sequencing methods were not precisely to a single-cell resolution. To observe the single-cell resolution of spatial transcriptome information, sci-Space placed the spatially arrayed hashing oligos on a glass slide and released them into cell nuclei in a tissue slide (Srivatsan et al. 2021). This method could measure much wider space in a tissue, and the whole E14 mouse embryo was analyzed. The authors found that the distribution of radial glia cells and neurons revealed gradients of developmental maturity in the brain. They also found neuronal migration patterns in different regions consistent with observations of previous studies (Tan et al. 2002; Watanabe et al. 2018). For the first time, they observed neurons in the midbrain migrating in both radial and tangential directions. In addition, sci-Space was supported as a spatial “scaffold” for the nonspatial single-cell transcriptome atlases since it recovered single-cell transcriptomes with spatial information (Srivatsan et al. 2021).
A further strategy for spatial transcriptomics is based on FISH. Single-molecule FISH (smFISH) is capable of detecting single transcripts, but it could only measure one to four genes simultaneously in its first application (Raj et al. 2008). By barcoding, the transcripts within individual cells through multiple rounds of hybridizations, sequential FISH (seqFISH) could detect over ten thousand genes in a tissue section (Eng et al. 2019; Shah et al. 2018). There are also other methods such as multiplexed error-robust fluorescence in situ hybridization (MERFISH) and spatially-resolved transcript amplicon readout mapping (STAR-map) (Moffitt et al. 2018; Wang et al. 2018). MERFISH used combinational labeling probes to identify each RNA species by multiple distinct signals. Up to one thousand genes in each single cell can be tested by using this method (Chen et al. 2015; Moffitt et al. 2018; Moffitt and Zhuang 2016). The cellular orchestra of the neural system in a spatial pattern has been directly presented by these techniques. For example, the mouse hypothalamic preoptic region was investigated by MERFISH (Chen et al. 2015), the primary visual cortex was mapped by STAR-map (Wang et al. 2018) and the subventricular zone and olfactory bulb in the mouse brain was profiled by seqFISH (Eng et al. 2019). Compared to the spatial barcode-based technologies, FISH based technologies could achieve subcellular resolution, but the detection ability relies on gene probes.
The spatial transcriptome methods discussed above have shed light on spatial atlases and spatial blueprints. In addition, they also promote research into discovering disease-related factors and crosstalk of different cell types in a microenvironment (Liao et al. 2021). Since many neural diseases were highly correlated with brain regions, these techniques should play an increasingly important role in future research.
TRANSCRIPTOME AND PROTEOME
As the terminal executors, proteins determine cell structure, metabolic reactions, and cell function (Labib and Kelley 2020). Direct proteome detection offers clues on the cellular responses to genomic, epigenomic, and environmental alterations. Investigating the proteome can help us to determine key functional markers and understand pathogenicity mechanisms in different diseases (Aslam et al. 2017). Further integrating the proteome with other omics layers such as the epigenome and transcriptome shows great potential in revealing novel insights into biological regulation (Zhang and Kuster 2019). Different from genome and transcriptome sequencing, which rely on the arrangement and combination of the four nucleotides A, T, C, G, proteins consist of 20 amino acids which are not detectable by the sequencers. The currently achievable way to resolve the proteome is partially by mass spectrometry. The essential chemical difference between the DNA/RNA sequence and the amino acid sequence sets a barrier to sequencing the proteome with other omics layers.
In recent years, researchers measured the abundance of specific proteins by transforming them to antibody-coupled DNA/RNA fragments. This way, these proteins could be detected simultaneously with the transcriptome within the same cell. Such techniques include PEA/STA (proximity extension assays/specific (RNA) target amplification), REAP-seq (RNA expression and protein sequencing assay), SPARC (single-cell protein and RNA co-profiling) and CITE-seq (cellular indexing of transcriptomes and epitopes by sequencing) (Genshaft et al. 2016; Peterson et al. 2017; Reimegard et al. 2021; Stoeckius et al. 2017). PEA/STA, REAP-seq and CITE-seq are only available for cell surface proteins while SPARC was developed for investigating the intercellular proteome. Each of those methods relied on the efficiencies of antibodies. This way, they could only detect limited protein types with accessible antibodies, while other proteins could not be studied by this method. Also, protein modifications including glycosylation, neddylation, SUMOylation and others can’t be distinguished through these methods.
Mass cytometry can also be used to detect proteins, such as PLAYR (proximity ligation assay for RNA), which can simultaneously detect more than 40 different protein epitopes and RNA targets in thousands of cells per second (Frei et al. 2016). PLAYR has the ability to detect both cell-surface and cytoplasmic proteins. But with a limited number of available tags, the proteins and RNA that can be detected in individual cells are also quite limited.
FUTURE PERSPECTIVES
The rapid development of single-cell multi-omics sequencing technology has enabled more accurate research into a series of biological processes, including disease and development. Fittingly, this technique was announced as ‘Method of the Year 2019’ by Nature Methods (https://doi.org/10.1038/s41592-019-0703-5). Even so, we must note the tradeoff between breadth and sensitivity in including expanded sets of omics data. That is, compared to single-cell mono-omics sequencing, multi-omics sequencing harvests more sparse data. SNARE-seq and Paired-seq detected less than one quarter of DNA reads compared to single-cell combinatorial indexing protocols for chromatin accessibility (sci-ATAC-seq) (Chen et al. 2019; Cusanovich et al. 2018; Zhu et al. 2019). The triple-omics method snmC2T detected fewer useful reads than the dual-omics method snmCT (Luo et al. 2019). On the other hand, different omics layers each have their own advantages, for example, the DNA methylome outperformed the 3D genome in distinguishing cell types (Lee et al. 2019), and chromatin accessibility could better elucidate the regulations of cell lineage specifications (Ma et al. 2020b). Therefore, it is necessary to weigh those advantages and disadvantages before selecting an optimal research method for a given question.
Single-cell analysis, especially single-cell multi-omics analysis, is vital to understanding systems of neural circuits as they comprise enormous numbers of neural cell types with complex regulation networks and experience-dependent plasticity (Armand et al. 2021). The development of technology has remarkably accelerated understanding of the cellular diversity of the nervous system and provided insights into the molecular regulatory networks regarding neural development and diseases. However, except for the integrated single-cell analysis of the transcriptome and chromatin accessibility or DNA methylation (Chen et al. 2019; Luo et al. 2019; Zhu et al. 2019), other multi-omics combinations have not shed light on their key roles in the neural system so far. For example, histone modifications and 3D genome structures are also essential for the nervous system (Hon et al. 2013; Li et al. 2018; Stadhouders et al. 2018, 2019), yet we still do not know how these epigenetic factors participate in neural fate commitment. Moreover, the key pathogenic mechanisms underlying neuropsychiatric diseases such as ASD and Alzheimer’s Disease still need further investigation (Breijyeh and Karaman 2020). Specific to neurons, omics information must also correspond to morphologies and electrophysiological characteristics under stimulation, thus defining the molecular determinants of neuronal function. The evolutionary relationships among neural populations within the same species and across species still need further investigation. In summary, there are still many open questions regarding the nervous system, many of which are best addressed by evolving technologies of single-cell multi-omics sequencing.