**The Role of Single-Cell Technology in the Study and Control of Infectious Diseases**

**Weikang Nicholas Lin 1,**†**, Matthew Zirui Tay 2,**†**, Ri Lu 3,**†**, Yi Liu 1,**†**, Chia-Hung Chen <sup>4</sup> and Lih Feng Cheow 1,5,\***


Received: 7 May 2020; Accepted: 5 June 2020; Published: 10 June 2020

**Abstract:** The advent of single-cell research in the recent decade has allowed biological studies at an unprecedented resolution and scale. In particular, single-cell analysis techniques such as Next-Generation Sequencing (NGS) and Fluorescence-Activated Cell Sorting (FACS) have helped show substantial links between cellular heterogeneity and infectious disease progression. The extensive characterization of genomic and phenotypic biomarkers, in addition to host–pathogen interactions at the single-cell level, has resulted in the discovery of previously unknown infection mechanisms as well as potential treatment options. In this article, we review the various single-cell technologies and their applications in the ongoing fight against infectious diseases, as well as discuss the potential opportunities for future development.

**Keywords:** single cell; infectious disease; pathophysiology; therapeutics; diagnostics

#### **1. Introduction**

Five months since the first reported infection cluster, COVID-19 has turned into a vicious worldwide pandemic that infected more than 3.6 million people and caused over 250,000 deaths [1]. The pandemic will also have large spillover effects in terms of economic damage both in the form of healthcare costs and in monetary losses from the disruption of global supply chains, with world trade expected to fall between 13% and 32% in 2020 [2]. The COVID-19 pandemic serves as a grim reminder that infectious disease is, and will always be, a major threat to the continued existence of mankind.

To date, there are about 1400microorganisms known to be pathogenic to humans. These pathogens can be broadly classified as viral, bacterial, fungal, and parasitic pathogens [3]. In particular, there have been 3 infectious diseases that have been persistently difficult to eradicate, namely, Human Immunodeficiency Virus and Acquired Immune Deficiency Syndrome (HIV/AIDS), tuberculosis, and malaria. AIDS, due to HIV, is responsible for nearly 1 million deaths per year [4]. The death toll from tuberculosis, caused by Mycobacterium tuberculosis (MTB) bacteria, is the highest amongst all infectious diseases, which is a problem that is exacerbated by the rise of antimicrobial resistance variants of the disease [4]. Malaria, a parasitic infection, has afflicted humans for thousands of years and continues to do so today [5]. In light of the above-mentioned examples, among other infectious diseases, further efforts have to be directed for the continued management of the global burden of these diseases.

The COVID-19 pandemic has highlighted many questions that are relevant in the context of infectious disease as a whole. Why are certain people more susceptible to infections? Why are some infected individuals asymptomatic or display only mild symptoms? Why are there differences in terms of disease progression and outcomes among patients? This diverse response to infection could be explained by the interactions of inherently heterogeneous populations of pathogens, host cells, and immune cells. However, discerning this heterogeneity is difficult in conventional bulk analyses, as they fail to recognize the following: (1) the genomic variability of pathogens, (2) the coexistence and interactions of infected host cells and bystanders, and (3) the diverse functional roles of immune surveillance participants. Aside from the limited resolving power in pathophysiological studies, bulk analyses often fall short in terms of the level of precision and the amount of derived information needed for early diagnostics and high-efficacy vaccine development against infectious diseases.

Just as how microscopy revolutionized our understanding of biology, the enhanced resolution, precision, and breadth of information offered by single-cell technologies has brought an exciting overhaul to our perception of infectious diseases in recent years. The use of single-cell genomics, transcriptomics, proteomics, and epigenetics (referred to as omics altogether in this article) has flourished in many areas of the battlefield against infectious diseases. Table 1 presents commonly used single-cell technologies in infectious disease studies alongside several of other non-single-cell systems. A scoring heatmap is used to represent the complexity of information in various aspects that they can provide (e.g., genetic, epigenetic, proteomic, spatial). The heatmap also provides an overall ranking of throughput, cost, and downstream assay compatibility amongst all the listed techniques. Hence, it serves as a general guide for future users to select the methods that match their desired outputs. For instance, if the primary target for the study is to collect genetic information (e.g., analysis of invading virus gene heterogeneity in single host cell), single-cell sequencing might be the best candidate. Likewise, if the focus is on proteomics orchestrating the immune responses, mass cytometry can furnish the most detailed insight. For infectious disease models that involve the interplay of genomics and proteomics, both CITE-seq and REAP-seq can be the suitable candidates. While flow cytometry has the highest throughput and lowest cost per experiment amongst the listed single-cell methods, it yields very limited aspects of information. Other single-cell assays capable of providing more complex data typically come at the expense of a decreased throughput and increased cost. It is also worth mentioning that microfluidics has made great success in boosting the throughput and cost efficiency of existing single-cell assays. One prominent example would be the transition of well plate systems into microchambers or microdroplets, which ultimately reduces the required amount of reagent required per experiment and in turn reduces costs.

In this review, we identified infection pathophysiology, therapeutic discovery, and disease diagnostics as three major areas in which single-cell omics has contributed substantially in the past decade. In pathophysiological studies of infectious disease, single-cell omics offer excellent spatial–temporal resolution that help to not only reconstruct the uneven subcellular distribution of pathogen across the entire host cell population, but also reveal the sequence of immune events accompanied by the change of immune cell profiles. Single-cell omics also extrapolates meaningful molecular details that describe the dynamic host–pathogen interplay and immune activation. Furthermore, single-cell omics identifies the rare molecules and cell subtypes that exhibit significant functionality in the pathogen–host immune interactions. Insights in fundamental pathophysiology naturally have spillover benefits for translational science, such as in vaccine development where single-cell omics has the capability to enhance the discovery of mechanistic correlates of protection through multi-parameter measurements of the immune state with respect to disease, and it enables precision quality control checkpoints to aid the evaluation of vaccine efficacy. In the field of antibody discovery, single-cell omics can simultaneously interrogate antigen specificity and recover the B cell receptor gene sequences, which in turn shortens the previously prolonged and labor-intensive research cycle in the search for effective therapeutic or diagnostic antibodies. In the application of infectious disease diagnostics, single-cell omics are on the verge of practical clinical deployment, as demonstrated

by some examples of automated and miniaturized devices. The diagnostic power of single-cell omics can be further enhanced by incorporating digital assays or integrating with other label free single-cell technologies. While there are many merits of single-cell analysis, we also discuss the new sets of challenges that need to be addressed in these systems. Finally, we will conclude with our insight on future prospects of single-cell research in infectious disease and highlight several emerging single-cell technologies that may further enrich our arsenal against infections.

#### **2. Uncovering Infection Pathophysiology**

Understanding the pathophysiology of infection is critical to the rational design of prophylactic and therapeutic strategies to tackle infectious diseases. The course of infection, determined by the encounter of pathogens and host cells, is often measured as population-averaged results, leaving the important cell-to-cell heterogeneity out of the picture. The heterogeneity arises from both the pathogens and the infected cells. For example, pathogen heterogeneity can be reflected in the case of viruses, as a mixture of mutated viral particles displaying different infection ability [6], or in the cases of bacteria, as a population of cells having different resistance to the same antibiotics [7]. Host cellular heterogeneity is a combined result of variances in metabolism, composition, activation status, cell cycle, or infection history [6]. Recent advances in single-cell analysis provide an attractive approach to probe the cellular population diversity and characterize infection pathophysiology at single-cell resolution. In this section, we will review how the recent advancement of single-cell technologies has helped deepen the understanding of pathogen and host cell heterogeneity and how the complex immune system reacts against infectious pathogens, with a focus on the contributions of single-cell sequencing.

#### *2.1. Pathogen Heterogeneity*

Pathogen heterogeneity can be inherent or as a result of heterogeneous host–pathogen interactions. It is a favorable feature for pathogens because varied genomic sequences or functional properties enable immune evasion, colonization in novel hosts, and drug resistance acquisition; therefore, they increase the possibility of survival. Besides, stochastic fluctuation in biochemical reactions may also contribute to cell-to-cell variability. Single-cell technologies provide high-resolution insights into different aspects of intracellular pathogen replication.

One area of virology that has benefited from the enhanced resolution of single-cell technologies is the study of variation in infection across single cells and the reasons for such variation. In the study by Heldt et al., cells were infected in a population, isolated into microwells, and incubated. The supernatant was subjected to viral plaques measurement, and viral RNA was quantified from lysed single infected cells [8]. It was shown that cells infected by influenza A virus (IAV) under the same conditions produced largely heterogenous progeny virus titers, ranging from 1 to 970 plaque-forming units (PFU) and intracellular viral RNA (vRNA) levels varied three orders of magnitude. Similarly, using scRNA-seq, another study determined the percentage of viral transcripts in the total mRNA generated from IAV-infected cells, and it revealed that while most cells contained less than 1% of viral transcripts, some cells generated more than 50%, demonstrating infection heterogeneity from the angle of viral load [9]. Reasons for this variation can be further explored through the use of high-throughput imaging technology. For instance, Akpninar et al. used virus expressing red fluorescence protein (RFP) to study the effect of defective interfering particles (DIP) on viral infection kinetics. DIP are noninfectious progeny particles lacking genes essential for replication, and they are commonly produced during infection due to the high mutation rate. When participating in infection along with viable viral particles, they compete for host cellular machinery and result in viral replication inhibition. In this study, cells in a bulk population were infected with a mixture of vesicular stomatitis virus (VSV) expressing RFP and VSV-DIPs, and they were either untreated or isolated by serial dilution. RFP expression was observed during incubation as a surrogate for viral replication levels. The results showed that DIP inhibited viral replication 10 times more on single cells, suggesting that the inhibition of viral replication is mitigated by cell–cell interactions when infection happens in a population [10].


**Table 1.**

Overview of commonly used single-cell technologies

 and their respective

characteristics:

 a higher color intensity corresponds

 to a higher score (e.g., higher

The genomic mutation of pathogens during infection can be also detected directly. The sequencing of transcriptome and viral genes in single infected cells showed that IAV is highly prone to mutation during infection [11]. Detected mutations can cause consequences include viral polymerase malfunction and failure to express the interferon (IFN) antagonist protein, which is correlated to heterogeneous immune activation among infected cells [11]. The sequencing of 881 plaques from 90 VSV-infected cells detected 36 parental single nucleotide polymorphism (SNP) and 496 SNP generated during infection (Figure 1A–E) [12]. Although extremely low multiplicity of infection (MOI) was adopted, resulting in 85% of the cells statistically infected with only one PFU, 56% contained more than one parental variant, indicating that pre-existing differences in viral genomes can be spread within the same infectious unit, in this case, the host cell population. Moreover, by measuring the viral titers produced by each infected cell, a significant correlation was found between the number of mutations in the viral progeny and the log yield of the initially infected cell.

**Figure 1.** Pathogen heterogeneity revealed by single-cell analysis. (**A**) Schematic of experimental setup for sequencing single-cell bottlenecked viruses. Cells were inoculated with vesicular stomatitis virus (VSV), and individual cells were transferred to separate culture wells with a micromanipulator. After overnight incubation, single, isolated plaques (viral progeny) from the supernatant were picked for massive parallel sequencing. The viral stock was subject to ultra-deep sequencing to detect the polymorphisms present in the inoculum (parental sequence variants). (**B**) The distribution of the number of non-parental single nucleotide polymorphisms (SNPs) found in the 7–10 plaques derived from each cell. (**C**) Distribution of the number of plaques derived from the same cell that contained a given non-parental variant. (**D**) Spectrum of nucleotide substitutions found after single-cell bottlenecks. (**E**) Correlation between the abundance of each type of substitution in single-cell-derived plaques and natural isolates. All panels adapted with permission from [12]. Copyright 2015, Elsevier.

Genomic variability also widely exists among bacteria populations. Fluorescence labeling enables the quantification of bacterial growth in single host cells [13–15], and by correlating the heterogenous growth with host response, it was found that the *Salmonella* population exhibits different induction levels of the PhoP/Q two-component system, which modulates lipopolysaccharides (LPS) on the surface of individual bacteria [14].

#### *2.2. Host Cell Heterogeneity*

To understand the pathophysiology of infectious diseases, it is important to study the identities of targeted cells. Mounting evidence has shown that even under identical conditions, individual host cells manifest differential susceptibility and responses to infection in a population. How does this preference arise? Do they share similar features that might be reasons for their susceptibility of infection? How do the states of infected cells affect pathogen replication and infection outcome? Furthermore, how are host cells' phenotypes influenced by infection individually and temporally? Answers to these questions are critical for the identification of target cells and individuals of novel pathogens, as well as for the understanding of infection pathophysiology.

Analysis of cells exposed to pathogens at single-cell resolution requires, first and foremost, strategies to distinguish infected cells from uninfected ones. Pathogen-specific proteins, such as viral glycoproteins embedded in the cell membrane, or intracellular proteins such as viral capsid or polymerases, as well as pathogen nucleic acids, including genomic DNA/RNA and transcripts, can serve this purpose. These microbial elements can be labeled with specific antibodies or oligonucleotide probes for detection and quantification. Alternatively, pathogen nucleic acids can be directly captured in deep sequencing. By combining tools for pathogen identification with host cell phenotyping assays, infected cells can be profiled at the single-cell level.

Xin et al. investigated the effects of host cell heterogeneity on both acute and persistent infection by foot-and-mouth disease virus (FMDV) [16]. By sorting single infected cells with FACS based on cellular parameters, and quantifying viral genome replication with RT-PCR, they showed that the host cell size and inclusion numbers affected FMDV infection. Cells with larger size and more inclusions contained more viral RNA copies and viral protein and yielded a higher proportion of infectious virions, which is likely due to favorable virus absorption. Additionally, the viral titer was 10- to 100-fold higher in cells in G2/M than those in other cell cycles, suggesting that cells in the G2/M phase were more favorable to viral infection or for viral replication. Such findings have also been reported for other viruses [9,17,18], revealing a general effect of heterogeneous cell cycle status in a population on virus infection.

Golumbeanu et al. demonstrated host cell heterogeneity using scRNA-seq: they showed that latently HIV-infected primary CD4<sup>+</sup> T cells are transcriptionally heterogeneous and can be separated in two main cell clusters [19]. Their distinct transcriptional profiles correlate with the susceptibility to act upon stimulation and reactivate HIV expression. In particular, 134 genes were identified as differentially expressed, involving processes related to the metabolism of RNA and protein, electron transport, RNA splicing, and translational regulation. The findings based on in vitro infected cells were further confirmed on CD4<sup>+</sup> T cells isolated from HIV-infected individuals. Similarly, enabled by scRNA-seq and immunohistochemistry, several candidate Zika virus (ZIKV) entry receptors were examined in the human developing cerebral cortex and developing retina, and *AXL* was identified to show particularly high transcript and expression levels [20,21].

scRNA-seq can also be used to identify potential target cells of novel pathogens and facilitate the understanding of disease pathogenesis and treatment. The spike protein of the virus SARS-CoV-2, the pathogen responsible for the COVID-19 pandemic, binds with the human angiotensin-converting enzyme 2 (ACE2) [22,23]. This binding, together with a host protease type II transmembrane serine protease TMPRSS2, facilitates viral entry [22,23]. By analyzing the existing human scRNA-seq data, it was identified that lung type II pneumocytes, ileal absorptive enterocytes, and nasal goblet secretory cells co-express *ACE2* and *TMPRSS2*, which suggests that they might be the putative targets of SARS-CoV-2 [24].

In the preparation of scRNA-seq library, standard poly-T oligonucleotide (oligo-dT) is commonly used to capture mRNA from single cells, which can also capture polyadenylated viral transcripts from DNA virus or negative-sense single stranded RNA virus. A simultaneous analysis of host transcriptome profiles and viral DNA/RNA offers information on the presence of the studied pathogen and its activities and allows a more accurate characterization on the dynamics of host–pathogen interactions.

Wyler et al. profiled the transcriptome of single human primary fibroblasts before and at several time points post-infection with herpes simplex virus-1 (HSV-1), and they described a temporal order of viral gene expression at the early infection stage [25]. More importantly, by simultaneously profiling the host and viral mRNA, they identified that transcription factor NRF2 is related to the resistance to HSV infection. The finding was verified with the evidence that NRF2 agonists impaired virus production. Steuerman et al. performed scRNA-seq of cells from mice lung tissues obtained 2 days after influenza infection [26]. FACS was applied to sort immune and non-immune cells based on CD45 expression. Nine cell types were clustered (Figure 2A), and viral load was determined by the proportion of reads aligned to influenza virus gene segments, with higher than 0.05% considered infected. The authors found that viral infection can be detected in all cell types, and the percentage ranges from 62% in epithelial cells to 22% in T cells. However, the high variability of viral load was only observed among epithelial cells, while the majority of infected cells of other cell types showed to have low viral load (less than 0.5%) (Figure 2B).

For positive sense RNA virus whose transcripts lack polyadenylation and cannot be captured by oligo-dT, a reverse complementary DNA oligo probe to the positive-strand viral RNA was employed. Zanini et al. described this method and correlated gene expression with virus level in the same cell to study the infection of dengue virus (DENV) and Zika virus (ZIKV). They identified several cellular functions involved in DENV and ZIKV replication, including ER translocation, N-linked glycosylation, and intracellular membrane trafficking [27]. Interestingly, by contrasting the transcriptional dynamics in DENV versus ZIKV-infected cells, differences were spotted in the specificity of these cellular factors, with a few genes playing opposite roles in the two infections. Genes in favor of DENV (such as *RPL31*, *TRAM1*, and *TMED2*) and against DENV infection (such as *ID2* and *CTNNB1*) was also validated with gain/loss-of-function experiments.

Analysis methods have been advancing for the detection of genetic variant-based scRNA-seq data [28–30]. They could contribute, in the study of infectious diseases, to the characterization of temporal changes in viral mutational prevalence [31]. Moreover, viral mutation can be correlated with host gene expression status at the single-cell level to further investigate their potential mutual effect on one another throughout the course of infection and reveal the dynamic host responses and pathogen adaptations in the progression of infection [32].

In spite of the above-mentioned examples characterizing virus presence with scRNA-seq, it is worth noticing that viral mRNA or genome occurrence is not necessarily equivalent to viral progeny, due to reasons such as missing essential genes caused by mutations. Experimental techniques enabling the joint analysis of host transcriptional responses and viral titers will be needed to reveal the underlying mechanisms of virus production levels and host cell heterogeneity. Another challenge of analyzing viral RNA data is distinguishing infected cells with intracellular viral transcription from uninfected cells acquiring exogenous viral RNA. Combining single-cell transcriptomics data with flow cytometry or mass cytometry by time-of-flight (CyTOF) to measure the intracellular viral protein may help overcome this issue.

**Figure 2.** Single-cell analysis of influenza-infected mice lung tissues demonstrated heterogeneous virus load and gene expression activation in infected cells. (**A**) Schematic illustration of the experimental workflow. Immune and non-immune single cells were isolated from the whole lung of control and influenza-treated mice for massively parallel single-cell RNA sequencing. Host and the viral mRNA were simultaneously measured, allowing the identification of infected as opposed to bystander cells, the quantification of intracellular viral load, and the profiling of transcriptomes. Nine cell types were distinguished based on their transcriptional identities (**B**) The single-cell heterogeneity of intracellular viral load during influenza infection. Percentages of low (yellow), medium (light brown), and high (dark brown) viral-load states (y axis) within the population of infected cells are shown for each of the nine cell types (x axis; total numbers of infected cells are indicated). (**C**) Host genetic responses across all cell types. Differential expression in influenza-treated and control mice (color bar) of nuclear-encoded genes (rows) across the nine major cell types (columns). Right column indicates membership in four type I interferon (IFN)-related categories. All panels adapted with permission from [26]. Copyright 2018, Elsevier.

#### *2.3. Host Immune Responses in Infection*

Immune responses activated by infection, since it is the innate immune responses that are primarily initiated in infected cells, or adaptive immune responses by lymphocytes carrying specific roles, are dynamic and complex, and they often happen in specific tissue microenvironments. Heterogeneity in immune responses is also a long-recognized phenomenon. For instance, the activation of antiviral responses in dendritic cells (DCs) by bacterial LPS starts with a small fraction of cells initiating the reaction, followed by the response by the rest of the population via paracrine responses [33]. Technologies that enable the simultaneous measurement of multiple parameters facilitate high-resolution characterization of transcripts and protein at the single-cell level and boost our understanding of how host immune responses are initiated and orchestrated against infection. Although pathogens usually dominate the war with host immune responses, hence the prevalence of infectious diseases, in-depth understanding of the interplay provides valuable information for the design of strategies to fight against infectious diseases. In this section, we cover the single-cell characterization of both innate immune responses from infected cells and adaptive immune responses activated in infected units.

Type I interferon (IFN), a key cytokine in innate immunity, orchestrates the first line of host defense against infection. Its production is initiated upon host cells sensing pathogen-specific molecules, and it turns on the antiviral state of host cells by activating the transcription of hundreds of IFN-stimulated genes (ISGs), some of which are crucial for coordinating adaptive immune responses. Many studies have shown a large variability of IFN expression among infected cells. In the case of influenza virus infection, this can be partially explained by the high mutation rate during replication, revealed by sequencing viral genes in single infected IFN reporter cells [11]. However, such viability was also found to exist in infected cells expressing unmutated copies of all viral genes, which might be a result of the stochastic nature of immune activation irrelevant to viral genotypes [11].

In another study, PBMCs from patients with latent tuberculosis infection (LTBI) or active tuberculosis (TB), and from healthy individuals were analyzed with scRNA-seq [34]. T cells, B cells, and myeloid cells were distinguished, and 29 subsets were clustered. The novel finding in this work is the consistent depletion of one natural killer (NK) cell subset from healthy individual samples to samples from LTBI and TB, which was also validated by flow cytometry. The discovered NK cell subset could potentially serve as a biomarker for distinguishing TB from LTBI patients, which is valuable for predicting disease outcome and developing treatment strategies. By analyzing scRNA-seq data of PBMCs derived from individuals before and at multiple time points after virus detection, Kazer et al. investigated the dynamics of immune responses during acute HIV infection [35]. After identifying well-established cell types and subsets in PBMCs, the authors examined how each cell type varies in phenotype during the course of infection. Genes involved in cell-type specific activities, including monocyte antiviral activity, dendritic cell activation, naïve CD4<sup>+</sup> T cell differentiation, and NK trafficking manifested similar changes with plasma virus levels: peaking closer to detection and gradually descending with time.

Phenotypic variations in bacteria populations were shown to influence host cell responses. Avraham et al. investigated macrophage responses against *Salmonella* infection with fluorescent reporter-expressing bacteria and scRNA-seq on host cells [14]. Transcriptional profiling revealed the bimodal activation of type I IFN responses in infected cells, and this was correlated with the level of induction of the bacterial PhoP/Q two-component system. Macrophages that engulfed the bacterium with a high level of induction of PhoP/Q displayed high levels of the type I IFN response, which was presumably due to the surface LPS level related to PhoP/Q induction. With a similar setup, Saliba et al. studied the *Salmonella* proliferation rate heterogeneity in infected macrophages [13]. The varied growth rate of bacteria, indicated by fluorescent expression by engineered *Salmonella* in single host cells, influenced the polarization of macrophages. Those bearing nongrowing *Salmonella* manifested proinflammatory M1 macrophages markers, similar with bystander cells, which were exposed to pathogens but not infected. In comparison, cells containing fast-growing Salmonella turned to anti-inflammatory, M2-like state, showing that bacteria can reprogram host cell activities for the benefit of their survival.

The above-mentioned strategy to simultaneously profile host cell transcriptome and viral RNA also plays an important role in characterizing immune responses against infection by identifying infected immune cells and analyzing the transcriptomes simultaneously. For instance, it was applied to study the heterogeneous innate immune activation during infection by West Nile virus (WNV) [17]. High variability was revealed for both viral RNA abundance and IFN and ISGs expression. Interestingly, the expression of some ISGs, with *Tnfsf10*, *Ifi44l*, and *Mx1* being the most prominent examples, was found to be negatively correlated with viral RNA abundance, which could be a direction for future studies on WNV-mediated immune suppression in infected cells. Similarly, Zanini et al. studied the molecular signatures indicating the development of severe dengue (SD) infection by analyzing single PBMCs derived from patients [36]. FACS was employed to sort PBMCs into different cell types (T cells, B cells, NK cells, DCs, monocytes), and then scRNA-seq was performed. The majority of viral RNA-containing cells in the blood of patients who progressed to SD were naïve immunoglobulin M (IgM) B cells expressing CD69 and CXCR4 receptors, as well as monocytes. Transcriptomic profiling data indicated that various IFN regulated genes, especially MX2 in naive B cells and CD163 in CD14+CD16<sup>+</sup> monocytes, were upregulated prior to progression to SD.

Comparison of the single-cell transcriptomes of lung tissue from health and influenza-infected mice revealed that 101 genes, among which the majority are ISGs and targets of antiviral transcription factors, were consistently upregulated among all nine identified infected cell types, including both immune and non-immune cells [26]. This finding suggested that antiviral innate responses against influenza infection generically exist (Figure 2C). Moreover, by contrasting the expression profiles among infected, bystander, and unexposed cells, it was shown that the non-specific IFN gene module is a result of extracellular exposure and responses of environmental signals.

While single-cell transcriptomics analysis provides an unbiased determination on host cell states, proteomics analysis offers direct characterizations of proteins expressed upon pathogen activation. Going beyond traditional flow cytometry, mass spectrometry, or cytometry by time-of-flight (CyTOF) offers vastly increased numbers of parameters that can be investigated simultaneously, exponentially increasing the depth of the dataset collected. For instance, to investigate the effect of a precedent dengue virus infection on the outcome of subsequent Zika infections, PBMCs derived from patients with either acute dengue infection or health individuals were incubated with dengue virus or Zika virus, and the treated PBMCs were assessed by multiparameter CyTOF [37]. CyTOF in this study allowed the simultaneous detection of changes in the frequency of immune cell subpopulations and quantification of functional activation markers and cytokines in distinct cell subsets. While secondary infection with dengue virus led to increases of CD4+ T cells and T cell subsets, which are involved in adaptive immunity, secondary infection with Zika virus induced the upregulation of several functional markers including IFNγ and macrophage inflammatory protein-1β (MIP-1β) in NK cells, DCs, and monocytes, indicating an intact innate immunity against Zika virus in the cases of possible concurrent dengue infection. Hamlin et al. compared two DENV serotypes (DENV-2 and DENV-4) in their infection in human DCs using CyTOF, which allowed simultaneous analysis on DENV replication, DC activation, cytokine production, and apoptosis [38]. The tracking of intracellular DENV proteins and extracellular viral particles showed different replication kinetics yet similar peak viral titers by these two serotypes, as well as the percentage of infected DCs. Moreover, DENV-4 infection was found to induce a higher expression of CD80, CD40, and greater production of tumor necrosis factor-α (TNFα) and interleukin-1β (IL-1β), compared to DENV-2 infection. Additionally, bystander cells, which were identified by the absence of intracellular viral proteins, were identified to produce less TNFα and IL-1β, but show more activation of interferon-inducible protein-1 (IP-1), which is a member of ISGs.

Besides CyTOF, host cell secretomes can also be measured with customized miniatured systems, and the level of multiplexing and flexibility of sample handling is often improved. For instance, Lu et al. showed the co-detection of 42 secreted proteins from immune effector cells stimulated with LPS [39]. In a similar setup, Chen et al. performed a longitudinal tracking of secreted proteins from single macrophages in response to LPS treatment [40]. These studies provide valuable insights into the dynamic and comprehensive responses to pathogen over time. Notably, such methods require microfabrication tools and skills, which is not always available and thus hinder their accessibility, compared with flow cytometery and CyTOF.

Epigenetic profiling at the single-cell level is also important, especially for elucidating the influence of host immune responses in chronic infection. The Assay for Transposase-Accessible Chromatin with high throughput sequencing (ATAC-Seq) utilizes Tn5 transposase to insert sequencing adapters into regions of open chromatin, in order to study genome-wide chromatin accessibility. Buggert et al. applied ATAC-seq and established the epigenetic signatures of HIV-specific memory C8<sup>+</sup> T cells resident in lymphoid tissue [41]. Yao et al. used chromatin immunoprecipitation followed by high-throughput sequencing (ChIP-Seq) to examine the histone modification of progenitor-like CD8<sup>+</sup> T cells from mice chronically infected with lymphocytic choriomeningitis virus (LCMV) [42]. They found that progenitor-like CD8<sup>+</sup> T cells showed distinct epigenomic features compared with memory precursor cells, exhibiting more abundant active histone markers (H3K37ac modification) at genes co-expressed with *Tox*, which encodes the thymocyte selection-associated high mobility group box protein TOX. This might promote the long-term persistence of virus-specific CD8<sup>+</sup> T cells during chronic infection.

In some cases, deep sequencing can be implemented together with other single-cell technologies for a comprehensive and systematic profiling of immune responses against infection. For instance, Michlmayr et al. performed 37-plex CyTOF on peripheral blood mononuclear cells (PMBCs), RNA seq on whole blood, and serum cytokine measurement of blood samples from patients with chikungunya virus (CHIKV) infection [43]. Moreover, samples collected at acute and convalescent phases were compared to study the disease progression. Such multidimensional analysis allows the large-scale, unbiased characterization of gene expression, cytokine/chemokine secretion, and cell subpopulation changes in response to infection. One important result of this study is revealing monocyte-centric immune response against CHIKV, with the frequency of two subsets both related to antibody titers and antiviral cytokine secretion. In addition, significant viral protein expression was found in two B cell subpopulations.

While multiple assays can be done on the same bulk sample to obtain different data parameters (e.g., transcriptomic, proteomic), such datasets are not able to correlate the data parameters at the resolution of a single cell. Newer advances allow the simultaneous collection of multiple types of parameters for the same cell. For instance, Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-seq) and RNA Expression and Protein Sequencing (REAP-seq) are techniques for the simultaneous collection of transcriptomic and high-dimensional information on specified proteomic targets. By using antibodies tagged with unique nucleotide sequences, the subsequent transcriptomic sequencing simultaneously sequences these tags to allow the quantification of the antibody targets. Corresponding transcriptomic and proteomic data at the single-cell level allows the opportunity to study the role of post-translational gene regulation in the immune response. The increased dimensionality of the information obtained may also allow more accurate machine learning to identify signatures of healthy or dysfunctional immune responses. For instance, using CITE-seq, Kotliarov et al. were able to identify a common signature of activation in a plasmacytoid dendritic cell-type I interferon/B lymphocyte network that was associated both with flares of systemic lupus erythematosus (SLE) and influenza vaccination response level [44].

#### **3. Therapeutics Discovery**

#### *3.1. Single-Cell Technology in Therapeutics Discovery and Clinical Application*

As noted above, the ability to study biological processes at the single-cell level gives an unprecedented to attribute bulk phenotypes in immunology and host–pathogen interaction to specific cell subpopulations, including rare cell populations, in a relatively unbiased fashion. Apart from basic science discovery, how do these insights affect clinical practice in infectious disease? Biomarker discovery is one obvious area of impact—the molecular differences found to underpin broader disease phenotypes can be used to diagnose or even predict disease. In particular, diagnosis is a notable problem in infectious disease, where identification of the causative pathogen can take days to weeks for culture-based systems, which may delay appropriate, targeted treatment [45]. Apart from biomarker discovery, single-cell technology is also revolutionizing the discovery of vaccines and therapeutics, which will be elaborated upon in the sections below. Other clinical uses of single-cell technology may require an increased uptake of such technologies within the hospital setting. For instance, one potential area of impact is antimicrobial resistance. The bulk genotype or phenotype of a pathogen population may not accurately identify its ability to become resistant to antimicrobials, since antimicrobial resistance can involve the selection of a previously rare, resistant population. Should single-cell technology become routinely used in hospitals, the increased resolution could enables the identification of such rare populations, which can inform the choice of antimicrobials prescribed. To generalize, this similarly applies to any disease phenotype that can be triggered by a rare host or pathogen cell population. The complexity of current single-cell technologies hinders their implementation in the clinic, and in the section titled Diagnostics, we highlight various steps that have been taken toward simplifying single-cell technology platforms to allow their clinical use.

#### *3.2. Vaccine Development*

The first step of the vaccine development pipeline would be to identify a promising disease antigen, which could be in the form of a recombinant protein or inactivated/attenuated virus. Unlike traditional vaccinology where vaccines were generated via pathogen growth and inactivation, the reverse vaccinology approach relies on predicting antigen features that are likely to trigger protective functions and engineering the antigen accordingly [46]. To predict these antigen features, two main approaches have been used: via whole genome sequencing and more recently, identifying and mapping the structural epitopes of neutralizing antibodies using the methods discussed later in this review [47].

After identifying a vaccine candidate, the next step would be to verify its efficacy. This efficacy is quantified based on its ability to bring about a set of specific immune responses which are specifically linked with protective functions, which are known as Correlates of Protection (CoPs) [48]. It is important to identify the CoPs for each vaccine for multiple reasons, including the following: (1) to understand the mechanisms of vaccine protection for improvement of vaccines, (2) to understand the mechanisms of vaccine protection for improvement of vaccines, (3) to determine the consistency of the vaccines produced, (4) to evaluate the levels of protection to patients before and after treatment, and (5) for the licensure of said vaccine [49]. Historically, most of the CoPs in commercial vaccines typically involve quantifying the titer of neutralizing antibody produced by antigen-specific memory B cells. In the past decade, better understanding of the in vivo vaccine response has led researchers to identify several relevant memory T-cell responses as CoPs, and these T-cell responses are usually quantified by measuring the expressed cytokines via techniques such as ELISpot, flow cytometry, and ELISA [50].

However, it remains difficult to define vaccine CoPs for a number of diseases. These include those diseases that cannot yet be eliminated by vaccine or infection-elicited immune responses (e.g., HIV-1 infection, tuberculosis), since a suitable end point of protection is not attainable. They also include those diseases for which vaccines do not yet exist but vaccine CoPs may be expected to differ from infection-related CoPs, including diseases for which natural clearance occurs via the innate immune response or early adaptive immune response (e.g., COVID-19). Even when immune parameters that correlate with disease risk are found, the causative mechanism of immune protection, or mechanistic CoP, may remain elusive if multiple immune parameters are elicited in parallel by a protective response. As seen in the excellent review by Plotkin [51], the CoPs may not always be as obvious or limited to humoral immunity, and since vaccines typically elicit multiple immune responses. This is especially true for the case of vaccines against complex pathogens such as HIV and malaria, where the resultant network of immune responses may not always be easily identifiable.

Single-cell approaches may define a greater space of immune parameters to be explored as CoPs. Furthermore, the increased breadth of data that can be obtained from a single sample is useful in increasing the number of hypotheses that can be probed, especially in longitudinal analyses, which are most useful for mechanistic immune studies but where the sample volume is often limited.

Furthermore, using a systems vaccinology approach via omics technology, researchers have begun to uncover these potential CoPs early in the vaccine development process [52]. In one of the earliest proof-of-concepts, Querec et al. successfully identified a CoP for vaccine efficacy on humans vaccinated against yellow fever. A gene marker present in CD8+ T cells which could predict for protection was discovered by using a multivariate analysis of the immune response via a combination of flow cytometry and microarray techniques [53].

With the rapid developments in single-cell omics technology, a deeper understanding of vaccine response can be obtained through an even more detailed mapping of the interactions between the various immune cell populations at the single-cell level, as well as identify the causes of heterogeneous vaccine response in individual immune cells [54]. This could be seen from the recent work by Waickman et al. [55] where a dengue vaccine elicited a highly polyclonal repertoire of CD8+ T cells that was identified using scRNA-seq. Combined with transcriptional analysis of the CD8+ T cells, the authors established a set of metabolic markers that could be potential CoPs for vaccine efficacy evaluation. Combining the simultaneous analysis of single-cell transcriptomic and TCR sequence

data, Tu et al. identified preferential transcriptional phenotypes among subsets of expanded TCR clonotypes. This is a strategy that may be highly valuable in assessing the functionality of T cells and their correlation to protection in vaccine responses [56].

#### *3.3. Antibody Discovery*

Antibodies are widely used in therapeutics and diagnostics due to their high specificity and generally low toxicity. Antibodies are capable of mediating protective functions against infectious diseases, including pathogen neutralization, antibody-mediated phagocytosis, antibody-mediated cellular cytotoxicity, and complement-dependent cytotoxicity. Antibody-containing sera remains in use for diseases where there are no other therapeutic options, including for viruses such as Hepatitis A or B, Rabies, Vaccinia, SARS-CoV-2 at the point of writing, and for toxins (e.g., snake venom). However, there are limitations to this approach: serum therapy from animal sources causes a risk of serum sickness due to immune reaction against animal protein, while pooled hyperimmune sera from humans is difficult to collect and standardize. Instead, the appropriate B cell clone that secretes antibody with protective activity can be isolated, and its antibody sequence can be obtained and expressed in culture to obtain monoclonal antibodies as therapeutics. Similarly, in diagnostics, monoclonal antibodies provide the specific recognition of pathogen antigens that allow the rapid diagnosis of infection.

In order to identify the correct B cell clone from thousands or millions of B cells, its antigen specificity and/or protective activity must be interrogated. This is classically done by cell immortalization (such as by hybridoma production or Epstein–Barr virus infection to generate B lymphoblastoid cell lines), followed by single-cell plating and expansion to obtain sufficient antibody from a single clone, and then the well-based screening of the antibody-containing cell supernatants. However, these techniques are low in throughput and efficiency, losing more than 99% of potential cells [57,58] for hybridomas, and 70–99% of potential cells for B lymphoblastoid cell lines [59,60]. Moreover, there remains a bottleneck in throughput at the subsequent stage of subcloning and screening the resulting clones to determine which clones are antigen-specific and functional for the desired purpose—even large experiments are limited to screening several thousand cells [61,62], or up to 100,000 cells for robot-assisted operations [63], whereas a single 30 mL human blood draw contains an order of magnitude more (approximately 900,000) candidate CD27+ IgD- class-switched memory B cells [64].

More recently, techniques that avoid the need for cell expansion have been developed—these speeds up the life cycle for monoclonal antibody discovery. Primary B cells expressing antigen-specific B cell receptors (BCRs) are labeled using fluorescent antigens, allowing flow cytometry-based single-cell sorting to isolate these antigen-specific B cells [65]. This technique is useful especially for the interrogation of memory B cells, which express the BCR on their surface. The interrogation of plasmablasts and plasma cells, which secrete antibodies but have low or no surface expression of the BCR, require other procedures such as the formation of an Ig capture matrix on the B cells [66], or alternative methods of screening that allow the physical separation of single cells such as droplets [67], nanowells [68–71], or microcapillaries [72]. Following the isolation of the desired B cells, they are lysed and their RNA is interrogated to recover the antibody heavy and light chain genes. With these techniques, both antigen interrogation and antibody gene recovery do not require large clonal cell populations, removing the need for inefficient and time-consuming cell expansion processes.

For the recovery of antibody genes, RT-PCR is commonly used, but recovery rates are typically low (<70% success rate for each pair of heavy and light chains) due to the large variability across the V gene families. Single-cell RNA-seq (Smart-seq2) is an alternative to RT-PCR, which results in improved recovery rates (>90%) [73]. BCR recovery can also be done in the same step as antigen-specific sorting via the use of DNA-barcoded antigens, such that both the antigen barcodes and BCR sequence are recovered simultaneously during single-cell NGS [74]. This has been used to successfully isolate broadly neutralizing HIV-1-specific antibodies and influenza-specific antibodies simultaneously from a single sample, although the resulting antibody candidates had variable neutralization functions, which required subsequent in vitro confirmation.

Another method for monoclonal antibody discovery is the use of phage display libraries, where phages expressing antibody genes are selected for using an antigen-coated surface in an iterative process of biopanning [75]. This has been a fast and effective method for monoclonal antibody discovery. The main limitation of phage library display is the random, largely non-native pairing of VH and VL genes, which may cause problems in subsequent antibody expression and production, and it may also have a higher likelihood of triggering anti-idiotypic allergic responses. More recently, a single-cell emulsion technique has been used for the interrogation of antigen specificity and high-throughput sequencing, allowing the interrogation of a yeast library utilizing natively paired human antibody repertoires [76]. Using it, rare broadly neutralizing antibodies against HIV-1 could be identified, albeit with the correct antigen required for identifying the desired B cell clones.

Antigen binding is the most common form of screening for monoclonal antibodies due to its compatibility with high-throughput methods including flow cytometry, biopanning, and nanowell-based ELISA. However, antigen binding may not correlate with functional activity against the intended target. For example, this may occur if the protein antigen used does not accurately mimic the native form of the antigen; monoclonal antibodies generated against the protein antigen may not be active against the native target [77–80]. Another example would be if functional activity requires binding in a specific orientation, such as virus neutralization requiring the monoclonal antibody to disrupt the receptor binding site [81,82]. Assays for monoclonal antibody function include assays for virus neutralization, opsonophagocytosis, antibody-dependent cellular cytotoxicity, and receptor agonism/antagonism [83]. Currently, these assays are typically done in bulk with relatively low throughput, creating a bottleneck in monoclonal antibody screening.

Microfluidic technologies, such as water-in-oil emulsions or nanowells, are being developed to increase the throughput of such assays. For instance, a high-throughput screen for enzyme antagonism using a droplet-based assay has been reported [84]. Using water-in-oil microdroplets, El Debs et al. co-encapsulated single hybridoma cells with an enzyme (ACE-1) and an enzyme substrate that emits a fluorescence signal upon enzyme hydrolysis, and they were able to sort out hybridomas secreting ACE-1-inhibiting antibodies through fluorescence-activated droplet sorting. Another group has recently also reported assays that are capable of assaying cellular internalization, opsonization, and the functional modulation of cellular signaling pathways [67], and several companies have also reported proprietary platforms that may be able to carry out some other functional assays [85]. However, the specificity and sensitivity of these assays have not been reported. The ability to immobilize single cells in nanowells allows repeated longitudinal profiling, which is a property that was utilized by Story et al. to obtain antibody–antigen binding curves that can classify related populations of B cells [71].

#### *3.4. Antibiotic Discovery and Antimicrobial Resistance*

The characterization of diverse bacterial populations, including microbiome studies, has traditionally been done at the bulk level. For instance, the selection of particular organisms out of a diverse population has been done by the plating and amplification of single colonies. However, this method is limited in throughput. The enhancement of throughput can be done via miniaturization—for instance, one study isolated antibiotic-resistant *E. coli* mutants by encapsulating and culturing single bacteria in nanoliter-scale droplets containing the antibiotic [86]. This approach can be applied to accelerate the identification of targets acted upon by antibiotics of unknown mechanisms.

Apart from being limited in throughput, traditional microbial selection systems also require the ability to culture the microorganism of interest in vitro. However, it is estimated that the bulk of microorganisms cannot be cultured and expanded in typical cell culture media [87]. One potential solution is to use microfluidic devices to physically separate and phenotype individual bacteria while immersing them in media derived from their natural environment. This method was adopted to identify a new antibiotic, teixobactin, from a previously unculturable β-proteobacteria belonging to a group of Gram-negative organisms not previously known to produce antibiotics [88].

#### **4. Diagnostics**

#### *4.1. Disease Monitoring and Clinical Diagnostics*

In the clinical setting, single-cell analysis techniques are currently rarely routinely used in infectious disease diagnostics and monitoring. It is still impractical to apply most of the other conventional single-cell analysis techniques for diagnostic applications due to the associated high costs, long workflow durations, and high degree of technical expertise required. One notable exception would be flow cytometry, where aside from its high initial equipment cost, its fast turnaround times, high sensitivity, and ease of operation make it a staple tool in clinical institutions worldwide [89]. Flow cytometry is mainly used to perform the immunophenotyping of blood cells against various disease-specific biomarkers [90–92]. The most prominent example would be in the routine monitoring of human immunodeficiency virus (HIV) progression by counting the number of CD4<sup>+</sup> T cells in a patient's blood sample [91].

The other single-cell technique that has seen some use in diagnostics against pathogens would be fluorescence in situ hybridization (FISH). As a diagnostic tool, FISH has numerous advantages that include low cost and complexity; its rapid turnaround time allows the diagnosis of fastidious bacteria and the ability to distinguish between mixed populations of pathogens at a single-cell resolution [93]. While FISH has been successfully used for the direct identification of panels of pathogens from blood samples [94,95], its reliance on image analysis as the readout limits the throughput of this technique, and the results are subject to user-to-user variation and bias [96]. To resolve these issues, a variant of the technique, FISH-flow, was developed. FISH-flow combines FISH with flow cytometry to achieve higher throughputs as well as automates the signal readout through the cytometric system [97], and it has been used to detect HIV reservoirs in T cells [98] as well as bacteria from blood [99].

#### *4.2. Toward Point-of-Care Applications*

While the ability to identify biomarkers at a single-cell resolution is certainly invaluable in the fight against infectious diseases, current flow cytometer systems are typically bulky and expensive, thereby limiting their use in a laboratory setting [100]. Fortunately, advancements in microfluidics and low-cost electronics have given rise to the development of portable platforms that can perform single-cell analysis in a point-of-care (POC) setting. Recent examples of portable cytometric systems that are relevant to infectious disease diagnosis include a miniaturized modular Coulter counter capable of label-free detection and the differentiation of particles of varying sizes [101], a low-cost and portable image-based cytometer for the quantification of malaria-infected erythrocytes [102] (Figure 3A), and a portable miniaturized flow cytometer that is capable of multi-channel fluorescence interrogation of whole blood samples [103].

The portability of such cytometers could mean faster turnaround test timings through on-site diagnostics and disease monitoring, hence expediting clinical decisions and improving healthcare outcomes in general [104]. In addition, the portability of such microfluidic systems lends to other practical applications of flow cytometry, especially in pathogen detection in water and food sources. Particularly, diarrheal diseases (a leading cause of death for children under the ages of 5) are closely linked to the consumption of contaminated water sources and could be mitigated via regular, on-demand pathogenic testing of drinking water [105].

However, adapting current single-cell technologies into a portable format holds its own set of unique challenges. Most of the existing literature surrounding such technologies still report separate sample enrichment or staining steps prior to cell analysis [106–108]; such additional preparatory steps increase assay complexity, which may not be desirable in a POC setting [109]. While a gamut of existing microfluidic technology has already been established for sample purification as well as for reagent addition and mixing, integrating the various modules into a single platform is typically not a trivial process [110]. For single-cell technology to make the successful transition from the lab to the bedside, such practicalities must be considered and successfully implemented.

#### *4.3. Digital Assays*

Digital assays are a relatively recent assay format comprised of the following steps: (1) the discretization of a single initial larger sample volume into multiple smaller volumes (typically via microwell, microvalve, or droplet emulsion partitioning techniques [111]), and (2) performing the chemical or biological assay on each individual volume to obtain a quantifiable signal [112]. Due to the ability to individually assay a large number of cells at the single-cell level, the digital assay format has been widely employed in single-cell omics studies [113]. In the field of infectious disease research, while most of the applications of digital assays have been centered on answering fundamental questions relating to pathophysiology, there are other single-cell diagnostic applications that can benefit tremendously from such an assay format.

An example mentioned earlier in the review would be rapid antimicrobial-susceptibility testing (AST) to address the surge of antimicrobial-resistant infections worldwide as a result of the misuse of antimicrobials. Phenotypic AST, which involves the culture of the pathogen in the presence or absence of antibiotics, may help guide treatment options, but existing conventional assays have low sensitivity and require a long time of 12–48 h for cell regrowth to achieve measureable assay outcomes [114,115]. Higher-sensitivity single-cell digital assays that have been recently reported can obtain measurable signals without requiring cell regrowth and could be the answer to reducing AST turnaround times (Figure 3B) [116–118]. Another application of digital assays could be in quantifying viral reservoirs in patients at a single-cell resolution. In HIV eradication studies, latent reservoirs are reactivated using latency-reversing agents (LRAs) for subsequent inhibition via antiretroviral therapy [119]. The ability to isolate and individually assay the patients' blood to obtain the distribution of reactivation states in the heterogenous cell population can give clinicians an idea of antiretroviral treatment efficacy in the future [120,121].

**Figure 3.** Single-cell platforms for infectious disease diagnostics. (**A**) Portable image cytometer capable

of performing the automated counting of cells containing malaria parasite. Image reproduced from reference [102] under a Creative Commons License. (**B**) Pump-free droplet emulsion generation system that is capable of performing antimicrobial-susceptibility testing (AST) of different species of bacteria with a turnaround time of ≈5 h. Image reproduced with permission from reference [117]. Copyright 2020 Royal Society of Chemistry. (**C**) Microfluidic impedance cytometry is able to differentiate between healthy and malaria-infected red blood cells at a single-cell resolution based on the difference in electrical impedance measured across two electrodes. Image reproduced from [122] under a Creative Commons License.

#### *4.4. Label-Free Analysis*

The development of label-free single-cell analysis techniques has been gaining considerable attention over the last decade with the advent of microfluidics because of their numerous advantages over their counterparts that require cell labeling. Some of the advantages include: (1) lower technical complexity and turnaround times in assay workflow because the preparatory step is omitted, (2) not requiring knowledge of cell biomarkers beforehand, making them suitable for assaying novel cell populations, and (3) by avoiding the use of labels that might affect the natural state of the cells, results might be more representative of actual in vivo cellular conditions [123]. Coupled with the precise fluid handling capabilities afforded by microfluidics systems, there is a burgeoning number of label-free single-cell analysis platforms that have been reported in recent years that are able to measure infection based on the inherent properties of the cells.

One prime example would be the identification of cells via their electrical properties, specifically electrical impedance. This impedance is derived from the change of voltage or current signal when single cells flow across a pair of miniaturized electrodes, and it has been shown to be able to differentiate between healthy and malaria-infected erythrocytes (Figure 3C) [122], as well as the viability and species of parasitic protozoa [124]. Another promising direction for label-free single-cell analysis is via measuring the inherent optical properties of cells. This has been shown in recent work such as the single-cell identification of parasites through their Raman spectra [125], the quantification of single-cell viral infection titer through Laser Force Cytology [126], and single bacteria detection via refractive index measurements [127]. Lastly, the mechanical and size properties of cells have also been exploited for identifying infected single cells. For example, using inertial microfluidics, white blood cells could be hydrodynamically isolated from lysed blood containing ring-stage malaria parasites as a result of the white blood cells' larger sizes [128]. Other recent works that demonstrate potential applications for single-cell label-free infectious disease analysis includes cell identification via their acoustophoretic responses [129], as well as their deformability and hydrodynamic resistance [130].

Evidently, there is a host of promising label-free single-cell analysis technology that could be translated to clinical diagnostic applications. In the near future, these technologies would be useful complement POC applications where labeling steps in the assay workflow would increase the technical complexity and hinder the transition from the lab to bedside.

#### **5. Considerations in Single-Cell Studies**

Among the methods discussed, scRNA-seq is the primary tool for single-cell studies. In the following section, we briefly cover some important points that needs to be considered when designing and conducting such experiments. For a more in-depth coverage of this subject matter, the reader is invited to read other excellent reviews from Luecken [131], See [132], and Lähnemann [133].

#### *5.1. Number of Cells and Sequencing Depth*

As covered earlier in this review, the main applications of scRNA-seq in infectious disease study comprise of the following: (1) studying effect of host cell heterogeneity on infection, (2) identifying host immune responses, and (3) antibody discovery. However, the number of sequenced cells and depth of sequencing ultimately depend on the end goal of each experiment as well as the amount

of financial resources at hand. The availability of a variety of commercial platforms for single-cell analysis with different throughput and sensitivity can provide users with different options to best suit the purpose of their studies [134].

For studies that involve identifying the cell types of a heterogeneous sample, a minimum of 50,000 reads per cell would be sufficient [135], while testing on a significantly large number of cells would ensure that rare subpopulations do not get missed out. One such application in infectious disease studies would be the systemic characterization of immune cell populations in response to an infection, wherein a large number of cells has to be screened in order to encompass the extensive diversity of B and T cells [136]. On the other hand, for studies which the main goal is to obtain a high resolution readout of the transcriptome for a small number of cells, 1,000,000 reads per cell would be a reasonable estimate [33].

#### *5.2. Reproducibility and Reliability of Data*

In typical bulk analysis, multiple biological and technical replicates can be performed in order to ensure the reproducibility of data. However, for single-cell experiments, particularly for scRNA-seq, there are two main issues to contend with. Firstly, measurements typically have high technical variability as replicate measurements cannot be performed on the same cell, which is lysed as part of the RNA extraction process. Secondly, the resulting single-cell data are typically noisy due to technical variations from the multitude of steps in scRNA-seq, as well as biological variation stemming from cell heterogeneity [137]. As such, great care has to be taken at each step of the scRNA-seq workflow (i.e., sample preparation, library preparation and sequencing, data analysis) to minimize such technical variability and batch effects.

One of the major sources of such variability arises from the initial sample preparation process. Regardless of how the cells are dissociated, purified, or enriched, cell expression is likely to change in response to the stress induced from these processes. To minimize such undesired changes which might affect downstream data analysis, the sample preparation protocol should be optimized iteratively for each cell type [138].

To reduce technical variability, one common method would be to spike = 0 in known quantities of synthetic RNA into the samples as controls to normalize read counts prior to data analysis [139]. A recent advancement in such RNA spike-in normalization methodology would be the BEARscc (Bayesian ERCC Assessment of Robustness of single-cell clusters), which generates simulated technical replicates based on the readout signal variation from spike-in measurements [140]. An alternative to RNA spike-ins would be the use of Unique Molecular Identifiers (UMIs) incorporated into the primers during reverse transcription, which essentially act as unique barcodes that allows the identification and subsequent tracking of transcribed mRNA. Then, the resulting data can be normalized against the UMI levels to account for amplification bias during the library generation step [141]. However, both RNA spike-in and UMI have their own set of limitations to consider; RNA spike-ins are unsuitable for protocols that utilize poly-T priming and template switching, and since they are typically used in large amounts relative to the endogenous RNA, they could potentially occupy a lot of reads. Protocols utilizing UMI need to ensure that library sequencing is sufficiently deep to cover all UMI transcripts; otherwise, there will be a risk of incorrectly quantifying the initial sample RNA [142].

Another source of error for scRNA-seq comes from batch effects, which are brought about by unavoidable variations between batches of experimental runs due to changes in environmental conditions, temperature, reagent lot, etc. In response, several computational methods have been developed to mitigate said batch effects from the scRNA-seq data. For example, one of the more commonly used batch-effect correction methods, ComBat, utilizes an empirical Bayesian framework that removes batch effects via a linear model, which factors in both the mean and variance of the scRNA-seq data [143]. For a more in-depth study on the comparative performance between the various batch-effect correction methods, we urge readers to consult a recent study by Tran et al. [144].

#### **6. Future Outlook**

#### *6.1. General Limitations of Current Single-Cell Platforms*

While single-cell platforms have indeed come a long way in the past two decades, the plethora of existing techniques still face a few general concerns that could present themselves as opportunities for development in the near future.

One of the inherent challenges in single-cell studies stems from the simple fact that the total amount of biological material present in a single cell is pretty limited and as a result, the resulting data are typically noisy from multiple biological and technical sources. Making sense of the data requires downstream data pre-processing and analysis, which are non-trivial components of the workflow that limits the accessibility of such studies to groups with the essential background. Additionally, with the increasing number and complexity of parameters at which single-cell assays are being performed, the curse of dimensionality is a pertinent problem that still requires further examination [133].

Another concern for single-cell platforms would be inter-experiment variability, as mentioned in the previous section. Single-cell technologies innately have high measurement sensitivity and thus are more susceptible to variations in results obtained from technical replicates, and methods to bioinformatically correct for such differences are required. Coupled with the fact that single-cell studies are typically expensive and therefore sample sizes are small, ensuring that results are comparable between each sample becomes an even more important issue.

Finally, the inability to maintain viable cells after analysis, particularly for high-throughput methods such as flow cytometry or scRNA-seq, gives rise to a couple of problems. Firstly, a majority of the conventional single-cell studies are limited to a single time point of study, following which the cells are discarded. Secondly, the irrecoverability of the cells makes it difficult to integrate back-to-back assays, which required measuring different parameters. As such, improvements in cell handling to improve cell viability would be invaluable in obtain multi-parametric datasets required for a more holistic understanding of cellular behavior.

#### *6.2. Techniques on the Horizon*

To date, the applications of single-cell technology have revolutionized our understanding of host–pathogen interactions. While many studies have focused on immune cells from the blood, the study of immune responses in the context of solid tissues or foci of infection (in both acute and chronic disease phases) is important to understand the local context of host–pathogen interaction. For this, techniques allowing the integration of spatial information with other single-cell technologies will be useful. For instance, imaging mass cytometry has been used to obtain quantitative information on 32 proteins at a spatial resolution of 1 μm [145]. This is done by systematically ablating a formalin-fixed tissue sample spatially line by line. The increased number of markers allows the fine distinction of cell subsets and activation states, providing valuable information on cellular roles in immune effector function or immunopathogenesis. It may even be possible to simultaneously obtain information on specific DNA and RNA targets via in situ hybridization. Similarly, several techniques have been recently developed to obtain simultaneous spatial and transcriptomic data, including multiplexed error-robust FISH (MERFISH) [146], laser capture microdissection sequencing (LCM-sequencing) [147], Tomo-seq [148], Slide-seq [149], and Spatial Transcriptomics [150]. These techniques may similarly be helpful in infectious disease to better define the interplay of immune cells, susceptible cells, stromal cells, and pathogens.

Another important gap that remains to be bridged is the ability to comprehensively access the state of a single cell across time. Both immune and infection processes are highly dynamic, but because most of the single-cell technologies listed above are destructive, changes over time must be assessed either by careful time-point studies, or by assuming the presence of a range of cells in a population that represent early and late stages of the process (e.g., cellular activation or infection stage). With the advent of microfluidic devices that can immobilize single cells for continued study, the same cell can be assessed at multiple points for longitudinal study. In addition to typical proteomic marker analyses and RNA or DNA in situ hybridization techniques with live cell imaging, it is already possible even to measure more complex phenotypes such as bioenergy metabolism [151]. Since microfluidic devices are also used for 3D organoid growth to simulate in vivo conditions, it is conceivable that future developments in technology will allow similar types of information to be collected in the context of organoids. This will represent one approximation toward high-dimensional in vivo data, which remains impossible with current methods.

Most of the single-cell studies reviewed in this article are based on end-point assays that are destructive and can therefore only measure a single time point of these single-cell targets. However, several recent studies outside the sphere of infectious disease have highlighted time as a prominent variable that influences the level of heterogeneity in host cells, immune cells, and pathogens. To that end, microfluidic platforms that enable the automated and precise control of media and reagents are ideal for performing such dynamic studies. For example, Wu et al. [152], using a customized microwell–microvalve system, performed a continuous measurement of A disintegrin and metalloproteinases (ADAMs) and matrix metalloproteinase (MMPs) secretions by single HepG2 cells upon a phorbol 12-myristate 13-acetate (PMA) challenge. Using their microfluidic platform, heterogenous changes in the secretion rates of ADAM and MMPs were observed in response to PMA stimulation, which may be used to predict HepG2 cell fates. In another recent study, a microfluidic platform that combined mutation visualization (MV) and microfluidic mutation accumulation (μMA) enabled real-time tracing of mutations of single bacteria [153].

Evidently, the utilization of microfluidic technology could enable high temporal resolution single-cell studies suited for uncovering the pathophysiology of infectious diseases. The advantages offered by microfluidics technologies include the efficient capture and compartmentalization of single cells, the precise control of fluid exchange, and ensuring a viable microenvironment for cell survival. These characteristics enable the dynamic study of large populations of single cells in parallel, which may eventually provide us with a more comprehensive understanding of the causes and effects of single-cell and single-pathogen heterogeneity.

#### **7. Conclusions**

Through the various applications of single-cell technology, we have gained a more thorough understanding of infectious disease pathophysiology at an unprecedented resolution. Revealing the heterogeneity within populations of pathogens has allowed a finer dissection of virulence factors, and similarly, heterogeneity within populations of infected cells has given us a deeper understanding of host immune defenses. In addition, the high-dimensional single-cell information that can be collected even from primary cells has allowed us to identify rare but important cell subtypes, and it has shed light on the complex interplay between the different cells of the immune system. With the advent of antibody and T cell-based therapeutics, and antibody-based diagnostics, the contributions of single-cell technology to the high-throughput identification of candidate B and T cell receptor sequences that are target-specific have also accelerated the development of new therapeutics and diagnostics for both newly emerging and existing diseases. The adoption of single-cell technologies is likely also to revolutionize clinical studies for both drugs and vaccines, given its immense potential for biomarker discovery. With the field of single-cell technology only just taking off in the last decade, there remain vast prospects in both the increased adoption of existing technologies and the development of new technologies.

**Author Contributions:** Y.L., R.L., M.Z.T., W.N.L., and L.F.C. contributed to the conception and design of the work. Y.L., R.L., M.Z.T., W.N.L., C.-H.C., and L.F.C. participated in the discussion and writing of the manuscript. All authors reviewed and approved the final version of the manuscript for submission.

**Funding:** This review was funded by IRG Grant [OFIRG17may105 to L.F.C.] from the National Medical Research Council in Singapore.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **A Single-Neuron: Current Trends and Future Prospects**

#### **Pallavi Gupta 1, Nandhini Balasubramaniam 1, Hwan-You Chang 2, Fan-Gang Tseng <sup>3</sup> and Tuhin Subhra Santra 1,\***


Received: 29 April 2020; Accepted: 19 June 2020; Published: 23 June 2020

**Abstract:** The brain is an intricate network with complex organizational principles facilitating a concerted communication between single-neurons, distinct neuron populations, and remote brain areas. The communication, technically referred to as connectivity, between single-neurons, is the center of many investigations aimed at elucidating pathophysiology, anatomical differences, and structural and functional features. In comparison with bulk analysis, single-neuron analysis can provide precise information about neurons or even sub-neuron level electrophysiology, anatomical differences, pathophysiology, structural and functional features, in addition to their communications with other neurons, and can promote essential information to understand the brain and its activity. This review highlights various single-neuron models and their behaviors, followed by different analysis methods. Again, to elucidate cellular dynamics in terms of electrophysiology at the single-neuron level, we emphasize in detail the role of single-neuron mapping and electrophysiological recording. We also elaborate on the recent development of single-neuron isolation, manipulation, and therapeutic progress using advanced micro/nanofluidic devices, as well as microinjection, electroporation, microelectrode array, optical transfection, optogenetic techniques. Further, the development in the field of artificial intelligence in relation to single-neurons is highlighted. The review concludes with between limitations and future prospects of single-neuron analyses.

**Keywords:** single-neuron models; mapping; electrophysiological recording; isolation; therapy; micro/nanofluidic devices; microelectrode array; transfection; artificial intelligence

#### **1. Introduction**

It would not be an exaggeration to state that the brain is the most complex structure present in the human body, with more than 100 billion neurons, ten times more glial cells, and hundreds of trillion nerve connections [1]. Neurons, the structural and functional unit of the nervous system, display a high complexity of cell diversity, and circuit organization rules. Rigorous research has demonstrated that the firing of even a single-neuron is sufficient to alter mammalian behavior or brain state [2]. Therefore, mapping individual neurons or sets of neurons with specifically distributed activity patterns displaying temporal precision is still an important and intriguing query. Single-neuron analyses in the mammalian brain requires crossing many technical barriers and involves four steps: (1) labeling individual neurons; (2) imaging at axon resolution levels in brain-wide volumes; (3) the reconstruction of functional areas or the entire brain via converting digital datasets of image stacks; (4) analysis to record morphological features of neurons with a proper spatial coordinate framework and also

extract, measure, and categorize biological characteristics, i.e., neural connectivity. Neuron morphology becomes a native illustration of type of neuron, replicating their input-output connections. The great diversity, huge spatial span, and troublesome dissimilarities of mammalian neurons present several challenges in labelling, imaging and analysis [3].

A major challenge in studying single-neuron anatomy is that many pathological factors like stroke, trauma, inflammation, infection, and tumors have not been recognized or deliberated to be an effect of the individual neurons. General clinical studies have almost neglected the role of a single-neuron in the absence of relevant technology and tools to do so. Additionally, conventional in vitro and in vivo assays predominantly measured an average response from a population of cells. Such information may be informative in most studies but is not enough in cases where subpopulation information determines the behavior of the whole population [4].

In the last two decades, the rapid advancement of micro- and nano-technologies and their integration with chemical engineering, chemistry, life science, and biomedical engineering has enabled the emergence of a new discipline, namely the lab-on-a-chip or micro-total analysis system (μ-TAS). The lab-on-a-chip can not only manipulate cells precisely but also provide an environment for single-cell analysis with little sample and reagent consumption. Precise single-cell analyses, including cultivation, manipulation, isolation, lysis as well as single-cell mechanical, electrical, chemical and optical characterization, can be conducted with relative ease using micro/nanofluidic devices [5,6]. These single-cell analyses can help us to understand different biological contexts, such as the functional mutation and copy number effects of genes, and cell–cell or cell–environment interactions. All of these analyses are crucial for the development of cellular therapy and diagnostics [3,6,7]. Because stimulating just one neuron can affect learning, intelligence, and behavior, conventional assays that mainly analyze the average responses from a population of neurons in the brain may not be sufficient to provide the required information. Through single-neuron analyses, relationships across neuron modalities, holistic representation of the brain state, and integration of data sets produced across individuals and technologies can be achieved, and would greatly benefit future precision medicine development. Single-unit recordings from human subcortical or cortical regions contribute significantly in enhancing the understanding of basal ganglia function and Parkinson's disease, neocortical function and epilepsy [8,9]. Single cell analysis was also employed in deciphering the neuronal signaling during epileptic form activity, owing to the alterations in metabolic state or level of arousal, and during normal cognition. After the first attempt at device implantation for single-cell recording in 2004, remarkable progress has been made. Based on the similar concept of single-cell recording and stimulation, the first intracortically directed two-dimensional (2D) cursor movements and simple robotic control were achieved by tetraplegia patients with an intracortical brain-computer interface. The studies conducted on patients with tremor medical condition using single-unit recordings helped in developing a better understanding of the role of individual basal ganglia and motor thalamic neurons, generating synchronized rhythmic firing in a tremor-associated manner [10].

The nervous system is composed of neurons and various supporting cells (oligodendrocytes, microglia, and astrocytes) of distinct morphology and neurochemical activity. Even a single sensory neuron activity can exactly predict the perceptions of animals [11]. Owing to the stochastic intercellular variation of the genome, epigenome, proteome and metabolome significantly cause variation in single-neuron response to therapeutics and the information is critical in precision medicine [12]. Therefore, isolation of distinct cells is a crucial step in single-neuron analyses, and the limitation factors associated with the process, such as efficiency or throughput, purity, and recovery, need to be improved.

This review article focuses on the latest developments in analytical technologies at the single-cell level in the nervous system. The technologies include modeling, isolation, mapping, electrophysiology, and drug/gene delivery (viral, optoporation, microinjection, and electroporation) at single-neuron levels. It also emphasizes therapeutic analysis and effect measurement using different micro/nanofluidic devices. Moreover, recent findings on the relationship between single-neurons and behavior and artificial intelligence will be summarized.

#### **2. Single-Neuronal Models**

Neuro-physiological research of single and multiple neurons has been carried out for centuries, yet the first mathematical model was established by Louis Lapicque in 1907 [13]. Based on the physical units of the interface, two categories of neuronal models were established. The electrical input–output membrane voltage model predicts the functional relationship between the input current and the output voltage. The other category, known as the natural or pharmacological input neuron model, relates the input stimulus (light, sound, pressure, electrical or chemical inputs) to the probability of a spike event. Even though many neuronal models were proposed, Hodgkin and Huxley's model (the H&H model) of the neuronal membrane is considered the classic neural model for computational neuroscience to date.

The base of the Hodgkin–Huxley (H&H) model lies in Bernstein's membrane theory, which was proposed in 1902 [14]. The H&H model established a relationship between the flow of ionic currents across the neuronal cell membrane and the voltage at the cell membrane. The major points from this theory were that the selective permeability of the cellular membrane allows only a particular concentration and type of ions to flow across the membrane. The voltage-current relationship was given by the formula:

$$\mathbf{C}\_{\rm m} \frac{\rm dV(t)}{\rm dt} = -\sum\_{i} \rm li \ (t, V) \tag{1}$$

where Cm denotes membrane capacitance, Ii is the current through a given ion channel, t is the time and V stands for voltage.

The Hopfield model discussed the distributive memory mechanism and the output firing rate [15]. The FitzHugh–Nagumo model is a qualitative and simplified two-dimensional model of the H&H model, in which regenerative self-excitation of a single-neuron was described [16].

Hindmarsh's and Rose's model is characterized by periodical or chaotical bursts of spikes. This model is used to model other neuron processes, which can be either be autonomic or cognitive [17].

The above-mentioned models involve several complex nonlinear differential equations; furthermore, the required simulation time is considerably more significant than the information about the neural circuit behavior. On the contrary, few models exist where the neuron is considered only an element and ignoring the complicated morphology of the dendrites and ionic mechanism inside the neurons and all the synapses were simplified as inputs with different weights. It considered only the input–output relationship of neurons as the simplified model. These models can be divided into two parts: (1) artificial neuron: this does not elucidate the mechanism of living neural circuits, rather a constructed artificial neural networks with some specific function to solve a practical engineering problem; (2) realistic simplified neuron model: though it is not based on subcellular mechanisms, yet its main assumptions are realistic and based on the available knowledge about the behavior of living neurons. Realistic models are further categorized into two classes based on the method of coding. Temporal coding, i.e., Louis Lapicque's integrate-and-fire model (1907) [18] and McCulloch and Pitts (1943) [19]; and neural coding considered as rate coding (output of a neuron is a continuous variable-firing rate of the frequency, for example, the Hopfield model (1994)) [16].

Briefly, the earliest model of a neuron, i.e., the Integrate and Fire model, represents neurons in terms of time. The firing frequency of a single-neuron was formulated as a function of constant input current, and it was given by frequency, f (I) = I/CmVth + tref I, where Cm denotes the membrane capacitance, Vm the membrane potential, I the membrane current, and tref the refractory period.

The drawback of this model was that sometimes when it received a below-threshold signal, the voltage boost of the model was retained until another firing occurred (i.e., lack of time-dependent memory). Thus, another model, the Leaky Integrate and Fire model, was proposed by adding a leak term to the membrane potential to resolve the memory problem. Since the cell membrane is not a perfect insulator, a membrane resistance that forces the input current to exceed the threshold (Ith = Vth/Rm) cause the cell to fire. The firing frequency with the membrane resistance (Rm) is given as

$$\mathbf{f} \text{ (I)} = \left\{ \begin{array}{c} \text{0 I} < \text{I}\_{\text{th}}\\ \left[ \text{t}\_{\text{ref}} - \text{R}\_{\text{m}} \text{C}\_{\text{m}} \log \left( 1 - \frac{\text{V}\_{\text{th}}}{\text{R}\_{\text{m}}} \right) \right]^{-1} \text{I} > \text{I}\_{\text{th}} \end{array} \right\} \tag{2}$$

where Ith and Vth denote threshold current and threshold membrane potential, respectively.

In summary, neurons can be considered as dynamic systems; therefore, nonlinear dynamical approaches are appropriate to justify the variation in their behaviors [20]. After going through all these models, the doubt becomes even more generic for deciding the basic unit of the nervous system: neurons or ion channels. Considering the variation in both neurons and ion channels, it would be justified to select either or some other entity as the basic unit of the nervous system in that particular or similar condition. Here, various single-neuron models and their categories and drawbacks are summarized in Table 1.


**1.**Single-neuronmodels,includingtheofmodelproposal,modelcategory,typeofmodel,keynotes,and

#### **3. Behavior and Single-Neurons**

It is accepted that behavior is the result of brain function and brain processes govern how we feel, act, learn, and remember [33]. The understanding of the performance and capacity of single cortical neurons on a perpetual task is a prerequisite for establishing the link between the brain and behavior [34,35]. Accumulating evidence in cortical research has shown that single-neurons match behavioral responses in discriminating sensory stimuli [36,37]. Cortical neurons show highly nonlinear responses as a result of probing by complex natural stimuli [38–42]. The first instance of stimuli-caused accurate discrimination was reported by Wang et al. using a songbird model to test the occurrence of natural behaviors involving complex natural stimuli [43,44]. In this context, the available sensory information in response to a song consists of a single spike train from all the neurons of the particular population. The quantification of all single spike trains helps in evaluating the contribution of single-neuron behaviors [45]. It can also be concluded that spike timing has a major impact on performance than spike rates and interspike intervals. Further temporal correlations in spike trains enhance the single-neuron performance in most cases [2].

Another study assessing the sensitivity in measurements of single middle temporal (MT) neurons towards the direction of discrimination suggests that a small number of neurons may account for a psychophysiological performance [46]. Nevertheless, sampling-based variation in the single MT neuron activity predicted a weak correlation with behaviors. The results suggest that the decision is dependent on the collective responses of several neurons [36]. Therefore Cohen et al. proposed two possible explanations for this paradox: (1) a long stimulation duration may overestimate neural sensitivity in comparison with psychophysical sensitivity; (2) mistaken assumptions due to insufficient data are possible when noise correlation level in MT neurons supports reverse directions. This quantitates the role of single-neurons in perception, dependent on the duration and the noise correlation [47]. Similarly, the variability of responses to visual stimuli in striate cortex neurons was analyzed, and the results showed that perceptual decisions on signals arise from a rather small number of neurons and are correlated across neurons [48]. The results also demonstrated the correlation between the pooled signals and neurons along with other neurons, and thus apparently the perceptual decision, generating high choice probabilities [49].

Similarly, Pitkow et al. predicted the role of single sensory neurons in behavior during discrimination tasks [50]. The notion is based on the limited sensory information from neural populations, either due to near-optimal decoding of a population with information-limiting correlations or by suboptimal decoding that is blind to correlations. Both possibilities involve different interpretations for the choice of correlations, i.e., the correlations between behavioral choices and neural responses. To assess this, experiments were conducted to record extracellular activities of single-neurons in the cerebellar nuclei (VN/CN), dorsal medial superior temporal (MSTd) and the ventral intraparietal (VIP) areas using epoxy-coated tungsten microelectrodes (FHC; 5–7 MΩ impedance for VN/CN, 1–2 MΩ for MSTd and VIP). The theoretical and experimental results shown in Figure 1 indicate the significance of noise correlations, which are governed by the response of the brain to these fundamental changes followed by processing sensory information [50].

**Figure 1.** Model for neural responses and decoding: (**a**) tuning curves f(s) showing the mean neural responses to a stimulus s (thin lines), curve from the von Mises functions (thick curves) model with parameters including the preferred stimulus sk (dots). (**b**) The relationship of two neurons generalizing to high-dimensional response spaces under varying stimulus s. (**c**) Linear decoding projects the neural responses, both noise and signal, towards a specific direction w for the estimation of s of the stimulus. ˆ (**d**) The phenomenon of showing neurons having similar tuning has higher correlated fluctuations. Noise correlation coefficients Rij between distinct neurons i and j are modeled as being proportional on average to the signal correlations Rijsig, with proportionality c0. (**e**) Two components to the noise covariance Σ: information-limiting correlations are distinguished; present along the signal direction f and therefore show covariance εf - f -<sup>T</sup> (front, matrix boxed in red), and the remaining noise with covariance Σ<sup>0</sup> (back, the matrix in the green box). The two types of noise show distinctive structures; apparent in the covariance matrices. The striations in the matrices correspond to the heterogeneous tuning curve amplitudes. Reprinted with permission from the authors of [50].

Single-neuron studies also illustrated the role of interval-selective neuron population for revealing changes in behavioral significance temporal patterns of presynaptic input. The behavioral sensitivity within the millisecond timescale in natural scallops was observed at the minimum, because of the midbrain neuron population acting as temporal filters intended for electrical communication signals. Variation of an order of interpulse intervals (IPIs) and the addition of even 1 ms jitter to natural scallops have the scope to affect both behavior and single-neuron responses even by different individuals. An amount of poorly decodable information is encoded in sensory and motor circuits via temporal patterns of spikes [51].

In most of the models, the precise control of the temporal input pattern onto temporally selective neurons in vivo is tough. Therefore, this limitation was overcome by the mormyrid electric fish model, where similar temporal patterns were found for presynaptic inputs against interval-selective central neurons and electrosensory stimuli [52]. Furthermore, based on single-neuron analysis, electric communication signals are tunable according to behavioral relevance. This shows that temporal patterns of presynaptic input onto interval-selective neurons can be tuned along with recording the responses of these neurons to input patterns, present while natural communication behavior. The results also show coherence between earlier findings of auditory and electrosensory pathways related to discriminating among scallops from different individuals. The neuron spikes of songbird field L neurons, grasshopper auditory receptors and higher-order neurons, and wave-type electric gymnotiform fish (which evolved their electric sense independently of mormyrids) hindbrain neurons would help in identifying conspecific signals by each individual. However, it is beyond the scope of single-neuron variation, reacting to natural signal differences to measure the power of single-neurons corresponding to specific temporal alterations [53].

#### **4. Single-Neuron Isolation**

Depending on the application, several techniques have been employed to isolate single-neurons. The pipette approach is the most commonly exploited single-neuron isolation method. Pipetting is a flexible approach that allows applications such as the functional electrophysiology, imaging, and transcriptomics of neurons to be achieved simultaneously [54]. The pipette isolation process is well equipped with video recording and image documentation facility and is thus suitable for post-capture quality control. Moreover, the protocol can be adjusted for isolating subcellular structures, such as dendrites and even biomolecules. Isolated ribonucleic acid (RNA) samples from single-neurons, allows the generation of transcriptomics data using either microarray or RNA sequencing (RNA-seq) techniques. Single-neuron transcriptomic analyses provide deep insight views into cell function and enable sorting out the global variations among single-neurons. The isolation of RNA from single cells in intact tissue and the subsequent handling of a large number of RNA samples require advanced instrumentation. Protocols, including the collection softwares, photoactivated localization microscopy (PALM) and laser capture microdissection techniques, have been developed for isolating single-neurons from cultures and tissue slices via pipette capture [55]. Laser capture microdissection is an indirect touch technique to isolate a single cell without altering or damaging the native morphology and chemistry of the sample as well as surrounding cells; this therefore makes this technique suitable for isolating cells for downstream processing, i.e., DNA genotyping and loss of heterozygosity (LOH) analysis, RNA transcript profiling, cDNA library generation, proteomics discovery and signal-pathway profiling [56]. The method employs a focused laser beam to melt the thin transparent thermoplastic film placed on a cap on the target cells. The melted film infuses with the underlying selected cell and allows the transfer of the attached targeted cells to a microcentrifuge tube for further downstream processing. Individual dopaminergic neurons or the ventral tegmental area are successfully isolated by the blend of infrared capture laser and the ultraviolet cutting laser exposure on polyethylene naphthalene membrane slides [56]. The support membrane maintains the integrity of the desired region while lifting during the sample collection.

Another approach to isolate single-neurons uses dielectrophoresis (DEP)-based microfluidic devices. Dielectrophoresis is an electro-kinetic phenomenon based on movement (trapping, alignment, and patterning) of polarizable particles (in this case, cells) under the influence of a non-uniform electric field. The technique employs minimal electric field intensity and therefore does not cause damage to neurons. Nevertheless, the low ionic strength buffer used in DEP may sometimes result in high susceptibility of the neurons towards the physiochemical environment (i.e., pH, temperature, humidity, and osmotic pressure) as well as transfection and transduction outcomes. Additionally, observation of morphology and activity of cultured neurons in DEP experiments under an inverted microscope may be limited due to non-transparent electrodes and substrate used in the devices [57]. The problem

was overcome, however, using a fully transparent DEP device fabricated with indium tin oxide (ITO) multi-electrode arrays and polydimethylsiloxane (PDMS). Such a device can be mounted on a microscope equipped with an incubator system to avoid contamination. The DEP electrode array traps and releases neurons (one at a time/electrode), as shown in Figure 20. The segregated single-neurons can be cultured and monitored over time, allowing the screening of various electrophysiological parameters and enabling detailed neurological studies [58].

#### **5. Single-Neuron Mapping**

The complex architecture of the human brain and how the billions of nerve cells communicate have perplexed great minds for centuries. However, in recent years, the rapid development of many new technologies is allowing neuroscientists map the brain's connections in ever-available detail. Brain navigation has become more accessible than ever and we are now able to fly through significant pathways in the brain, perform comparison among circuits, scale-up the exploration of cells comprising the region, and the functions depending on them. The Human Connectome Project (HCP), targets creating a complete neuron map involving structural and functional connections in vivo, within and across individuals, providing an unparalleled compilation of the neural data.

From each synapse to single-neurons to long-range neural networks, combining individual maps could create a "meta-map" that provides something closer to a full, detailed computer simulation of brain networks. The use of high-end brain mapping technology CLARITY, in addition to light microscopy, has allowed researchers to draw limited maps for specific neurons of interest, even in large brains [59]. The CLARITY is a technology to transform intact biological tissue into a hybrid form where tissue component removal and replacement takes place with exogenous elements for better accessibility and functionality. The light microscope is not competent to decipher all at the nanometer scale—thin wires and synapses, connecting neurons—only electron microscopy (EM) possess the power to do that. "The wires define the computations that are possible by the circuits", says Albert Cardona, a group leader at the Howard Hughes Medical Institute's Janelia Research Campus. The subjects studied in connectome research range from living individuals to the preserved brains of tiny animals such as worms and flies. The investigative technologies are also diverse, ranging from light and electron microscopy to Magnetic Resonance Imaging. Regardless of the approaches, painstaking efforts have to be exerted to build an atlas, even with the aid of powerful computation tools. Although the roles of single-neurons in brain functioning have not been fully elucidated, a high-resolution neural connectome map that precludes redundancy to facilitate clear messaging is essential to understand the brain. At first, charting and understanding the full wiring diagram of the brain seems to be an impossible task, yet recent technological advancements make it optimistic without requiring decades to complete. Such an ambition also prompts efforts to overcome major challenges in robustness and reproducibility during sample preparation, handling, and analysis. Technologies concerning automatic image data acquisition and efficient data storage and analysis tools also need to be developed. This section will briefly discuss these challenges and possible solutions, together with novel imaging techniques to meet the challenge of single-neuron mapping in the nervous system [60].

Kebschull et al. highlighted the importance of understanding the fundamental neural wiring network to figure out how the brain works [61]. Similarly, Professor Toga pointed out that brain mapping is similar to traditional cartography that shows even the footpaths and steppingstones of individual neurons and synapses at resolutions of a few nanometers [62]. Neuronal cell types are the nodes of the neural circuit regulating the information flow through long-range axonal projections in the brain. Single-cell and sparse-labeling techniques have been employed to reconstruct long-range individual axonal projections in various parts of brain, i.e., the basal ganglia, neocortex, hippocampus, olfactory cortex, thalamus, and neuromodulatory systems, with limited reliability and throughput of axonal reconstruction due to labeling restrictions executed on one or very few neurons within a single brain. The manual tracking of individual distinct segments among consecutive slices generally gets deformed or damaged during standard histological processing techniques. Although the reliable

and efficient reconstruction of long-range axonal projection can be achieved by visualizing neurons in continuous whole-brain image volumes. The serial two-photon (STP) tomography-based fast volumetric microscopy provides high-resolution imaging in complete three-dimensional space in a large volume of tissue, thus minute axonal collaterals may be unambiguously tracked to their targets [63]. Along with using this technique, high intensity sparse neuronal labeling, the new tissue clearing method, and bioinformatics tools to process, handle, and visualize huge imaging data lead to a suitable platform to efficiently reconstruct the axonal morphology. This was demonstrated by reconstructing the extensive, brain-wide axonal arborizations of diverse projection neurons present in the motor cortex within a mouse brain, as shown in Figure 2 [63].

**Figure 2.** Complete reconstruction of axonal morphology. (**a**) Complete reconstruction of the five projection neurons, superimposed on a horizontal (left) and sagittal (right) position while imaging the mouse brain. The subset comprises pyramidal neurons in layer II (blue, purple), layer V (red, black), and layer VI (green). (**b**) Axonal and dendritic reconstruction of the layer, five pyramidal cells (colored red in (**a**) presented in the coronal plane. The black dashed line depicts the profile of the coronal section at the rostrocaudal position of the cell body. Colored segments highlight axonal arbors initiating from common branch points. Reprinted with the permission of the authors of [63].

*Cells* **2020**, *9*, 1528

Manual intervention of the dataset remains a major bottleneck for neuronal reconstruction. A specialized custom reconstruction software generally takes 1–3 weeks to reconstruct a complete complex cortical neuron from precisely stitched brain volumes [3]. To increase the throughput of single-neuron mapping, an RNA sequencing-based method was developed [61]. Zador et al. implemented a Multiplexed Analysis of Projections by Sequencing (MAPseq) method, based on speed and the parallelization of high-throughput sequencing for brain mapping [64]. Multiplexing can be achieved in MAPseq by short, random RNA barcodes for unique and distinct labeling of individual neurons [64–66]. Barcodes are important as their diversity grows in an exponential manner as per the sequence length, overpowering the restricted resolvable color range. For example, the 30 nt sequence has a potential diversity of generating 430–1018 unique barcode identifiers, way more than what is needed to distinguish 10<sup>8</sup> neurons in a mouse brain [67]. As fast and inexpensive high-throughput sequencing can differentiate the barcodes, the MAPseq has the potential to identify the projections of millions of individual neurons in a brain simultaneously. In MAPseq, neurons are uniquely labeled by injecting a viral library encoding an assorted group of barcode sequences in a source region (see Figure 3). The highly expressed barcode mRNA is transferred to the axon terminals located at distal target projection regions. Later, the barcode mRNA is extracted from the injection site or target area and sequenced to read out the single-neuron projection pattern, as shown in Figure 3. The target should be precisely dissected to achieve higher spatial resolution. Like green fluorescence protein (GFP) tracing, MAPseq is unable to trace fibers of passage, therefore leaving out large fiber bundles while dissecting the target areas is critical in the study. This method takes less than a week to determine the brain-wide map of projections of a particular area, allowing efficient single-neuron circuit tracing [61].

**Figure 3.** Multiplexed Analysis of Projections by Sequencing (MAPseq) procedure for mapping single-neuron projections. (**A**) Various underlying projection patterns develop identical bulk mapping. (**B**) Random labeling of single neurons with barcodes. (**C**) The expected fraction of uniquely labeled cells is given by F = (1-1/N)(k-1), where N is the number of barcodes and k is the number of infected cells, assuming a uniform distribution of barcodes. (A1, primary auditory cortex; Ctx, neocortex). (**D**) In MAPseq, neurons are infected at low MOI with a barcoded virus library. Barcode mRNA is expressed, trafficked, and can be extracted from distal sites as a measure of single-neuron projections. Reprinted with the permission from [61].

As described earlier, along with the limitation of spatial resolution due to micro-dissection, MAPseq might show inherent sensitivity. Therefore, neuronal reconstructions based on microscopy ensure the gold standard for deciphering connections as well as the spatial organization of axonal projections. Optical imaging approaches, in combination with genetic tools and computational techniques, are starting to enable such global interrogations of the nervous system [68]. Haslehurst et al. employed a custom-built light fast sheet microscope (LFSM) using synchronized galvo-mirror and electrically tunable lens. The high-speed image acquisition facilitated the dendritic arborization of a living pyramidal neuron for 10 s in mammalian brain tissue at configurable depth. Post-hoc analysis represented localized, rapid Ca2<sup>+</sup> influx events occurring at various locations and their spread or otherwise through the dendritic arbor [69]. Prior to this, Ahrens et al. used high-speed light-sheet microscopy for image the neurons in intact brain of larval zebrafish with single-neuron resolution. They could image as many as 80% of neurons at single cell resolution, while the brain activity was being recorded once every 1.3 s by genetically encoded calcium indicator GCaMP5G. The indicator is expressed under the influence of the pan-neuronal elavl3 promoter. The SiMView light-sheet microscopy framework plays a key role in volumetric imaging during this fast, three-dimensional recording from an entire larval zebrafish brain, mostly consisting of ~100,000 neurons [70]. The chemically cleared fixed brain tissues were also imaged with single-cell resolution using light sheet microscopy and the reconstructions of dendritic trees and spines in populations of CA1 neurons in isolated mouse hippocampi was performed [71].

Multiple variants of super-resolution microscopy, including structured illumination microscopy (SIM), stimulated emission depletion microscopy (STED), and photoactivated localization microscopy (PALM)/stochastic optical reconstruction microscopy (STORM), each with special features, have overcome the drawbacks of conventional microscopy and have helped remarkably in neuroscience to decipher mechanisms of endocytosis in nerve growth and fusion pore dynamics, and also describe quantitative new properties of excitatory and inhibitory synapses [72,73]. Though most recently, a super-resolution microscopy approach was developed to unravel the nanostructure of tripartite synapses with direct STORM (dSTORM) using conventional fluorophore-labeled antibodies. As a result, the reconstruction of the nanoscale localization of individual astrocytic-glutamate transporter (GLT-1) molecules surrounding presynaptic (bassoon) and postsynaptic (Homer1) protein localizations in fixed mouse brain sections was achieved [74].

Economo et al. imaged the whole brain with a sub-micrometer resolution with the help of serial two-photon tomography. The sensitivity of the method also allowed manual tracing of fine-scale axonal processes through the entire brain, as shown in Figure 4 [63,75].

Further improvement has been made to develop a semi-automated, high-throughput reconstruction method to reconstruct >1000 neurons in the neocortex, hippocampus, hypothalamus, and thalamus. Figure 5 shows the schematic representation of reconstruction for 1000 projection neurons. The reconstructions are made available in an online database MouseLight Neuron Browser with a wide visualization and inquiry window [76]. The findings discovered new types of cells and established innovative organizational doctrines which handle the connections among brain regions [77].

**Figure 4.** Axonal arbor for three cortical projection neurons of layer five of the motor cortex collapsed in the sagittal plane (**a**) and coronal plane (**b**). Intratelencephalic neurons shown in yellow and green color are projected to other cortical areas and the striatum with a higher level of projection heterogeneity. Pyramidal tract neurons (red) are connecting the motor cortex with hindbrain and midbrain. Reconstructions are retrieved from MouseLight Neuron Browser [76]. Total axonal lengths of shown neurons are 44.7, 30.1 and 13.4 cm for yellow (ID: AA0100), blue (ID: AA0267), and red (ID: AA0180), respectively. Reprinted with the permission of [75].

**Figure 5.** Schematic representation of 1000 projection neurons reconstruction deciphering new cell types and long-range connectivity organization present in mouse brain. Reprinted with the permission of [77].

#### **6. Electrophysiological Recording**

The electrical nature of neurophysiology was first identified by Italian scientist Luigi Galvani in 1794 [78]. The first recording of extracellular action potentials was carried out using a tungsten electrode of sub microns diameter tip sizes by Hubel [79]. The study of individual neurons provides high spatiotemporally resolved activities, which help us to study the inner working function of the brain [80]. In 1977, Gross et al. designed a two-dimensional multi-microelectrode system to study the single-unit neuronal activity. The microelectrode system, as shown in Figure 6, and it was fabricated by a photoetching process followed by galvanic plating of gold to produce a high-density gold electrode array. The 12 μm wide and 2 μm thick gold conductor de-insulated at the tip with a single laser shot. The de-insulated conductor had an impedance at 1 kHz of approximately 4 MΩ for a smooth gold surface and 2 MΩ for a rough gold surface facilitating electrophysiological recordings from more than 30 neurons [81].

**Figure 6.** The microelectrode array consists of 36 photoetched microelectrodes with electrode holders, culture ring, and contact strips to study the single-unit neuronal activity. Reprinted with the permission of [81].

Traditionally, a technique called stereotrode was designed in 1983 to record the extracellular action potentials of the nervous system—the ratio of the distance between the cells and two electrode tips governs the spike-amplitude ratios—while recording via both the channels. For this study, the electrode pair fabricated from Teflon-insulated platinum-iridium wires of 25 μm diameter, with an impedance of 1 MΩ at 1 kHz was used. The recordings provided a study on the statistical interaction among the spike trains of a local set of neurons, which improves the quality of the chronic unit recordings [82]. In 1999, a neurochip with a 4 × 4 array of metal electrodes recorded and stimulated electrical activity in individual neurons with no crosstalk between channels. By using this device, the action potentials recorded from individual neurons were detected with a signal-to-noise ratio of 35–70:1. But the chip showed the survival of neuron rarely beyond 7 days [57].

Considering the scope and limitations of this review paper, the electrophysiological recordings from single-neuron level are categorized into two parts: in vitro recording and in vivo recording. The in vivo part also includes single-neuron recordings from brain slices and ex vivo.

#### *6.1. In Vitro Recording*

With the advancement of technology, multielectrode platforms have been developed with thousands of electrodes for the stimulation and recording of cell activity. In vitro single-neuron recording can be carried out using a 64 × 64 microelectrode array consisting of a total of 4096 microelectrodes with high spatial (21 μm of electrode gap) and temporal resolution (0.13 ms to 8 μs for microelectrodes of 4096 and 64 respectively), as depicted in Figure 7a,b. With high neuronal populations, the possibility to study an individual neuron is difficult; hence, low neuronal culture populations are preferred for single unit activity study. Also, single-pixel electrodes were selected to record signals from single-neurons and were interpreted to identify spiking and bursting events [83]. Mitz et al. conducted experiments on the frontal pole cortex of macaque monkeys to record the single-unit activity and neurophysiology of single cells. The recordings were performed by inserting 4–13 moveable microelectrodes, and their position was confirmed by magnetic resonance imaging. The monkeys experiments were conducted to perform three tasks out of which two were strategy tasks, and one was the control task, and the activity of isolated neurons was recorded [84]. Similarly, microelectrode arrays with 59,760 platinum microelectrodes [85], a complementary metal–oxide–semiconductor (CMOS) multielectrode array (MEA) chip with 16,384 titanium nitride electrodes [86], and 26,400 bidirectional platinum electrodes [87] also exist for in vitro electrophysiological recording with single-neuron resolution. The results depicted the activation of single-neuron arrays via intracellular stimulations. Electrophysiological recording shown the potential of tracing spiking neurons within neuronal populations, which is helpful to reveal the connection and activation modalities of neural networks [88]. Further, for better electrical interfacing with the aim of minimizing neuronal membrane deformation during the intracellular access, a vertical nanowire multi electrode array (VNMEA) was developed. This platform is capable of neuronal activation with the spatially/temporally confined effect along with recording its activity [89]. Next-generation non-invasive electrophysiology recording platforms are developed in the form of a thin-film, 3D flexible polyimide-based microelectrode array (3DMEA), facilitating the formation of 3D neuron networks. The array consists of 256 recording or stimulation channels. The action potential spike and burst activity were recorded for human-induced pluripotent stem cell (hiPSC)-derived neurons and astrocytes entrapped in a collagen-based hydrogel and seeded onto the 3DMEA, over 45 days in vitro [90].

**Figure 7.** (**a**) Electrophysiological platform integrated with a complementary metal–oxide–semiconductor (CMOS) microelectrode array chip, the interface board, and a workstation. (**b**) Immunofluorescence imaging of single-neurons on the chip and the electrophysiological activity of three selected neurons. Reproduced from [83] with the permission of the Royal Society of Chemistry.

#### *6.2. In Vivo Recording*

Further, to study in vivo single-unit activity, stereoelectroencephalography probes with a parallel batch of polyimide-platinum cylindrical microelectrodes of 800 μm diameter were used. The configurations of up to 128 electrode sites were set up to study the single-neuronal activity when various tasks were performed [91]. Similarly, the stereoelectroencephalography probes with 18 platinum microelectrodes of 35 μm diameter with an impedance of about 255 kΩ at 1 kHz were designed to measure the single-neuron activity to study focal epilepsy [92]. The dendritic integration of neurons can be studied only if the inhibitory and excitatory synaptic inputs of individual neurons are measured. For this measurement, an extracellular high-density microelectrode array of 11,000 electrodes were fabricated with firing at microsecond resolution. The presynaptic potentials were measured for a patched single-neuron with high reliability by eight randomly selected electrodes from the array [93]. In a study, the electrophysiological recordings of single-neurons were carried out by the patch-clamp technique followed by RNA sequencing to reveal the physiological and morphological properties of an individual neuron [94]. Single-neurons that were electrically transfected with plasmid DNA using micropipettes were studied for electrophysiological recordings. The membrane potential of the transfected and non-transfected neurons was examined to check whether there was any discrepancy, and was found to be −72 mV and −71 mV, respectively. Also, the electrophysiological properties of transfected and non-transfected neurons in brain slices were recorded and it was noted that the electroporation process did not affect the characteristics of the individual neurons [95].

Multielectrode array can record the two-dimensional range of action potential propagation in single-neurons via averaging the signals recorded extracellularly, which were detected by multiple electrodes. Here, medium-density arrays with an electrode pitch of 100 ± 200 μm were used to detect action potentials from single-axonal arbors. This non-invasive extracellular recording helped to identify the spiking of an individual neuron and it can be used to observe variations because of degeneration and in disease-models [96]. Electrophysiological recordings of single-neurons in the cortical and

subcortical of mammalian animals were conducted using various conformations of microelectrode matrices. Microelectrodes were made from Teflon coated stainless steel with 50 μm diameter with two parallel rows of eight microwires each. They were inserted as chronic implants in rat primary (SI) somatosensory neurons to perform recording in the ventral posterior medial nucleus of the thalamus and sub-nuclei of the trigeminal brain stem complex with a configuration consisting of eight or 16 microwires. The advantage of this neuro technique is that the neural recordings may help to reconstruct neural engrams [97].

Qiang et al. developed a transparent microelectrode array to simultaneously record electrophysiological study as well as imaging by using the two-photon technique, as shown in Figure 8. The transparent microelectrodes were made from the Au nanosphere, and polyethylene oxide (PEO) was used for close packing of nanospheres. A 32-channel microelectrode array with 80 μm in diameter and an impedance of 12.1 kΩ was used with high spatial distribution and resulted in high uniformity neural recordings. This transparent microelectrode arrays provided high temporal and spatial resolution with high sensitivity and selectivity for recording single-neuronal signals, as shown in Figure 8d [98]. To measure the single-neuron membrane potential, simultaneous multi patch-clamp and multielectrode array recordings were combined. This system consisted of a 60-electrode array with 30 μm electrode diameter and a pitch of 0.5 mm. The multielectrode array provides spontaneous firing activity to the neurons, and the system can record simultaneously extracellular and intracellular activities of the patched neuron [99].

**Figure 8.** (**a**) Position of multielectrode array (MEA) on the mouse brain and the cranial window. (**b**) Implantation of the MEA in the mouse brain. (**c**) Epifluorescence of the brain and the surrounding areas. (**d**) Simultaneous electrophysiological recording, arousal, and two-photon imaging with single-neuron Ca++ activity. Reprinted with the permission of [98].

Direct interfacing with the nervous system may facilitate the extraction of millions of millisecond-scale information from single-neurons that will greatly benefit the personal diagnosis and follow-up treatment. Even though modern techniques have been developed to achieve good spatial resolution, such as structural and functional MRI, and temporal resolution, such as electroencephalography and magnetoencephalography, the measurement of the action potential and firing pattern in single-neurons have not been completely resolved. Hence, numerous animal models are still being used in the study to understand the physiological activities of small populations of individual neurons. In 1971, the first single-unit activity recording in epilepsy patients was performed by inserting an electrode with fine wire through the center of the brain. This study found that when the seizures were approaching the neuronal action potentials were periodic with the frequency associated with the time and phase of the gross waves. This can be related to the changes in the interaction between groups of neurons in neuronal networks [100]. After two decades, Fried et al. in 1999 described a technique that measured extracellular neurochemicals by cerebral microdialysis along with simultaneous measurement of electroencephalographic recordings and single-unit activity of neurons in the selected target. They conducted this study in 42 patients with a total of 423 electrodes, and the number of electrodes for each person varied from six to 14. These electrodes for single-unit neuron activity recording have four to nine 40-μm microwires that were made of a platinum alloy. The tests were conducted at 5–10 min intervals during seizures, cognitive tasks, sleep-waking cycles, and the release of amino acids and neurotransmitters for the evaluation of patients with a head injury, epilepsy, and subarachnoid hemorrhage [101]. Another single-neuronal recording platform, known as the Utah array, consisted of etched silicon array of 100 probes and was developed to record the patterns from individual neurons. A Utah array with 96 microelectrode contacts has been placed in the center of the brain to record the neuronal activity and hence monitor the symptoms of Parkinson's disease [102,103]. The Utah array has also been implanted intracortical, directed by two-dimensional cursor movements to record the single-unit activity in epilepsy patients. In these studies, the Utah arrays recorded signals from different single units rather than from different layers of the brain. One interesting finding obtained from this epilepsy study was that there was an interplay between multiple classes and types of neurons, but the seizures did not propagate to the outside regions [104,105].

Furthermore, a relationship between single-neuron spiking and interictal discharges was established by analyzing the spiking rates of neurons that were recorded between seizures and during the seizures. A total of 90 neurons were recorded extracellularly from 17 awake patients, and it was noted that few neurons showed increased spiking rates during epileptic activity [8]. The drug-resistant focal epilepsy can be treated with stereoelectroencephalography probes by studying the single-unit activity recorded during epileptic seizures. The trials were conducted on a monkey by inserting three polyimide platinum cylindrical probes with varying electrodes sites [32,64] and the recordings were made. The single-unit activity of the neurons measured from the device was used to improve the precision of epileptic focus detection [91,92]. Various experiments were conducted on 36 patients with advanced Parkinson's disease, who underwent microelectrode-guided posteroventral pallidotomy. The microelectrodes were placed to measure the single-unit recording and this was analyzed under various firing patterns, frequencies, and the response of movement-related activity. Magnetic resonance imaging was carried out to examine the size and location of the lesions [106].

The rabies virus is a genetically modifiable virus that allows high-level expression of a specific gene in synaptically coupled neurons. The property is well suited for single-neuron analysis. A two-plasmid system has been utilized: one encoded replication-defective rabies virus RNA with the glycoprotein gene truncation and the other encoded only the glycoprotein. When electroporated into a single-neuron, the virus that assembled in one neuron lost its ability to replicate after it moved trans-synaptically (Figure 9). Analysis of the viral protein expression pattern would help to understand not only the pathogenesis of the rabies virus but also the neural connectivity in a dynamic fashion [107].

**Figure 9.** Testing of viral spread in the pre- and post-synaptic cells. (**a**) Image of the slice and recording pipettes with cells. (**b**) Merger of the fluorescent image. (**c**) Transfected cell with viral spread. (**d**) Transfected cell with TVA (cellular receptor for subgroup A avian leukosis viruses (ALV-A)) and rabies-virus glycoprotein (**e**) coinciding postsynaptic currents and action potentials in the cell with a monosynaptic connection. Reprinted with the permission of [107].

#### **7. Single-Neuron Transfection Methods**

The delivery of biomolecules into cells is an important strategy to investigate cell behaviors as well as the development of therapeutics. Conventional biological and chemical transfection agents, such as viral vectors [108], calcium phosphate, basic proteins [109], and cationic polymers [110], can deliver different biomolecules into cells and are suitable for general usages. However, most of these techniques are cell-type-specific bulk delivery and are often limited to low delivery efficiency and cell viability [111,112]. For example, certain viral vectors may be mutagenic to the transfected cells and can trigger immune responses and cytotoxicity [113]. Genes delivered via cationic polymers may be targeted to endolysosomes and result in endocytic degradation [114]. On the other hand, physical transfection methods use physical energy to create temporary pores on the cell membrane that allow foreign biomolecules into the cells by simple diffusion [115–117]. In the last two decades, due to the rapid development of micro- and nano-technologies, many physical techniques can deliver different sized biomolecules in different cell types (at a single-cell level) with high transfection efficiency and high cell viability [3,6,7]. The most commonly used physical transfection methods include microinjection [118–120], electroporation [121–124], optoporation [125–128], sonoporation, magnetoporation [129–132], and biolistic gene delivery [133–135]. The advantages and limitations of different single-neuron cell therapies and analyses are discussed below.

#### *7.1. Microinjection*

Microinjection is a versatile transfection method, suitable for almost all cells. The technique involves direct insertion of a hollow microneedle into a subcellular location of the membrane and delivers a precise amount of biomolecules into cells irrespective of their size, shape, and chemical nature [136]. The approach is quite labor-intensive and occasionally causes substantial stresses due to disruption of the plasma membrane, resulting in decreased survival rates of transfected neurons. Despite these drawbacks, microinjection has successfully delivered exogenous proteins, cDNA constructs, peptides, drugs, and particles into transfection-challenged individual neurons. One such example is the delivery of active recombinant enzymes (caspase-3, -6, -7, and -8) into individual primary neurons. The neurons displayed caspase-specific responses, including prolonged time-dependent apoptosis by caspase-6 (>0.5 pg/cell) [137]. The selectively toxic of Aβ1–42 via activation of the p53 and Bax proapoptotic pathway to only neurons was also proved by microinjecting

Aβ1–40, Aβ1–42, and control reverse peptides Aβ40–1 and Aβ42–1 or cDNAs expressing cytosolic or secreted Aβ1–40 and Aβ1–42 in primary human neuron cultures, neuronal, and non-neuronal cell lines [138]. The mechanistic dissection of single-neural stem cell behavior in tissue was further evaluated by microinjection. The microinjection set-up consisted of a phase-contrast microscope with epifluorescence, trajectory, and micromanipulator [139]. Although current imaging techniques are equipped to monitor such behavior, the genetic manipulation tools are still devoid of achieving a balance between the gene expression and timescale for the singular gene product. Microinjection in mouse embryonic brain organotypic slice culture targeting individual neuroepithelial/radial glial cells (apical progenitors) avoided these shortcomings. The apical progenitor microinjection acutely manipulated the single-neural stem, and progenitor cells within the tissue and the cell cycle parameters otherwise indecipherable to apical progenitors in utero, go-through self-renewing divisions and neurons were produced. The microinjection of recombinant proteins, single genes, or complex RNA blends stimulated acute and distinct modifications in the behavior of apical progenitor cells and also changed the destiny of progeny [140]. Further, the role of two essential genes in mammalian neocortex expansion, namely the human-specific gene *ARHGAP11B* [141] and Insm1 [142] was assessed via microinjection.

Another study highlighted the fast and efficient CRISPR/Cas9 (Clustered regularly interspaced short palindromic repeats- associated protein 9) technology for the disruption of gene expression involved in neurodevelopment [143–146]. The technology eradicates the restrictions of transgenic knockouts and RNAi-mediated knockdowns. A radial glial cell (RGCs) in telencephalon slice of heterozygous E14.5 *Tis21*:: GFP mice were microinjected as shown in Figure 10a, to distinguish the progeny cells from the microinjected aRGCs. The microinjection cargo included recombinant Cas9 protein with either gRNA (gLacZ) or gGFP control. In this experiment, dextran 10,000-Alexa 555 (Dx-A555) acted as a fluorescent tracer for the aforementioned identification. Microinjection mainly aims single aRGCs in the G1 phase of the cell cycle, and therefore facilitates the monitoring of the CRISPR/Cas9-mediated disruption effect of gene (under observation) expression in the same cell cycle of the microinjected neural stem cell, as depicted in Figure 10b–d [147]. The microinjection mediated CRISPR technology provides new prospects for functional screenings and to determine the loss-of-function in the individual cell.

Kohara et al. performed simultaneous injection of DNAs of green fluorescence protein tagged with brain-derived neurotrophic factor (BDNF) and red fluorescence protein (RFP) into a single-neuron (Figure 11). Thereafter, they visualized the expression, localization, and transport of BDNF in the injected single-neuron. This co-expression of two fluorescent proteins revealed the activity-dependent trans-neuronal delivery of BDNF [148]. Shull et al. recently developed a robotic platform for image-guided microinjection of desired volumes of biomolecules into single-cell. In this study, they delivered exogenous mRNA into apical progenitors of the neurons in the fetal human brain tissue. For the autoinjector, the injection pressure was set between 75 and 125 m bar, and it was microinjected from the ventricular surface to the depths of 10, 15, and 25 μm with the efficiency of 68%, 22%, and 11%, respectively. Thus, the autoinjector can deliver exogenous materials into targeted cells to the cluster of cells with high control and at single-cell resolution [119].

A variant of microinjections has been formulated combining electrophysiology recordings, electrical micro-stimulation, and pharmacological alterations in local neural activity, most commonly used in monkey. The combination of the above-mentioned activities helps in providing a better way of explaining neural mechanisms [149]. Therefore, targeting simultaneous drug delivery, neurophysiological recording, and electrical microstimulation, various groups have developed "microinjectrode" systems. Sommer et al. established the primary connection between corollary discharge and visual processing via injectrode and segregating single cortical neurons. The results showed that spatial visual processing impairs if the corollary discharge from the thalamus is disturbed [150]. Crist et al. developed a microinjectrode which contains a recording electrode in addition to an injection cannula, facilitating simultaneous drug delivery and extracellular neural

recording in monkeys. But the recording wire of the syringe typically recorded multi-unit activity, with frequent single-cell isolation [151]. Subsequently, modified injectrodes were introduced to achieve better recording quality and the ability to alter both neuronal activity and behavior in animals, an example being shown in Figure 12 with single-neuron recording, electrical microstimulation and microinjection in the frontal eye field (FEF), along with recorded single-neuron waveforms [84,149,152,153].

**Figure 10.** CRISPR/Cas9 (Clustered regularly interspaced short palindromic repeats- associated protein 9) -induced disruption of green fluorescence protein (GFP) expression in the daughter cells of single microinjected aRGCs in organotypic slices of the telencephalon of Tis21::GFP mouse embryos. (**a**) Scheme of the Cas9/gRNA complex microinjection. (**b**) Reconstruction of optical sections with maximum intensity projections for daughter cells of single aRGCs microinjected with either Cas9/control gRNA (top) or Cas9/gGFP (bottom) revealed by Dx-A555 immunofluorescence (magenta); cell 1, aRGC daughter; cell 2, BP daughter. Dashed lines depict ventricular surface. Scale bars, 20 μm. (**c**) Single optical sections of cells 1 and 2 shown in (**b**), showing the effects of Cas9 and control gRNA (top) or gGFP (bottom) on GFP expression. Scale bars, 5 μm. (**d**) Quantification of the proportion of daughter cells (Dx-A555+) of microinjected cells showing GFP expression 24 h after control (Con, white) or gGFP (black) microinjection. (\* *p* < 0.05, Fisher's test) Reprinted with the permission of [147].

**Figure 11.** Cortical neurons expressing brain-derived neurotrophic factor (BDNF): (**a**) with green fluorescence protein after 24 h of delivery; (**b**) stained with anti-BDNF antibody; (**c**) merge image of both green fluorescence protein and anti-BDNF antibody. Reprinted with permission from [148].

**Figure 12.** Microinjectrode system and its application. Briefly, a thin microelectrode passes through a 32 G cannula (OD: 236 m) which is connected to a T-junction via a ferrule. The electrode goes into a T-junction and a polyimide-coated glass tube with the terminal soldered to a gold pin. The polyimide tubing, gold pin, and ferrule are all pasted together. The middle part shows cross-sections through different parts of microinjectrode, i.e., the top ferrule, middle T-junction and bottom the cannula. An enlarged view of the microelectrode and cannula tips shows their relative position and size. A sample experiment is also displayed with single-neuron recording, electrical microstimulation and microinjection being performed in the frontal eye field (FEF). The single-neuron waveforms (black traces) segregated from background (gray traces) are also presented. Reprinted with the permission of [149].

#### *7.2. Electroporation*

Contrary to microneedles, single-cell electroporation displays better performance in specificity, dosage, cell viability, and transfection efficiency. Single-cell electroporation (SCEP) uses electric field application surrounding or a localized area of the single cell, with inter-electrode distance in the range of a micrometer to nanometer scale [154,155]. The application of a high external electric field in the vicinity of cell membranes increases their electrical conductivity and permeability owing to structural deformations occurring at the membrane for creating transient hydrophilic membrane pores and deliver biomolecules inside single-cell by simple diffusion process [156]. These transient pores are developed from the initial form of hydrophobic pores and therefore facilitate electroporation. The electric field can be applied in various ways, as shown in Figure 13: (a) non-uniform electric field distribution (higher field at poles and lower field at equators); (b) membrane area-dependent density of pores formation on single-cell due to non-uniform electric field application; and (c) nano-localized electric field application using nano-electrodes and biomolecular delivery [154,156].

**Figure 13.** (**a**) Schematic showing distribution of electric field facilitating single-cell electroporation (SCEP); the induced transmembrane potential is found to be highest at the cell pole and decreases towards the equator. (**b**) Microfluidic SCEP with cell trapping. Reprinted with permission from [154]. (**c**) Localized SCEP with electric field (**b**) membrane area dependent density of pore formation and distribution due to non-uniform electric field application (**c**) nano-localized single-cell nano-electroporation. Reprinted with permission from [156].

The cell membrane surface subjected to electroporation is dependent on the nanochannel opening with diameter generally <500 nm and it could be constituted in the form of an array. The above-mentioned various types of set-up porate a small patch on the cell membrane, electrophoretically pushing polarized macromolecules inside the cell via the nanoscale pores [123]. Haas et al. originally used electroporation for studying the role of genes in the morphological development and electrophysiology of neurons in Xenopus laevis tadpole brain. They electroporated individual cells using electrical pulses from a DNA-filled micropipette. Single-cell electroporation was preferred due to the uniqueness of the individual neuron's axonal and dendritic processes without any intervention from neighboring neurons' processes. They also highlighted the role of gene expression on the transfected cell, and are either cell-autonomous or secondary because of interplay with transfected neighbors [123,157]. The most effective current for SCEP lies between 1 and 4 mA and the co-transfection rate for pGFP and pDsRed are greater with SCEP (96%), in comparison to whole-brain electroporation. Earlier dendritic growth of single-cell electroporated neurons in the

tadpole brain can be examined only over six days [123]. Now it has been advanced to the level of the intact developing brain, where live two-photon fluorescence imaging shows the SCEP of a fluorescent dye or plasmid DNA into neurons within the intact brain of the albino Xenopus tadpole in the timescale of seconds to days without altering the neighboring tissues [158].

Electroporation has been employed for the transfection of the spinal cord. The technique was initially amended for the transfection of single cells or small sets of cells inside the axolotl spinal cord, in the vicinity of the amputation plane. However, now it has attained advancements to allow the transfection of the labeled spinal cord cells, overcoming the requirement of transgenic knockouts or RNAi-mediated knockdowns [124]. Further, Echeverri and Tanaka tracked the explicit cell fate of neural progenitors present in the spinal cord via electroporation in tiny and transparent axolotls, transparent skin allows imaging of differentiating neurons with epifluorescence using differential interference contrast microscopy. As shown in Figure 14, the timeline of the growth of the regenerating spinal cord is as follows: progenitor cells recruitment from mature tissue to the regenerating part (day 2–4), cell-division (day 4–15), and cell-clones spreading along the A/P axis (day 7–15) [124].

Further in vitro electroporation and slice culture was performed for the interpretation of gene function in the mouse embryonic spinal cord owing to the low transfection efficiency of in utero spinal cord electroporation. The expression of the external gene in the embryonic spinal cord is governed by in utero electroporation. The axonal projections are unanimously directed from inside to the lateral side of the spinal cord. In comparison to neurons present in vivo, a single-neuron growing in the slice culture owns an extra number of complete neurites and therefore offers ease in the study of structural and behavioral alterations in individual neurons [159].

Electroporation has been shown to overcome the issues related to intracellular pressure resulting from injection or iontophoresis. Single-cell electroporation is simple, reproducible, highly efficient, and capable of introducing a variety of molecules, including ions, dyes, small molecular weight drugs, peptides, oligonucleotides, and genes up to at least 14 kb, into cells. The electrophysiological recording and anatomical identification by electroporation have been performed in a number of cells (CHO, HEK293, α-TN4 cells, etc), primary cultures of chicken lens epithelial cells [160] and retinal ganglion cells [161], using microelectrode and a few volts supplied from a simple voltage-clamp circuit. Graham et al. have demonstrated single-cell manipulations using a whole-cell patch type electrode, which can adapt to obtain electrophysiological responses easily using an amplifier that allows both a recording and stimulation mode [161]. Moreover, time-lapse in vivo electrical recordings of contralateral and ipsilateral, sensory-evoked spiking activity of individual L2/3 neurons from the somatosensory cortex of mice was also facilitated by using electroporation [162]. On-chip electroporation performed using micrometer-sized gMμE (an array of gold mushroom-shaped microelectrodes) device that enabled membrane repair dynamics and transient in-cell recordings [121]. Several additional devices with miniaturized and integrated microneedle electrodes or microchannels have been fabricated to perform single-cell electroporation [163]. These devices, consist of a wave generator, a biochip containing an array of microelectrodes, and a control system, permit the transfer of signals to a pre-selected single microelectrode of the biochip achieving the transfection of Cos-7 cells and single-neurons with oligonucleotides [164,165].

Further, optogenetic probes are also precisely targeted on individual neurons via single-cell electroporation. A targeted optogenetic expression among precisely grouped neurons helps in assessing the relation between neuron count, uniqueness, and spatial organization in circuit processing [165]. A similar approach will also help in the analysis of calyx-type neuro-neuronal synapses of the embryonic chick ciliary ganglion (CG) via single-axon tracing, electrophysiology, and optogenetic techniques. In vivo electroporation manipulated presynaptic gene and later 3D imaging was performed for single-axon tracing in isolated transparent CGs, followed by electrophysiology of the presynaptic terminal, and an all-optical approach using optogenetic molecular reagents [166] Long-term in vivo single-cell electroporation was conducted using Two Photon Laser Scanning Microscopy (2-PLSM) of synaptic proteins, combined with longitudinal imaging of synaptic structure and function in L2/3

neurons of the adult mouse neocortex. This result also expresses and longitudinally image SEP-GluR1 dynamics, suggesting a difference in spontaneous activity of synapses, and consequently, constitutive insertion through GluR1 receptors takes place [167].

**Figure 14.** (**a**–**j**) Cell transfection is shown with cytoplasmic DsRed2-N1 and nuclear green fluorescent protein plasmids (**b**,**c**). The merged fluorescence and differential interference contrast (DIC) images after 2 days of amputation depict both the cells in the spinal cord with a distance of approximately 250–300 μm from the amputation plane (**c**). In the next 2 days, the cells undergo division and recruitment to the regenerating spinal cord (**e**,**f**). (The panels show only regenerating tissue.) The cell division continues and spinal cord growth continues rapidly (**g**–**j**). (**j**) A composite image of DIC images merged with the fluorescent image (15 days). Here, the initial two cells give rise to approximately ten cells on both the dorsal and ventral sides of the midportion of the developing spinal cord. The cell group is present over 560 μm length along the anterior/posterior axis. The original amputation plane is depicted by an arrow sign. Scale bar 100 μm in (**j**) (applicable to **a**–**j**). Reprinted with permission from [124].

Tanaka et al. performed single-cell electroporation and small interfering RNA (siRNA) delivery for gene silencing against the green fluorescent protein into GFP-expressing Golgi and Purkinje cells in cerebellar cell cultures. The temporal alterations in the GFP fluorescence (in the same electroporated cells) were observed for 4–14 days via repeated imaging (Figure 15). Furthermore, they checked the

dependency of concentration for specific gene silencing and the non-specific off-target effects of siRNA inserted through this method, showing that the effects were present at least up to 14 days, yet differed between neuronal cell types [122,168].

**Figure 15.** Immunostaining images of single-cell electroporated Purkinje cells small interfering RNA (siRNA) against calcium/calmodulin-dependent protein kinase ß (CaMKIIß) or 14-3-3η). SCEP was done at 11 days in vitro (DIV). The cell fixation was performed on day 7 (**a**,**c**) or day 14 (**b**,**d**) post electroporation (18 or 25 DIV, respectively) and double fluorescent immunostaining against CaMKIIß (green in **a**,**b**) and calbindin-D-28 K (CBD28K) (red in **a**,**b**) or 14-3-3η (green in **c**,**d**) and IP3R (red in **c**,**d**) was performed. Therefore, 1, 2 and 3 correspond to green, red and merged stains respectively. CaMKIIß or 14-3-3η signals decreased in electroporated Purkinje cells (arrows), but not in nearby non-electroporated Purkinje cells (asterisks). It is noteworthy that CaMKIIß and 14-3-3η expression was present for both Purkinje cells and granule cells. Scale: 20 μm. Reprinted with permission from [168].

Apart from the above-mentioned routes, single-neuron electroporation was performed on the cultured cortex to transfect gene encoding yellow fluorescence protein. Analysis of the dynamic of axon morphology indicated that electroporation had not affected developmental aspects [169]. Electroporation was also tested on an organotypic culture of hippocampal slices to introduce plasmids into single-neurons [170]. The approach has been used to demonstrate synthetic oligonucleotides delivery to identify duplex RNA and antisense oligonucleotide activators of human frataxin expression [171]. Using fluorescent Ca2<sup>+</sup> indicator-loaded brain slices and in vivo samples, the morphology of the apical dendrites of several pyramidal neurons was found to be normal, indicating that the neurons had recovered from the electroporation procedure [172]. Single-cell electroporation accompanied by virus-borne genetically encoded Ca2<sup>+</sup> sensors also allowed functionally trans-synaptic tracing in targeted single cells [173]. Single-cell electroporation was also used to identify and selectively label active homeodomain transcription factors mnx negative neurons in embryos of the double-transgenic line Tg(elavl3:GCaMP6f)Tg(mnx1:TagRFP-T) via two-photon confocal microscopy imaging [174].

#### *7.3. Optical Transfection*/*Optogenetics*

Antkowiak et al. designed a technique with an image-guided, three-dimensional laser-beam steering system for transfecting specified cells (Figure 16). Channelrhodopsin-2 (ChR2) was successfully introduced into a large number of cells in a neural circuit individually in a sequential manner as shown in Figure 16b,c. This technique enabled the transfection of selective cells on a large-scale basis and performed rapid genetic programming of neural circuits [175]. Barrett et al. successfully phototransfected primary rat hippocampal neuron with a Ti-sapphire p laser using 100 fs pulses with 30 mJ power, and 1–5 ms pulse duration. Successful transfection of the neuron could be observed after 30 min of laser exposure [176].

Optogenetics are now widely used for activation and silencing of neuron populations defined by their molecular and activity profiles and projection patterns [177]. Some commonly used tools are light-gated ion channels (e.g., channelrhodopsin-2, or ChR2) and ion pumps (halo-rhodopsin or archaerhodopsin-3). These molecules, combined with a suitable optical method, can trigger their function to control neuronal activities. Owing to low channel conductance of ChR2, single-cell optical stimulation has not been feasible previously [178]. Simultaneous activation of a large number of channels can help to achieve sufficient depolarization up to a space of tens of μm2. Nevertheless, conventional one-photon and two-photon scanning imaging systems addressing this issue inevitably activate neurons in an untargeted fashion. Though these studies showed high spatial resolution, yet the required activation time for large area appropriate for firing action potentials was approximately 30 ms. The two-photon temporal focusing (TEFO) technique developed earlier in this decade may realize the demand. The system has an independent axial beam profile from lateral distribution and simultaneous excitation of multiple channels on individual neurons, resulting in strong (up to 15 mV) and fast (≤1 ms) depolarizations. The techniques may allow quasi-synchronous activation of neurons along with specific cellular compartments. The TEFO with a conventional dual galvanometer-based scanning system repositions the excitation spot in a rapid manner typically <0.2 ms to any point in a 100 μm field. The precise spatial and temporal control of firing activity performed with a single or preferred several single cells, particularly while combining with selective ChR2 expression of specific population of cells. This technique highlights the scope for detailed, high-throughput analysis of connections and neural network dynamics and evaluation of the functional significances of their activation both in vitro and in vivo [179].

To overcome the limitation of the requirement of high opsin expression and complex stimulation techniques, Packer et al. used a new red-shifted chimeric opsin C1V1T formed by combining ChR1 and VChR1 (Figure 17a). This technique involved a spatial light modulator, in which the laser beam was split and targeted to several positions in a neuron, allowing simultaneous optogenetic activation of selected neurons in three dimensions. The method also showed the possibility to optically map

short-term synaptic plasticity. Figure 17b shows the effect of a single 150 ms TF stimulation pulse (red bar) via two-photon highest intensity projections of Alexa 594 in the form of fluorescence and current responses for patched and dye-filled pyramidal cells in acute slices expressing targeted (T) and nontargeted (N) ChR2 [180].

**Figure 16.** Optical transfection system using femtosecond laser (**a**) Schematic of the optical transfection system. (**b**) Side view of the Petri dish containing a single-neuron for transfection. (**c**) Irradiation patterns (red dots) superimposed on phase-contrast images of cortical neurons. Reprinted with permission from [175].

To avoid undesired neuron labeling and studies, a combined temporal focusing with the spatial confinement of ChR2 expression to the neuronal cell body and proximal dendrites were also tested. This was based on the Kv2.1 potassium channel, which has a particularly unique localization to clusters at the neuronal soma and proximal dendrites. As shown in Figure 17b, the action potential was evoked in individual neurons, and peak generation took place with GCaMP6s, and functional synaptic connections with patch-clamp electrophysiological recording could be determined at a single-neuron resolution [181]. Another study also presented a conventional optogenetic two-photon mapping method in mouse neocortical slices by activating pyramidal cells with the red-shifted opsin C1V1, while recording postsynaptic responses in whole-cell configuration. The use of temporal-focused excitation or holographic stimulation, as in earlier method, limits the problem of dendritic activation, yet the current method is simple and fast [182].

**Figure 17.** (**a**) Two-photon activates of individual neurons present in mouse brain slices with C1V1T. (**i**) The experimental scheme shows the opsin *C1V1T* and *EYFP* genes encoded by Adeno-associated virus (AAV) are inserted in the somatosensory cortex of the mouse. Brain slices were prepared at a designated time point from the infected region. (**ii**) Two-photon fluorescence image of a living cortical brain slice expressing EYFP (940-nm excitation, 15 mW on the sample, 25×/1.05-NA objective; scale bar, 100 μm). (**iii**, **iv**) Magnified images from (**b**) show cells with C1V1T-expression present in higher (**iii**) and lower (**iv**) layers (scale bars, 20 μm (**iii**), and 10 μm (**iv**) Reprinted with permission from [180]. (**b**) Illustrative two-photon highest intensity projections of Alexa 594 fluorescence and current responses against a single 150 ms temporal focusing (TF) stimulation pulse (red bar) for patched and dye-filled pyramidal cells present in acute slices expressing targeted (T) and nontargeted (N) ChR2. Scale bar = 100 mm. Reprinted with permission from [181].

Contrary to the above-mentioned single-cell resolution optogenetics, sometimes neurons own high expressing opsins so that even two-photon (2P) stimulation of a single-neuron soma is sufficient to excite opsins present on crossing dendrites or axons along with stray excitation of neighboring neurons. Therefore, the localization of a novel short amino-terminal peptide segment of the kainate receptor KA2 subunit 18 fused with high-photocurrent channelrhodopsin CoChR19 in neuron soma avoided crosstalk and facilitated selected handling of CoChR to neuron soma in mammalian cortex. The combined holographic 2P stimulation using low-repetition fiber laser optogenetically stimulated single cells present in mammal brain slices. The use of light pulses with subtle powers lead to zero-spike crosstalk with neighboring cells and a shown temporal resolution of <1 ms. It also implemented protein fusion known as somatic CoChR (soCoChR), along with parametrized 2P stimulation enabled probing of various circuit neural codes and computations. The 2P computer-mediated holography sculpts light for simultaneously lighting many neurons in a network while maintaining the standard temporal precision to precisely stimulate neural codes [183]. The expression of some opsins is restricted genetically within the somatic part of the neurons; it offers a crucial feature of eliminating spurious activation of nontargeted cells while causing excitation of multiple neurons. Also, parallel illumination of conventional ChR2 and slow opsins such as C1V1 and ReaChR have fired up to 20–30 Hz spike with susceptibility to spike duration changes and the generation of spurious extra spikes. The problems are due to the limited kinetics of opsins. Certainly, high-frequency, light-driven action potential (AP) trains need opsins with rapid off kinetics maintaining fast membrane repolarization and inactivation recovery after every spike. All these facts postulate one hypothesis, that the in-depth optical regulation of neuronal firing with high spatiotemporal precision is dependent on 2P parallel photostimulation of fast opsins. Therefore, 2P action spectrum and kinetics of the fast opsin Chronos with holographically shaped light pulses were characterized. It was demonstrated that efficient current integration with 2P parallel illumination, enabled AP generation with sub-millisecond temporal precision and neuronal spike frequencies up to 100 Hz. The use of a fiber amplifier and high-energy pulse laser decreased the average illumination power many-folds. The outcome suggested mimicry of a broad range of physiological firing patterns with sub-millisecond temporal precision, as it is critical for understanding the relationship between behavior and pathological states in terms of particular patterns of network activity [184].

Another research article computationally predicts the power of external regulation of the firing times of a cortical neuron following the Izhikevich neuron model. The Izhikevich neuron model helps to follow the membrane potential values and firing times of cortical neurons efficiently and in a biologically possible way. The outside regulation is a simple optogenetic model including an illumination source, which stimulates a saturating and decaying membrane current. Here, the firing frequencies are assumed to be significantly lower for the membrane potential to achieve resting potential after firing. The model fits neuron charging and recovery time along with peak input current, to derive lower bounds on the firing frequency, achievable without significant distortion [185].

#### **8. Micro**/**Nanofluidic Devices for Single-Neuron Analysis**

In the last two decades, the rapid development of micro/nanotechnologies and their integration with chemistry, chemical engineering, and life science have encouraged the emergence of lab-on-a-chip devices or micro-total analysis systems (μ-TAS), which are powerful tools used to perform a variety of cellular analyses. The devices are capable of performing precise single-cell and subcellular analyses with minimal sample consumption. Micro/nanofluidic devices can create optimal microenvironments for growing cells and guiding their growth direction, especially for neurons. Microenvironments within micro/nanofluidic devices can enhance the axonal growth and can dissolve molecules and can create contact-mediated signaling from guided cells and cellular matrix [186].

The neurochip with microwells and microchannels along with planar multielectrode arrays to confine single-neuronal cells were designed and used to study the cell electrophysiological activity. A PDMS film with varying microwell sizes for cell patterning was fabricated on glass substrates with 40 μm wide ITO electrodes. The cell patterning structures restricted the movement of soma by allowing only the neurites to extend through the microchannel. Thus, one-to-one neuron electrode interfacing was established along with patterned structures and planar multielectrode arrays [187].

This study was further extended by integrating a substrate with a multielectrode array for recordings purpose from extended neurites in individual microchannels, as shown in Figure 18. The activity of extended neurite from the microwell was recorded by 18 electrodes, and a density analysis of single-cell current was carried out. By using this technique, the electrical stability of the electrode-neuron interface was enhanced, in comparison with the other using a planar multielectrode array [188].

**Figure 18.** Schematic showing a microelectrode device fabricated by photolithography with microwells and microchannels on a planar multielectrode array, in which neurons were individually positioned in microwells, view from top (**a**) and side (**b**). Redrawn from [188].

A biochip with asymmetrical channels was developed to study the polarized axonal growth in neural circuitry. This device consisted of microwells connected by numerous micro tunnels, which served as a guidance for developing axons to reach target neurons. A laser-based cell deposition system was used to place single cells into specific microwells in the device. The design of asymmetric channels improved the polarity as well as connectivity of the individual neurons [189]. Another asymmetric microchannel platform consisting of independent cell culture chambers, separated by axonal diodes, which helped to achieve required directionality for growing single-neurons. The neuronal cells were cultured in a way that the cell somas were retained in the microchamber, while the axon of a single-neuron extended to the other chamber through the axon diode. The axon diode had a decreasing cross-section from the culture chamber of 15 μm to the target chamber of 3 μm, hence enhancing the directionality and synapse formation. This device helped to study neuronal development and synaptic transmission and hence it can be developed further to study neurodegenerative diseases such as Alzheimer's, Parkinson, and Huntington diseases [190]. A similar type of device including symmetric but smaller microfluidic channels also showed unidirectional extension of axons. By using this device, degeneration and regeneration of individual axons were studied by injuring the extended axons along the microchannels with the help of femtosecond laser. It was noticed that even after

the injury, the axons tend to extend to the target chamber, and hence this device enabled a better understanding of neuronal response to injury [191].

A silicon-based device with a patch-clamp microchannel array that acts as a cell-trapping platform has been designed for the electrical recording of single-neurons. The device consisted of two fluidic compartments with a cell injection chamber at the top layer and six independent microchannels and microholes at the bottom compartment. The local perfusion of single-neurons was obtained by controlling pressure in the microfluidic compartments. The device had a successful trapping rate of approximately 58%, which facilitated further analysis of the trapped cells in electrical recording and drug screening applications [192]. Figure 19 shows a microfluidic device with a complementary metal–oxide–semiconductor microelectrode array, which was designed to study the axonal signal behavior of single-neurons. This device consisted of two neuronal culture chambers connected by 30 microchannels with 12 μm width and about 10–50 microelectrodes were fabricated along each channel. This study revealed that the electrical activity of soma could be related to its axons, and the single action potential propagating along the long length of individual axons with high spatial resolution can be recorded [193].

**Figure 19.** (**a**) The geometry of the microfluidic device on the microelectrode array. (**b**) Image of the packaged chip with the device on the top. (**c**) Magnified image of the electrodes and the channels; channels are highlighted with red, and the scale bar is 10 μm. (**d**) Cross-section of chip depicted with dimensions. (**e**) The images of the channels are highlighted in red; the scale bar is 20 μm. (**f**) The device with a small chamber and channels with an array marked inside the black box. Reprinted with permission from [193].

The first neurochip was a silicon-based micromachined device with a 4 × 4 array of metal electrodes, which allowed growth and monitor neuronal cell individually. In the neurochip, neurowells were designed to capture the soma, while the neurites extend to gold electrodes, which was fabricated on the bottom of the chip. This device was designed to mechanically trap a neuron near an extracellular electrode of the multielectrode array with electrodes surrounded by micro tunnels. When an individual neuron was trapped onto an electrode site, the cell soma was captured inside it, allowing only the neurites to propagate along micro tunnels. The biochip yielded high neuronal cell viability and the action potential of each neuron was detected by each electrode, and there was no crosswalk between the channels [194]. These micro tunnels help the neurites from different neurons to form neural connections and can be recorded to study synaptic connections [195].

A compartmentalized microfluidic device integrated with microelectrode array was designed to study activity-dependent dynamics in single-neurons and synaptic networks. The device has three microfluidic chambers, presynaptic, synaptic, and postsynaptic chambers (each) with axonal, reference, and postsynaptic electrodes to record the activity of single projecting axon. These presynaptic axons were recorded selectively by placing electrodes under the presynaptic chamber, and this study was further extended to study calcium dynamics [196]. Figure 20 depicts a microfluidic DEP device consisted of a PDMS microfluidic chip with ring-shaped indium tin oxide microelectrode array. In this device, the single-neuron was selectively trapped into the electrode, and the other neurons in the vicinity of the electrode were repelled by the DEP force. The amplitude and frequency of alternating current used to trap cells on the electrodes were 8 Vpp and 10 MHz, respectively. The trapped neuron was recorded, and its morphological changes were tracked with the assist of a phase-contrast microscope. Thus, this device enabled us to study multiple single-neurons at the same time and also electrical communication between them [58].

**Figure 20.** (**a**) Image of the microfabricated device and bright-field microscopic image of the electrode array. (**b**) Recorded images of single-neuronal cell manipulation on the array of ring-shaped traps. Incoming neuron (**I**) entering the 1st trap. (**II**) The neuron gets immobilized in the 1st trap electrode against a fluid flow. (**III**) When a neuron is trapped, the repelled particle keeps on moving in the flow

of media. (**IV**) The released neuron gets trapped in the 2nd trap. (**V** and **VI**) The neuron is trapped in the 3rd and the 4th ring trap in turn. (**c**) The images show bouncing motion of the neuron subjected to a repulsive force. When the target neuron gets trapped in the desired electrode, the incoming neuron faces repulsion due to DEP force. At the end, when the incoming neuron reaches the outside of the electrode, the repulsive force pushed the neuron out of the ring. Reprinted with permission from [58].

#### **9. Artificial Intelligence and Single-Neuron**

With advances in technology and instrumentation sensitivity, huge data is generated, but variation between batches in inevitable with enhanced susceptibility. In spite of the application of several correction models, the result is dependent on the actual magnitude of the effect [197]. Therefore, artificial intelligence is being employed to stimulate the learning processes otherwise occurring in humans, i.e., neural networks. Accelerated brain research initiatives are relying on AI-based tools, despite the different approaches, emphases and routes of neural studies. In spite of different research domains in the field of neuroscience employing different approaches and methodology, all have same objective of developing the next generation of AI-based tools [198,199]. For example, the Brain Research through Advancing Innovative Neurotechnologies (BRAIN) Initiative is moving forward to bring revolution in machine learning through neuroscience. As per the scope of this review paper, single-neuron analysis comes with several challenges, i.e., the curse of dimensionality, sparsity, degree of noise, batch errors, and data heterogeneity, which often hinder the performance of conventional computational approaches to scale up as data complexity and size grow, making the platform for contemporary deep learning algorithms. Processing and interpreting such high-dimensional single-cell information increasingly challenges conventional computational informatics calling for powerful and scalable deep learning models for dropout imputation, cell-subtype clustering, phenotype classification, visualization, and multi-omics integration. Iqbal et al. developed a fully automated AI-based method for whole-brain image processing to Detect Neurons in different brain Regions during Development (DeNeRD—Detection of Neurons for Brain-wide analysis with Deep Learning). This method to detect neurons labeled with various genetic markers is based on the state-of-the-art in object detection networks called the Faster Regions with Convolutional Neural Network (Faster RCNN) [200]. Further, a deep learning platform was developed for the identification and segmentation of active neurons. The core component consists of 3D CNN named STNeuroNet.re derived employing the two-sided Wilcoxon rank sum test. STNeuroNet was conceptualized on the basis of DenseVNet, a deep learning platform consisting of 3D convolutional layers, for the segmentation of active neurons from two-photon calcium imaging data. The STNeuroNet is equipped to extract relevant spatiotemporal features from the imaging data without prior modelling [201]. The next area in which deep neural networks have been employed for single-neuron analysis is single-cell RNA sequencing (scRNA-seq) data. An accurate, fast and scalable DeepImpute "Deep neural network Imputation" imputes single-cell RNA-seq data; outperforming the efficiency of other methods like mean squared error or Pearson's correlation coefficient as the dataset size increases [202]. During scRNA-seq, sometimes noise due to amplification and dropout may obstruct analyses, therefore the need for scalable denoising methods arise. A quality, high speed deep count autoencoder network (DCA) was proposed to denoise scRNA-seq datasets. This takes the count distribution, overdispersion and sparsity of the data into account using a negative binomial noise model with or without zero-inflation, and nonlinear gene-gene dependencies are captured. It is possible to work with datasets from millions of cells owing to the linear scaling with the number of cells [203]. Another single cell-based model scDeepCluster was developed to overcome the statistical and computational challenge during the Clustering transcriptomes profiled by scRNA-seq to reveal cell heterogeneity and diversity [204]. Therefore, lately, deep learning is an ideal choice for big data integration or testing. However, a major concern around deep learning methods is the "black-box" nature of the models and their un-interpretability due to the huge number of parameters and the complex approach for extracting and combining features. While the data science community is active in enhancing interpretability of deep neural networks, further research in biomedical contexts is required to understand clinically or biologically relevant patterns in data raised to accurate predictions,

and to improve the users' trust ensuring that the model decides based on reliable reasons rather than artifacts in data.

#### **10. Limitations and Future Prospects**

The current review includes the merits and limitations of single-neuron analysis. As discussed, the single-neuron-at-a-time methodology amalgamated with complementing technologies allowing recordings or imaging groups of neurons helps build a better understanding of complex neural networks. A recently developed, sophisticated electrophysiology and connectivity tool, named Patch-seq, associated with neuronal activity visualization and manipulation platform can assist in outlining the connections and functions of each neuronal type. Similarly, another technique, named scRNA-seq, elucidates the cell types in the brain via single-cell sequencing methods, single-cell genomics, epigenomics (including methylation, mapping, sequencing, DNA accessibility, and chromosome conformation), and multi-omics. These tools help in the decoding development stages, epigenetics, and functionality of the brain at single-cell resolution. But sometimes, the connecting RNA techniques require in few micrograms, corresponding to several cells, presenting the scope in this front. Moreover, the difference in spatial positions, temporal points, and poor health stages may cause variation in the analysis.

In recent years, advanced techniques facilitate automatic and high-throughput single-cell trapping followed by sequencing along with analyzing large datasets. All of these techniques motivate and strengthen the upcoming research activities in the direction of preparing an all-inclusive human brain cell atlas. But the rate of data production raises a challenge to process and make sense of it. Based on the processing of data, many scientists can make discoveries daily by employing new computational methods. On the other hand, droplet-based sequencing can produce scRNA-seq datasets covering <sup>&</sup>gt;<sup>5</sup> <sup>×</sup> <sup>10</sup><sup>5</sup> single cells, comprehends speed, and memory adeptness to state-of-the-art tools.

As stated earlier, multiple studies reported differences in cell types, number, cell cycle stage, extracellular matrix, and cell networking in different parts of the brain. Herewith, efforts are needed to integrate cell types from various studies. Hence, the biggest problem occurs on the level of scale in different reports. Additionally, from the aspect of multiplexity, current multiplexing is still not enough for whole proteomics detections (>10,000 proteins in a single cell). During standard bulk analysis, data reproducibility can be controlled owing to multiple biological and technical replicates. Single-cell experiments, particularly for scRNA-seq, contain the inability to replicate measurements on the same cell, and single-cell data is generally full of noise owing to technical variations occurring in multiple-step processes. Also, the biological variations arise as a result of cellular level heterogeneity, therefore increasing the sensitivity of scRNA-seq workflow at multiple levels, ranging from sample preparation, library preparation and sequencing and data analysis towards technical inconsistency and batch effects.

Apart from biological differences, experiment methodology, processing, handling as well as data processing workflows make it difficult to get a comparable result from the same model of the diseased or normal brain at any scale, i.e., organ, tissue, or at the level of an individual neuron. Therefore, the experimental protocols and computational outlines based on including and comparing scRNA-seq data from various platforms would overcome this issue. Lately, linked inference of genomic experimental relationships (LIGER) has proven useful in integrating multi-omics single-cell sequencing data [205]. Finally, single-cell multi-omics is going to gain huge success for brain studies by integrating data from various platforms. The classification of retinal bipolar cells has been the best-suited example of this set-up. [206]. The classification took cues from different techniques as well, i.e., structure and morphology (electron microscopy), electro-physiology (calcium imaging), and molecular biology (scRNA-seq) data. For better consideration of network organization and functioning of the brain, there is a great need for the unprejudiced, methodical assembly of molecular, morphological, physiological, functional, and connectivity data.

Overall, the knowledge of the brain is still in its infancy, but the rapidly growing single-cell sequencing technologies have already gathered ample data for future assessment and presented a never before seen map of the brain with single-cell resolution. Therefore, despite a range of complications and challenges, overwhelming progress is anticipated in the upcoming decade.

#### **11. Conclusions**

This review provides a broad perspective to the readers about the recent advances in single-neuron activity, neural circuit designing, and their sensitivity. We also emphasize in detail the current progress and future trends of single-neuron behavioral analysis, including the models, isolation, mapping, and electrophysiological recording. So far, isolation of single-neurons and maintaining their viability is still a challenging task. The advanced imaging and manipulating tools would continue to decipher the rise of thoughts and actions in the human brain. The details of single-neuron manipulation, isolation, sequencing, transfection, and analysis were elaborated using recent developed micro/nanofluidic devices as well as some physical methods, such as microinjection, electroporation, and optogenetics. The single-neuron optogenetics reveal the fundamental information about the sparseness of representations in neural circuits. Mapping neural connection at single-cell resolution would encourage planning systematic physiological experiments, probing connectivity between hundreds or thousands of neurons. Alongside this, deep learning is a promisingly potent machine learning technology, and the ongoing research in this field is expected to reign over the recent "big bang" of single-neuron data, just like it has been doing in other fields. The amalgamation of sophisticated visualization hardware, software, and huge neuro-anatomy data has supported the interpretation of decades of cumulative knowledge into a human axonal pathway atlas, which would be key for educational, scientific, or clinical investigations in future. However, we have made remarkable achievements in the field of human neuroscience, always accompanied by real-world problems.

**Author Contributions:** Conception and design: P.G., N.B. and T.S.S.; drafting the manuscript and final editing: P.G., T.S.S., H.-Y.C., F.-G.T.; review of the literature: P.G., N.B.; critical revision of the manuscript: T.S.S., H.-Y.C., F.-G.T. All authors have agreed to the published version of the manuscript.

**Funding:** This research was funded by the DBT/Wellcome Trust India Alliance Fellowship under grant number IA/E/16/1/503062 and the Indian Institute of Technology, Madras Institute post-doc grant (ED19IPF03).

**Acknowledgments:** We acknowledge all authors and publishers who provided copyright permissions.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Cells* Editorial Office E-mail: cells@mdpi.com www.mdpi.com/journal/cells

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18