Bacterial phenotypic heterogeneity through the lens of single-cell RNA sequencing

ABSTRACT

Bacterial transcription is not monolithic. Microbes exist in a wide variety of cell states that help them adapt to their environment, acquire and produce essential nutrients, and engage in both competition and cooperation with their neighbors. While we typically think of bacterial adaptation as a group behavior, where all cells respond in unison, there is often a mixture of phenotypic responses within a bacterial population, where distinct cell types arise. A primary phenomenon driving these distinct cell states is transcriptional heterogeneity. Given that bacterial mRNA transcripts are extremely short-lived compared to eukaryotes, their transcriptional state is closely associated with their physiology, and thus the transcriptome of a bacterial cell acts as a snapshot of the behavior of that bacterium. Therefore, the application of single-cell transcriptomics to microbial populations will provide novel insight into cellular differentiation and bacterial ecology. In this review, we provide an overview of transcriptional heterogeneity in microbial systems, discuss the findings already provided by single-cell approaches, and plot new avenues of inquiry in transcriptional regulation, cellular biology, and mechanisms of heterogeneity that are made possible when microbial communities are analyzed at single-cell resolution.

Introduction

As single-celled organisms, bacteria are well-suited model systems in which to answer basic questions concerning transcription. Early work characterizing transcription in bacterial systems used biochemical characterization of DNA elements and proteins to identify the bacterial transcriptional machinery [1–4] and DNA elements that make up promoters and regulatory elements [5,6]. Further analysis used a variety of genetic, biochemical, and structural approaches to determine the interactions of these components and many other transcriptional regulatory mechanisms including transcriptional attenuation, riboswitches and mechanisms for transcriptional promoter escape and termination (for review see [7]). With the invention of microarrays in the late 1990s (for example [8,9]), it became possible to simultaneously correlate the gene expression of thousands of genes in a single experiment, providing the first genome-wide view of transcriptional regulation and systematic inference of gene expression networks [10,11]. In addition, Chromatin Immunoprecipitation assays (ChIP-chip and ChIP-seq [12]) provided genome-wide maps of transcription factor binding, while RNA-sequencing and related sequencing technologies provided genome-wide maps of start sites and many additional insights into transcriptional regulation and adaptation in bacteria [13,14]. While much has been learned from these techniques, they all work by combining signals at the population level, which effectively averages the input of many cells. Therefore, these approaches obscure transcriptomic patterns when gene expression is heterogeneous in a population of cells, when, for example, a small subset of cells are undergoing a unique physiological process like sporulation. In such cases, the presence of distinct cell states and transcriptional patterns can be mistaken as universal shifts in expression across the community. Furthermore, it is difficult to unravel regulatory patterns of transcriptional regulation and cell–cell interactions if those patterns cannot be observed with single-cell resolution. Single-cell bacterial RNA sequencing alongside other single-cell techniques such as fluorescence in-situ hybridization and transcriptional reporters can help us to answer fundamental questions concerning transcriptional regulation in bacteria:

• What are the regulators and mechanisms that allow for heterogeneity in isogenic microbial populations?

• How do bacteria determine the optimal cell state for a given environment, and how is that decision-making process influenced by the decisions of neighboring cells?

• How are transcriptional rates (measured by transcriptional reporters or inferred by RNA-velocity[15]) used to regulate expression states?

• Does the presence of other microorganisms restrict or expand the range of heterogeneity observed in isogenic populations?

These are exciting lines of questioning, and they represent only a fraction of the intellectual frontier opened by reliable single-cell analytics in bacterial systems.

Technical inspirations and considerations

In the past decade, single-cell analysis of eukaryotic samples was used to address many open questions in eukaryotic gene expression. Single-cell RNA sequencing of eukaryotic samples has become routine, with hundreds of papers using mammalian single-cell RNA sequencing published every month and large gene-expression atlases uncovering previously unknown cell types in both healthy and diseased tissue [16,17]. While the vast majority of eukaryotic RNA-seq focuses on human and mouse tissues, the approach has been extended to profile cells from a very large number of eukaryotic systems including fungi [18,19], protozoa [20], plants [21], insects [22], fish [23], birds and mammals [24], ranging from studies of cells in laboratory culture conditions [25–27] to entire organisms [28,29]. The most commonly used mammalian single-cell RNA-seq approaches profile transcript abundance by counting the 3’ ends [30] but there are techniques that cover reads across exons and introns, and techniques that recover full-length transcripts, providing information on splice variants and RNA processing rates that can inform more careful estimates of transcriptional dynamics (RNA-velocity) [30].

In addition to the single cell sequencing of transcripts, numerous techniques that interrogate the mechanisms of transcriptional regulation have been applied to eukaryotic systems. Several techniques that quantify physical properties of DNA and binding proteins have been developed for single-cell analysis. These include single-cell ATAC-seq [31] to measure chromatin accessibility, single-cell bisulfite sequencing [32] to measure DNA methylation and single-cell CutnTag [33] which provides information on transcription factor occupancy. These techniques can be used simultaneously with single-cell RNA sequencing (for example [34]) to provide a snapshot of the transcriptional output provided by different regulatory inputs. Additionally, there are studies in which genetic variations are linked to transcriptomes in single cells. These include single-cell platforms such as Perturb-seq [35,36] that link the deletion of regulators to changes in transcriptional output, as well as single-cell DNA and RNA multi-omic sequencing efforts [37–39] that link mutants with their transcriptional phenotypes. In the past decade, the use of these powerful platforms has been able to identify many new cell types and regulatory mechanisms in eukaryotic systems. Within the complex body plan of truly multicellular organisms, distinct epigenetic programs reinforce the transcriptomic state of cells and can be remodeled to allow for differentiation and development.

Single-cell bacterial transcriptomics faces a variety of technical and conceptual challenges absent from similar work in eukaryotes. Many technical difficulties are related to the wider diversity and complexity of cellular envelopes in bacteria, which can make it a challenge to permeabilize cells and access mRNA in the cytoplasm. Once the cell is permeabilized, preferential isolation of mRNA is complicated by the general lack of a poly-adenosine tail on transcripts, and the overwhelming abundance of rRNA in the overall RNA content of the cell (~95% in some cases) [40]. Furthermore, microbial mRNA transcripts degrade rapidly as compared to those in eukaryotes. While half-life estimates vary anywhere from 10 minutes to less than 1 minute [40–42], the general consensus is that transcripts are degraded much more rapidly in bacteria than in mammalian cells. As a result, it is easy to miss relevant transcriptional bursts in any one cell when sampling bacterial populations due to a lack of transcripts, either from rapid degradation or low rates of transcription. This means that transcriptional signatures captured in single cells provide only a short snapshot in time, as illustrated in Figure 1. To mitigate some of the difficulties with short-lived bacterial transcripts, all single-cell bacterial RNA-seq methods include a fixation step to stabilize transcripts [43–46]. In addition, there is hope that measurements of transcript degradation, akin to RNA-velocity in mammalian cells [15], can provide information on transcript dynamics. Stochastic pulses (transcriptional bursting) should not be coordinated and therefore should not result in the same cellular clusters that are often seen in long-lived cellular fates, but could potentially provide valuable information about common regulators which may pulse with different tendencies. Finally, temporal sampling may reveal information about oscillations, especially in cases in which a population can be synchronized. Despite these technical and conceptual challenges, the field of single cell bacterial transcriptomics is maturing rapidly, and a variety of techniques to perform single-cell RNA sequencing on bacterial populations have been presented in the literature. Recent technical developments in this field are not the focus of this review, as they have been addressed in detail in other recent manuscripts [13,47,48]. Therefore, for our purposes we will only note that multiple different approaches can produce reliable data, and that technical concerns have been set aside as a barrier to single cell transcriptomic analysis in bacteria (Table 1).

Figure 1.

Single cell sequencing provides a snapshot in time. In any given cell a transcriptional readout at a specific time can be caused by different underlying regulation. For example, an increase in signal can be due to the upregulation of a promoter, a promoter that is starting to become downregulated, an oscillation in gene expression due to temporal cycling like the circadian rhythm or (in limited examples in bacteria) cell cycle, or a stochastic pulse in activity.

Table 1.

Overview of bacterial single-cell RNA sequencing techniques.

Bacterial scRNA-seq Method	Key Technical Features	Notable Observations of Heterogeneity
PETRI-seq [49]	96-well plate combinatorial indexing for transcript capture	Heterogenous prophage induction in Staphylococcus aureus
microSPLIT [44]	96-well plate combinatorial indexing for transcript capture	Heterogenous myo-inositol catabolism in Bacillus subtilis
MATQ-seq [50]	Single-cell isolation using FACS CRISPR-based rRNA depletion Reverse transcription for transcript capture	Heterogenous expression of small regulatory RNAs in Salmonella enterica serovar Typhimurium
M3-seq [46]	Droplet-based single-cell encapsulation rRNA depletion Reverse transcription and combinatorial indexing for transcript capture	Heterogenous acid tolerance as a bet-hedging strategy in Escherichia coli
BacDrop [51]	Droplet-based single-cell encapsulation rRNA and DNA depletion Combinatorial indexing using both 96-well plates and droplet barcodes	Heterogenous transcriptional response to antibiotic stress in Klebsiella pneumoniae
proBac-seq [45]	Droplet-based single-cell encapsulation Probe-based transcript capture	Heterogeneity in arginine production in Bacillus subtilis and Escherichia coli

Sources, benefits, and known examples of bacterial phenotypic heterogeneity

As opposed to eukaryotic organisms, in which heterogeneous cell types are clearly defined, morphologically distinct, and heavily epigenetically regulated, mechanisms of heterogeneity in bacteria often result in phenotypically distinct subpopulations that are more transient and more difficult to identify [52–54]. A primary source of heterogeneity in bacterial systems is the noise inherent to a system of gene expression governed by interactions on a molecular scale, manifesting as stochastic expression of genes even in clonal populations in homogeneous environments. This noise takes two forms: intrinsic noise, caused by the inherent randomness that governs reactions between enzyme and substrate, and extrinsic noise, caused by the fluctuating and often scarce supply of enzyme, substrate, and other cofactors necessary for transcription [55]. Certain forms of phenotypic heterogeneity are a functional consequence of this noisy gene expression, as environmental pressures and positive regulatory feedback loops turn small variations in expression into large subpopulations of transcriptionally distinct cells. Furthermore, bacterial transcription occurs in bursts, wherein certain cells experience periods of rapid transcription while others are transcriptionally quiet, resulting in heterogeneous levels of transcript and protein between cells in a population. This hypothesis is supported by the demonstrated non-correlation between levels of mRNA transcript and levels of expressed protein in an Escherichia coli model [56]. Transcriptional bursts in bacteria have been mechanistically linked to several processes, including gyrase activity in the case of highly expressed genes in E. coli, in which DNA supercoiling governed by transient gyrase association with the chromosome prevents access to the genome by transcriptional machinery [57]. Both noise and burstiness in transcription create ample opportunities for variations in gene expression across populations of isogenic cells, resulting in subpopulations of transcriptionally distinct cell types.

Heterogeneity can also be induced along spatial gradients. Bacteria respond to spatial variations in nutrient and essential metabolite availability by sensing these gradients and differentially expressing genes to optimize fitness for their current microenvironment. A common example of this type of variation is oxygen availability. In the classic Winogradsky column, oxygen concentrations are highest at the top and lowest at the bottom, resulting in a higher concentration of aerophilic and facultatively aerobic microorganisms toward the top, and higher concentrations of anaerobic bacteria toward the bottom [58]. Even in an isogenic population of a facultatively anaerobic bacterium grown in a test tube, cells near the top of the liquid culture that have more access to oxygen will upregulate genes for aerobic respiration, while cells with lower access to oxygen will instead upregulate fermentation genes, resulting in aerobic and anaerobic subpopulations [59,60]. Similar spatial transcriptional patterns can be observed in biofilms, where the extracellular matrix prevents a uniform distribution of oxygen and metabolites throughout the community, resulting in metabolic heterogeneity (driven by heterogeneous gene expression) and cross-feeding between transcriptionally distinct subpopulations [61,62]. Spatial transcriptional heterogeneity can also be induced by cell–cell interactions, such as quorum-sensing (QS), in which the accumulation of signaling molecules to a certain concentration in a bacterial population results in changes in gene expression and behavior across the population. If signaling molecule concentrations vary, then different cells will experience different degrees of quorum, resulting in differential activation of quorum-regulated genes and the development of heterogeneous subpopulations [63,64].

Alongside spatial gradients, phenotypic heterogeneity can be induced over temporal progressions and across cellular lineages. We will discuss two examples of this temporal phenomenon: cellular development cycles and history-dependent behavior (HDB), also known as cellular memory. The former is essential to the eukaryotic cell life cycle, but has only been well characterized in a few highly specialized bacterial species. Caulobacter crescentus, a freshwater marine organism, cycles between two phenotypically and morphologically heterogeneous cell types; an attached stalk cell, and a sessile swarmer cell, which buds off of a stalked cell during cellular division [65,66]. The swarmer cell can then attach to a new surface, differentiate into a stalked cell, and eventually bud off a new swarmer cell. Pioneering transcriptomic analysis using microarrays demonstrated that this life cycle is perpetuated by temporally synchronized oscillations in gene expression [8]. In the cyanobacteria Synechococcus elongatus, circadian oscillations in the phosphorylation state of the KaiC regulatory protein synchronize growth in the population to day-night cycles [67,68]. While heterogeneity is suppressed in the cyanobacteria circadian cycle, it demonstrates the power of oscillation in regulatory architectures to tightly control gene expression across bacterial populations. Finally, the experiences of certain cells in a population can result in distinct and heterogenous transcriptional patterns in the progeny of those specific cells at a later point in time. This results in a form of cellular memory, often termed history-dependent behavior, that allows bacterial populations to more quickly adapt to environmental conditions that they have previously encountered [69]. Bacterial memory has been observed in a variety of organisms and environments. In the face of dynamic nutrient availability, E. coli can more quickly induce metabolism of a specific carbon source if it has encountered that carbon source in the past, a form of memory driven by intergenerational transmission of metabolic proteins and by activation of regulatory networks that persists beyond the presence of the regulatory activator [70]. Pseudomonas aeruginosa demonstrates heritable HDB in the early stages of biofilm development, as cells that transiently attach to a surface and reenter a planktonic state more quickly attach and form biofilms when they later encounter another surface [71]. Variations in membrane potential induced by light exposure across a Bacillus subtilis biofilm persisted for hours after cessation of light exposure, resulting in an electrically mediated form of HDB [72].

Bacteria rely on diverse regulatory mechanisms in order to convert noise levels and spatial and temporal signals into coherent cellular states (Figure 2). The first observation of a well-defined cellular state in bacteria, sporulation, was independently reported in the late 1800s by both Robert Koch [73] and Ferdinand Cohn [74]. However, the mechanisms that regulate the intricate expression cascades that control spore formation in B. subtilis were only characterized many decades later using meticulous biochemical and genetic approaches [75], and recent work has linked the decision to sporulate to stochasticity in an extrinsically noisy phosphorelay governing expression of the spo0A master sporulation regulator [76]. The use of a phosphorelay with multiple discrete phosphorylation steps allows for a timed progression through the sporulation process, providing cellular checkpoints in the commitment process before a cell transitions into a subsequent developmental step in the cascade (Figure 2).

Figure 2.

Sporulation initiation converts cellular noise into a stepwise phosphorylation cascade. Genetic competence in Bacillus subtilis is initiated by small fluctuations in comS, an inhibitor of the mecA – clpP/C proteolysis complex that controls the presence of the master competence transcriptional regulator comK [73,74]. Phase variation uses transposition of DNA, typically in the promoter region, to turn transcription on in the desired direction. This step is stochastic and reversible, with flagellar regulation in E. coli serving as a classic example. In some cases, pulses in bacterial regulons can be directed by the temporary competition for a limited shared resource, such as RNA polymerase. In this scenario, termed “molecular time sharing”, the regulator that has temporary control of the transcriptional machinery can direct its own regulation to extend or limit the duration of the pulse [81].

In B. subtilis genetic competence is another heterogeneously manifested cell state driven by noise. In this system a transient cell state is generated when noise changes the activity of a proteolytic enzyme [77], allowing an important master transcriptional regulator (comK) to evade degradation. This interrupted negative feedback loop, which contains a hypersensitive switch and a delay mechanism to restore the proteolytic regulator, can control the onset and reset of cells pulsing in and out of the competent state [78]. While heterogeneous cells in the examples above remain genetically clonal, there are mechanisms in which cells use stochasticity to make genomic changes that lead to physiologically distinct cell types. In a well-characterized example E. coli and Salmonella strains use phase switching, a reversible transposition of regulatory DNA, to tune transcription of genes involved in pili or flagellar synthesis [79,80]. Similar phase switching mechanisms occur in other organisms and processes, like toxin-expression in the Gram-positive pathogen Clostridium difficile [81], and allow a stochastic switching rate that turns on or reverses cellular phenotypes in a manner that can be regulated by expression of the transposition machinery [82] or other cues including the growth environment and the length of the DNA repeats [83]. In addition to the use of dedicated machinery and DNA structure to control cellular entry into physiological states, there are reports that small fluctuations in protein-level extrinsic noise in bacteria can toggle cells between different states by virtue of competition for a common resource like RNA polymerase. In B. subtilis, for example, small fluctuations in sigma factor levels lead to upregulation of the sigma-factor itself and result in distinct, temporary pulses in the expression of the regulon controlled by the extrinsically noisy sigma factor [84]. Under some conditions, it appears that the competition for RNA-polymerase and sigma factor binding works as a “molecular time-sharing” scenario in which cells can pulse different sigma factor-driven gene sets one at a time as these factors get a turn at the time-sharing competition for RNA polymerase [85].

No matter the source, phenotypic heterogeneity provides a variety of benefits to bacterial populations. In the face of rapidly changing environmental conditions, the transcriptional patterns and behaviors necessary for optimal fitness can change faster than any individual cell can adapt. The presence of transcriptionally heterogeneous subpopulations in such conditions acts as a form of bet-hedging [86], since a subpopulation that is less fit in the current environment might be more fit in a changed environment in the future. That subpopulation can then flourish in the new conditions to which it is better adapted, ensuring that some portion of the overall population survives. By hedging bets against future shifts in selective pressures through phenotypic heterogeneity, bacterial populations can improve their chances of long-term survival without energetically expensive overhauls of their transcriptome and proteome and without making genetic (evolutionary) changes that lock the population into one strategy that may have long-term costs. HDB often acts in service of a bet-hedging strategy, as an organism is more likely to hedge bets against a certain environmental condition if that environmental condition has been previously encountered by ancestor cells [69]. Another benefit provided by transcriptional heterogeneity is the division of labor, in which cells can segregate the production of metabolically incompatible or energetically expensive gene products into a subpopulation of cells [87]. The most distinct example of such functional compartmentalization can be observed in the multicellular cyanobacteria Anabaena, which grow in long chains of single cells. These organisms perform both photosynthesis and nitrogen fixation, which are chemically incompatible, as photosynthesis requires oxygen and nitrogen fixation cannot occur in the presence of oxygen. To solve this problem, Anabaena differentiate their cells into either vegetative cells, which are more common and perform photosynthesis, and heterocysts, a less common cell state which maintains an anoxic intracellular environment and fixes nitrogen [88]. Vegetative cells provide heterocysts with glucose while heterocysts provide vegetative cells with fixed nitrogen, allowing the organism as a whole to thrive through heterogenous metabolism. The division of labor is often associated with public goods dynamics, wherein subpopulations of bacteria secrete useful products that are accessible to all cells in a population, even cells that did not produce any product themselves. In marine bacterioplankton populations, siderophores, which sequester environmental iron and are subsequently taken up by bacterial cells, are produced as a public good. Non-siderophore producers were shown to selectively lose genes for siderophore production while retaining genes for siderophore uptake, indicating that division of labor dynamics in microbial communities are robust to the point of influencing evolutionary dynamics [89].

Use of scRNA-seq to identify regulators of heterogeneity and memory in bacteria

As discussed earlier and highlighted in Figure 2 several distinct mechanisms for the regulation of heterogeneous transcription and cellular differentiation have been identified in bacteria, but these mechanisms are generally difficult to identify. In mammalian systems single cell transcriptomic data can identify transcriptional regulatory mechanisms both when used independently (for example [26,90,91]) or when used in single cell multi-omic studies (for example [92,93]). In bacterial systems the evolutionary diversity between species offers both challenges and opportunities for the study of transcriptional mechanisms that regulate heterogeneity. On the one hand, transcriptional states that are heterogeneously expressed in one species or strain may be homogeneous in another, making it difficult to identify conservation in features and well-annotated cell types common in higher organisms. On the other hand, we predict that this diversity will ultimately allow for comparative genomics approaches that can identify genetic regulatory features that govern both the onset of cellular heterogeneity and also modulate frequency (percentage of the population expressing the genes) or amplitude (level of gene expression) in heterogeneous gene expression. An illustration of such a comparative genomics approach is shown in Figure 3. In this scheme, once a cell state is found to be heterogeneous in a target strain, single-cell data for related strains can be compared to determine differences in the presence and extent of heterogeneity across genetic diversity. When closely related strains that either lack heterogeneous gene expression or have significantly altered frequency or amplitude are identified, the genomic features common to strains with similar phenotypes can be compiled into a candidate list and tested genetically to confirm their hypothesized role. Such comparative genomics approaches can potentially be coupled with lab-based evolution experiments that evolve strains with different phenotypic heterogeneity outcomes or, as done in mammalian systems, with mutant libraries such as those used in Perturb-seq [35].

Figure 3.

Comparative genomics can potentially identify molecular regulators of heterogeneity. A genomics approach that takes advantage of the genetic diversity in bacteria can potentially identify regulatory features of heterogeneity. In this two-step approach potential regulators are first identified by comparing the heterogeneity of different strains. Once target genes that are expressed differentially in subsets of cells in a given strain are identified the corresponding genomic regions in strains undergoing different levels of heterogeneity are compared to identify potential regulatory elements (panel a). In the second step of the approach each candidate genetic element is introduced into strains with the opposite phenotype to determine their ability to influence heterogeneity (panel b).

In addition to comparative genomics approaches, the short half-life of bacterial transcripts may, in principle, provide computational opportunities to infer temporal events in a manner similar to pseudotime analysis in mammalian cells [94]. If technical advances achieve single cell transcriptomic readouts throughout the entire transcript or operon length it would be interesting to see if the short bacterial transcript half-life allows for analyses that determine whether given transcripts are being preferentially degraded within a cell cluster. Such analyses, if sufficiently robust, could predict whether a cell is entering a new state by transcriptional activation or by degradation of active transcripts (in essence a downregulation of genes).

Single-cell transcriptomics in the context of microbial communities

While it is common to work with pure cultures of a single bacterial species in the laboratory, microbes in nature almost always exist in diverse communities composed of a large array of bacterial taxa. Be it an oil pipeline, a deep-sea hydrothermal vent, or the human intestinal tract, bacteria must be considered in the context of their neighbors and the environment in which they live. Genomic heterogeneity, i.e., taxonomic diversity, is a primary metric with which to evaluate microbial communities, and significant gains in our understanding of microbial community dynamics have come from identifying which microbes are present within a given environment. The function imparted by the expression of genomes present in the environment is less well characterized and must be understood with the acknowledgment that isogenic, transcriptionally heterogeneous subpopulations exist and can serve distinct and essential roles within the community. As the field of single-cell bacterial transcriptomics matures, it is thus essential that an analysis of transcriptional heterogeneity is integrated into the study of microbial communities to better comprehend their immense, emergent complexity. Here, we present potential avenues of inquiry to accomplish this goal.

The simplest external influence on a microbe(s) behavior is its surrounding abiotic environment, and the conditions, nutrients, and stressors that it encounters. In E. coli, stress response to heat was observed to be heterogeneous, and mathematical modeling suggested the presence of heat-sensitive and heat-resistant subpopulations that coexist at any given time [95]. Later work identified pulsing in regulatory genes associated with multiple stress response pathways, including heat stress, that resulted in stress-resistant E. coli subpopulations [96], suggesting a bet-hedging strategy for stress response in this context. Given that stress response is a fundamental capability of microbial organisms, these findings suggest the potential for widespread heterogeneity and bet-hedging in response to adverse conditions (Figure 4(a)). Abiotic conditions can influence microbial community composition in a similar manner. Freshwater lake-associated microbiomes showed significant variation in genomic heterogeneity along temperature and oxygen gradients, which result in a wide variety of niches that are filled by a diverse set of bacteria [97]. We predict that strains identified across all sampled lakes might act as generalists [98], and use phenotypic heterogeneity to occupy multiple distinct niches if given the chance. Given the significant environmental differences between lakes, it is unlikely that the same transcriptional profile results in robust fitness across all conditions, and phenotypic heterogeneity has the capacity to help potential generalists adapt to specific abiotic niches.

Figure 4.

Phenotypic heterogeneity enables stress response and niche-filling in microbes and microbial communities. a) Bet-hedging using heterogeneous subpopulations of heat-resistant cells (red) allows for overall survival of a bacterial population after a sudden temperature shift kills heat-susceptible cells (light blue). b) A biofilm matrix (lower panel, purple) prevents uniform diffusion of a quorum-sensing signaling molecule (yellow circle) that induces pili formation in a bacterial species. Green cells sense the quorum and express pili, beige cells do not sense the quorum. c) Phenotypic heterogeneity allows a pathogenic bacterium to infect and occupy a variety of niches within its human host. Each cell color indicates a unique heterogeneous cell state corresponding to the occupied niche.

Within microbiomes, interactions between both isogenic cells and cells of different species result in significant transcriptional modulation. Quorum-sensing is a well-characterized mechanism for signaling and behavioral coordination in bacteria of the same species. In these systems, bacteria constitutively secrete a signaling molecule that, when sufficient concentration is reached (a quorum), induces a transcriptional response in all bacteria that can sense the quorum. Given that QS signaling is dependent on spatial distribution of the signaling molecule, there is inherent potential for heterogeneous quorum induction based solely on signaling molecule gradients. If the signaling molecule is limited in its diffusion by abiotic or biotic factors, such as the matrix of a biofilm or free-floating bacterial aggregate (Figure 4(b)), then it is possible that multiple “mini-quorums” are induced among cells that are extremely close to one another. The end product would be pockets of quorum-activated cells that are especially close to one another interspersed with quorum-silent cells, resulting in phenotypic heterogeneity. QS-induced phenotypic heterogeneity has been observed in a variety of bacteria and has been linked to antibiotic persistence phenotypes in human pathogenic species [63,64], and single-cell transcriptomic approaches hold the potential to identify new regulatory effects associated with this phenomenon. Microbiome constituents also interact with each other through metabolic competition and exchange, and these dynamics result in similar niche filling to that observed as a result of abiotic factors [99]. It therefore stands to reason that phenotypic heterogeneity might allow microbiome members to participate in opportunistic niche filling, allowing them to outcompete other members of the community. Technical development in bacterial single-cell transcriptomics has not yet progressed to the point of profiling complex microbiomes, but achieving that goal will help to answer outstanding and fascinating questions: how widespread is phenotypic heterogeneity within microbiomes? Is phenotypic heterogeneity inversely correlated with genomic heterogeneity, i.e., does an organism occupy fewer niches if there are more species in the community to fill those niches?

A more specialized influence on phenotypic heterogeneity in microbes and microbial communities is the interaction between microbes and their host organism, such as that between a pathogen and the host immune system. The stressors and selective pressures applied by an immune response have resulted in unique and highly differentiated pathogenesis mechanisms, some of which take advantage of phenotypic heterogeneity to increase fitness in the face of immune opposition. Borrelia burgdorferi, the tick-transmitted causative agent of Lyme disease, and Neisseria gonorrhoeae, a common sexually transmitted infection, both make use of antigenic variation to evade the host immune response [100,101]. By varying the structure and composition of external antigens, cells can differentiate into phenotypically distinct subpopulations to avoid recognition by the adaptive immune system and induce more severe infection. Salmonella enterica serovar Typhimurium is an intracellular pathogen that makes use of transcriptionally distinct cell types throughout its pathogenic life cycle, beginning with the development of virulent and avirulent subpopulations upon inoculation into the host. Virulent cells invade the epithelial lining of host tissues and induce an inflammatory response that benefits avirulent cells, allowing the latter to flourish and more readily transmit to other potential hosts [102]. Beyond this initial differentiation, Salmonella Typhimurium invades a variety of immune cells including M cells, dendritic cells, and macrophages [103] (Figure 4(c)). Upon invasion of the latter, Salmonella Typhimurium further differentiates into actively growing or persister subpopulations, with active cells perpetuating infection and persister cells forming a reservoir from which to initiate secondary infection at a later time [104]. Persister cells also have the capability to influence macrophage polarization away from an antibacterial pro-inflammatory state and toward a tolerant anti-inflammatory state through type 3 secretion system effectors, dampening the host immune response and prolonging their survival within the macrophage [105]. Interestingly, Salmonella Typhimurium pathogenicity can be inhibited by other members of the human-associated microbiome, specifically the gut microbiome. Short-chain fatty acids, a common microbial metabolite in the intestinal tract, were shown to inhibit expression of type 3 secretion systems in virulent Salmonella Typhimurium subpopulations [106]. This resulted in lessened virulence through inhibition of phenotypic heterogeneity in an invading pathogen on the part of the gut microbiome. Single-cell transcriptomic analysis of this interaction might serve to identify markers for anti-pathogenic commensal members, unlocking alternative approaches for the treatment of bacterial infections.

Conclusions and outlook

Single-cell bacterial transcriptomics is a rapidly maturing and relatively untapped experimental approach, and the prevalence and importance of phenotypic heterogeneity across the domain of Bacteria provide many fascinating and as-yet unanswered research questions that can be addressed with these new techniques. We would be remiss not to highlight the discoveries already made using bacterial scRNA-seq. Several studies identified previously unknown cellular states including metabolic subpopulations [44,45], acid tolerance [46], toxin expression [45] and others. In the context of phage activation and cellular response to stress several studies identified cellular heterogeneity [46,49,51]. Functionally, single-cell analysis was able to predict perturbations that can modulate the size and activity of a subpopulation of cells that produce toxins [45] and to link transcriptomic cellular states to antibiotic tolerance in the pathogen Klebsiella pneumoniae [51]. Finally, Homberger et al. recently observed heterogeneity in small regulatory RNA expression in Salmonella Typhimurium using fluorescence-activated cell sorting (FACS) combined with CRISPR-based rRNA depletion [50]. These insights represent the first steps toward a better understanding of the role of phenotypic heterogeneity in microbes, the detailed mechanisms of fundamental transcriptional regulation allowing variable expression in subsets of cells, and the intricate dynamics underpinning bacterial behavior in complex microbial communities. We eagerly await the next forays into this exciting frontier.

Funding Statement

A.W. and A.Z.R. acknowledge funding from the UNC School of Medicine and the CGIBD institute at UNC including by a grant from the National Institutes of Health, [P30 DK034987].

Disclosure statement

No potential conflict of interest was reported by the author(s).