Emerging tools for uncovering genetic and transcriptomic heterogeneities in bacteria

Abstract

Bacterial communities display an astonishing degree of heterogeneities among their constituent cells across both the genomic and transcriptomic levels, giving rise to diverse social interactions and stress-adaptation strategies indispensable for proliferating in the natural environment (Ackermann in Nat Rev Microbiol 13:497–508, 2015). Our knowledge about bacterial heterogeneities and their physiological ramifications critically depends on our ability to unambiguously resolve the genetic and phenotypic states of the individual cells that make up the population. In this short review, I highlight several recently developed methods for studying bacterial heterogeneities, primarily focusing on single-cell techniques based on advanced sequencing and microscopy technologies. I will discuss the working principle of each technique as well as the types of problems each technique is best positioned to address. With significant improvements in resolution and throughput, these emerging tools together offer unprecedented and complementary views of various types of heterogeneities found within bacterial populations, paving the way for mechanistic dissections and systematic interventions in laboratory and clinical settings.

Introduction

Bacteria occupy every conceivable ecological niche of our planet, and their physiological and social behaviors are inextricably linked to those of all other living organisms on Earth. Although it has long been recognized that population heterogeneities exist within most microbial communities (Kroll et al. 1988; Novick and Weiner 1957; van Ham et al. 1993), their pervasiveness and degree of complexity did not fully manifest themselves until researchers began to go beyond the bulk level and examine the identities and physiology of individual cells within each population. It is now known that naturally harvested and host-derived microbial populations usually comprise at least hundreds of different species (Anantharaman et al. 2016; Turnbaugh et al. 2007) and that even lab-grown clonal cultures often harbor distinct subpopulations with temporally varying compositions (Elowitz et al. 2002; Ozbudak et al. 2002). Such diversities usually contribute fitness and survival advantages to the population as a whole, especially under adverse growth conditions or host environments (Dewachter et al. 2019; Reyes Ruiz et al. 2020). It is thus of no exaggeration to say that the individual characteristics of each bacterial cell, however miniscule and subtle when analyzed in ensemble, may bear profound ecological and biomedical consequences on a grander scale.

Although single-cell techniques have now been widely used to study population heterogeneities in eukaryotic systems, discovering and characterizing heterogeneities for bacterial populations have proven more challenging due to the unique physical and biological properties of bacteria. Nonetheless, recent years have witnessed the advents of several high-resolution, high-throughput techniques capable of profiling genomic and phenotypic diversities of bacterial cells, even for complex environmental and clinical samples. Here, I discuss how these recently developed techniques could help to obtain deeper functional insights on the collective behaviors of bacterial populations by illuminating on their genetic and transcriptomic heterogeneities with previously inaccessible level of details.

The following discussion is structured by categorizing various techniques into three subgroups: (i) single-microbe sequencing techniques for resolving genetic and transcriptomic heterogeneities, (ii) spatial genomics and transcriptomics tools for bacteria, and (iii) techniques for studying bacterial temporal heterogeneity. It should be noted that “heterogeneity” is a general term used to describe a broad range of genetic and phenotypic variations, both between and within species, and this presentation is organized based on the primary intended use for which each technique was originally developed. However, many techniques are inherently versatile and, with minor procedural adjustments, are capable of revealing bacterial heterogeneities of different natures. For example, by profiling both 16S rRNA and mRNA, it is possible to simultaneously characterize the taxonomic makeup of a multi-species community, as well as variations in gene expression patterns among cells of the same species or strain. In addition, while this short review is focused on heterogeneities at the genetic and transcriptomic levels, significant variations in epigenetic and metabolic states among even clonal cells are also prevalent in bacteriology (Davidson and Surette 2008). For excellent review articles covering recent progresses in studying epigenetic and metabolic heterogeneities in bacteria, please refer to Beaulaurier et al. (Beaulaurier et al. 2019) and Evans et al. (Evans and Zhang 2020).

New single-cell sequencing tools for bacteria

Multi-species bacterial communities are ubiquitous in nature, and their phylogenetic compositions have strong and diverse impacts on their immediate neighborhoods of inhabitance and beyond. For example, it was shown that the presence of a mixture of nitrogen-fixing bacterial species in the soil helps to enhance the survival and biomass production of plants (van der Heijden et al. 2006) and that imbalance in the gut microbiota composition is linked to a range of chronic diseases in human beings (Arumugam et al. 2011) such as obesity (Ley 2010), autoimmune disorders (De Luca and Shoenfeld 2019), and cardiovascular diseases (Jie et al. 2017).

Furthermore, even within a single species, there usually exists multiple strains with subtle differences in their genetic makeup. Not only are allelic variations the principal ingredients of evolution for any given species, they also shape how individual cells within the community interact with each other through cooperation, competition, and horizontal gene transfer (Rakoff-Nahoum et al. 2016; Smillie et al. 2011). In addition, strain-level genetic variations can provide protections to clonal populations under stressful conditions. For example, resistance-conferring mutations can sporadically arise within clonal isolates and then spontaneously vanish after stressor removal either due to the intrinsic genetic instability (e.g., tandem gene amplifications) or high fitness-cost of the mutation (Andersson et al. 2019; Nicoloff et al. 2019). Thus, understanding the behaviors of microbial communities demands going beyond the species-level to quantify genetic diversities at the strain-level, which was difficult to achieve with conventional shotgun metagenomic sequencing due to the prevalence of ambiguous sequencing reads when the sample contains closely related species or strains, as is generally the case for microbes collected from natural and host environments.

Droplet-based microfluidics whole-genome sequencing offers a promising solution to resolve both inter- and intra-species genetic differences among individual bacterial cells. An early endeavor in this direction is the development of the single-cell genomic sequencing (SiC-seq) technique (Lan et al. 2017). SiC-seq uses a co-flow droplet generator to encapsulate single bacterial cells in molten agarose microdroplets, which, upon solidification, permits cell lysis and DNA extraction to be performed at the bulk level by keeping individual cells physically separated. Afterwards, the microgels are re-encapsulated and the genomic DNA enclosed within are subject to tagmentation before being pooled for Illumina sequencing. A subsequent modification of SiC-seq, known as SAG-gel (Chijiiwa et al. 2020), performs cell lysis, genomic DNA purification, and the first round of whole-genome amplification all in the agarose matrix. Agarose beads containing amplified DNAs were fluorescently sorted into 96-well plates where another round of whole-genome amplification was performed to generate the single-amplified genomes (SAG) library. Using SAG-gel, the authors obtained 346 SAGs from the mouse gut microbiota and identified potential candidates involved in dietary fiber metabolism.

Instead of relying on agarose microgels, Zheng et al. developed a water-in-oil droplet microfluidics platform for high-throughput sequencing of complex microbial samples at the single-cell level (Zheng et al. 2022). This workflow, known as Microbe-seq, streamlines single-cell droplet encapsulation, cell lysis, whole-genome amplification, tagmentation, and droplet barcoding on an integrated, four-stage microfluidic platform before pooling the library for sequencing (Fig. 1A). As with other droplet-based microfluidics approaches, each sequenced read can be traced back and assigned to a single bacterium from the original sample, substantially alleviating ambiguities in sequencing reads assignment that plague many previous approaches. In addition, this unified microfludics platform should lead to better automation and scalability compared to earlier methods. Using Microbe-seq to profile human gut microbiome samples, the authors were able to obtain over 20,000 SAGs from 76 species, many of which were non-culturable. Despite having an average coverage of only ~ 10% for each SAG, comparing single-nucleotide polymorphism (SNP) patterns between all SAG pairs allowed the authors to resolve different strains, including non-culturable ones, of the same bacterial species, and to construct a consensus genotype for each.

Fig. 1.

Newly developed single-microbe sequencing techniques. A Workflow of Microbe-seq, which builds upon previously developed droplet-based single-cell sequencing techniques and streamlines single-cell isolation, cell lysis, and sequencing library preparation with an integrated microfluidic platform. B Total RNA-seq with MATQ-seq. Single-cell isolation is achieved with FACS sorting. Alternatively, ten-pooled samples can be allocated to each well of the microplate to improve the number of detectable genes. C General workflow for PETRI-seq and microSPLiT. In microSPLiT, there is an extra step of mRNA enrichment after cell permeabilization, where mRNAs are polyadenylated in situ with E. coli poly(A) polymerase I (PAP). Barcoding of transcripts from different cells is achieved through three rounds of split-pool, where transcripts from any two cells can be distinguished if these cells travel to different wells of the microplates at least once during the split-pool, as in the case for cells 1 and 2 here

Microfluidics-based sequencing techniques as such create new opportunities to profile the genomic diversities and population abundances of microbial communities at the sub-species level. Nonetheless, they still need to overcome challenges associated with low recovery of genomic DNA, short sequencing read lengths, and biases introduced in the amplification step. It is possible that with optimizations of cell lysis protocols and the introduction of long-read sequencing technologies into the workflow (Liu et al. 2021), the quality of reads and consequently the coverage and accuracy of Microbe-seq and alike can be further improved to approach bonafide single-cell resolution.

Even within a bacterial population that is genetically identical, gene expression patterns among individual cells can differ significantly. Heterogeneities at the transcriptomic level underlie a variety of adaptation and survival strategies widely adopted across the bacterial domain, including division of labor during biofilm development or infection and the coexistence of cells with different stress-tolerance levels within a population (Dhar and McKinney 2007; van Gestel et al. 2015). It is not uncommon to find the behavior of an entire bacterial population shaped by that of a tiny proportion of its constituents. For example, the fraction of multidrug-tolerant persister cells within a bacterial culture may be as low as one per million; they can nevertheless render antibiotics ineffective by preventing the complete eradication of their community and reseeding another population after drug removal (Bigger 1944). In such a scenario, the development of high-resolution transcriptomic tools is key for understanding how the presence of distinct and possibly rare phenotypes within a bacterial population modulates the behavior of the collective whole.

Transcriptomic profiling of individual cells by single-cell RNA sequencing (scRNA-seq) has now become routine practice for eukaryotic systems (Papalexi and Satija 2018; Wang and Navin 2015), whereas the development of corresponding tools for bacteria was significantly lagging behind for several reasons. First, the amount of RNAs in a bacterium, typically in the femtogram range, is usually at least two orders of magnitude lower than that of a eukaryotic cell (Milo and Phillips 2015). On average, bacterial messenger RNAs are present with exceedingly low abundances (< 1 copy per cell) (Markson et al. 2013; Taniguchi et al. 2010) and are degraded on much faster timescales, with typical half-lives ranging from seconds to minutes, in comparison to their eukaryotic counterparts which can remain stable for hours or even days (Nierlich and Murakawa 1996; Rauhut and Klug 1999; Sharova et al. 2009). Probably most importantly, most functional bacterial transcripts are non-polyadenylated, precluding the usage of oligo(dT) primers for mRNA enrichment and concomitant depletion of rRNAs (Sarkar 1997). In addition, due to the small physical dimensions of bacterial cells and their thick cell envelops, performing single-cell encapsulation and lysis on microfluidic chips is also technically more demanding. Despite these challenges, high-throughput scRNA-seq techniques previously developed for eukaryotic cells have recently been successfully adapted to profile bacterial transcriptomes.

Building upon MATQ-seq (multiple annealing and dC-tailing-based quantitative scRNA-seq), Imdahl et al. devised a workflow for total RNA sequencing in bacteria (Imdahl et al. 2020) (Fig. 1B). In this protocol, fluorescence-activated flow cytometry (FACS) was used to deliver single or ten-pooled Salmonella and Pseudomonas cells into individual wells of PCR plates, where cell lysis, reverse transcription, cDNA amplification, and tagmentation subsequently took place. The indexed libraries were then pooled and sequenced on Illumina platforms. The use of random hexamer-containing MALBAC primers during reverse transcription in principle allows all transcripts within the cell to be targeted and sequenced, including low-abundance species (Sheng et al. 2017). The reads from mRNAs and small non-coding RNAs can then be filtered from those of tRNAs and rRNAs via bioinformatic analysis. The inclusion of the highly abundant rRNAs in the workflow, which account for > 90% of all sequencing reads, does substantially raise the cost per experiment, but nonetheless retains valuable information regarding the taxonomic makeup of the population, which should be particularly appealing for studies involving complex environmental samples. As a proof-of-concept, the authors compared the transcriptome of Salmonella measured with MATQ-seq under anaerobic and salt shock conditions to the existing database SalCom, and found good correlations between the two.

Given that gene expression patterns in bacteria can be highly dynamic and context dependent, and that bacterial mRNAs are generally labile, cell sorting and isolation prior to fixation have the potential risk of missing transient transcriptional events that could be physiologically significant. Two bacterial scRNA-seq methods, PETRI-seq (prokaryotic expression profiling by tagging RNA in situ and sequencing) (Blattman et al. 2020) and microSPLiT (microbial split-pool ligation transcriptomics) (Kuchina et al. 2021) obviate live-cell encapsulation and thus bypass potential complications associated with time delays and local environmental perturbations to cells before fixation (Fig. 1C). These two methods are technically similar and were developed based on SPLiT-seq (split-pool ligation-based transcriptome sequencing), a scRNA-seq method initially invented for transcriptomic profiling of the mouse central nervous system through combinatorial barcoding (Rosenberg et al. 2018). In both microSPLiT and PETRI-seq, reactions for sequencing library preparation occur within each individual bacterium itself via three rounds of split-pool barcoding. Specifically, after fixation and permeabilization, bacterial cells are split into a 96-well microplate, where cells from each well of the microplate are barcoded with a well-specific random hexamer via reverse transcription. Then, cells from all wells are pooled and re-split into another 96-well plate, where a second microplate well-specific barcode is ligated to the cDNA of each cell adjacent to the first barcode. Repeating the split-ligate-pool procedure once more yields a total of nearly one million (96³) different barcode combinations for cell identification, after which cells are lysed and their cDNA library prepared and sequenced on Illumina platforms.

The feasibility of microSPLiT and PETRI-seq has been demonstrated in both Gram-positive (B. subtilis and S. aureus) and Gram-negative species (E. coli), with both approaches capable of capturing up to a few hundred (~ 10%) mRNAs per cell, allowing for separation of condition-dependent growth states and discovery of rare developmental states such as genetic competence and prophage induction within isogenic populations. Recently, PETRI-seq was used to interrogate the global transcriptional patterns in exponential-phase E. coli and Staphylococcus aureus cell populations (Pountain et al. 2023), which revealed, in addition to local correlations in gene expression within operons, an unexpected genome-wide correlation presumably due to the passage of the replication fork. This study further demonstrates the power and immediate applicability of scRNA-seq methods in revealing concealed patterns in bacterial physiology.

In most scRNA-seq experiments, native transcripts are captured and reverse-transcribed to cDNA by binding to barcoded oligo(dT) primers at the poly(A) tails. Most bacterial RNAs, however, are not polyadenylated. Adopting existing scRNA-seq techniques to bacterial cells thus necessitates developing efficient and unbiased polyadenylation methods, and substantial progress has been made in this regard. For example, Smart-seq-total is a sequencing library preparation method capable of capturing and quantifying RNAs regardless of their polyadenylation status or size (Isakova et al. 2021). RNAs extracted from cell lysis were polyadenylated by the E. coli poly(A) polymerase (PAP) and subsequently processed through the Smart-seq2 protocol (Picelli et al. 2013), which, through the activities of the Moloney murine leukemia virus reverse transcriptase and the template switch oligo, allows for the generation of full-length and high-quality cDNA from single cells. In doing so, it becomes feasible to examine the identities and quantities of both coding and non-coding RNAs at the single-cell level. For example, Smart-seq-total was applied to reveal previously unknown relationships between the levels of various miRNAs and protein-coding genes in HEK293T cells. With the recent introduction of VASA-drop (Salmen et al. 2022), a scRNA-seq method that utilizes droplet-based microfluidics to automate sequencing library preparations, including the addition of poly(A) tails, the throughput of single-cell total RNA-seq has been further improved by at least one order of magnitude.

Although many poly(A)-independent total RNA-seq methods, including Smart-seq-total and VASA-drop, were originally developed for investigating non-polyadenylated RNAs in eukaryotic systems, the polyadenylation and subsequent enrichment procedures employed in these methods could potentially provide new directions to improve existing techniques specifically tailored for profiling bacterial transcriptomes.

Emerging spatial genomics and transcriptomics techniques for bacteria

The geographical arrangement of microbes within a community is known to have major influences over the stability of the population and also how it interacts with the surrounding environment (Kastrup et al. 2008; Lee et al. 2013; Nagara et al. 2017; Nava et al. 2011; Pasarkar et al. 2021). Methods for delineating the relative abundances of taxa within microbial communities, including shotgun metagenomic sequencing and 16S rRNA amplicon sequencing, are well-established and routinely used nowadays (Hasan et al. 2014; Jovel et al. 2016). However, these methods come with a cost of losing the knowledge about the spatial placement of different microbial taxa within the sample. Conventional imaging-based techniques such as fluorescence in situ hybridization (FISH), though can preserve spatial information for a small number (typically ≤ 4) of species (Amann and Fuchs 2008), cannot be readily scaled up to profile samples with even moderately complex taxonomic makeup due to spectral overlap.

To improve the ability to spatially resolve phylogenetic heterogeneities within microbial communities, Sheth et al. developed a multiplexed sequencing method named metagenomic plot sampling by sequencing (MaPS-seq) (Sheth et al. 2019) that is capable of delineating the microbiome biogeography at micrometer-scale resolution (Fig. 2A). In MaPS-seq, microbiome samples are first embedded within a polyacrylamide matrix containing universal reverse primers for the species-specific 16S rRNA, and the matrix is then fractured into cell clusters (gel particles) with diameters in the 10–30-µm range. Cell lysis and immobilization of cellular DNA occur within each of these clusters, preserving the local spatial relationship between different taxa. Using microfluidics, the cell clusters are then co-encapsulated with gel beads containing barcoded forward primers so that each cluster can later be uniquely identified. Both the forward and the reverse primers contain photocleavable linkers, allowing for release and subsequent amplification and deep sequencing of the barcoded genomic DNA. Processed sequencing reads can then inform about the identities and abundances of different bacterial taxa present within each cluster. How a taxon varies in abundance across clusters gives a measure of its spatial heterogeneity within the microbiome sample, and comparing the abundances between taxa both within and across clusters can reveal the strengths of co-associations between taxa both locally and globally, which serves as a valuable guide for discovering functional interactions between species within the sample.

Fig. 2.

Emerging spatial genomics and transcriptomics techniques. A Schematic diagram of sample preparation steps of MaPS-seq, along with illustrations of chemical modifications of the polyacrylamide matrix and the gel bead which allow for barcoding of clusters and PCR-amplification of the 16S V4 region. B The two-step labeling procedure of HiPR-FISH, along with an explanatory diagram showing how four different bacterial species (species 1, 2, 3 and N) can be resolved with HiPR-FISH. In this example, the 16S rRNAs of species 1 are bound in equal proportions to three types of probe pairs, yellow/blue, red/yellow, and purple/blue, giving a unique spectral image that can be distinguished from those of species 2 (which only produce blue emissions), species 3 (in which half of 16S rRNAs fluoresce in purple and the other half in blue/green), and species 4 (only produce red and blue). C µExM, which is based on the expansion microscopy (ExM), relies on the difference in cell wall expandability to resolve bacteria of different species. D Schematic illustration of seqFISH, showing how fluorescent signals from three different transcripts are turned on and off in sequence via hybridizations between the gene-specific primary probes and the fluorescently labelled readout probes. In the example given, cell 1 expressed all three transcripts, whereas cell 2 did not express transcript 2. Fluorophores of different colors can be used to visualize multiple genes each round, and rRNAs can be targeted in addition to mRNAs to enable par-seqFISH. E STRS introduces an extra step, in situ polyadenylation, to the commercially available Visium protocol to enable the capture and spatial mapping of total RNA within the sample, including bacterial and viral RNAs which are not natively polyadenylated

The applicability of MaPS-seq in charting the biogeography of complex microbial samples was demonstrated in a case study with the mouse gut microbiome, where heterogeneous distributions of various taxa and their co-association patterns were analyzed in three separate gastrointestinal regions. The findings revealed diet-induced spatial re-organization of the microbiota in the distal colon. In practice, MaPS-seq should be particularly well suited for metagenomic studies of microbiomes with large physical dimensions because only a subset of cell clusters is needed to infer the taxonomic organizations of the entire original sample, and it is a relatively cost-efficient approach since only the V4 regions of 16S rRNAs rather than the whole bacterial genome are sequenced. On the other hand, due to the inevitable loss of cell clusters during sample preparation, spatial context other than co-occurrence information is lost, and it is not possible to unambiguously map out the identities of all cells and their respective locations within the original intact sample.

An alternative imaging-based approach, named high-phylogenetic-resolution microbiome mapping by fluorescence in situ hybridization (HiPR-FISH) (Shi et al. 2020), can instead map out the identities of individual bacterial species within a microbial community while preserving its original spatial structure (Fig. 2B). Based on fluorescence in situ hybridization (FISH), HiPR-FISH uses a two-step combinatorial labeling strategy to overcome the spectral limitations of the conventional FISH. In the first step, the 16S rRNA of each bacterial species present in the sample is hybridized to its specific encoding probe. The encoding probe is flanked on both sides with readout sequences, each of which then hybridizes to a secondary, fluorescently tagged readout probe in the second step. It would still be far from enough to resolve the hundreds of species present in a typical microbiome if each rRNA were merely identified by the combination of two fluorophores, given the limited number of spectrally distinct fluorophores to choose from.

To address this limitation, in HiPR-FISH the rRNA molecules in a given cell were instead hybridized to encoding probes that are flanked by different combinations of readout sequences, which allows the identity of each rRNA and its corresponding bacterial species to be uniquely determined by the presence or absence of each of the ten fluorophores via spectral imaging. This technique takes advantage of the fact that rRNAs are present in high abundances in each bacterial cell, which is usually an undesirable feature for sub-cellular RNA imaging and counting due to spatial crowding. In the present case, the high copy numbers of rRNAs makes it possible to for them to be labeled with multiple fluorescent probes of different colors within the same cell, yielding a distinct spectral signature for each bacterial species. With ten fluorophores in use, HiPR-FISH can resolve a total of $2^{10}-1=1023$ bacterial species with a single-round of imaging (~ 5 min per field of view) while preserving the native arrangement of cells within the sample, and multiple rounds of hybridization and imaging could be implemented to increase the throughput further, making HiPR-FISH an attractive approach to profile the spatial phylogenetics of microbial communities at the single-cell level. Applying HiPR-FISH to study the mouse gut microbiome and human oral biofilms, the authors were not only able to quantify the abundances of the constituent species in each case, but also mapped out the distances between them despite the spatial and biological complexities of the host environment.

Instead of using spectrally distinct fluorophores to distinguish between species within a heterogeneous microbial community, another imaging-based technique based on the expansion microscopy, μExM (expansion microscopy of microbes), was developed to approach this task from a different angle (Lim et al. 2019). In the original expansion microscopy (Chen et al. 2015a), specimens embedded in a swellable hydrogel are physically and isotropically stretched as the embedding hydrogel expands upon contact with water. When examined under a microscope, this physical enlargement effectively achieves a higher magnification and improved spatial resolution for the sample. Rather than achieving a uniform expansion across the entire sample, μExM attempts to differentially expand cells to various degrees in a species-specific manner (Fig. 2C). This is possible due to the different physical structures and biochemical compositions of the bacterial cell walls between species. These differences impart distinct mechanical properties and expandability to cells, which can serve as a contrast agent for resolving the identities of species within a microbial community. This contrast, as quantified by the expansion ratio between species, can be further enhanced through partial cell wall digestion by lysozyme to achieve even better resolving power.

The authors demonstrated the feasibility of μExM by successfully identifying nine commensal species from the human gut, including both Gram-positive (e.g., Bifidobacterium breve) and Gram-negative (e.g., Bacteroides ovatus) ones. Because the extent to which cells are expanded depends on the cell wall characteristics, it was shown that μExM could also be used to resolve certain phenotypic heterogeneities within the same species, such as variations in cell wall damage resulted from antibiotic treatment. Similar to HiPR-FISH, μExM is compatible with bacterial strains that are not amenable to genetic manipulations or not even culturable. μExM does not rely on expensive optics or fluorescent markers either and is thus more budget-friendly. Nonetheless, its performance with more complex microbial communities, such as those found in the human microbiota containing hundreds to thousands of species, still awaits further assessment.

Besides heterogeneities at the genetic level, local variations in chemical compositions and physical properties within a bacterial community inevitably create conditions for the emergence of spatial heterogeneities in gene expression. Probing transcriptomic heterogeneities in intact microbial samples while retaining spatial information has been challenging, as sequencing-based methods typically require spatial homogenization of the bacterial community during sample preparation. Consequently, we still know very little about how cells of different transcriptomic states are distributed and associated with each other within a bacterial community, and what physiological functions these microscale organizations serve.

To address this limitation, seqFISH (sequential fluorescence in situ hybridization), previously used for spatial transcriptomic profiling of eukaryotes (Lubeck et al. 2014; Shah et al. 2016), has recently been adapted to quantify mRNA abundances in individual bacterial cells (Fig. 2D), including both planktonic cells as well as those embedded within biofilms (Dar et al. 2021). In seqFISH, the sample is first fixed and permeabilized, and each mRNA of interest is then hybridized with 12 to 20 single-stranded non-fluorescent DNA probes known as the primary probes. Each primary probe is flanked on both sides by short sequences uniquely assigned to each gene. The flanking regions of the primary probe then hybridize to short fluorescently tagged readout probes to enable visualization of the specific transcript. While all mRNAs under study are hybridized to their respective primary probes at once, the hybridization of the readout probes and thus the switching-on of fluorescence signals for each gene occurs sequentially. Using fluorophores of different colors, transcripts of up to three different genes can be visualized during each round of hybridization. The short readout probes are readily stripped afterwards while all primary probes remain bound to the transcripts, allowing for subsequent rounds of readout probe hybridization and visualization of the remaining transcripts. The abundance of each mRNA per cell can be obtained by directly counting the number of fluorescent spots inside the cell, or by converting the total fluorescent intensity per cell to mRNA copy numbers based on the known intensity profile of a single transcript in cases where individual fluorescent spots cannot be spatially resolved due to signal crowding. Using this approach, the authors were able to quantify the abundances of 105 mRNAs in individual Pseudomonas aeruginosa cells in a single experiment, comparable to what is currently achievable with scRNA-seq approaches, and also to reveal the spatial distribution patterns of these mRNAs within biofilms.

In addition to mRNAs, rRNAs can also be labeled to enable parallel seqFISH (par-seqFISH). In such case, fluorescence signals from labeled rRNAs can be used to distinguish between different bacterial species present within the same population, or to pool samples collected from different growth conditions or time points. In other words, a single par-seqFISH experiment can simultaneously reveal the transcriptomic and taxonomic (or temporal) profiles of an entire bacterial population at the single-cell level while preserving the original spatial relationship between individual cells within the sample. For example, the authors performed par-seqFISH with both young (10 h) and aged (35 h) biofilms, revealing how aerobic and anaerobic pathway genes vary in expression strength and spatial distribution as the cell density increases. Spatial correlations in gene expression patterns can provide valuable insights regarding how different physiological processes are coordinated both locally and globally across the population, and may reveal unexpected functional relationships between different genes and/or species, which may not manifest when decontextualized from the specific biogeographical structure of the environment in which the cells reside.

Very recently, an orthogonal approach named STRS (spatial total RNA-sequencing) was developed for spatial mapping of total RNAs in sectioned tissues (McKellar et al. 2023). STRS is based on the commercially available Visium spatial technology from 10 × Genomics, where RNAs released from fixed and permeabilized tissue sections are captured by poly(dT) probes on the slides patterned with positional barcodes (Ståhl et al. 2016). Spatially barcoded cDNAs are then synthesized from captured RNAs to generate the final sequencing library. STRS extends the applicability of the Visium spatial technology by incorporating yeast poly(A) polymerase-catalyzed polyadenylation into the workflow (Fig. 2E).

This simple addition allows RNAs that are not endogenously polyadenylated to be captured on Visium slides and sequenced, thereby paving ways for spatial mapping of non-coding RNAs such as lncRNAs, snRNAs, and miRNAs within tissues as well as microbial RNAs. The versatility of STRS was demonstrated with two separate case studies in mice, which revealed spatial patterns of non-coding RNAs involved in muscle regeneration and viral RNAs associated with myocarditis that had been missed by earlier methods. Furthermore, as STRS captures both viral and host RNAs, it allows for direct interrogation of the relationship between the viral load and the transcriptional response of the host. Given its flexibility and ease of implementation, it is likely that STRS will also become a widely adopted spatial transcriptomic tool for investigating bacterial heterogeneity.

Compared to FISH-based methods, STRS does not require the identities of the target species or their sequences to be known a priori. It also avoids the setup of an automated fluidics system for delivering hybridization probes and buffers to the sample. In addition, STRS should perform better in identifying smaller RNA molecules as these may not be as efficiently labeled by the hybridization probes due to the limited number of hybridization sites available. On the other hand, par-seqFISH provides sub-cellular spatial resolution as well as single-molecule sensitivity, which is particularly relevant for bacterial samples given their tiny size and generally low mRNA abundances. It is conceivable to combine the strengths of both techniques to unlock spatial transcriptomic tools with even greater discovery powers. For instance, employing STRS for the initial screening of potentially important RNA species and subsequently using par-seqFISH to map their distributions and abundances could be a powerful approach.

Emerging techniques for studying heterogeneities in bacterial cellular dynamics

Within a bacterial population, both genetic mutations and gene expression noise can induce variabilities in cellular dynamics, even when cells are grown in constant conditions or triggered by the same stimulus. For example, a single point mutation in the cyanobacterial circadian clock protein is enough to cause the oscillation period to deviate from its usual ~ 24-h period (Ito-Miwa et al. 2020). On the transcriptional level, stochastic gene expressions are known to drive E. coli and B. subtilis, respectively, into the drug-tolerant persister cell state (Fisher et al. 2017; Lewis 2007) and the genetically competent state (Maamar et al. 2007; Suel et al. 2006) in an asynchronous manner. Temporal heterogeneity bestows a mechanism for bacteria to achieve division of labor and to maximize chances of survival when confronted with uncertainties in the external environment.

The combination of live-cell time-lapse fluorescence microscopy and genetically fused fluorescent proteins has been highly successful in discovering transient or asynchronous dynamic behaviors in bacterial cells. The single-cell time traces collected through this approach do not suffer from ambiguities associated with inferring temporal dynamics from static snapshots manually collected from a small number of predetermined time points, as is the case with single-cell sequencing and FISH-based methods. However, due to difficulties with simulating realistic growth conditions on the microscope, it has always been difficult to associate many microscopy-captured phenotypes to specific environmental conditions either found in nature or in other bulk-level experiments. For example, due to cell crowding and non-trivial requirements for fluidics handling, it is difficult to follow the fates of individual bacteria as they transition from the exponential phase to stationary phase and then back with a typical microscope.

Using a sophisticated custom-built fluidics system, Bakshi et al. devised a high-throughput platform which can monitor the time trajectories of bacterial physiology at the single-cell level despite cells are growth in a bulk culture (Bakshi et al. 2021). This platform, referred as the dual incubator setup, is mainly composed of a bulk culture incubator (flask on a shaker with optical density monitoring and medium exchange capabilities), and an incubator modified from a microfluidic device known as the mother machine (Wang et al. 2010) that traps cells into its channel-shaped structures for long-term time-lapse imaging (Fig. 3A). These two incubators are connected by an insulated duct through which cells in the bulk culture can be transferred to the mother machine at various growth stages or with different growth media. The flow of the media from the batch culture to the mother machine exposes cells from both incubators under the same growth condition at all times, and thus, the physiology of the cells from the mother machine, which can be tracked at the single-cell level over many generations, essentially mirrors that of the cells from the batch culture. This allows for researcher to delineate cellular dynamics of the cells grown in the bulk culture. Using this setup, the authors were able to monitor cell morphology and gene expression states (via fluorescent reporters) as cells in the bulk culture repeatedly transition between the exponential growth phase and the stationary phase, which is not possible with conventional time-lapse microscopy. The design of the mother machine also allows this setup to achieve an impressive throughput, by screening up to 16 strains in parallel and up to 10⁸ cells daily.

Fig. 3.

Techniques for decoding heterogeneities in bacterial cellular dynamics. A The dual-incubator system couples bulk culture to mother machine microfluidics for real-time tracking of cellular dynamics at the single-cell level. This setup constantly monitors the optical density (OD) of the bulk culture and can adjust the growth condition accordingly. At the same time, bacteria from the bulk culture are flown to the microfluidic device, carrying the growth medium with them, which essentially synchronizes the growth condition for cells both grown in the flask and in the mother machine channels. Cellular morphology and gene expression time traces (via fluorescent reporters) of cells from the mother machine can thus serve as a surrogate for those of cells in the bulk culture under various growth phases (e.g., exponential and stationary) and conditions. B High-throughput, imaged-based screening of pooled genetic-variant libraries in bacteria. Each bacterial cell from the pool expresses a plasmid containing a unique combination of a genetic variant and a barcode. The one-to-one correspondence between each genetic variant and its barcode is known a priori by sequencing. Following phenotype characterization by microscopy, the identity of each cell can be revealed by reading out their barcode using sequential fluorescent in situ hybridization. C In SIFT, bacteria are first grown in mother machine channels for phenotype characterization. Afterwards, the push-down valves separating the growth channels and the collection lane are opened, and cells deemed interesting by the user are transported to the collection lane using an optical trap, allowing for reseeding of colonies and various downstream analyses including next-generation sequencing

While microscopy has always been indispensable for studying biology, it has long remained a challenge to connect the observed cellular dynamics to the underlying genetic and transcriptional landscapes of cells in an efficient, high-throughput way. Below, I highlight three techniques recently developed to tackle this problem. All of these techniques leverage the idea of capturing cellular dynamics using microscopy and using deep sequencing to establish a link between observed phenotypes and the genetic or transcriptional state of the corresponding cells, providing potential solutions to pinpoint the biological origins of temporal heterogeneity in a large-scale, systematic manner.

The first two methods, developed by Emanuel et al. (Emanuel et al. 2017) and Lawson et al. (known as DuMPLING) (Lawson et al. 2017) respectively, are both based on the idea of genotyping cells in situ via barcode imaging (Fig. 3B). Specifically, pooled libraries were constructed where the one-to-one correspondence between each genotype (e.g., a specific deletion or mutation) and a genetic barcode is established in advance through sequencing. Following phenotype characterization by imaging, cells are fixed and the identities of the barcodes are decoded by sequential FISH. With DuMPLING, bacterial cells are trapped in the microfluidic channels of the mother machine. Lawson et al. took this approach to image a pooled CRISPRi library of hundreds of different strains and investigated the effects of various gene knockdowns on the dynamics of DNA replication and cell cycle progression in E. coli (Camsund et al. 2020). The conceptually similar high-content fluorescence imaging developed by Emanuel et al. were employed to screen in E. coli for better versions of the yellow fluorescent protein YFAST that are brighter and more photostable, but the same approach can be readily adopted to study how various genetic perturbations alter the dynamic behaviors of bacterial cells.

An alternative approach, developed by Luro et al., abrogates the need to insert genetic barcodes for genotyping cells, but instead combines a microfluidic device and an optical trap to isolate cells of interest for downstream genomic and transcriptomic analyses (Luro et al. 2020). This technology, named SIFT (single-cell isolation following time-lapse imaging), adds a secondary fluidic pathway to the canonical mother machine design for cell retrieval after phenotype screening (Fig. 3C). Access to this collection lane from the growth channels is controlled by a set of push-down valves, which only open after the completion of time-lapse imaging. Cells deemed worthy of further analysis are transported to this collection lane by optical trapping, and subsequently collected into dedicated bins for downstream off-chip analysis. As the entire retrieval procedure occurs without contamination or loss of cell viability, isolated cells can be used to reseed new colonies or directly subjected to next-generation sequencing. Such flexibility is a major advantage over barcode-based approaches, which require cells to be fixed on-chip after time-lapse imaging in order to decode their genotypes. The authors used SIFT to enhance the performance of two well-established synthetic gene oscillators, the dual-feedback oscillator (Stricker et al. 2008) and the repressilator (Elowitz and Leibler 2000; Potvin-Trottier et al. 2016), by screening among half a million mutants for the variants which displayed favorable dynamical properties. For example, the authors successfully isolated a variant of the dual-feedback oscillator that eliminated the need for drug inducers required by the original design and also exhibited improved oscillatory accuracy.

The versatility of these phenotype-genotype screening methods is evident, and each holds significant potential for teasing out the biological ingredients of temporal heterogeneity within a population. The barcode-based approaches do not necessarily require the use of a microfluidic device, which may not be readily available in many biology labs. In fact, it can be technically challenging to efficiently load and remove hybridization probes into and from the microfluidic chips due to the hydrophobicity of the PDMS device. Also, as genotypes are read out directly in situ, the spatial information is preserved, meaning these techniques could potentially be leveraged to probe spatiotemporal heterogeneities within a population. SIFT on the other hand, has several unique advantages such as bypassing the need for barcode insertion and the ability to capture live cells. This provides greater flexibility in downstream analysis on a small number of cells manually selected based on their phenotypes. Given the wide range of sequencing methods available, SIFT should be well-suited for investigating the factors giving rise to temporal heterogeneities, whether they stem from genetic, epigenetic, or transcriptional origins, whereas barcode-based methods are currently restricted to probing the effects of genetic perturbations on cellular dynamics. In addition, SIFT may also help to reduce the cost per experiment by bypassing expenses associated with constructing and visualizing the barcodes and conducting deep sequencing on the entire mutant library.

In terms of throughput, SIFT is able to screen the phenotypes of approximately half a million variants in a single experiment, whereas Emanuel et al. and Camsund et al. screened for about 60,000 YFAST variants and 235 CRISPR interference knockdowns, respectively. While the reported number of variants screened is lower for barcode-based methods, they allow for phenotype-genotype mapping for all screened variants rather than just a subset. By expanding the diversity of barcodes and hybridization rounds, it should be feasible to increase the throughput of barcode-based method even further. In general, the distinct principles and workflows of SIFT and barcode-based methods make them complementary approaches, and it is reasonable to expect both approaches will contribute to unlock of the mysteries of temporal heterogeneities within bacterial populations.

Conclusion and outlook

The consequences of genomic and transcriptomic heterogeneities within planktonic and sessile bacterial populations are manifold, ranging from the regulation of Earth’s biogeochemical cycle (Polerecky et al. 2021) to the onset and progression of human diseases in clinical settings (Andersson et al. 2019; Dewachter et al. 2019; Van den Bergh et al. 2017). For a long time, the roles of inter-cellular variabilities in decision-making and fate-shaping of bacterial populations were largely concealed due to the lack of experimental tools to genotype and phenotype bacteria with sufficient throughput and resolution. Such constraints are especially pronounced when analyzing environmental and host-derived samples, of which the taxonomic makeup is often dominated by nonculturable strains and the collective behaviors steered by underrepresented phenotypic traits. The recent developments of high-throughput, high-resolution genomic and transcriptomic tools for bacteria, including those highlighted above, represent a key step forward in our ability to discover bacterial heterogeneities and to subsequently formulate the mechanistic and teleological explanations thereof. For example, the field of microbial experimental evolution has benefited tremendously from advances in sequencing technologies. As a first endeavor, polony sequencing was applied to E. coli cells evolved for 200 generations and discovered polymorphisms and deletions that gave rise to population heterogeneity (Shendure et al. 2005). A year later, whole-genome sequencing of experimentally evolved Myxococcus xanthus cells identified single-nucleotide mutations responsible for the evolutionary transition between the predominantly cooperative mode of social interaction and the “cheating” mode (Velicer et al. 2006). Since these early studies, sequencing technologies have found numerous applications in experimental evolution studies and helped to elucidate how various environmental factors may shape the evolutionary trajectories and population heterogeneities of bacterial cells. Recent progresses in microbial experimental evolution studies are covered in reviews by McDonald (McDonald 2019) and Koonin et al. (Koonin et al. 2021).

The types of bacterial heterogeneity each method featured in this article was designed to address are listed in Table 1. Given their diverse technical foundations and workflows, it is unlikely that any of these techniques will absolutely prevail over the others. Besides the inherent pros and cons of each method, the technical expertise of the personnel conducting the experiment and budgetary constraints will also play a role in determining the method of choice.

Table 1.

Applicability of each of the techniques featured in this short review

Technique	Tx	G	Tr	Sp	D	Special technical requirement
Microbe-seq	●	●				Custom-made microfluidic chips
MATQ-seq for microbes	○	○	●
PETRI-seq and microSPLiT	○	○	●
MaPS-seq	●	○	○			Custom-made microfluidic chips
HiPR-FISH	●			●		A microscope with spectral imaging capability
µExM	●			●
par-seqFISH	●		●	●		Requires an automated fluidic delivery system
STRS	○		●	●
Dual incubator setup	○	●	●		●	Custom-made microfluidic chips and an automated fluidics system
DuMPLING and Emanuel’s high-content imaging	○	●		○	●
SIFT	○	●	●		●	Custom-made microfluidic chips with pneumatic valves, as well as lasers and various optical components for optical trapping

Tx inter-species genetic and taxonomic heterogeneity; G intra-species genetic heterogeneity; Tr intra-species transcriptional heterogeneity; Sp spatial information; D temporal dynamics information

Empty circles (○) denote heterogeneity types for which the corresponding technique should in principle be feasible given some adjustments to the protocol (e.g., sequencing the entire genome or transcriptome vs. a few pre-selected targets such as 16S rRNA)

In the coming years, we will certainly see refinements of existing techniques and the emergence of new ones. It is expected that major improvement in performance can be attained by combining multiple existing methods. For example, the coverage of SAG-based methods should benefit significantly through integration with long-read sequencing technologies. Similarly, the throughput of seqFISH can be substantially improved with multiplexed labeling strategies such as MERFISH (Chen et al. 2015b), which should be feasible if bacterial cells are first expanded using ExM to alleviate signal overcrowding. Time-lapse imaging of genetically barcoded strains could be done using valve-based PDMS devices that contain thousands of segregated microchambers to enable parallel screening for combinations of genetic variants and growth conditions on a large-scale.

Of particular biomedical relevance is to make these techniques simultaneously compatible with bacteria and eukaryotes, especially mammalian cells. Progresses have been made in this direction with the invention of techniques such as single-cell dual sequencing (scDual-seq) which can profile the transcriptomes of both the host and the infecting pathogen for individual mammalian cells via RNA-seq (Avital et al. 2017). Spatial transcriptomics tools for simultaneous mapping of bacterial compositions and the host transcripts, are starting to become available. One such example is the recently developed SHM-seq (spatial host-microbiome sequencing), which captures and sequences both polyadenylated RNAs of the host as well as hypervariable regions of the microbial 16S rRNA (Lötstedt et al. 2022). However, as of now, there still lacks a technique that can reliably resolve heterogeneities among infecting bacteria within a single host. Methods that can efficiently permeabilize both bacteria and host cells and then amplify transcripts with minimum biases despite large discrepancies in mRNA abundance and stability between species still await further development. Overcoming these challenges may entail substantial revamp of established protocols for various stages of sample preparation (e.g., microfluidic chip design, cell permeabilization), but it is nevertheless the inevitable next step if one were to go beyond phenomenological characterizations to systematically search for the dominant genetic factors and regulatory pathways that modulate bacteria-host interactions. For example, many of the highly successful human pathogens, such as Helicobacter pylori and Haemophilus influenzae, produce phase variants among the population with altered cell surface structures to evade immune clearance (Bergman et al. 2006; van der Woude 2006; Weiser and Pan 1998). Simultaneously profiling the genotypes of these bacterial subpopulations and the transcriptional activities of the host cell may reveal which signaling pathways of the host cell give rise to strain-specific susceptibility and heterogeneities in immune responses during infection.

It is important to recognize that the potential of any newly developed technology can only be maximized if it can be readily adopted by other research labs. Thus, any effort in simplifying existing protocols to make them more accessible and economically feasible will likely influence which technique will eventually become the go-to technique for most researchers. In general, methods based on streamlined protocols and commercially available reagents will be more accessible than the ones involving laborious and potentially error-prone sample preparations, as evidenced by the quick adoption of PETRI-seq by another research lab to investigate transcription-replication interactions in E. coli (Pountain et al. 2023). An another example, OIL-PCR, a one-step protocol for linking plasmid-encoded genes with bacterial 16S rRNA (Diebold et al. 2021), was recently developed by combing reaction steps of the established epicPCR protocol (Spencer et al. 2016) into one. This simplified workflow should be especially valuable for studying horizontally acquired multi-drug resistance genes in bacteria, by providing a user-friendly and cost-effective solution. Similarly, techniques like STRS and SHM-seq, which are built upon the widely used Visium spatial technology, are expected to find faster adoption by typical biology labs compared to those relying on specialized devices such as customized microfluidics systems. Compared to methods based on the popular Illumina sequencing platform, microscopy-based techniques often involve more individualized hardware configurations. Optics-free strategies such as the DNA microscopy (Weinstein et al. 2019) might serve as an appealing alternative for spatial transcriptomics and genomics once its feasibility with bacteria is demonstrated. In any case, reducing the dependence on specialized imaging hardware, standardizing, and open-sourcing data analysis pipelines will enable wider accessibility to these techniques. In addition, streamlining and integrating microscopy with sequencing-based techniques will truly allow for multi-scale investigations of bacterial heterogeneity.

Finally, with further improvement in throughput and versatility, both the size and the dimensionality (number of features) of the datasets produced will increase considerably. Consequently, data mining will play an increasingly pivotal role in maximizing the full potential of these techniques. The performance of many state-of-the-art data analysis pipelines, especially those based on deep neural networks, benefit tremendously from large datasets for model training. The sheer amount of high-quality data generated by these tools can thus serve as valuable training datasets for various computational models, which in return can help better identify hidden patterns of biological interest that are too often obscured by confounding factors present within the heterogeneous sample. Moving forward, with continual developments and synergistic integrations of different experimental and computational techniques, the mechanistic basis and functional roles of bacterial heterogeneities will become increasingly clear, opening up opportunities for manipulating the collective behavior of a bacterial population by targeting a selected few.

Author contributions

YL conducted literature search and wrote the manuscript.

Data availability

Not applicable.

Code availability

Not applicable.

Declarations

Ethics approval

Not applicable.

Consent to participate

Not applicable.

Consent to publish

Not applicable.

Competing interests

The author declare no competing interests.