Biofilms, contributing to approximately 80% of chronic and recurrent microbial infections in the human body (Costerton et al., 1999), are complex microbial ecosystems characterized by a diverse array of bacterial cells existing in various physiological states (Costerton et al., 1999; Evans et al., 2023; Stewart and Franklin, 2008). This heterogeneity within biofilms is influenced by multiple factors, including bacterial interactions (such as competition, symbiosis, and parasitism) that contribute to stable community structures (Flemming et al., 2023; Shokeen et al., 2021), environmental factors (like host environment, aquatic conditions, and nutrient concentrations), and spatial organization. Bacteria occupying different positions within the biofilm perform distinct roles (Qian et al., 2022): some mediate material exchange between cells and with the extracellular matrix, others facilitate complex communication systems between bacteria and with the host through signaling molecules, while certain bacteria participate in various energy conversion processes. This intricate division of labor not only contributes to bacterial heterogeneity but also enhances the biofilm’s overall resistance to various stresses (Momeni, 2018). The resulting cellular and functional diversity reflects the complex nature of the biofilm ecosystem, allowing it to adapt and persist in challenging environments. However, the study of biofilms faces significant limitations, primarily stemming from challenges in investigating heterogeneity within a bacterial population (Cheng et al., 2023; Spormann, 2008). Single-cell RNA-seq emerges as a promising avenue for addressing this heterogeneity (Shapiro et al., 2013; Tang et al., 2009; Blattman et al., 2020; Imdahl et al., 2020; Kuchina et al., 2021; Ma et al., 2023; Wang et al., 2023; McNulty et al., 2023; Lu et al., 2023). Expending on established protocols for cell fixation and permeabilization which facilitate in-cell barcoding while avoiding cell lysis, combinatorial barcoding-based bacterial scRNA-seq techniques, such as prokaryotic expression profiling by tagging RNA in situ and sequencing (PETRI-seq) (Blattman et al., 2020) and microbial split-pool ligation transcriptomics (microSPLiT) (Kuchina et al., 2021), have been developed. Nevertheless, these methods encounter challenges in terms of low transcript recovery rates due to overwhelmingly abundant rRNA, restricting the comprehensive analysis of within-population heterogeneity. In comparison to mammalian cells (Maynard et al., 2020), the absence of mRNA polyadenylated tails in bacteria necessitates an alternative approach for isolating mRNA (~5%) from the significantly more abundant rRNA (~95%). Here, by integrating a ribosomal RNA-derived cDNA depletion protocol (RiboD) into a PETRI-seq, we developed RiboD-PETRI-seq that efficiently eliminates rRNA reads, thereby significantly improving mRNA detection rates and enabling exploration of within-population heterogeneity.

In the RiboD protocol, we designed a set of probe primers that spans all regions of the bacterial rRNA sequence (Supplementary file 1). The core principle behind our probe design is twofold: the 3'-end of the probes is reverse complementary to the r-cDNA sequences, allowing for specific recognition of r-cDNA, while the 5'-end complements a biotin-labeled universal primer. This design enables the probes to be bound to magnetic beads, facilitating the separation of r-cDNA-probe-bead complexes from the rest of the library. Following template switching and RNaseH treatment on the barcoded cDNA from lysed cells to eliminate hybridized RNA, the library of probe primers and biotin-labeled universal primers is introduced to facilitate adequate hybridization. Pre-treated Streptavidin magnetic beads are then added to the hybridized rRNA-derived cDNA. The mRNA-derived cDNA remains in the supernatant and is collected for subsequent library construction and sequencing (Figure 1A). To assess the efficiency of single-cell capture in RiboD-PETRI, we calculated the multiplet frequency (Blattman et al., 2020) using a Poisson distribution based on our sequencing results (see details in Materials and methods). The multiplet frequency for RiboD-PETRI ranges from 1.16% to 3.35% (Supplementary file 2), indicating the technique’s capability to effectively capture transcriptomes at the single-cell level.

To assess the performance of RiboD-PETRI, we designed a comprehensive assessment of rRNA depletion efficiency under diverse physiological conditions, specifically contrasting exponential and stationary phases. This approach allows us to understand how these different growth states impact rRNA depletion efficacy. Additionally, we included a variety of bacterial species, encompassing both gram-negative and gram-positive organisms, to ensure that our findings are broadly applicable across different types of bacteria. By incorporating these variables, we aim to provide insights into the robustness and reliability of the RiboD-PETRI method in various biological contexts. The results highlight a substantial enhancement in rRNA-derived cDNA depletion, with mRNA ratio increases from 8.2% (Ctrl, the PETRI-seq we performed) to 81% (ΔΔ, RiboD-PETRI) for E. coli from exponential phase, from 10% (Ctrl) to 92% (ΔΔ) for S. aureus from stationary phase, and from 3.9% (Ctrl) to 54% (ΔΔ) for C. crescentus from exponential phase (Figure 1B; Supplementary file 3). Additionally, we compared our findings with other reported methods (Figure 1B; Supplementary file 4). The original PETRI-seq (Blattman et al., 2020) protocol, which does not include an rRNA depletion step, exhibited an mRNA detection rate of approximately 5%. The MicroSPLiT-seq (Kuchina et al., 2021) method, which utilizes poly A polymerase for mRNA enrichment, achieved a detection rate of 7%. Similarly, M3-seq (Wang et al., 2023) and BacDrop-seq (Ma et al., 2023), which employ RNaseH to digest rRNA post-DNA probe hybridization in cells, reported mRNA detection rates of 65% and 61%, respectively. MATQ-DASH (Homberger et al., 2023), which utilizes Cas9-mediated targeted rRNA depletion, yielded a detection rate of 30%. smRandom-seq utilizes a CRISPR-based rRNA depletion technique, reduced the rRNA proportion from 83% to 32%, increasing the mRNA proportion from 16% to 63% (Xu et al., 2023). BaSSSh-seq’s employs a rational probe design for efficient rRNA depletion, though specific efficiency was not reported (Korshoj and Kielian, 2024). Among these, RiboD-PETRI demonstrated superior performance in mRNA detection while requiring the least sequencing depth. With equivalent sequencing depth, RiboD-PETRI demonstrates a significantly enhanced unique molecular identifier (UMI) counts detection rate compared to PETRI-seq alone (Figure 1C). This method recovered approximately 20,175 cells (92.6% recovery rate) with ≥15 UMIs per cell with a median UMI count of 42 per cell, which was significantly higher than PETRI-seq’s recovery rate of 17.9% with a median UMI count of 20 per cell (Figure 1—figure supplement 1A and B), indicating the number of detected mRNA per cell increased prominently. Notably, this enhancement was achieved while maintaining mRNA profiles consistent with non-depleted samples (r=0.93; Figure 1D) and show a significant correlation with profiles from the traditional bulk RNA-seq method (r=0.84; Figure 1E).

We subsequently investigated the transcriptome coverage of RiboD-PETRI across different physiological states and bacterial species. For exponential phase E. coli cells, we sequenced a library with 60,000 cells, recovering approximately 30,004 cells (50% recovery), each with ≥15 UMIs (Figure 2A, Figure 1—figure supplement 1C). This analysis revealed 99.86% transcriptome-wide gene coverage across the cell population. The method achieved an average of 128.8 UMIs per single cell, with a median UMI count of 102 per cell. Further examination of high-quality cells showed varying levels of detection: the top 1000, 5000, and 10,000 cells exhibited median UMI counts of 462, 259, and 193, respectively (Figure 2B), and median gene detection of 362, 236, and 188, respectively (Figure 2C). These high-performing cells demonstrate the upper limits of the method’s capabilities. For stationary phase S. aureus cells, we sequenced a library with 30,000 cells, recovering approximately 9982 cells (33.3% recovery), each with ≥15 UMIs (Figure 2—figure supplement 1A). Analysis showed 99.96% transcriptome-wide gene coverage across the cell population. At the single-cell level, we observed an average of 153.8 UMIs and a median of 142 UMIs. Top high-quality cells exhibited the following median UMI counts: 378 (top 1000 cells), 207 (top 5000 cells), and 167 (top 8000 cells) (Figure 2—figure supplement 1B). These cells also demonstrated median gene detection of 308, 194, and 158 genes, respectively (Figure 2—figure supplement 1C). For exponential phase C. crescentus cells, we sequenced a library with 30,000 cells, recovering approximately 13,897 cells (46.3% recovery), each with ≥15 UMIs (Figure 2—figure supplement 1G). Analysis showed 99.64% transcriptome-wide gene coverage across the cell population. At the single-cell level, we observed an average of 439.7 UMIs and a median of 182 UMIs. Top high-quality cells demonstrated the following median UMI counts: 2190 (top 1000 cells), 662 (top 5000 cells), and 225 (top 10,000 cells) (Figure 2—figure supplement 1H). These cells also exhibited median gene detection of 1262, 529, and 219 genes, respectively (Figure 2—figure supplement 1I). These results underscore RiboD-PETRI’s ability to capture a wide range of transcripts across varying cell qualities and species, providing a comprehensive view of gene expression at the single-cell level.

Our results affirm RiboD-PETRI’s reliability in capturing the bacterial single-cell transcriptome, providing ample coverage and sensitivity for various species. To provide a thorough evaluation of our sequencing depth and library quality, we performed sequencing saturation analysis on our sequencing samples. The findings reveal that our sequencing saturation is greater than 90% (Figure 1—figure supplement 1D–F), indicating that our sequencing depth is sufficient to capture the diversity of most transcripts.

We further investigated its ability to consistently identify within-population heterogeneity across different bacterial species and growth conditions. In the exponential phase of E. coli, we recovered 1464 cells and identified three major subpopulations (Figure 2D), with 17 cells (1.2%) in a unique subpopulation characterized by pentose and glucuronate interconversions (Figure 2E and F) and the marker genes of cluster 2 included yfgM, glpG, scpA, elyC, ptsA, carB, actP, and pgpB (Figure 2G). For the expression levels of marker gene shown in Figure 2E, violin plots have been created to offer a more comprehensive view of the distribution across different cell populations (Figure 2—figure supplement 2). In stationary phase S. aureus cells, we recovered 9386 cells and found six major subpopulations (Figure 2—figure supplement 1D), with 437 cells (4.7%) in a distinct subpopulation named cluster 4. The marker genes of cluster 4 included KQ76-13335, KQ76-00740, and KQ76-11725 (Figure 2—figure supplement 3). In the stationary phase of C. crescentus cells, we recovered 5728 cells and identified four major subpopulations (Figure 2—figure supplement 1J), with 603 cells (10.5%) in a unique subpopulation named cluster 3. The marker genes of cluster 3 included CCNA-00259, CCNA-03402, CCNA-02361, and CCNA-03119 (Figure 2—figure supplement 4). These findings highlight RiboD-PETRI’s consistent ability to unveil within-population heterogeneity across different cell physiology and bacterial species (Figure 2H and I, Figure 2—figure supplement 1E, F, K, L), crucial for understanding bacterial population complexity. While RiboD-PETRI consistently detects potential heterogeneity, further experimental validation would be required to confirm the biological significance of the observations.

We next focused on exploring biological heterogeneity of a biofilm at the early stage of development by utilizing the static biofilm system (Merritt et al., 2011). E. coli cells were cultured in microtiter dishes overnight, adhered cells were fixed for RiboD-PETRI processing in duplicate experiments. For these two replicates, we sequenced libraries containing 20,000 and 40,000 cells, recovering 5244 and 11,344 cells, which corresponded to recovery rates of 26% and 28%, respectively. The correlations between detected reads and UMIs were found to be 0.87 and 0.90 for the two replicates, respectively (Figure 3—figure supplement 1A). The median UMI counts for the recovered cells were 34 and 52 (Figure 3A). After screening, the final datasets comprised 1621 and 3999 cells for the two replicates, respectively. While no significant batch effects were observed, we applied batch correction as a precautionary measure (Figure 3—figure supplement 1B–D). In replicate 1, each cell was sequenced with an average of 1563 reads, while in replicate 2, the average was 2034 reads (Supplementary file 5), yielding median UMI counts of 283.5 and 239 per cell, respectively (Figure 3B). For gene detection, the median counts were 219 and 193 genes per cell for the respective replicates (Figure 3C). Additionally, UMAP visualization was employed to illustrate the distribution of cellular UMI numbers, revealing heterogeneity among populations that was independent of UMI counts (Figure 3—figure supplement 1E and F). Unsupervised clustering analysis identified four major subpopulations in each replicate, with a consistently identified rare subpopulation (2.6%/2.1%) as cluster 2, driven by cell envelope genes (Figure 3D–F). Marker genes for this cluster included yffO, lptE, rdgB, pdeI, sstT, fixA, yjjG, rlmI, accC, and yaiA (Figure 3G, H and Figure 3—figure supplement 2).

PdeI, identified among marker genes, was predicted as a phosphodiesterase enzyme hydrolyzing c-di-GMP, a vital bacterial second messenger (Yu et al., 2023; Li et al., 2023; Figure 4A and B). However, our comprehensive structural analysis revealed a more complex and novel role for PdeI. While PdeI contains an intact EAL domain typically associated with c-di-GMP degradation, it also possesses a divergent GGDEF domain, generally linked to c-di-GMP synthesis (Figure 4—figure supplement 1). This dual-domain architecture suggested potential complex regulatory roles. To validate PdeI’s function, we created a PdeI-BFP fusion construct under the native pdeI promoter, integrated with a ratiometric c-di-GMP sensing system (Vrabioiu and Berg, 2022) in E. coli. Confocal microscopy revealed PdeI as a membrane protein (Figure 4C). Single-cell level monitoring showed cell-to-cell variability in c-di-GMP levels and PdeI expression, with a positive correlation observed (Figure 4D), indicating PdeI upregulated c-di-GMP synthesis rather than degradation. This finding was confirmed by high-pressure liquid chromatography-tandem mass spectrometry (HPLC-MS/MS), which showed an approximately 11-fold increase in c-di-GMP concentration in the PdeI overexpression strain compared to the control strain (Figure 4E). These results align with previous studies showing that a point mutation (G412S) in PdeI’s divergent GGDEF domain in a strain lacking PdeH, the major phosphodiesterase in E. coli, resulted in decreased c-di-GMP levels (Reinders et al., 2016). Our additional experiments with a PdeI(G412S)-BFP mutation strain showed constant c-di-GMP levels despite increasing BFP fluorescence, serving as a proxy for PdeI(G412S) expression levels (Figure 4D). These results, combined with the presence of a CHASE (cyclases/histidine kinase-associated sensory) domain in PdeI, strongly suggest that PdeI functions as a membrane-associated sensor that integrates environmental signals with c-di-GMP production under complex regulatory mechanisms. This discovery challenges the initial prediction of PdeI as solely a phosphodiesterase and highlights its novel role as a c-di-GMP synthetase, contributing significantly to our understanding of bacterial signaling pathways. It’s worth noting that while the other marker genes in this cluster are co-expressed, our analysis indicates that they do not have a significant impact on biofilm formation or a direct relationship with c-di-GMP or PdeI.

Confocal laser scanning microscopy provided further insights into the spatial distribution of PdeI-positive cells within the biofilm structure. In the PdeI-BFP fusion strain, PdeI-BFP-positive cells, characterized by elevated c-di-GMP levels, were predominantly located at the bottom of the static biofilm (Figure 4F). This localization corresponds to the region of cell-surface attachment, aligning with our hypothesis that PdeI functions as a membrane-associated sensor integrating environmental signals with c-di-GMP production through complex regulatory mechanisms (Lacanna et al., 2016). In contrast, in the control strain where BFP was expressed alone under arabinose-induced promoter, BFP-positive cells were observed to be distributed throughout the entire biofilm community (Figure 4G). This distinct spatial distribution pattern between the PdeI-BFP fusion and the BFP-only control strains provides compelling evidence for PdeI’s specific role in biofilm formation, particularly at the biofilm-surface interface. The concentration of PdeI-positive cells at the bottom of the biofilm suggests that PdeI may be especially crucial in the initial stages of biofilm formation, potentially responding to surface-associated cues to modulate c-di-GMP levels and promote attachment. The uniform distribution of BFP in the control strains suggests that the localization of PdeI-BFP is not affected by BFP labeling. These observations further underscore the complex and nuanced role of PdeI in bacterial signaling and biofilm development, highlighting the importance of considering cell-to-cell heterogeneity in understanding the function of regulatory proteins in microbial communities.

To investigate the association of the PdeI-high cluster with bacterial drug tolerance in the early stages of biofilm development, we isolated PdeI-high cells using flow cytometry (Figure 4H, I) and subjected them to an ampicillin antibiotic killing assay to determine their persister frequency. Our results revealed that the PdeI-high population produced a significantly higher ratio of persister cells (~7.3%) compared to the whole biofilm population (~0.6%). Notably, cells expressing high levels of BFP alone or PdeI(G412S)-BFP showed no increase in persister ratios (Figure 4J). This finding suggests that the increased persistence is specifically linked to PdeI activity. Time-lapse imaging during the antibiotic killing process consistently demonstrated that persisters primarily originated from PdeI-GFP-positive cells (Figure 4K and Video 1). These PdeI-GFP-positive cells, displaying characteristics of dormancy, survived ampicillin treatment for 6 hr without visible growth or division. Upon antibiotic removal and replacement with fresh growth medium, the PdeI-GFP-positive persister cells resumed activity, elongating, dividing, and forming new microcolonies (Figure 4K and Video 1). This dynamic behavior provides direct visual evidence of the persister phenotype associated with PdeI-high cells. These findings strongly suggest that c-di-GMP, a molecule whose intracellular levels are upregulated by PdeI, plays a significant role in generating a persister subpopulation during the early stages of biofilm development. The mechanism by which elevated c-di-GMP levels contribute to antibiotic tolerance may involve modulation of cellular metabolism or activation of stress response pathways, leading to a state of dormancy that enables survival under antibiotic stress. This discovery not only enhances our understanding of the link between biofilm formation and antibiotic tolerance but also identifies PdeI as a potential target for strategies aimed at combating persistent bacterial infections.

In this study, we introduce RiboD-PETRI, an enhanced bacterial scRNA-seq method that offers a cost-effective (Supplementary file 6), equipment-free, and high-throughput solution. By incorporating a probe hybridization-based rRNA-derived cDNA depletion protocol, our approach efficiently removes rRNA reads and significantly improves mRNA detection rates, enabling a more comprehensive exploration of within-population heterogeneity. At $0.0049 per cell, RiboD-PETRI is substantially more economical than the original PETRI-seq ($0.056 per cell), making it an attractive option for budget-conscious researchers. The method demonstrates improved mRNA detection, versatility across various bacterial species and growth conditions, preservation of transcriptome profiles consistent with non-depleted samples (r=0.93) and traditional bulk RNA-seq methods (r=0.84), high transcriptome coverage (>99%), and robust single-cell resolution with median UMI counts ranging from 102 to 182 per cell across different species and conditions.

The application of RiboD-PETRI to investigate biofilm heterogeneity exemplifies its potential for exploring complex biological systems. Our analysis of early-stage biofilm development uncovered a rare subpopulation (2.1–2.6%) characterized by cell envelope genes, including the previously uncharacterized gene pdeI. Further investigation revealed PdeI’s novel role as a c-di-GMP synthetase rather than a phosphodiesterase, challenging initial predictions and contributing significantly to our understanding of bacterial signaling pathways. Moreover, we demonstrated that PdeI-high cells exhibit increased antibiotic tolerance, with a significantly higher proportion of persister cells compared to the general biofilm population. This finding establishes a link between elevated c-di-GMP levels, regulated by PdeI, and the generation of antibiotic-tolerant subpopulations during early biofilm development. The spatial distribution of PdeI-positive cells at the bottom of the static biofilm, corresponding to the cell-surface attachment region, supports our hypothesis that PdeI functions as a membrane-associated sensor integrating environmental signals with c-di-GMP production. This localization pattern suggests PdeI’s crucial role in the initial stages of biofilm formation, potentially responding to surface-associated cues to modulate c-di-GMP levels and promote attachment. While other marker genes in this cluster are co-expressed, our analysis indicates they do not significantly impact biofilm formation or directly relate to c-di-GMP or PdeI.

In conclusion, RiboD-PETRI represents a significant advancement in bacterial scRNA-seq methodology. Its ability to uncover hidden variations within bacterial populations, as demonstrated in our biofilm analysis, underscores its potential impact on advancing our understanding of microbial behavior and population dynamics. By providing a cost-effective and efficient tool for exploring bacterial heterogeneity, RiboD-PETRI opens new avenues for research in microbiology, potentially leading to novel insights into antibiotic resistance, biofilm formation, and other critical areas of bacterial biology.