Deploying smFISH for accurate quantitation of cell-scale gene expression in bacterial biofilms

Understanding how individual bacteria residing in biofilm communities adopt location-specific cell fates requires measurements of spatiotemporal gene expression at cell-scale resolution. Traditional fluorescent reporters are inadequate for this task because their output is dampened by the low oxygen levels that exist in biofilms, which prohibits proper reporter maturation [36]. Indeed, in V. cholerae biofilms, signal from multiple fluorophores rapidly declines over time, well before biofilms reach maturity (S1A Fig), making fluorescent reporters poor proxies of gene expression in this context. One solution to this problem employed recently is to grow biofilms under flow, and thus supply a constant source of oxygen [37,38]. While this strategy eliminates fluorescent signal decay, it prevents the natural accumulation of small signal molecules, for example QS autoinducers, that are required to properly regulate the biofilm lifecycle.

Compounding issues with traditional fluorescent reporters, biases inherent to confocal imaging preclude accurate quantitation of spatial fluorescence patterns in biofilms [35]. z-positional bias exists because captured fluorescence signal declines as a function of distance from the objective. Radial (r)-positional bias is likely caused by the biofilm’s dome shape. Specifically, the center of the biofilm is thicker and contains more layers of cells than does the edge, so background fluorescence from out-of-focus z planes is higher at the biofilm center than at the periphery. Indeed, thinner biofilms with fewer z planes display less dramatic r-positional bias. Consistent with these imaging biases, we show that signal from DAPI stained biofilms, which is expected to be spatially uniform, decreases from the bottom to the top and from the inside to the outside of V. cholerae biofilms (S1B Fig).

With the aim of overcoming the above limitations and developing a robust tool for measuring cell-scale gene expression in bacterial biofilms, we employed smFISH technology. smFISH has been successfully used in planktonic bacteria to accurately measure single-cell gene expression at ranges from <1 to 100 mRNA molecules per cell [39,40]. smFISH does not require an oxygen-dependent maturation step, making it suitable for use in bacterial biofilms. Indeed, sequential smFISH (seqFISH) was recently used to measure the expression of over 100 genes in individual Pseudomonas aeruginosa biofilm cells [41]. Below, we combine smFISH with confocal imaging and a biofilm-specific downstream analysis pipeline to accurately quantify cell-scale gene expression over space and time in V. cholerae biofilms. Our approach provides a powerful tool for such analyses, including for transcripts that display only low-level expression and for proteins that are recalcitrant to tagging.

Our first goal was to assess the imaging artifacts associated with our smFISH strategy independent of any biological gene-expression patterns. To do this, we measured the smFISH fluorescence signal from a gene expressed uniformly across the biofilm. We grew V. cholerae biofilms containing arabinose-inducible m-neon green (mNG) integrated into the chromosome. Previous reports demonstrated uniform cell growth across the biofilm under our conditions [35], and thus, mNG expression levels should track with the exogenously supplied arabinose inducer concentration. To ensure equivalent arabinose penetration throughout the biofilm, it was added at the start of the experiment, prior to biofilm formation. Importantly, while V. cholerae cells are able to import arabinose, they are not known to metabolize it, and thus arabinose concentrations should remain spatially uniform and constant over the course of growth [42]. We grew biofilms in the presence of high and low concentrations of arabinose and measured mNG transcripts with smFISH probes labeled with each of the three fluorescent dyes used throughout the remainder of this work. The fluorescence signal from the mNG protein does not survive the fixation and smFISH hybridization process, and thus does not add to nor interfere with quantitation of smFISH fluorescence signal. This strategy allowed us to ascertain fluorescence signal biases across entire biofilms for each fluorescent probe, and over a range of transcript levels. mNG expression levels assessed by each fluorophore in biofilms paralleled those in planktonic cells grown with the corresponding inducer concentration (S2A Fig). The high and low mNG expression values are comparable to the highest and lowest expression levels we observed for genes targeted in this work.

We quantified smFISH fluorescence signal at cell-scale resolution (2.32 µm) by dividing biofilms into cell-sized cubes [43] (here forward, referred to interchangeably as “cell cubes” or “cells” for simplicity) with 2.32 µm sides and calculating the volume-normalized smFISH signal in each cell cube (see S2B, S2C Fig, Methods: smFISH image analysis). We refer to resulting values henceforth as ‘per-cell-volume fluorescence signal’. Of note, per-cell-volume fluorescence signal is reported here and throughout this work in arbitrary units (AUs). These values can be converted to transcript numbers by identifying the fluorescence distributions of individual puncta reporting on transcripts that are produced at low levels, where the signal from most puncta is predicted to arise from an individual mRNA molecule (S2D Fig and [39]). In our experiments, individual mRNA puncta intensities cluster around 1 and differ by no more than 2-fold across fluorophores and genes. Thus, 1 AU corresponds to approximately 0.5–2 mRNA transcripts.

To quantify spatial smFISH fluorescence signal patterns, we grouped cells across replicate biofilms by their r and z positions. We calculated the average per-cell-volume fluorescence signal for each group of cells relative to the average per-cell-volume fluorescence signal of a group of cells located near the center of the biofilm (S2E Fig, Methods: Analysis of spatial gene-expression patterns at cell-scale resolution). The r and z positions of each group, and the corresponding average per-cell-volume fluorescence signals relative to that in the reference group of cells with position r, z = 3.5, 3.5 µm, can be visualized as a heatmap, as in Fig 2A. Relative expression is used to highlight spatial differences and aid in comparison of spatial patterns across genes and conditions where expression levels differ. To compare both expression patterns and absolute expression simultaneously, we also binned cells by their distance from the center position r, z = 0, 3.5 µm and calculated the average per-cell-volume fluorescence signal as a function of this distance, as in S3A Fig (Methods: Analysis of spatial gene-expression patterns at cell-scale resolution).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Quantitation of cell-scale gene expression in V. cholerae biofilms.

(A) Heatmaps showing relative, non-corrected spatial mNG per-cell-volume smFISH fluorescence signal data as a function of position r, z in biofilms of V. cholerae carrying pBad-mNG. Biofilms were grown in the presence of 0.2% or 0.0375% arabinose, denoted High and Low Expression, respectively. In all cases,100 µM Nspd was included. mNG expression was measured by smFISH using probes labeled with one of three fluorophores, as indicated. Cells are grouped into bins with size Δr, Δz = 2.32, 2.32 µm. Gene expression in each bin is represented as a relative value, where the average per-cell-volume fluorescence signal across all cells in a bin is normalized to the average per-cell-volume fluorescence signal across all cells in the bin denoted with the asterisk (bin r, z = 3.5, 3.5 µm). White boxes represent zero values. (B) Heatmaps as in (A) showing the relative, corrected per-cell-volume fluorescence signal values. All data represent n = 10–12 biofilms. The data underlying this figure can be found in S1 Table.

https://doi.org/10.1371/journal.pbio.3003187.g002

Absent fluorescence signal biases, and given uniform expression of mNG, one would expect unvarying fluorescence signal across all cells in the biofilm. This is not the case. Rather, spatially-dependent fluorescence signal differences are observed, in which average per-cell-volume fluorescence signal is highest in cells closest to the center of the biofilm, and lowest in cells most distant from the center (Figs 2A, S3A). The pattern is consistent with the known z and r positional biases of confocal imaging described above. Crucially, the fluorescence signal pattern is the opposite of what would be predicted if the arabinose inducer and/or mNG probes could not penetrate the biofilms. In both of those cases, one would expect lower per-cell-volume fluorescence signal in the center of the biofilm and higher per-cell-volume fluorescence signal at the biofilm periphery. While we cannot eliminate the possibility that the obtained fluorescence pattern arises from higher arabinose concentrations in the biofilm center than at the periphery, arabinose accumulation at the biofilm core is unlikely given that V. cholerae does not metabolize arabinose. We know of no other a priori reason for why arabinose concentration would differ across the biofilm. Thus, the spatial fluorescence signal pattern observed here most likely stems from constitutive gene expression and arises entirely from the expected confocal imaging artifacts.

To account for these artifacts, we developed a mathematical model to quantify the above fluorescence signal positional biases for each fluorophore and across different gene-expression levels. We can use this method to correct the mNG expression data to achieve uniformly measured average mNG per-cell-volume fluorescence signal throughout biofilms, consistent with our expectation for a constitutively expressed gene (Figs 2B, S3A). The average per-cell-volume fluorescence signal appears less uniform following correction in biofilms that display low mNG expression as a consequence of higher expression variation between cells. The average per-cell-volume fluorescence signal is most noisy in bins far from the surface, the region with the least fluorescence signal. However, 97% of bins across biofilms with low mNG expression have a z-score between −0.25 and 0.25, i.e., they deviate from the average per-cell-volume fluorescence signal of the center bin by no more than one quarter of a standard deviation, demonstrating that there are minimal fluorescence signal output differences between groups of cells. The model is not only able to correct spatial fluorescence signal biases in the large biofilms used to generate the model, but can also be directly applied without modification to accurately correct biases from smaller biofilms grown with identical concentrations of arabinose inducer (S3B, S3C Fig).

To assess the applicability of our model, we measured the expression of two constitutively expressed housekeeping genes, gyrA, encoding DNA gyrase subunit A, and polA, encoding DNA polymerase I, using smFISH. We observe consistent expression of both genes over the course of biofilm development (S4A Fig) at levels comparable to those previously reported (the ratios of gyrA:polA expression measured by smFISH here and measured in planktonic V. cholerae cells by RNA-seq in [44] are 3.5 and 2.8, respectively). Following correction using our model, we measure uniform expression of gyrA and polA, with expression varying less than 2-fold across all biofilm positions (S4B–S4E Fig). These data highlight the power of our model to correct for imaging biases irrespective of the specific transcript being measured, and they demonstrate how the correction can be used to obtain accurate cell-scale spatial gene-expression patterns in biofilms for, in principle, any gene of interest. These data also support our assumption that confocal imaging biases, and not regional differences in arabinose concentration, underlie the spatial pattern used to generate our model. If non-uniform arabinose-driven mNG expression occurred in biofilms, our model would not accurately correct the spatial expression profiles of other uniformly expressed genes.

Full details of how we developed the model and how it is applied are provided in Methods. All main text figures show data that has been corrected using our model. All companion plots generated from the non-corrected data are provided in S1 Data, aside from select examples that are included in supplementary figures as noted. Descriptions of per-cell-volume fluorescence signal and expression levels henceforth are also based on corrected data, unless otherwise noted. Using our model to correct for spatial fluorescence signal imaging biases overcomes a major limitation of confocal imaging and provides us with a V. cholerae biofilm-specific analysis pipeline that delivers accurate spatial and temporal gene-expression patterns at cell-scale resolution.

smFISH accurately measures QS gene expression in biofilms

We validated that our smFISH technique could accurately quantify cell-scale gene expression of genes controlling the biofilm lifecycle in V. cholerae biofilms across space and time. Here, we focus on the genes encoding the QS LCD and HCD master regulators, qrr4 and hapR, respectively, and the downstream QS-controlled target gene vpsL, the first gene in the vpsII operon that encodes VPS biosynthetic enzymes. HapR represses vpsL at HCD (Fig 1). Regarding the strategy of using the qrr4 gene as the readout for the QS LCD state, the aphA gene encoding the LCD master transcription factor would have been the obvious choice (Fig 1). However, AphA is a small protein encoded by a short transcript, and despite multiple attempts, we were unable to design smFISH probes against the aphA transcript that yielded measurable fluorescence signal. Attempts to construct AphA fusions to increase transcript length destabilized the mRNAs and they could not be detected by smFISH nor qRT-PCR. The sRNA Qrr4 is also, by definition, a small transcript. Thus, to monitor the QS LCD state, we used smFISH probes against mNG encoded under the pqrr4 promoter that has previously been demonstrated to accurately report qrr4 levels [45,46].

To validate our approach, we compared the fluorescence signal detected by smFISH in biofilms to that from an established technology, smFISH probing planktonic cultures. We know that QS gene expression in V. cholerae biofilms can be dramatically different from that in planktonic cells [31,47]. Thus, for this series of experiments, we exploited a V. cholerae strain in which we could control its QS status. Specifically, the V. cholerae strain contains only a single QS receptor, LuxPQ, which detects the autoinducer AI-2. The strain also contains a deletion of the gene encoding the AI-2 synthase, luxS. Therefore, the level of expression of QS-regulated genes is set exclusively by exogenously supplied AI-2, the amount of which we can precisely control. From here forward, we call this strain ‘the AI-2 sensor strain’. We grew planktonic and biofilm cultures of the AI-2 sensor strain without AI-2 or with a saturating AI-2 concentration (10 µM) [48]. We supplied the AI-2 at the start of the experiment to ensure a constant concentration in the shaking culture and across the biofilm at all developmental stages. This approach enables comparison of gene expression between biofilm and planktonic cells because it forces the strain into the LCD and HCD states, respectively, and in turn drives maximal/minimal qrr4 and minimal/maximal hapR expression. Moreover, because QS gene expression at each AI-2 concentration is fixed and unvarying in the AI-2 sensor strain, we could assess the accuracy of our measurements over time and space in biofilms. Specifically, we confirmed temporal accuracy by measuring expression in biofilms grown without or with AI-2 over a range of biofilm developmental stages, and we confirmed spatial accuracy by measuring the pattern of gene expression across entire biofilms. Important for our strategy, and mentioned in the Introduction, is that V. cholerae forms biofilms at LCD and disperses from biofilms at HCD. Thus, in all cases, the AI-2 sensor strain was grown in the presence of the polyamine norspermidine (denoted Nspd) that prevents biofilm dispersal at HCD [49].

First, regarding the QS regulators qrr4 and hapR, there is good agreement in the AI-2 sensor strain between expression measured in biofilms and planktonic cells by smFISH (Fig 3A). Specifically, high qrr4 expression and low hapR expression occur in the −AI-2 (LCD) samples, and low qrr4 and high hapR expression occur in +AI-2 (HCD) samples (Fig 3A–3C). Importantly, the average per-cell-volume expression of qrr4 and hapR measured both relative to each other and across the LCD and HCD conditions are consistent between planktonic and biofilm cells (Fig 3A, slope of best fit line on a log–log scale for qrr4 and hapR points is 0.88), confirming the accuracy of our smFISH strategy for measuring gene expression in biofilms. Moreover, for any given condition and gene studied, the global average per-cell-volume fluorescence signal differed by no more than 2-fold over time and across a range of biofilm sizes (Fig 3B). Following application of our spatial correction model to these data (S5A–S5D Fig), we find consistent and appropriate high or low expression of each gene under each condition across the biofilm (Figs 3C–3E, S5E, S6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. smFISH accurately quantifies QS gene expression at cell-scale resolution in V. cholerae in biofilms.

(A) Left panel: Average per-cell-volume qrr4 (mNG), hapR, and vpsL smFISH fluorescence signal in V. cholerae planktonic and biofilm cells. Values for planktonic samples represent the means of n = 3 biological replicates. Values for biofilm samples represent the means across all biofilm sizes, as shown in (B). The slope of the best fit line as shown is 0.80. The slope of the best fit line excluding vpsL is 0.88. Right panel: values for biofilm samples calculated as the means across n = 9−10 replicate biofilms with the smallest biovolumes shown in (B). The slope of the best fit line is 0.99. (B) Average per-cell-volume qrr4 (mNG), hapR, and vpsL smFISH fluorescence signal across replicate V. cholerae biofilms grown without (left) or with (right) 10 µM AI-2 as a function of the average biofilm biovolume. (C) First and third rows: Representative confocal microscopy images showing DAPI, and qrr4 (mNG), hapR, and vpsL smFISH fluorescence signals in biofilms with the largest biovolumes shown in (B) grown without (first row) or with (third row) 10 µM AI-2. Images represent maximum projections of the first four in-focus z slices. Scale bars represent 5 µm. Second and fourth rows: Heatmaps showing average qrr4 (mNG), hapR, and vpsL per-cell-volume signal across n = 10 replicate biofilms as a function of x, y position (with replicate biofilms aligned around their center positions) for biofilm cells with z positions between 0 and 4.68 µm in biofilms with the largest biovolumes shown in (B) grown without (second row) or with (fourth row) 10 µM AI-2. Scale bars represent 5 µm. (D) Relative average per-cell-volume fluorescence signal as a function of distance from the periphery of the biofilm for the data shown in (C). Values are relative to the average per-cell-volume fluorescence signal of cells located at a distance of 2 µm from the biofilm periphery for each condition and gene. (E) Heatmaps of relative qrr4 (mNG), hapR, and vpsL average per-cell-volume fluorescence signal relative to the average per-cell-volume signal in the bin denoted with an asterisk are shown as in Fig 2B for V. cholerae biofilms grown without (left) and with 10 µM AI-2 (right), for n = 10 replicate biofilms with the largest biovolumes shown in (B). Error bars denote standard deviations, which are in some cases smaller than the sizes of the symbols used in the plots. Best fit lines were calculated by fitting log₂ transformed data. All biofilms were grown with 100 µM Nspd. All data have been corrected using the model described in the text (See Methods: Spatial correction model). The data underlying this figure can be found in S1 Table.

https://doi.org/10.1371/journal.pbio.3003187.g003

The data for expression of vpsL in the AI-2 sensor strain differs markedly from that of the two upstream QS regulators in V. cholerae biofilms. vpsL exhibits changes in both its temporal and spatial expression patterns. We come back to the molecular mechanism underlying these results below, but briefly, the patterns occur because vpsL is regulated by both QS and by the small molecule second messenger cyclic diguanylate (c-di-GMP) [30], levels of which vary in space and time in biofilms of the AI-2 sensor strain. A few key points pertinent to the validation of our approach are relevant. First, to confirm the accuracy of measurements of vpsL expression, we compared vpsL expression levels in planktonic cells to those in the smallest biofilms in our analyses. In nascent biofilms, cells have only recently attached to the surface and begun the transition from the planktonic to the biofilm lifestyle. Thus, the biological state of V. cholerae cells in immature biofilms should closely resemble that of planktonic cells. Indeed, there is good agreement between vpsL expression levels in planktonic and early-stage biofilm cells under both LCD and HCD conditions (Fig 3A, right panel). Additionally, there is high correlation between expression levels for all three measured genes in planktonic and early-stage biofilm cells (Fig 3A, right panel, slope of best fit line on a log–log scale is 0.99). Thus, our approach also allows comparisons of expression levels across genes probed with different fluorophores. Second, unlike qrr4 and hapR, at LCD (−AI-2) vpsL is expressed over 2-fold more highly at the biofilm periphery than in the biofilm core, as quantified in all analyses of spatial expression (Figs 3C–3E, S5E). This pattern cannot be an artifact of our spatial correction, as the pattern can be observed in raw images (Fig 3C) and is present in the non-corrected data (S5A–S5D Fig). The pattern is also independent of the cube side length used to segment the biofilms (S7 Fig). Again, we explain the biology underlying formation of this pattern below. Taken together, our measurements of hapR, qrr4, and vpsL expression demonstrate that our smFISH strategy for probing cell-scale gene expression in V. cholerae biofilms can accurately quantify gene expression, both temporally and spatially.

V. cholerae cells transition from the LCD to the HCD QS state as biofilms mature

Equipped with our smFISH strategy, we measured the spatiotemporal gene-expression patterns of key biofilm regulators and structural components in wildtype (WT) V. cholerae that undergoes the full biofilm lifecycle, from attachment to dispersal. We focused first on QS signaling and used qrr4 and hapR as readouts of the LCD and HCD QS states of cells, respectively. To capture different biofilm developmental stages, we grew biofilms for different lengths of time, ranging from 6 to 10 h, before fixing the samples and conducting smFISH. Consistent with previous data [32], as biofilms mature and biovolumes increase, expression of qrr4 decreases and expression of hapR increases (Fig 4A), reflecting the transition from the QS LCD- to the HCD-state. Indeed, in the largest biofilms assayed, hapR levels are comparable to those measured in biofilms of the AI-2 sensor strain grown with saturating AI-2 (Fig 3A, 3B, +AI-2). We interpret these results to mean that over the course of WT V. cholerae biofilm development, endogenously produced autoinducer concentrations accumulate to the level required to fully engage the QS system.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. QS signaling over V. cholerae biofilm development.

(A) Average per-cell-volume qrr4 (mNG) and hapR fluorescence signal across n = 10 replicate WT V. cholerae biofilms as a function of the average biofilm biovolume. (B) Average per-cell-volume qrr4 (mNG) and hapR fluorescence signal across n = 9–10 replicate V. cholerae biofilms formed by the LCD-locked luxO D61E strain as a function of the average biofilm biovolume. (C) Representative images showing DAPI, and qrr4 (mNG), and hapR smFISH fluorescence signals and heatmaps of per-cell-volume fluorescence signal as a function of x, y position as in Fig 3C for n = 9–10 replicate biofilms formed from WT (left) and LCD-locked (right) V. cholerae with the largest biovolumes shown in (A) and (B), respectively. Scale bars represent 5 µm. (D) Relative average per-cell-volume fluorescence signal as a function of distance from the periphery of the biofilm as in Fig 3D, for the data shown in (C). (E) Heatmaps of qrr4 (mNG) and hapR average per-cell-volume fluorescence signal relative to the average per-cell-volume fluorescence signal in the bins denoted with asterisks are shown as in Fig 2B for n = 9–10 replicate WT (left) and LCD-locked (right) V. cholerae biofilms with the largest biovolumes shown in (A) and (B), respectively. We note that while z-dependent differences in hapR signal are observed here, given that the hapR signal decreases by less than 2-fold from the surface to z locations far from the surface and no such pattern exists according to other spatial analyses (panels C, D, F), the differences likely stem from experiment-to-experiment variations in imaging biases that are not fully corrected by our spatial model. (F) Average per-cell-volume fluorescence signal versus distance to the biofilm center, as in S3A Fig, for the data shown in (E). Error bars denote standard deviations, which are in some cases smaller than the sizes of the symbols used in the plots. All data have been corrected using the model described in the text (See Methods: Spatial correction model). The data underlying this figure can be found in S1 Table.

https://doi.org/10.1371/journal.pbio.3003187.g004

To verify the above results, we performed control analyses in which we measured qrr4 and hapR levels over time in biofilms of V. cholerae carrying the constitutively active LuxO~P phosphomimetic allele, luxO D61E. LuxO D61E locks cells into the LCD QS state [26]. We refer to such samples as ‘LCD-locked’. As expected, given that the LCD-locked strain cannot undergo the QS transition to HCD, expression of qrr4 and hapR remained consistent across biofilm development (Fig 4B). Furthermore, qrr4 expression was higher and hapR expression was lower in biofilms formed by the LCD-locked strain than in WT biofilms. These results confirm that the qrr4 and hapR expression changes that occur in WT V. cholerae biofilms are a consequence of QS. We note that hapR levels in the biofilms formed by the LCD-locked strain are identical to those measured in biofilms of the AI-2 sensor strain grown in the absence of AI-2 (i.e., when the AI-2 sensor strain is in the LCD QS state), however, qrr4 levels are lower in biofilms formed by the LCD-locked strain than in biofilms formed by AI-2 sensor strain under the no AI-2 condition (Fig 3A, 3B, −AI-2). We do not know the mechanism underlying this discrepancy. There exist multiple feedback loops in the QS pathway [21,45,46]. We speculate that feedback onto qrr expression occurring in one of these two mutant strains but not in the other could be responsible for the difference.

QS autoinducers are spread uniformly across mature V. cholerae biofilms

A question that has long perplexed the biofilm field is whether QS autoinducers are present at uniform concentrations across a biofilm, or alternatively, if there exist regions of high and low autoinducer concentrations that could drive regional differences in gene-expression patterns. It is not yet possible to directly measure autoinducer concentrations in biofilms. However, the spatial accuracy that our smFISH method delivers in quantitation of gene expression in biofilms allows us to probe this longstanding question, at least in the case of V. cholerae biofilms. Specifically, by comparing expression levels of QS-controlled genes throughout biofilms, we can infer whether autoinducers are uniformly present or vary regionally. As expected, in biofilms containing the LCD-locked strain, spatial expression of both qrr4 and hapR differs by less than 2-fold across the biofilm (Fig 4C–4F). Similar constant, location-independent expression patterns of qrr4 and hapR occur in WT biofilms, suggesting that autoinducers can freely diffuse to achieve uniform and/or saturating concentrations across the community of cells (Fig 4C–4F and see text in the legend to Fig 4E regarding the hapR signal).

Although qrr4 and hapR do not display regional expression heterogeneity in V. cholerae biofilms, there is nonetheless location-independent expression variability across individual cells (S8A Fig). To investigate the source underlying this heterogeneity, we measured the relationship between qrr4 and hapR expression across cells (see Methods). Our logic is that, if location-independent non-uniformities in autoinducer concentration exist across the biofilm, the consequence would be differences in the QS states of individual cells. In that case, we would expect qrr4 and hapR expression to be negatively correlated in WT biofilms, but not in biofilms of the LCD-locked strain that is impervious to autoinducers. We do not detect any correlation between qrr4 and hapR expression in either strain, a finding consistent with intrinsic noise driving all observed heterogeneity (S8B, S8C Fig) [50]. We also measure negative correlations between expression variation (measured by the coefficient of variation, CV) and expression levels for both qrr4 and hapR across samples, another feature of intrinsic noise in expression (S8D Fig) [50]. This correlation is consistent across biofilms formed from the WT, LCD-locked, and AI-2 sensor strains, and between biofilm and planktonic samples, further suggesting that autoinducer fluctuations, which would be distributed and detected differently among these conditions, do not drive gene-expression heterogeneity.

vpsL is preferentially expressed in cells located at the periphery of mature V. cholerae biofilms

Two important changes occur as V. cholerae cells transition from the planktonic lifestyle to the biofilm mode: (1) alteration in expression of hundreds of genes that enable biofilm formation and that adapt cells to the new lifestyle, and (2) organization of cells into a stereotypical yet non-uniform architecture. We hypothesized that discrete spatiotemporal patterns of expression of biofilm structural genes could arise, providing a link between gene-expression changes and specific biofilm architectural transitions. To explore this possibility, we focused first on vpsL, which, as mentioned, encodes a biosynthetic enzyme required for extracellular matrix polysaccharide production. vpsL is regulated by both QS and c-di-GMP signaling (Fig 1), allowing us to probe the contributions of both pathways to spatiotemporal gene-expression patterning. We probed biofilms formed by WT and LCD-locked V. cholerae to, respectively, quantify the combined contributions of QS and c-di-GMP signaling (WT) and the individual contribution of c-di-GMP signaling (LCD-locked) to establishment of the spatiotemporal vpsL gene-expression pattern. While QS and c-di-GMP signaling are the only known regulators of vpsL expression, we note that additional regulators may exist that could contribute to the regulation we observe in biofilms formed from both WT and LCD-locked cells.

Overall vpsL levels decrease more than 30-fold during maturation of WT biofilms (Fig 5A). We expected this temporal pattern of expression because HapR is a repressor of vpsL, and hapR is more highly expressed in mature HCD WT V. cholerae biofilms than in mature biofilms of the LCD-locked strain (Fig 4A, 4B). Temporal reductions in vpsL also occur in biofilms formed from the LCD-locked strain, although to a lesser extent than in WT biofilms. This temporal pattern is also observed in the AI-2 sensor strain (Fig 3B). Both the LCD-locked strain and the AI-2 sensor strain are QS inert, thus, residual changes in vpsL expression that occur in such biofilms cannot be a consequence of QS, but rather, likely stem from changes in c-di-GMP signaling. We conclude that contributions from both QS and c-di-GMP signaling drive the temporal patterning of vpsL expression in WT biofilms. The reduction in vpsL expression that occurs as biofilms mature is likely important for terminating VPS production to allow cells to disperse from mature biofilms.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. vpsL expression over V. cholerae biofilm development.

(A) Average per-cell-volume vpsL fluorescence signal across n = 10 replicate WT biofilms or biofilms of the LCD-locked strain as a function of the average biofilm biovolume. Error bars represent standard deviations, which are in some cases are smaller than the sizes of the symbols used in the plots. (B) Representative images showing DAPI and vpsL smFISH fluorescence signals and heatmaps of per-cell-volume fluorescence signal as a function of x, y position as in Fig 3C for n = 10 replicate biofilms formed from LCD-locked V. cholerae with the two largest biovolumes shown in (A). Scale bars represent 5 µm. (C) Relative average per-cell-volume fluorescence signal as a function of distance from the periphery of the biofilm as in Fig 3D, for the data shown in (B). Light symbol coloring represents the smaller biofilms (left in (B)) and dark symbol coloring represents the larger biofilms (right in (B)). For smaller biofilms, when cells are grouped by distance from the periphery, all but 3 bins contain fewer cells than our sample size cutoff and are therefore excluded from analysis (see Methods: Analysis of spatial gene-expression patterns at cell-scale resolution). (D) Heatmaps of vpsL average per-cell-volume fluorescence signal relative to the average per-cell-volume fluorescence signal in the bins denoted with asterisks are shown as in Fig 2B for n = 10 replicate biofilms with the two largest biovolumes shown in (A). (E) Average per-cell-volume fluorescence signal versus distance to the biofilm center, as in S3A Fig, for the data shown in (D). Light symbol coloring represents the smaller biofilms and dark symbol coloring represents the larger biofilms. All data have been corrected using the model described in the text (See Methods: Spatial correction model). The data underlying this figure can be found in S1 Table.

https://doi.org/10.1371/journal.pbio.3003187.g005

Our analyses show that vpsL expression declines to approximately 1 mRNA per 50 cells in mature WT V. cholerae biofilms (Fig 5A). Consequently, while we were able to monitor overall changes in vpsL levels by averaging over many biofilms and thousands of cells, we are unable to quantify spatial differences in vpsL expression in WT V. cholerae biofilms because, in local biofilm cell subpopulations that can contain fewer than 50 cells, we often detect zero vpsL mRNA transcripts. Fortuitously, mature biofilms of the LCD-locked strain, because they express overall higher levels of vpsL per cell than do mature WT biofilm cells, made it possible for us to probe spatial vpsL expression. In early-stage biofilms, vpsL is uniformly expressed across the biofilm (Fig 5B–5E). In mature biofilms, vpsL transcript levels are higher in cells at the biofilm periphery than in cells located at the biofilm core (Fig 5B–5E). This pattern mirrors that observed in biofilms of the AI-2 sensor strain in the absence of AI-2 (Figs 3C–3E, S5E), with the difference in the magnitude of the pattern likely attributable to differences in the size, and thus developmental state, of the biofilms assayed.

Our findings regarding QS and regulation of vpsL spatiotemporal expression in biofilms are as follows: QS is the major driver of temporal changes in vpsL expression over V. cholerae biofilm development, and QS is responsible for terminating vpsL expression at biofilm maturity. While QS-independent vpsL expression reductions occur over time in biofilms composed of QS-deficient strains, they are modest compared to those that occur in WT biofilms. We attribute the QS-independent temporal vpsL expression changes to declining c-di-GMP levels that occur concurrently with biofilm maturation, as discussed below. Regarding spatial vpsL regulation: QS cannot establish regional differences in vpsL expression due to invariant QS signals across biofilms (Fig 4C–4F). Nonetheless, vpsL develops a distinct spatial expression pattern with higher expression in peripheral biofilm cells than in interior cells. This pattern persists in QS-inert strains (Figs 3C–3E, S5E, 5B–5E), further confirming the lack of a role for QS in establishing this spatial pattern. Thus, any spatial heterogeneity in vpsL expression that is present in biofilms must be established independently of QS activity. As c-di-GMP is the only other known regulator of vpsL, regional differences in c-di-GMP signaling across the biofilm are most likely responsible for setting up local patterns of vpsL gene expression. We further validate this assertion below.

Spatiotemporal expression patterns of genes encoding matrix structural proteins parallel those of vpsL and are controlled by c-di-GMP signaling in V. cholerae biofilms

To further investigate how spatiotemporal gene-expression patterns are established during V. cholerae biofilm formation, we assessed the expression patterns of additional biofilm regulators and structural genes. Specifically, having considered QS regulation above, we now primarily focus on the regulators VpsR and VpsT that link alterations in c-di-GMP levels to changes in gene expression [15]. VpsR and VpsT are transcription factors that bind c-di-GMP and are active in their c-di-GMP-bound states. Once liganded, they promote expression of their downstream target genes. Transcription of vpsT is activated by the VpsR-c-di-GMP complex and repressed by HapR at HCD. This regulatory arrangement links vpsT levels to both c-di-GMP and to QS (Fig 1) [30].

We measured expression of vpsR and vpsT and genes encoding the three matrix proteins RbmA, RbmC, and Bap1 that are downstream targets of VpsR and/or VpsT (Fig 1). Our smFISH technology limits us to probing three genes in a given biofilm, so we measured genes in pairs; genes assayed in the same biofilms are displayed together in the figures. We again probed both WT biofilms and biofilms formed from the LCD-locked strain to disentangle contributions from QS from those of c-di-GMP-mediated regulation. We note that the while the regulatory roles of VpsR and VpsT in biofilms are known, the timing with which and location(s) at which they act during biofilm development, to our knowledge, have not been defined.

We first established baseline expression values for rbmA, rbmC, bap1, vpsR, and vpsT in immature biofilms (Fig 6A). Each gene displayed a similar basal expression level in young WT biofilms and young biofilms of the LCD-locked strain. One could anticipate that the two strains would have similar expression levels given that cells in immature biofilms are in the LCD QS state. rbmA, rbmC, and bap1 are all expressed at similar levels, which are approximately 5-fold higher than vpsL levels. Presumably, vpsL, encoding an enzyme, is not required at the quantities needed for matrix structural proteins, at least early in biofilm development. The two regulators also display striking differences in expression, with vpsR expressed >15-fold higher than vpsT.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Expression of genes encoding matrix components and regulators over V. cholerae biofilm development.

(A) Average per-cell-volume fluorescence signal of rbmA, rbmC, bap1, vpsL, vpsR, and vpsT across n = 3–7 replicate WT biofilms or biofilms formed by the LCD-locked strain, for biofilms with the smallest biovolume shown in (B). Error bars represent standard deviations. (B) Relative average per-cell-volume fluorescence signal of vpsL, rbmA, rbmC, bap1, vpsR, and vpsT across n = 3–7 replicate WT biofilms or biofilms formed by the LCD-locked strain as a function of biovolume. Values are relative to the average per-cell-volume fluorescence signal in biofilms with the smallest biovolume for each background and gene. Pairs of plots displayed next to each other (vpsL/rbmA, rbmA/bap1, and vpsR/vpsT) represent pairs of genes measured in parallel in the same biofilms. (C) Representative images showing DAPI, and rbmA, rbmC, bap1, and vpsR smFISH fluorescence signals and heatmaps of per-cell-volume fluorescence signal as a function of x, y position as in Fig 3C for n = 5 replicate WT V. cholerae biofilms with the largest biovolumes shown in (B). Scale bars represent 5 µm. (D) Relative average per-cell-volume fluorescence signal as a function of distance from the periphery of the biofilm as in Fig 3D, for the data shown in (C). (E) Heatmap of rbmA, rbmC, bap1, and vpsR average per-cell-volume fluorescence signal relative to the average per-cell-volume fluorescence signal in the bin denoted with an asterisk are shown as in Fig 2B for n = 5 replicate WT biofilms at the largest biofilm size shown in (B). (F) Average per-cell-volume fluorescence signal versus distance to the biofilm center, as in S3A Fig, for the data shown in (E). All data have been corrected using the model described in the text (See Methods: Spatial correction model). The data underlying this figure can be found in S1 Table.

https://doi.org/10.1371/journal.pbio.3003187.g006

We hypothesized that, similar to what occurs for vpsL (Fig 5A), expression of the three matrix-protein encoding genes would decrease over biofilm development. Indeed, decreases do occur in each case as biofilms mature, although the timing and magnitudes of the reductions differ between the three genes (Fig 6B). In all three cases, the changes that occur in biofilms composed of the LCD-locked strain mirror those that occur in WT biofilms, indicating that their expression timing is regulated by c-di-GMP signaling, and potentially other yet-to-be-identified regulatory pathways, but not by QS. rbmC and bap1 are controlled by c-di-GMP though VpsR; they are not regulated by VpsT and they are not known to be controlled by QS. rbmA is controlled by c-di-GMP through VpsR and by both c-di-GMP and QS through VpsT. Consistent with regulation of vpsT by both QS and c-di-GMP, larger decreases in vpsT occur in WT biofilms than in biofilms of the LCD-locked strain (Fig 6B). Given that VpsT controls rbmA, it is therefore surprising that the decrease in expression of rbmA that occurs is equivalent in the WT and LCD-locked strains (Fig 6B). We account for this unexpected result as follows: we calculate basal expression of vpsT to be <1 mRNA copy per cell in both WT biofilms and in biofilms of the LCD-locked strain (Fig 6A). Likely, there is insufficient VpsT present in each biofilm cell, irrespective of the strain background or biofilm development stage, for it to act as a potent regulator of rbmA. Indeed, the role of VpsT in control of V. cholerae biofilm development has been primarily studied using a rugose variant that possess higher basal c-di-GMP levels, and consequently higher vpsT levels than WT and LCD-locked V. cholerae strains [15,47,51]. Thus, our examination of WT V. cholerae that possesses native c-di-GMP levels suggests that the role of VpsT in connecting c-di-GMP and QS changes to rbmA expression is largely inconsequential. For this reason, below we focus exclusively on the role of VpsR in connecting changes in c-di-GMP levels to changes in target gene expression in biofilms.

The reductions in rbmA, rbmC, and bap1 that occur over biofilm development could be a consequence of decreases in c-di-GMP concentrations, decreases in vpsR expression, or both. vpsR expression remains constant over the course of biofilm formation in biofilms formed by both the WT and the LCD-locked strain (Fig 6B). These results suggest that, over the time of V. cholerae biofilm development, decreases in c-di-GMP levels occur that lower VpsR activity. Loss of active VpsR, in turn, results in reduced expression of the three downstream biofilm matrix genes.

One prediction stemming from our assertion that differences in c-di-GMP levels across biofilms drive the regionally-distinct pattern of vpsL expression is that rbmA, rbmC, and bap1, which we find are regulated by c-di-GMP and not QS, should behave similarly. To test this hypothesis, we measured the spatial expression patterns of the three matrix-protein encoding genes. Fortuitously, rbmA, rbmC, and bap1 are more highly expressed than vpsL, enabling us to probe their spatial expression patterns in WT V. cholerae biofilms, which as mentioned, we could not do for vpsL. A clear pattern, with higher expression in peripheral cells than in cells at the core, exists for all three matrix genes in WT biofilms (Fig 6C–6F). The magnitude of the spatial variability is distinct for each gene, presumably a consequence of differences in the strength of c-di-GMP-mediated regulation to which each gene is subject. The same spatial pattern of expression occurs in biofilms formed by the LCD-locked strain (S9A–S9E Fig), consistent with c-di-GMP regulation and no QS input into control of these three genes in either time or space in biofilms.

In addition to unvarying temporal vpsR expression (Fig 6B), uniform vpsR expression also occurs across space in WT biofilms (Fig 6C–6F), demonstrating that the regional patterns of expression that exist for the three VpsR target genes are not driven by regional differences in vpsR expression. Together, the results support our model that spatial differences in c-di-GMP levels exist across biofilm cells and, by modulating VpsR activity, establish specific spatial gene-expression patterns of target matrix genes. Further supporting the claim that differences in c-di-GMP concentration exist across cells, a positive correlation exists between expression of VpsR target genes, namely between rbmA and vpsL and between rbmC and bap1, in biofilms formed by both the WT and the LCD-locked strain (S9F, S9G Fig). These relationships are unlikely to result from extrinsic noise driving global cell-to-cell differences in gene expression, as no such correlation exists for other pairs of genes under study (S8B, S8C Fig). Rather, we suggest that the correlation stems from cell-to-cell differences in c-di-GMP concentrations driving cell-to-cell differences in VpsR activity and, in turn, expression of target genes.

Perturbing c-di-GMP signaling disrupts the spatial patterning of vpsL expression in V. cholerae biofilms

To bolster our hypothesis that variations in c-di-GMP levels across V. cholerae biofilms are responsible for patterning matrix gene expression, we perturbed the endogenous c-di-GMP levels and investigated whether that altered matrix gene-expression patterns. We modulated c-di-GMP levels in biofilms by two mechanisms. First, we grew V. cholerae biofilms in the presence of 100 µM Nspd, which, as discussed in the Introduction, increases c-di-GMP by modulating the activity of MbaA (Fig 1). Second, we exploited a mutant vpvC gene, denoted vpvC^W240R and often referred to as the rugose mutation, that confers constitutive VpvC DGC activity [52]. Both Nspd addition and the vpvC^W240R mutation have been shown previously to increase cytoplasmic c-di-GMP levels, and correspondingly, matrix gene expression, with the vpvC^W240R mutation being the more potent activator of c-di-GMP production [19]. Consistent with these earlier findings, we show that vpsL expression is highest in biofilms formed by the LCD-locked strain containing the vpvC^W240R mutation, displays intermediate expression in biofilms of the LCD-locked strain to which Nspd has been administered, and is lowest in biofilms of the LCD-locked strain with unperturbed, endogenous c-di-GMP levels (Fig 7A). The spatial pattern in which cells at the biofilm periphery display higher vpsL expression than cells at the core remains present in biofilms of the LCD-locked strain following treatment with Nspd (Fig 7B–7E). However, further increasing c-di-GMP levels through introduction of the vpvC^W240R mutation eliminates the vpsL spatial pattern (Fig 7B–7E). These results suggest the following model: Globally increasing c-di-GMP levels overrides the naturally occurring differences in c-di-GMP concentrations that exist in V. cholerae cells in biofilms. Consequently, the c-di-GMP-controlled spatial pattern of vpsL gene expression is lost.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Effect of perturbation of c-di-GMP signaling on the vpsL spatial gene expression pattern.

(A) Average per-cell-volume fluorescence signal of vpsL across n = 5 replicate biofilms formed by the LCD-locked strain, LCD-locked strain grown with 100 µM Nspd, or LCD-locked, vpvC^W240R strain. Error bars represent standard deviations. (B) Representative images showing DAPI and vpsL smFISH fluorescence signals, and heatmaps of per-cell-volume fluorescence signal as a function of x, y position as in Fig 3C for n = 5 replicate biofilms formed by the LCD-locked strain, LCD-locked strain grown with 100 µM Nspd, and LCD-locked, vpvC^W240R strain. Scale bars represent 5 µm. (C) Relative average per-cell-volume vpsL smFISH fluorescence signal as a function of distance from the periphery of the biofilm as in Fig 3D, for the data shown in (B). (D) Heatmaps of vpsL average per-cell-volume fluorescence signal relative to the average per-cell-volume fluorescence signal in the bins denoted with asterisks are shown as in Fig 2B for n = 5 replicate biofilms formed by the LCD-locked strain (left), LCD-locked strain grown with 100 µM Nspd (center), or LCD-locked, vpvC^W240R strain (right). (E) Average per-cell-volume vpsL smFISH fluorescence signal versus distance to the biofilm center, as in S3A Fig, for the data shown in (D). All data have been corrected using the model described in the text (See Methods: Spatial correction model). The data underlying this figure can be found in S1 Table.

https://doi.org/10.1371/journal.pbio.3003187.g007

Differences in spatial vpsL expression patterns in biofilms of the AI-2 sensor strain grown in the presence of Nspd and in the absence or presence of AI-2 confirm the above mechanistic interpretation for how the vpsL biofilm spatial pattern is established. Regarding mechanism, HapR activates transcription of nspS-mbaA encoding the Nspd receptor-effector pair. Consequently, at HCD (i.e., + AI-2), when HapR is abundant, Nspd potency increases due to increased NspS-MbaA-mediated detection. Under this condition, MbaA DGC activity increases to a higher level than when Nspd is supplied in the absence of AI-2 [49]. Consistent with this mechanism, in biofilms of the AI-2 sensor strain grown with Nspd but without AI-2, vpsL is preferentially expressed in peripheral biofilm cells. By contrast, cells of the AI-2 sensor strain in biofilms grown with Nspd and AI-2 display uniform vpsL expression across space (Figs 3C–3E, S5E). Again, these results show that globally increasing c-di-GMP in V. cholerae biofilms abolishes the c-di-GMP-directed vpsL spatial gene-expression pattern. Moreover, the mechanism by which the c-di-GMP levels are altered is not relevant.

Bacterial strains and growth conditions

The parent V. cholerae strain used in this study is O1 El Tor biotype C6706str2 ∆vpsS ∆cqsR (BB_Vc416). All strains used in this study are listed in S2 Table. V. cholerae biofilms were grown as follows: Cultures were seeded from single colonies into 1 mL LB and grown at 37 °C for 3 h with shaking, to an approximate OD₆₀₀ = 0.1. Cultures were back diluted to an OD₆₀₀ = 6 × 10⁻⁶ in M9 medium (1× M9 salts, 100 μM CaCl₂, 2 mM MgSO₄, 0.5% dextrose, and 0.5% casamino acids). A total of 200 µL of cultures was dispensed into No. 1.5 glass coverslip-bottomed 96-well plates (MatTek, Ashland, MA, USA), and grown statically at 37 °C. When indicated, 0.0375%, or 0.2% arabinose, 100 µM norspermidine, and/or 10 µM AI-2 was added to LB and M9 medium. When AI-2 was provided, media also contained 0.1 mM boric acid. In every case in which data for planktonic cultures are reported, the cultures were grown in the identical medium as that used for biofilm growth and the same cultures used to seed biofilm growth were used to seed planktonic cultures. Cultures were back diluted to OD₆₀₀ = 0.0002 in 10 mL M9 medium and grown with shaking at 37 °C.

Single-molecule RNA FISH (smFISH) probe design and hybridization

Custom Stellaris RNA FISH probes were designed against genes of interest using the Stellaris RNA FISH Probe Designer v.4.2 (LGC, Biosearch Technologies). Probes were labeled with Quasar 670, CAL Fluor Red 590, or FAM (see S3 Table for a complete list of probe sequences and associated fluorophores). For biofilm samples, probes were hybridized following the manufacturer’s instructions for adherent cells in 96-well glass bottom plates, with minor modifications. The full protocol is available online at www.biosearchtech.com/stellarisprotocols. Briefly, following growth, the medium was decanted, and adhered biofilms were fixed for 10 min at room temperature with 200 µL 1× PBS and 3.7% formaldehyde. The fixed biofilms were washed four times with 200 µL 1× PBS. To permeabilize cells, the samples were incubated overnight, or up to 24 h, at 4 °C in 200 µL 70% EtOH. Samples were incubated for 5 min at room temperature with 200 µL RNA FISH wash buffer A (Biosearch Technologies, prepared following the manufacturer’s instructions) containing 10% formamide. Probes were hybridized with 75 µL RNA FISH hybridization buffer (Biosearch Technologies) containing 10% formamide and 1.5 µL of each probe stock (12.5 µM), followed by overnight incubation at 37 °C. Samples were washed twice for 30 min at 37 °C with 200 µL RNA FISH wash buffer A containing 10% formamide. Cells were stained with 50 µg mL^–1 DAPI in wash buffer A for 20 min at 37 °C. Following a final wash with 200 µL Stellaris RNA FISH wash buffer B, 50 µL VectaShield mounting medium was added, and the samples were immediately imaged. Regarding planktonic cells, probe hybridization was performed as previously described [39], with minor modifications as detailed in [57].

Confocal imaging

Imaging was performed with a Nikon Eclipse Ti2 inverted microscope equipped with a Yokogawa CSU-W1 SoRa confocal scanning unit. Samples were imaged with a CFI Apochromat TIRF ×60 oil objective lens (Nikon, 1.49 numerical aperture) with excitation wavelengths of 405 (DAPI), 488 (mNG and FAM), 561 (mScarlet and CAL Fluor Red 590), and 640 nm (Quasar 670) and with 0.5 µm z-steps. Images were captured through a 2.8x SoRa magnifier.

smFISH image analysis

Images were processed using Nikon NIS-Elements Denoise.ai software. Denoised images of planktonic cells were segmented using Fiji software [58] from maximum z-projections of the DAPI channel. A gaussian blur was applied to images, and cells were identified using an intensity threshold. Cells that could not be accurately resolved from neighboring cells were excluded from downstream analyses. Using the DAPI channel, denoised images of biofilm cells were segmented with BiofilmQ into cubes with side lengths of 2.32 µm [43]. The center x, y, z coordinates of each cell cube and the volume fraction of each cube occupied by cell mass were exported for use in downstream analyses. We interchangeably use “cell cube” and “cell” to refer to this resulting cube-based segmentation. The boundaries of individual biofilms in denoised images were manually defined using the DAPI channel. Cell cubes outside of these boundaries were excluded from downstream analysis.

smFISH data were analyzed using custom Python scripts. Briefly, images were convolved with a Gaussian function to remove noise. Puncta were detected as local maxima with intensities greater than a threshold that was established based on a set of negative control images specific to each probe set. For hapR and mNG probes, negative control images were acquired using biofilms containing cells lacking hapR and mNG (BBVc_801). For vpsL probes, mature WT biofilms (BBVc_416), i.e., at a stage when vpsL is no longer expressed, were hybridized with vpsL to acquire negative control data. The vpsR, vpsT, rbmA, rbmC, and bap1 genes could not be deleted, because without them, biofilms do not form. Likewise, polA and gyrA are essential genes and cannot be deleted. Thus, as a negative control for these probe sets, WT (BBVc_416) biofilms, i.e., biofilms lacking mNG, were hybridized with mNG-targeting probes labeled with the fluorophore matching each probe set of interest. Each punctum was fitted with a 3D Gaussian function to determine the integrated, background-subtracted punctum intensity. In instances with multiple puncta residing in close proximity, a 3D multi-Gaussian fit was performed. Puncta were assigned to cells (planktonic samples) or to cell cubes (biofilm samples) and the total fluorescence signal per cell or cube was calculated as the sum of all puncta intensities. For biofilm samples, this fluorescence signal was further normalized by the Cube_VolumeFraction, which is the fraction of the cube volume occupied by biomass, calculated by BiofilmQ.

Spatial correction model

Spatial correction factors were calculated using data from the largest biofilms formed by the parent strain carrying ∆vc1807::pBad-mNG (BB_Vc813) imaged under each condition. Unique correction factors were derived for each of the three fluorophores, at both high and low expression levels of mNG, and for all positions (r, z). As biofilms are not perfect half spheres, the r position of each cell cube was calculated in a z-dependent manner. Specifically, a unique center position was determined for each z bin as the volume-weighted average x, y coordinate across all cells within that z bin. For z bins with z ≤ 391 pixels (15.1 µm) and z > 391 pixels (15.1 µm), this procedure shifted the calculated center position relative to the x, y center of the biofilm as a whole by an average of 4.8 µm (SD = 2.6 µm) and 13.0 µm (SD = 5.6 µm), respectively. The r position for each cell cube was subsequently calculated as the distance between the cell cube’s x, y location and the center x, y position corresponding to that cell cube’s z location. The correction factors use experimental data to describe both z and r positional biases, as detailed below. To calculate z bias, the average per-cell-volume smFISH fluorescence signals from cell cubes with radial position r = 311 pixels (12.0 µm) were calculated as a function of cell cube z position and normalized to the average per-cell-volume smFISH fluorescence signal of cell cubes with z, r = 91, 311 pixels (3.52, 12.0 µm). These data were fit with the ad-hoc Equation (1) to determine the correction factor as a function of z:

(1)

To calculate r bias, the average per-cell-volume fluorescence signal of cell cubes with z position z = 91 pixels (3.52 µm) were calculated as a function of cell cube r position. We observed that the average smFISH fluorescence signal decreased linearly as a function of r before plateauing at r = r_i, with the position of r_i differing for each sample. For each of the six samples, we selected an r_i value by visual inspection of the data. The average smFISH fluorescence signal was normalized to the average smFISH fluorescence signal of cell cubes with r = r_i, z = 91 pixels (3.52 µm) and fit to the ad-hoc Equation (2):

(2)

Using the specific fit parameters for each fluorophore and for high and low gene-expression levels, the correction factor for all positions (r, z) was calculated for each condition as the product of Equations (1) and (2). These values form the basis for calculating interpolative equations used to correct all datasets, as described below.

In order to acquire appropriate correction factors for smFISH fluorescent signals from genes with any expression level, not only those matching the mNG high and low expression conditions, and also for biofilms of any size, we developed a generalized interpolation model as follows: for a given fluorophore, for each position (r, z), a linear equation Y(X) connecting points (X₁, Y₁) and (X₂, Y₂) was calculated, in which Y₁ and Y₂ are the correction factors derived above from the low- and high-gene-expression conditions, respectively, and X₁ and X₂ are the log values of the average summed smFISH fluorescence signals across the biofilms that were used to calculate the above correction factors in the low- and high-gene-expression conditions, respectively. With this model, for any given biofilm or group of replicate biofilms, unique correction factors can be determined for each position (r, z) based on the total smFISH fluorescent signal in the biofilm(s). Specifically, for each position (r, z), the log of the average summed smFISH fluorescence signals across replicate biofilms was taken as an X value and introduced into the corresponding (r, z)-dependent linear equation Y(X) generated above to calculate an (r, z)-specific correction factor for those biofilms. To use these calculated correction factors, for each cell cube within the biofilm, the per-cell-volume smFISH fluorescence signal, calculated as described above, was divided by the correction factor based on its position (r, z). The resulting calculated values were used for all subsequent analyses, unless otherwise noted. Further validation of the linear interpolation of the model is included as S2 Data. Values for all fit parameters, (r, z)-dependent correction factors, and resulting (r, z)-dependent interpolative linear equations can be found in S4 Table. The code detailing generation and application of the model, and accompanying sample plots of fit parameters, can be found at https://doi.org/10.5281/zenodo.12687319.

Analysis of spatial gene-expression patterns at cell-scale resolution

Spatial gene-expression patterns in individual cells were assessed in several ways. First, to visualize radial patterns (as in Fig 3C), cell cubes at the bottom of the biofilm, defined as cells with z positions between 0 and 4.6 µm, across replicate biofilms were grouped by their x and y positions. For each cell cube, the x, y position was set relative to the center x, y position of the biofilm. The per-cell-volume signal for each group of cells was calculated and plotted as heatmaps using bins with ∆x, ∆y = 100, 100 pixels (3.9, 3.9 µm). Second, the distance of each cell cube at the bottom of the biofilm to the periphery of the biofilm was calculated as the shortest distance from the cell to the biofilm perimeter, which was manually defined. Based on this distance, cells across replicate biofilms were grouped into 1 µm sized bins and the relative average per-cell-volume fluorescence signal per bin was calculated relative to cells with distance 2 µm from the biofilm periphery. Resulting data were plotted as a function of the left boundary of each bin (as in Fig 3D) for bins with greater than n = sample size × 10 cells. Third, cell cubes across replicate biofilms were grouped by their center r and z positions. Radial positions for each cell cube were calculated as described above. The average per-cell-volume fluorescence signal for each group of cells was calculated and normalized to the group of cells with center position r, z = 91, 91 pixels (3.5, 3.5 µm). Resulting data for groups containing ≥ 25 cells were plotted as heatmaps using bins with Δr, Δ z = 2.32, 2.32 µm (as in Fig 2). Fourth, the distance of each cell cube from the center of the biofilm was calculated from the center r, z position of that cell cube to the center position r, z = 0, 91 pixels (0, 3.5 µm). Based on this calculated distance, cells across replicate biofilms were grouped into 2.5 µm sized bins, and the average per-cell-volume fluorescence signal per bin was calculated. Resulting data were plotted as a function of the left boundary of each bin (as in S3A Fig).

Correlating gene expression in cell groups

The relationship between gene-expression levels measured in parallel across cells was assessed as follows: For a given reference gene, all cells with zero signal corresponding to that gene were omitted from analysis. The remaining cells were ordered by expression level of the reference gene and grouped into bins of 500 cells each. The average per-cell-volume fluorescence signal was calculated for each bin for both the reference gene and a second comparison gene, and these two average signal values were plotted against one another for each bin.

Quantitation and statistical analyses

Standard deviations were calculated using custom Python scripts from log₂ transformed data.

z-scores of average per-cell-volume fluorescence signal in r, z bins were calculated using the following formula, where represents the mean per-cell-volume fluorescence signal across cells in the bin, is the standard deviation of the per-cell-volume fluorescence signal across cells in the bin, and is the mean per-cell-volume fluorescence signal in the bin with r, z = 91, 91 pixels (3.5, 3.5 µm).