Quantitative Visualization of Gene Expression in Mucoid and Nonmucoid Pseudomonas aeruginosa Aggregates Reveals Localized Peak Expression of Alginate in the Hypoxic Zone

A goal for microbial ecophysiological research is to reveal microbial activities in natural environments, including sediments, soils, or infected human tissues. Here, we report the application of the hybridization chain reaction (HCR) v3.0 to quantitatively measure microbial gene expression in situ at single-cell resolution in bacterial aggregates. Using quantitative image analysis of thousands of Pseudomonas aeruginosa cells, we validated new P. aeruginosa HCR probes. Within in vitro P. aeruginosa aggregates, we found that bacteria just below the aggregate surface are the primary cells expressing genes that protect the population against antibiotics and the immune system. This observation suggests that therapies targeting bacteria growing with small amounts of oxygen may be most effective against these hard-to-treat infections. More generally, this proof-of-concept study demonstrates that HCR v3.0 has the potential to identify microbial activities in situ at small spatial scales in diverse contexts.

D espite decades of research that has elucidated mechanisms of bacterial virulence, antibiotic tolerance, and antibiotic resistance, many infections remain impossible to eradicate. Phenotypic heterogeneity likely plays an important role in the failure of drugs and the immune system to clear chronic infections. Chronic Pseudomonas aeruginosa lung infections in people with cystic fibrosis (CF) are a prime example. Within individual lobes of the CF lung, genetically antibiotic-susceptible and -resistant P. aeruginosa sibling bacteria coexist (1). This likely affects treatment, because resistant bacteria can protect susceptible bacteria when mixed together in vitro (2,3). Likewise, CF lung mucus contains steep oxygen gradients, and anoxic conditions reduce antibiotic susceptibility (4)(5)(6)(7). While we know that bacterial genetic diversity and infection site chemical heterogeneity exist, tools to measure bacterial phenotypes in situ are lacking. Here, we applied the third generation of the hybridization chain reaction (HCR v3.0) to quantitatively measure gene expression in P. aeruginosa in an in vitro aggregate model system. Our findings suggest that HCR v3.0 could prove to be a useful tool for analyzing in situ bacterial gene expression in virtually any species in any context.
Validation of HCR v3.0 probes for quantifying P. aeruginosa gene expression. As a first step, we tested HCR probe specificity and quantitation. HCR is a fluorescent in situ hybridization method that includes a quantitative, enzyme-free signal amplification step to help visualize low-abundant RNAs (8,9). We previously used single HCR v2.0 probes to detect bacterial taxa in CF sputum samples (10), and HCR v2.0 was also used by Nikolakakis et al. to detect host and bacterial mRNAs in the Hawaiian bobtail squid-Vibrio fischeri symbiosis (11). We chose to apply HCR v3.0 as a tool to quantify bacterial gene expression in situ because of its improved selectivity over HCR v2.0. HCR v3.0 requires pairs of probes to hybridize to adjacent binding sites on the target RNA in order to colocalize a full HCR initiator and trigger growth of an HCR amplification polymer (Fig. 1A), ensuring that individual probes will not generate amplified background even if they bind nonspecifically in the sample (8,9). We designed and validated HCR v3.0 probe sets that could be used to (i) differentiate species by targeting rRNAs and (ii) measure gene expression by targeting mRNAs.
To test specificity, we designed one HCR v3.0 probe set to detect 16S rRNA in all eubacteria (1 probe pair) and another HCR v3.0 probe set (3 probe pairs) to detect P. aeruginosa specifically, using our previous HCR v2.0 probe sets as starting points (10). P. aeruginosa, Pseudomonas fluorescens, Escherichia coli, and Staphylococcus aureus were grown in vitro in LB broth, fixed with paraformaldehyde, incubated with the eubacterial and P. aeruginosa-specific probes, washed to remove excess probes, incubated with fluorescent HCR amplifiers, washed to remove excess HCR amplifiers, and imaged via confocal microscopy. Single-cell fluorescence intensities were quantified computationally with Imaris (see methods in the supplemental material). When only one probe from each probe pair was used, no fluorescence was observed, demonstrating the background suppression feature of HCR v3.0 probes (see Fig. S2). As expected, the eubacterial probe set detected all four organisms ( Fig. 1B and C and S1) while the P. aeruginosa probe set produced a visible signal only for P. aeruginosa. Quantifying single-cell fluorescence for the P. aeruginosa probe set, intensities were an order of magnitude lower for P. fluorescens (the most closely related of the three off-target strains) and 2 orders of magnitude lower for Escherichia coli and Staphylococcus aureus (Fig. 1D), demonstrating high selectivity for the intended bacteria.
Next, we tested the utility of HCR v3.0 for performing mRNA relative quantitation for single P. aeruginosa cells within a population. According to Trivedi et al. (12), we performed a 2-channel redundant detection experiment in which the algD mRNA was simultaneously detected using two probe sets that trigger different HCR amplifiers carrying spectrally distinct fluorophores (each probe set comprised 10 HCR v3.0 probe pairs that bind to different subsequences along the mRNA). Because the HCR signal scales approximately linearly with the number of target mRNAs per cell (12), we expected a 2-channel scatter plot of single-cell fluorescent signal to yield an approximately linear distribution. In this setting, accuracy corresponds to linearity with zero intercept and precision corresponds to the scatter around the line (12). We generated a range of algD single-cell expression levels by overexpressing algD from the arabinoseinducible expression plasmid pMQ72 in a P. aeruginosa ΔalgD mutant (13,14) and analyzed the 2-channel scatter plot for more than ϳ8,000 cells in confocal micrographs. As expected, the single-cell signals were highly correlated between the two channels (median Pearson r ϭ 0.93 for N ϭ 5 micrographs) ( Fig. 1E and S3), and the distribution was approximately linear with a small intercept and tight scatter around the line. These results are consistent with previous validation studies demonstrating that HCR v3.0 enables accurate and precise relative mRNA quantitation in cells and embryos (8,9). Note that the precision increases with probe set size (12), and so it is beneficial to Each probe within a pair carries one-half of an HCR initiator so that cognate binding to the target colocalizes a full HCR initiator. Fluorophore-labeled HCR amplification hairpins are kinetically trapped so that they do not polymerize until they encounter a full HCR initiator, generating a fluorescent amplification polymer tethered to a specifically bound probe pair. As a result, using HCR v3.0 reagents, individual probes and hairpins that bind nonspecifically in the sample do not generate amplified background. (B) High selectivity using HCR v3.0 to detect rRNAs across bacterial species. Micrographs show that the Eub338 probe pair (turquoise) binds P. aeruginosa, S. aureus, P. fluorescens, and E. coli rRNA as intended, while the P. aeruginosa Paerug probe set (red) selectively binds P. aeruginosa rRNA. Scale bars, 2 m. (C) Quantification of single-cell Eub338 fluorescence intensities shows the intended broad selectivity (violin plots summarize data from 3 micrographs per organism). (D) Quantification of single-cell Paerug fluorescence intensities shows that Paerug probes are selective to Pseudomonas spp., with ϳ10-fold higher signal intensity produced by P. aeruginosa than by P. fluorescens and even higher selectivity against the more distantly related E. coli and S. aureus (violin plots summarize data from 3 micrographs per organism). (E) Representative two-channel single-cell HCR analysis using two algD probe sets. Highly correlated signal (r ϭ 0.93, Pearson correlation, n ϭ 1,808 cells analyzed); linear distribution and small intercept characterize accuracy, and scatter around the line characterizes precision (9,12). Ten probe pairs were used for each algD probe set to target algD mRNA in the PAO1 ΔalgD/pMQ72::algD strain after algD expression induction with 0.10% L-arabinose. Scale, 10 m. Graph is representative of five analyses of replicates using 0.10% L-arabinose to induce algD expression; median Pearson correlation, r ϭ 0.93 (see Fig. S3 in the supplemental material for other replicates). See also Fig. S1, S2, and S3. HCR reveals alginate gene expression in hypoxic zones of P. aeruginosa aggregates. As a case study, we chose to measure P. aeruginosa alginate (algD) and nitrate reductase (narG) gene expression in aggregates formed by mucoid (FRD1) and nonmucoid (PA14) strains. This approach was chosen for several reasons. First, measuring algD expression in situ is of interest because alginate is overproduced by mucoid strains in CF lung infections (15,16), and mucoid strains are associated with worsened lung function (17). Second, as a technical control, the algD gene should be more highly expressed in the mucoid than in the nonmucoid strain and produce a stronger HCR signal (15). Third, previous research suggests that alginate may be expressed under hypoxic and anoxic conditions in mucoid isolates (7,(18)(19)(20), yet the precise location of alginate gene expression in both mucoid and nonmucoid aggregate biofilms was unknown and alginate expression had not been studied at the single-cell level. We could also quantify algD expression relative to narG, a gene induced under hypoxic and anoxic conditions (18,21), a useful reference point for mapping algD expression in aggregates relative to oxygen availability.

Quantitative Visualization of Gene Expression In Situ
Using the agar block biofilm assay (ABBA) (22), we grew mucoid and nonmucoid aggregates suspended in an LB agar medium supplemented with 5 mM nitrate for 16 h, fixed the aggregates in paraformaldehyde, and measured narG and algD gene expression with HCR (see supplemental methods). As expected, the mucoid strain expressed algD more highly than the nonmucoid strain ( Fig. 2A and C to E). Spatially, algD expression was highest in the zones within the first 200 m below the air-agar interface ( Fig. 2A and C to E). In this same region, rRNA levels were highest and aggregates were largest, suggesting a higher growth rate in this region (Fig. S4). This is consistent with previous studies by Sønderholm et al. (23) and Wessel et al. (24) that found growth rates were higher in P. aeruginosa aggregates growing near air-agar surfaces, correlating with oxygen availability. Interestingly, narG was also expressed more highly in the mucoid than in the nonmucoid strain (Fig. 2D) and was expressed more evenly in aggregates at various depths below the agar surface ( Fig. 2A to D and S5). Analysis of individual aggregates in the ABBA experiments showed a ring-like pattern of 16S rRNA gene, algD, and narG expression, correlating with decreasing oxygen availability from the surface to within the aggregate, as would be expected for aggregates of this size (4,24). Within individual mucoid and nonmucoid aggregates, algD expression was detected in cells ϳ5 to 15 m below the aggregate surfaces but was not detected in the innermost cells within ϳ10 m of the aggregate centers ( Fig. 2F and G), and this was generalizable to nearly all aggregates expressing algD. In contrast, the innermost cells highly expressed narG, but cells within ϳ0 to 10 m of the aggregate surface did not express narG ( Fig. 2F and G). This led us to hypothesize that algD was being expressed by cells experiencing hypoxia just below the aggregate surface and not by cells in the innermost, presumably anoxic, regions of the aggregates.
To test where cells were expressing algD relative to oxygen availability, we used a microelectrode to measure oxygen concentrations from 0 to 600 m below the nitratesupplemented LB agar surface in mucoid and nonmucoid ABBA experiments after 16 h each scatter plot (**P Ͻ 0.005, unpaired two-tailed t test). (I) Mean algD and narG expression per mucoid ABBA aggregate (left y axis) plotted with mean oxygen concentrations measured (right y axis). Middle pink arrow indicates oxygen concentration at which peak algD expression was detected, bottom and top pink arrows indicate minimum and maximum oxygen concentrations at which algD expression was detected. Error bars in panels H and I indicate SEMs for the oxygen concentrations. (J) Expression of algD is restricted to hypoxic regions, while narG is detected in hypoxic and anoxic regions. Oxygen profiles (yellow) overlay 3D micrographs showing rRNA, algD, and narG HCR signals in mucoid ABBA samples. This shows algD gene expression is only detected in regions where oxygen is also detected, while narG and rRNA are detected in regions where oxygen is not measurable. Oxygen profiles are plotted multiple times using perspective at different xz planes along the y axis. In panels A to G, I, and J, data are shown from a representative ABBA experiment. In all micrographs, the power for each individual laser used to excite the different fluorophores and the gain for each detector were kept consistent from one experimental replicate to another. Thus, gene expression for individual genes (e.g., algD in FRD1 or PA14 strains) can be compared from one experiment to another as well as across space within individual experimental replicates (e.g., 0 versus 100 m from the surface). However, conclusions regarding differences in expression of different genes (e.g., narG versus algD) cannot be drawn due to differences in lasers and detectors used for each gene being analyzed. Results from a replicate experiment are shown in Fig. S5.
Quantitative Visualization of Gene Expression In Situ ® of growth. Unexpectedly, the mucoid strain showed a modest increase in its oxygen consumption rate compared to that of the nonmucoid strain (Fig. 2H). However, as we predicted, the mucoid strain expressed algD highest in hypoxic regions (5 to 200 M oxygen) of the agar, from 0 to 350 m below the agar surface, and maximally at ϳ75 M oxygen ( Fig. 2I and J). In regions with less than 5 M oxygen, algD expression plummeted to Ͻ1% of the maximum value detected (Fig. 2I and J). This was surprising because in planktonic cultures, we found that anoxia most strongly induced algD expression compared to that under oxic and hypoxic conditions (Fig. S6), similar to previous research (19). Thus, alginate gene expression patterns differ between planktonic and aggregate cells: in aggregate cells, algD expression is greatest under hypoxic rather than anoxic conditions.
Conclusions. These data show that nonmucoid and mucoid P. aeruginosa strains express alginate genes in hypoxic zones, which goes against the general thought that alginate gene expression is constitutive in mucoid P. aeruginosa. Furthermore, these experiments demonstrate the utility of HCR v3.0 for analog quantitation of bacterial gene expression in situ at spatial scales relevant to microbial assemblages. Going forward, it will be exciting to combine HCR with tissue clearing methods such as MiPACT (10) to determine whether the expression patterns observed in these in vitro studies similarly characterize aggregate populations in more complex environments, such as in the study of pathogens in vivo. Direct insight into how pathogen physiology develops in infected tissues, or any other context where spatial observation of microbial activities is important, promises to yield insights that will facilitate more effective control of these communities. Many applications of HCR v3.0 can be envisioned, such as using this visualization tool to analyze microbes after therapeutic interventions to identify bacterial subpopulations that either resist or succumb to treatment; identification of the subpopulations that survive a specific perturbation could be used to guide the development and implementation of future therapeutics. Importantly, the potential utility of HCR 3.0 transcends applications in the realm of pathogenesis and stands to aid the study of microbial activities at the single-cell level in diverse contexts.

ACKNOWLEDGMENTS
We thank Will DePas, Ruth Lee, Niles Pierce, Maayan Schwarzkopf, and the Programmable Molecular Technology Center at the Caltech Beckman Institute for technical assistance and advice. Confocal microscopy was performed in the Caltech Biological Imaging Facility at the Caltech Beckman Institute, which is supported by the Arnold and Mabel Beckman Foundation.