AirGel: A tissue–engineered airway infection model

We grew AirGels from primary human bronchial epithelial (HBE) cells, which expand to confluence on the cylindrical cavity of the ECM scaffold (Fig 1B). An elastomeric microfluidic chip maintains AirGels and allows for luminal access. The ECM geometry guides epithelial architecture, enabling morphological customization of AirGels. Here, we designed and optimized AirGels to enable high–resolution fluorescence microcopy to monitor infection dynamics at the single bacterium level in live tissue. Maintaining an ALI in the lumen promotes epithelial cell differentiation and reproduces the physiological conditions encountered in the airway. We therefore optimized the matrix formulation so that AirGels remain stable at the ALI upon removal of culture medium from the lumen, thereby biologically and physically replicating the airway environment (Fig 1C).

AirGel epithelia are tight and impermeable (Fig 1D). Single–plane illumination microscopy images show that mature AirGels form tubular epithelial tissue, recapitulating the architecture and dimensions of a human small bronchus (Fig 1E) [36,37]. We characterized HBE cell differentiation in 34–day–old AirGels. Immunofluorescence highlighted an abundant population of mucus–producing goblet cells and ciliated cells (Fig 1E). To quantify the abundance of each cell type, we performed single–cell RNA sequencing (scRNA–seq) of mature AirGels. We identified 5 main clusters (Figs 1F and S1): basal cells (8%), ciliated cells (41%), secretory cells (34%), as well as immature ciliated (7%) and immature secretory cells (which are also sometimes defined as suprabasal cells) (10%). AirGels therefore reproduce the cellular composition and histological signature of human airway epithelia [38–41] and more specifically the distal human airway [42].

Given its prominent function in host–microbe interactions, we carefully characterized the architecture of mucus in AirGels. Immunofluorescence against the airway gel–forming mucins MUC5AC and MUC5B showed the presence of extracellular mucus in the form of thick luminal filaments (Figs 2A and S2). We also observed similar fiber–like mucus architecture in live AirGels by staining with the fluorescently–labeled lectin jacalin [21]. These fibers recapitulate the mucus architecture observed in porcine and murine tracheal explants [18,20,21]. We then characterized AirGel mucociliary clearance functions. Measurements of cilia beating frequency (CBF) in AirGels were indistinguishable from previous ex vivo measurements (Fig 2B and S1 Video) [9,43–45]. In addition, AirGel cilia generated a directional flow whose clearance velocity was in the lower end of the physiological range (Figs 2C and S3) [9,18,23]. We attribute the discrepancies to different experimental conditions. For example, since AirGels are curved, our in–plane visualizations image particles that are at different heights from the epithelium surface. As a result, only a fraction of these particles lies in the appropriate range for maximal clearance. Consequently, median velocities we measured tend to underestimate of the actual clearance speed. However, we measured that maximum velocities are similar between Transwells and AirGels (S3B Fig). Humidity is another potential cause for these differences, since it is known to impact mucociliary clearance in mouse lungs [23]. Overall, though, AirGels reproduce biological, physical, and dynamic parameters of the human airway including its tube–shape, all in a system allowing for live imaging of host–pathogen interactions at high resolution.

Download:

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

Fig 2. Mucociliary function of AirGels.

(a) Extracellular luminal mucus in AirGels in methacarn–fixed (left) and live configurations (right). Stainings were done with antibodies against MUC5AC and MUC5B gel–forming mucins, as well as the fluorescent lectin jacalin (which targets glycans), respectively. (b) CBFs of 5 different AirGels, measured by tracking the oscillations of fluorescent beads attached to cilia. The kymograph on the left shows the trajectory of such a bead during one second. (c) Mucociliary clearance in AirGels. (i) Trajectories of fluorescent microparticles in the lumen of 1 undifferentiated and 1 differentiated AirGel. (ii) Corresponding velocity distributions. Black lines indicate the median velocity. (iii) Median particle velocities for 3 differentiated and undifferentiated AirGels show the contribution of cilia beating in clearance. Each data point corresponds to the median in each experiment; the gray bar shows the median of triplicates. Statistics: independent samples Student t test with Bonferroni correction (p < 10⁻⁷). The data underlying this figure can be found in S2 Data. CBF, cilia beating frequency.

https://doi.org/10.1371/journal.pbio.3002209.g002

P. aeruginosa rapidly forms mucus–associated biofilms in AirGels

To visualize biofilm formation in a realistic airway mucosal context, we inoculated P. aeruginosa constitutively expressing the fluorescent protein mScarlet in the lumen of AirGels maintained at the ALI. After 13 h of incubation, we observed that bacteria had extensively colonized the mucosal surface. P. aeruginosa formed interconnected bacterial clusters colocalized with mucus within the airway surface liquid (ASL) between epithelial cells and the ALI (Fig 3A). In dynamic visualizations, bacteria remained attached to mucus despite movements induced by beating cilia (S2 Video). Since P. aeruginosa takes days to form biofilms in vitro, we were surprised to see these communities form only within a few hours in AirGels [46]. We therefore went on to investigate the mechanisms of biofilm formation on mucus.

Download:

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

Fig 3. P. aeruginosa rapidly forms mucus–associated biofilms.

(a) P. aeruginosa infection of a 62–day–old AirGel. Confocal images were acquired 13 h after inoculation. P. aeruginosa constitutively expresses the fluorescent protein mScarlet. The plasma membrane of epithelial cells was stained with CellMask Deep Red (pink). Mucus was stained with jacalin (green) shortly before infection. (b) (i) Maximal intensity projection images show P. aeruginosa biofilm formation within hours. (ii) Mean biofilms cluster area for 3 AirGels. The bar indicates the range between the maximum and minimum of the 3 means. The circle represents the mean of the means. (iii) Percentage of clusters that were larger than 100 μm² in each replicate (black dots). The bars represent the mean across replicates. Statistics: paired samples Student t test with Bonferroni correction (p = 0.051 and p = 0.01). (c) P. aeruginosa rapidly colonizes mucus surfaces. Images show maximal intensity projection of confocal stacks at t = 0 and t = 5.5 h post–inoculation. The graph quantifies the proportion of mucus not occupied by bacteria. Gray bars show the mean of triplicates. Statistics: paired samples Student t test with Bonferroni correction (p = 0.02). The data underlying this figure can be found in S3 Data.

https://doi.org/10.1371/journal.pbio.3002209.g003

We infected AirGels with exponential P. aeruginosa cultures (multiplicity of infection approximately 10) and imaged biofilm biogenesis at the single–cell level using confocal spinning disk microscopy. P. aeruginosa already formed aggregates a few hours after inoculation (Figs 3B and S4). While the mucus surface was initially largely devoid of bacteria, half of it was covered by P. aeruginosa multicellular structures after 5.5 h of infection (Fig 3C). Bacterial clusters with the same architecture also formed in the absence of jacalin staining, confirming these biofilms do not form through labeling artifacts (S5 Fig). To confirm the pivotal role of mucus in biofilm formation, we infected a non–differentiated AirGel that does not produce mucus. In the absence of a protective mucus layer, epithelial cells were more vulnerable to P. aeruginosa infection (S6 Fig). Bacteria breached through the epithelial barrier and invaded the underlying ECM. P. aeruginosa did not form 3D multicellular structures in the ASL. This further demonstrates the role of mucus hydrogel as a substrate for biofilm formation in differentiated AirGels, and at the same time highlights its protective function.

Our data suggests that P. aeruginosa forms biofilms in the airway by attaching to mucus at early stages of infection. To further explore the biophysical mechanisms of biofilm formation, we harvested mucus to perform ex situ visualizations. However, we could not observe the formation of P. aeruginosa biofilms on mucus extracted from HBE cultures (S7 Fig and S3 Video). We attribute this discrepancy to perturbations in mucus mechanical integrity when extracted from the epithelium and immersed in buffer. This difference highlights the importance of investigating microbe–mucus interactions in a native mucosal context such as the one established in AirGels.

To understand how biofilms form on native mucus, we therefore inspected the different steps of their formation in AirGels. To nucleate in vitro biofilms, P. aeruginosa cells navigate the surface of abiotic materials using twitching motility, which promotes the formation of aggregates [47]. Fast imaging of single cells shows that P. aeruginosa moves with twitching–like trajectories at the surface of mucus fibers (S4 Video). Single cells aggregated into small clusters within 2 h of colonization (Fig 3B). These small multicellular clusters subsequently moved along mucus fibers to eventually fuse into larger biofilms (Fig 4A). This caused a cascade of cluster fusion events that sped up biofilm formation (Fig 4A and S5 Video). We tracked aggregate fusion in kymographs highlighting the correlation between mucus and bacterial displacements (Fig 4B). The size of individual clusters remains approximately constant during motion and fusion, showing aggregate fusion predominates over bacterial growth. After 6 h of aggregation and fusion, dense biofilms are formed.

Download:

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

Fig 4. Dynamics of biofilm formation on mucus.

(a) Dynamic visualization of P. aeruginosa cluster fusion on mucus (t₁ = 6.2 h). The blue and pink arrowheads show 2 aggregates that fuse within the first 15 min. The resulting cluster is indicated by a yellow arrowhead. This new cluster then moves closer to the one indicated by the white arrowhead. All images are maximal intensity projections from z–stacks. (b) Kymographs showing the displacement of 2 clusters along their axis of motion. The bacterial aggregate and underlying mucus traveled together at an approximate speed of 0.5 μm/min. (c) Time course visualization of a mucus patch in a colonized by WT P. aeruginosa (not displayed). Reference features and their trajectories are indicated by colored arrowheads and dashed lines (t₁ = 1.2 h). (d) Mucus contraction was quantified by tracking the distances over time between N = 7 reference features in the mucus patch. The distances were normalized to the initial time point. They decrease over time, indicating contraction of the mucus patch. t₁, t₂, t₃ refer to the time points shown in panel c. Black circle: mean. Black line: standard deviation. The data underlying this figure can be found in S4 Data. WT, wild type.

https://doi.org/10.1371/journal.pbio.3002209.g004

We found that during biofilm formation, the mucus surface area tends to decrease compared to an uninfected control (Figs 4C and S8). The distances between landmarks in a mucus patch decreased over time (Fig 4D), demonstrating that mucus contracts during biofilm formation. We therefore hypothesized that mucus contraction speeds up biofilm formation by bringing P. aeruginosa cells closer to each other. Ultimately, these cells would become contiguous to form small aggregates. By carrying on mucus contraction, these aggregates would then fuse to each other. Since laboratory P. aeruginosa strains can differ substantially from strains infecting human individuals [48], we infected an AirGel with P. aeruginosa isolated from the bronchoalveolar lavage fluid of a CF patient. This strain colonized and contracted mucus within hours, in a manner similar to PAO1 (S9 Fig and S6 Video). To substantiate the physical contraction mechanism, we investigated how P. aeruginosa could restructure mucus during attachment and colonization.

P. aeruginosa forms biofilms on mucus using T4P

We envisioned 2 mechanisms for bacteria–induced mucus deformations: degradation or direct mechanical contraction. P. aeruginosa secretes mucinases capable of degrading gel–forming mucins [49]. Enzymatic mucus degradation could release polymers that generate entropic depletion forces promoting bacterial aggregation or that generate osmotic forces compressing mucus [14,50]. To test whether mucus degradation could drive contraction, we colonized AirGels with a mutant in the type II secretion system locus xcp that is necessary for mucin utilization [49,51]. The Δxcp mutant however formed biofilms similar to wild type (WT), ruling out the hypothesis of polymer–induced forces driving the formation of multicellular structures (Fig 5A).

Download:

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

Fig 5. Type IV pili retraction promotes mucus contraction.

(a) Biofilm formation of PAO1 mutants unable to degrade mucus or to generate force (t = 5.5 h). Both the Δxcp mutant (lacking type II secretion system necessary for secretion of mucinases) and the ΔfliC mutant (lacking flagella) form biofilms similar to WT. In contrast, the ΔpilA mutant lacking T4P was unable to form biofilms in AirGels. (b) Finite element simulations of mucus deformation during surface contraction. (i) Schematic representation of P. aeruginosa applying contractile force on mucus by retracting T4P. (ii) Finite element simulation of deformation of a mucus cylinder at rest (top) and under active surface stress (bottom). Colormap indicate displacement of surface elements. (iii) Relative mucus area as a function of surface contractile modulus. As the surface contractile modulus increases, the relative area of mucus decreases. (c) T4P retraction is necessary for biofilm formation. Images compare biofilms from WT P. aeruginosa and from the ΔpilT mutant unable to retract T4P (t = 5.5 h). ΔpilT cluster area and percentage of large clusters is significantly smaller than WT (N = 3). Statistics: independent samples Student t test with Bonferroni correction (p = 0.035 and p = 0.015). (d) Mucus does not rearrange during ΔpilT colonization (t = 5.5 h). Most of the mucus surface remains free of bacteria during ΔpilT colonization (N = 3). Statistics: independent samples Student t test with Bonferroni correction (p = 0.01). The data underlying this figure can be found in S5 Data. WT, wild type.

https://doi.org/10.1371/journal.pbio.3002209.g005

Could P. aeruginosa remodel mucus by directly and actively applying force on the surface? P. aeruginosa can generate extracellular forces using flagella and T4P, motorized filaments that also play a role during in vitro biofilm biogenesis. In addition, T4P and flagella mediate single–cell interactions with mucins [12,15,52–54]. To investigate their functions in the context of biofilm formation on mucus, we infected AirGels with P. aeruginosa mutants lacking flagella (ΔfliC) and T4P (ΔpilA). The ΔfliC mutant formed biofilms that were indistinguishable from WT (Fig 5A). By contrast, ΔpilA cells did not form multicellular structures, indicating T4P play a role in mucus–associated biofilm formation. Since T4P may bind to glycans present on mucins [53,54], weaker cell attachment to mucus could cause a decrease in aggregation of ΔpilA. Yet, colocalization shows that the ΔpilA mutant is still able to attach efficiently to mucus (S10 Fig and S7 Video). We therefore envisioned a mechanism where T4P generate retractile forces that contract mucus from the surface, ultimately speeding up P. aeruginosa aggregation and cluster fusion.

To physically explore this scenario, we ran nonlinear finite element simulations wherein mucus is treated as a hyperelastic material [55]. The mechanical action of P. aeruginosa T4P at the mucus surface is considered through the introduction of an active surface stress. The simulations recapitulated the experimental observations of mucus hydrogel contraction during P. aeruginosa colonization (Fig 5B). Simulations also predict that the steady–state mucus area decreases with the magnitude of the surface contractile modulus. This suggests that the more T4P retract, the more P. aeruginosa compresses mucus. To experimentally validate this model, we visualized AirGels colonization by a ΔpilT mutant that produces T4P that cannot retract, mimicking conditions of zero contractile modulus. P. aeruginosa ΔpilT could still associate with mucus and form a few small clusters, but clearly failed to form mucus–associated biofilms to the same extent as WT (Fig 5C and 5D), which was coherent with simulations. These results show that T4P retraction is necessary for biofilm formation on mucus and is consistent with a scenario where retraction compresses the mucus substrate.

To further support the surface contraction model, we tested the prediction that deformations increase further with surface contractility. We imaged AirGel colonization by the hyperpiliated P. aeruginosa mutant ΔpilH, whose T4P retraction frequency is approximately twice the one of WT (S11 Fig). ΔpilH formed biofilms more rapidly than WT: we observed dense aggregates as early as 2 h, while we only did after 4 h for WT (Fig 6A and 6B). In addition, ΔpilH induced more rapid mucus contraction than WT (Fig 6C and S8 and S9 Videos), consistent with simulations. After 5.5 h, WT and ΔpilH biofilms had similar morphologies and size, suggesting biofilm fusion reaches a physical limit most likely due to packing at the mucus surface. To control that the observed differences did not arise from growth rate variation between the strains, we quantified colony–forming units (CFUs) of WT, ΔpilT, and ΔpilH before infection and after 5 h of growth at the airway mucosa (S3 Table). All strains grew at indistinguishable rates.

Download:

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

Fig 6. Increased T4P retraction speeds up biofilm formation and mucus contraction.

(a) Increased T4P activity speeds up biofilm formation on mucus. Comparison of biofilm formation by the ΔpilH mutant with hyperactive T4P vs. WT, at t = 2 h and t = 5.5 h after inoculation. ΔpilH already forms small biofilms after 2 h. (b) (i) Kinetics of biofilm size for WT, ΔpilT, and ΔpilH. For each strain, we infected 3 AirGels from a healthy donor. Bars represent the range between the maximum and minimum of the means from triplicates, circles represent the overall mean. (ii) Comparison of percentage of large clusters for WT, ΔpilT, and ΔpilH, over time. Statistics: one–way ANOVA for each time point, followed by a post hoc Tukey test if the null hypothesis was rejected. At t = 4 h, the differences between WT and ΔpilH (p = 0.003) and between ΔpilH and ΔpilT (p = 0.001) were significant. At t = 5.5 h, the differences between WT and ΔpilT (p = 0.02) and between ΔpilH and ΔpilT (p = 0.001) were significant. (c) ΔpilH dramatically contracts mucus. (i) Timelapse images showing an event of mucus contraction by ΔpilH. (ii) Orthogonal views of the bacteria–covered mucus cluster at t₁ + 85 min, showing that PAO1 ΔpilH cells surround mucus. (iii) Relative mucus area changes measured during 30 min for WT and ΔpilH; since ΔpilH starts aggregating and remodeling mucus earlier than WT, the starting points of the recording differed (ΔpilH: 2.5 h, WT: 6.2 h, negative control: 8.1 h). Images are maximal intensity projections of z–stacks throughout the figure except for the orthogonal projection in H. The data underlying this figure can be found in S6 Data. WT, wild type.

https://doi.org/10.1371/journal.pbio.3002209.g006

Overall, our results support a model where P. aeruginosa contracts the surface of mucus by active T4P retraction. Single cells initially twitch on mucus to form small aggregates. The static aggregate collectives generate forces from T4P that are sufficient to deform their substrate, driving large–scale mucus contraction. By contracting, mucus brings aggregates closer. They eventually fuse and form biofilms (Fig 7).

Download:

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

Fig 7. Proposed model for the formation of mucus–associated biofilms by P. aeruginosa.

(i) Single bacterial cells attach to mucus. (ii) P. aeruginosa navigates the mucus surface using T4P–dependent twitching motility. (iii) T4P retraction locally contracts mucus. (iv) Surface exploration promotes encounters between single cells. This initiates aggregation and the formation of small clusters. These clusters remain static on mucus. As single cells in aggregates use T4P to pull on and contract mucus, they generate surface contractile forces. (v) Collective retraction of T4P from many cells compacts bacterial aggregates. (vi) Still under the action of retractile T4P forces, aggregates further contract mucus to initiate fusion into biofilms.

https://doi.org/10.1371/journal.pbio.3002209.g007

Ethics statement

No consent from the ethics committee was required for the clinical specimen as only anonymized health–related data was used.

AirGel chip fabrication

Cell culture

Live staining

To label mucus in live AirGels, we used jacalin conjugated to fluorescein (Vector Laboratories). We prepared a 50 μg/ml solution and loaded it in the lumen. We stored the chips for 30 min in a cell culture incubator before aspirating all fluid from the lumen. In addition, in order to assess epithelial integrity, we loaded a 4 kDa fluorescent dextran solution in the lumen of an 11–day–old chip and incubated it for 30 min. We then verified that all signal was localized in the lumen of the chip.

Immunofluorescence

All steps were performed at room temperature. First, differentiated cells in AirGels were fixed with either 4% paraformaldehyde (PFA, Electron Microscopy Sciences) or methacarn, when we wanted to better preserve extracellular mucus. Methacarn was made fresh before every use as follows: 1 part glacial acetic acid (Sigma–Aldrich), 3 parts chloroform (PanReac AppliChem), 6 parts anhydrous methanol (Sigma–Aldrich). Regardless of the chemical used, the fixation step lasted for 15 min. PFA–fixed cells were then permeabilized with a 0.2% Triton X–100 solution (VWR Life Science) for 20 min. Then, we exposed all cells (i.e., PFA–and methacarn–fixed) to a blocking solution consisting of 1% bovine serum albumin (Sigma–Aldrich) during 45 min. Afterwards, we added solutions of primary antibodies to each AirGel and incubated them for 1 h. In case of PFA–fixed cells, we used rabbit anti–MUC5AC (1:100, Abcam) and mouse anti–acetylated alpha tubulin (1:250, Sigma–Aldrich); for methacarn–fixed cells shown in Fig 2A, we used the same anti–MUC5AC, together with rabbit anti–MUC5B (1:100, Sigma–Aldrich); for methacarn–fixed cells shown in S2 Fig, we replaced the rabbit anti–MUC5AC antibody with mouse anti–MUC5AC (1:100, Sigma–Aldrich). After incubation with the primary antibodies, we labeled the samples with secondary antibodies during 1 h in the dark. More specifically, we used goat anti–rabbit IgG H&L Alexa Fluor 488 (1:200, Abcam) and goat anti–mouse IgG H&L Alexa Fluor 594 (1:200, Thermo Fisher). Finally, nuclei were counterstained for 10 min with DAPI (1:1,000, Sigma–Aldrich); in addition, in PFA–fixed cells, actin was stained with Phalloidin Atto 655 (1:40, Sigma–Aldrich) for 10 min.

Sample preparation before lightsheet imaging

To perform lightsheet microscopy on AirGels, we needed to extract the ECM gel and cells from the PDMS chip. After fixation and staining, we filled in the lumen with a 1% low–melt agarose solution in order to ensure structural integrity of the airway. We let it solidify; using tweezers, we could then carefully detach the PDMS from the glass; indeed, since the surface of the 3D printed mold was not perfectly smooth, plasma bonding was not irreversible, which we could leverage for ECM extraction. We then used a scalpel and a spatula to release the piece of ECM from the PDMS chip and later embedded it in 1% low–melt agarose. While the agarose was still liquid, we aspirated the whole gel into a 1–ml syringe (Omnifix–F, B. Braun), whose tip had previously been cut out. After the agarose solidified, we could then use the plunger to freely push the fixed AirGel in and out of the syringe, in order to image it with selective plane illumination microscope (SPIM).

Microscopy

To image AirGels at low magnification (Fig 1C and 1D), we used a Nikon TiE epifluorescence microscope equipped with a Hamamatsu ORCA Flash 4 camera and either a 10× objective with N.A. of 0.25 or a 4× objective with N.A. of 0.1. For full channel cross–sectional imaging (Fig 1E), we used a Zeiss Lightsheet Z1 dual–sided SPIM. It was equipped with PCO Edge 5.5 cameras and a 5× magnification objective with N.A. of 0.16. All the other visualizations were acquired with a Nikon Eclipse Ti2–E inverted microscope coupled with a Yokogawa CSU W2 confocal spinning disk unit and equipped with a Prime 95B sCMOS camera (Photometrics). We either used a 20× water immersion objective with N.A. of 0.95 or a 40× water immersion objective with N.A. of 1.15. We used Imaris (Bitplane) for 3D rendering of lightsheet z–stack pictures and Fiji for the display of all the other images [66].

Single–cell RNA–seq

Quantification of cilia beating frequency (CBF)

We filled the lumens of AirGels with a 1:500 solution of yellow–green carboxylated fluorescent beads with a 2 μm diameter (FluoSpheres, Life Technologies). We incubated them for 1 h; in this time interval, despite the flow generated by ciliary beating, some beads were able to settle down and attach to cilia. We then removed all fluid from the lumen and brought the chips to the spinning disk confocal microscope. We then selected small regions of interest (50 × 50 pixels, i.e., 27.5 μm × 27.5 μm) around individual beads and recorded videos at 100 frames per second. Using Matlab R2016b (Mathworks), we computed the mean intensity of each frame over time. We computed the fast Fourier transform of each mean intensity signal, which we then used to obtain single–sided power spectra. We only kept frequencies between 1 and 30 Hz, thereby getting rid of artifacts. We finally looked for the frequency with maximal amplitude in the power spectrum, which corresponded to the CBF.

Quantification of mucociliary clearance

Like for CBF quantification, we loaded a 1:500 solution of 2 μm FluoSpheres in the lumen of AirGels (or at the surface of Transwells, for S3 Fig). We immediately visualized them with the spinning disk confocal microscope. We recorded 10 s videos at a rate of 10 frames per second. Then, we tracked the trajectory of individual beads with the Fiji plugin TrackMate [70,71], using the built–in simple LAP tracker. We wrote a script in a Jupyter Notebook to compute the velocity (track displacement over track duration) of each particle [72].

Bacterial strains, plasmids, and culture conditions

We used P. aeruginosa PAO1 (WT or mutants, listed in S1 Table) for all the infection experiments. To obtain the clinical strain AP1889, bronchoalveolar lavage fluid of a CF patient was sent to the routine lab of the Institute of Medical Microbiology in Zurich, Switzerland. The clinical sample was cultured on different media types and P. aeruginosa was isolated from a selective Mac Conkey agar (Biomérieux, France) after incubation for 18 h at 37°C. Species identification was performed by MALDI–TOF MS biotyper (Bruker Daltonics, Germany) as described elsewhere [73]. No consent from the ethics committee was required for the clinical specimen as only anonymized health–related data was used. Most strains were made to constitutively express the fluorescent protein mScarlet following a published protocol using the plasmids listed in S2 Table [74]. The backbone plasmid pUC18t–Mini Tn7 with gentamycin resistance was purchased from Addgene and isolated from E. coli XL10 Gold by GeneJET Plasmid Miniprep Kit (Thermo Fisher). The isolated plasmid was digested with the restriction enzymes HindIII and BamHI. The Ptet promoter was amplified by PCR using P. aeruginosa PAO1 genomic DNA and the mScarlet gene was amplified from a preexisting plasmid. The Ptet promoter and mScarlet was then fused via Fusion PCR by overlapping extension. The resulting extended product was digested with HindIII and BamHI, and then ligated to the digested pUC18t–MiniTn7 Gm backbone. Since this plasmid included a gentamycin resistance cassette, we grew the fluorescent PAO1 strains overnight in LB medium with 30 μg/ml gentamycin. The next morning, we diluted the stationary cultures 1:1,000 in plain LB and let them grow 3 to 4 h before infecting AirGels.

Infection of AirGels

The night before infection, AirGels were stained with the plasma membrane dye CellMask Deep Red (Life Technologies). Aside from the infection assay shown in Fig 3A and S2 Video, which was performed in a CF AirGels, all infections were run with cells from healthy donors. The dye was diluted to 5 μg/ml and loaded in both the apical and basal compartments. The next morning, the lumen was again exposed to air for 3 to 4 h. Mucus was stained with jacalin as described above, and all luminal fluid was then aspirated. Finally, we infected AirGels with mScarlet P. aeruginosa. We measured the optical density of our exponential bacterial cultures and centrifuged them for 2 to 3 min at 5,000 rpm. We discarded the supernatant and resuspended the pellet in D–PBS to reach an optical density value of approximately 3. We then loaded 0.5 μl of bacterial culture in the lumen of AirGels (this small volume allowed for ALI maintenance). The resulting multiplicity of infection was approximately 10 (approximately 10⁶ bacteria for approximately 10⁵ airway cells). For the infection shown in Fig 3A, we started with a stationary P. aeruginosa culture that we diluted in D–PBS to an optical density of approximately 0.035. We then dipped a sterile toothpick in the culture and lightly touched the inlet of an AirGel with it in order to deposit bacteria while maintaining the ALI. This second method may mechanically compromise the epithelium with the toothpick; we therefore opted for the first one in most experiments.

The chips were then placed in an UNO–T–H–CO2 stage–top incubator (Okolab) for temperature, humidity, and CO₂ control. The environment was maintained at 37°C and 5% CO₂ and connected to a bottle of Milli–Q water for humidification. Since condensation frequently appears on the PDMS chip during imaging, we placed pieces of Kimtech Science Kimwipes (Kimberly–Clark Professional) in the inlet ports of AirGels; this prevented dripping water from disrupting the ALI conditions. We visualized the infection progress over time with the aforementioned spinning disk confocal microscope. For WT, ΔpilT, and ΔpilH, we repeated the infections to reach N = 3 replicates per condition. The AirGels for all 3 replicates were all made from the same healthy donor and were between 33– and 38–day–old at the time of infection.

Quantification of bacterial growth on airway mucus

To compare the growth rates of the WT, ΔpilT, and ΔpilH strains on mucus, we grew all 3 strains and concentrated them in D–PBS as indicated above to reach a starting optical density of 3. We loaded 0.75 μl of the resulting culture into NHBE cells growing at the ALI on Transwell membranes (Corning) with a 0.4 μm pore size. We infected N = 3 wells with each strain. In parallel, we also plated the inoculum to count CFUs. After 5 h incubation of the Transwells in a cell culture incubator (set to 37°C, 5% CO₂), we dissociated the infected airway tissue using a 0.1% Triton X–100 (VWR Life Science) solution and mechanical disruption by pipetting and vortexing vigorously. We plated the extracted bacteria for a CFU count on LB–agar and incubated overnight at room temperature. We finally calculated the number of generations n_gen as follows:

Colonization of extracted mucus

We isolated mucus from 8.5–month–old NHBE cells grown on 0.4 μm pore size polyester Transwell membranes (Corning). To do so, we immersed the apical side of the membranes in a jacalin–fluorescein solution (50 μg/ml in D–PBS) and we placed them in a cell culture incubator for 30 min. We then collected all fluid from the apical side with a pipette and dispensed 12.5 μl into 4 mm PDMS gaskets bonded to a glass–bottom dish (1.5 coverslip, glass diameter 20 mm, MatTek). We filled the space around the PDMS gasket with D–PBS to prevent dehydration of the mucus. We then centrifuged an exponential P. aeruginosa mScarlet culture at 5,000 rpm for 3 min and resuspended them in D–PBS before loading 15 μl on the labeled mucus. We then placed the dish in the stage–top incubator and imaged the bacteria and mucus with the confocal spinning disk microscope described above.

Twitching motility on mucus

We infected a 4–month–old HBE Transwell with P. aeruginosa mScarlet as follows. We loaded 3.3 μl of early exponential culture (approximately 10⁵ CFUs) on the apical side of the HBE culture, which had been labeled with fluorescent jacalin. We carefully took the Transwell insert out of the cell culture plate using sterile tweezers and we placed it on a glass–bottom dish (1.5 coverslip, MatTek). We then placed the dish in a stage–top incubator and recorded time–lapse videos of twitching bacteria with our spinning–disk confocal microscope. Because of the lack of culture medium in the visualization setup, recordings could not last long and would dehydrate within minutes.

Biofilm image acquisition and analysis

We acquired z–stack of infected AirGels over a 35–μm deep range at different time points (t = 0 h, 2 h, 4 h, and 5.5 h ± 0.5 h). All the image analysis steps were done in Jupyter Notebooks [72].

Since the AirGel surface is curved, for all subsequent steps, we projected images in 2D using the maximal intensity projection tool in Fiji in order to facilitate downstream analysis. We started by quantifying the sizes of bacterial clusters. First, we visually inspected the pictures: if there were large intensity variations (e.g., in case of a mix of dim single cells and bright clusters), we saturated bright pixels to 1.5 times the mean intensity of the picture. We then segmented the pictures using Otsu thresholding (from the “opencv” Python package [75], version 4.5.4.60) and visually assessed the result. In the rare cases where the segmentation was not deemed satisfactory (i.e., if some features were not detected properly or if there was too much noise), a simple threshold was manually selected. The pictures were then closed and filtered; more specifically, we removed any object smaller than approximately 6 μm² (20 pixels), which approximately corresponds to the area of a single cell. We then obtained the area of each cluster using the function “regionprops” (“scikit–mage” Python package [76], version 0.19.2), which calculates properties of segmented objects in binary pictures. We calculated the mean cluster area for each replicate; then, for each condition, we plotted the maximum, minimum, and mean of the means (e.g., Fig 3B). We also computed and plotted the proportion of aggregates larger than 100 μm² (S4 Fig).

We then quantified colocalization between mucus and bacteria. The segmentation and filtration of mucus pictures was identical as for bacterial clusters. Then, using the logical “&” function, we identified the pixels that were common between the binary pictures from the bacterial and mucus channels. With “regionprops,” we obtained the areas of these common zones and we normalized them to the total area of mucus. Thus, we could find the proportion of mucus that was covered in bacteria. We finally calculated the proportion of mucus devoid of bacteria as follows: 1 –(proportion of mucus covered in bacteria).

To quantify the contraction of a patch of mucus, we first canceled the effects of drift by registering the images in Fiji using the “Correct 3D drift” plugin. We then manually tracked the displacement of N = 7 reference features with the Fiji plugin “Manual Tracking.” We loaded the trajectories in a Jupyter Notebook and calculated the distances between each pair of positions over time. We finally normalized the resulting data to the initial distances and plotted them, along with the mean and standard deviation at each time point (Fig 4D).

Finally, we also measured mucus shrinkage over time for WT, ΔpilH, and an uninfected control AirGel. To do so, we used images from 30 min timelapses (the starting point of the timelapses differed: 6 h 10 min for WT, 2 h 30 min for ΔpilH, and 8 h 5 min for the negative control). We segmented and quantified mucus areas as described above for each time point, and normalized it to the initial area (Fig 6C and S8 Video).

iSCAT–based quantification of type IV pili retraction frequency

P. aeruginosa were grown as previously described [77]. Briefly, an overnight culture was obtained from a single colony and grown in LB at 37°C with 290 rpm shaking. The overnight culture was diluted 1:500 or 1:1,000 and grown for 2 to 3 h to obtain a mid–exponential phase culture. For surface–grown cells, 100 μl of the mid–exponential phase cell suspension were plated on LB 1% agarose, grown for 3 h at 37°C and harvested in 500 μl LB by gentle scraping. Cells were diluted to OD600 0.02 to 0.05 prior to visualization. Both liquid–and solid–grown cells were either loaded on 500 μm × 140 μm PDMS microchannels or in 6 mm PDMS gaskets. Cells sticking to the surface were visualized without flow with iSCAT and movies were recorded at 10 fps for either 2 min, 1 min, or 30 s. Raw iSCAT images were processed as described previously [59,77]. Individual movies were manually analyzed using Fiji [66] by counting the total number of TFP in each cell as well as the number of TFP retractions represented by tensed TFP. The residence time of each cell on the coverslip was also recorded. For each cell, we computed the retraction frequency by dividing the total number of retractions by the residence time of the cell on the coverslip. Finally, we computed a bootstrap median retraction frequency and 95% confidence interval by pooling the data obtained by all 2 to 3 biological replicates. Data analysis was performed using Matlab R2020a (Mathworks).

Statistical analysis

All statistical tests were run in Python using Jupyter Notebooks [72]. Independent or paired–samples Student t tests were performed with Bonferroni correction using the function “add_stat_annotation” from the statannot package [78] (version 0.2.3). One–way ANOVAs were run using the function “f_oneway” in the “stats” module from SciPy [79] (version 1.7.3). When the ANOVA result rejected the null hypothesis, we followed up with a post hoc Tukey test using “stats.multicomp.pairwise_tukeyhsd” from the “statsmodels” module [80].

Computational model of mucus remodeling by T4P

We refer the reader to our previous work [55,81] for the general theory on the kinematics of the surface and the volume of a 3D soft body, focusing on cell–laden microtissues. Specific considerations on the implementation of this work are introduced in the following formulations.

1. Kinematics

Let V be a fixed reference configuration of a continuum body ℬ. We use the notation χ: V→ℝ³ for the deformation of body ℬ. A motion χ is the vector field of the mapping x = χ(X), of a material point in the reference configuration X∈V to a position in the deformed configuration x∈v. The kinematics of a material point are described by the following: (S1) where u(X, t) is the displacement vector field in the spatial description. The kinematics of an infinitesimal bulk element are described by the following: (S2) (S3) where F(X, t) and F⁻¹(x, t) are the deformation gradient and inverse deformation gradient, respectively. Note that J(X, t) = dv/dV= det F(X, t) is the Jacobian determinant defining the ratio of a volume element between material and spatial configuration.

A motion of an arbitrary differential vector element can be mapped by the deformation gradient F. However, a unit normal vector N in the reference configuration cannot be transformed into a unit normal vector n in the current configuration via the deformation gradient [82], motivating us to develop the kinematics of an infinitesimal surface element [83]. Note that we utilize to denote the surface quantity bounded by outer surface denoted as ∂Ω₀. (S5) (S6) where and are the deformation gradient and inverse deformation gradient, respectively. Note that and are the mixed surface unit tensors, where I and i are the unit tensors, and N and n are the outward unit normal vectors, in reference and current configuration, respectively. Note that is the Jacobian determinant defining the ratio of an area element between material and spatial configuration.

2. Equilibrium

The total potential energy functional W(χ) is defined as: (S9) where Ψ and are strain energies in bulk and on surface, and B is the reference body force and T is the surface traction. An equilibrated configuration is obtained by minimizing this functional considering all admissible deformations δχ. It is important to note that the strain energies (Ψ, ) can be varied depending on the bacterium and mucus models so that the following sections can be written in a single formulation for brevity, and the specific strain energies are to be defined in section 4.

Following the derivation presented in [84], we can finally arrive at a set of localized force balance equations. Neglecting the inertial effect, the local form of linear and angular momentum balances for bulk and surface are defined by the following: (S10) (S11) (S12) where is the prescribed displacement on the boundary S_u, is the bi–normal vector to the boundary curve, and ⟦•⟧ indicates summation over surfaces intersecting on boundary curves [83].

3. Weak form

For the finite element implementation, we need to obtain the weak form for our problem. By adding the constraint that the first variation of the total potential energy must be equal to zero δW(χ) = 0, we obtain a weak form statement as (S13) where δu is the admissible deformation field.

We employed the open–source platform FEniCS [85], to implement the finite element simulation. We used the Scalable Nonlinear Equations Solvers (SNES) from the open–source toolkit PETSc [86], which provides numerical computations of a Newton–type iterative procedure to solve the nonlinear variational problem. Note that the value of γ should be ramped from zero to its prescribed value for numerical stability as the problem is highly nonlinear.

4. Constitutive relations

To relate the active stress with deformation, we must specialize our choice for the strain energies in the bulk and on the surface. For the deformation of compressible tissue, we consider a passive bulk energy Ψ^p that captures the permanent elasticity of the collagen network, and for the contribution of bacterium contractility, we can consider the active surface energy that accounts for the action of the bacterium on the surface of mucus tissue.

4.3. Energy penalization.

As at high level of contraction the bacteria are bound to jam, barring the additional contraction of the mucus (even if the material itself can accommodate it), we have to enforce this jamming transition. Assuming that we know the initial surface concentration of bacteria, we enforce the kinematic constraint via energy penalization. From the experimental observation, we enforce the surface area of deformed mucus tissue cannot be smaller than a ratio () of the initial surface area. An appropriate energy penalization for enforcing the prescribed surface condition is given by (S16) where P is the penalty parameter (energy per unit area), and 〈•〉 is the Macaulay brackets that used to describe the ramp function.

(S17)

5. Finite element simulation

The reference (undeformed) state corresponds to a state where the active contractile moduli are set to zero. Experimentally, this reference state corresponds to the initial state of the mucus right after the mixing of mucus and bacterium and before the application of forces by encapsulated bacterium. The reference configuration for the finite element simulations represents the geometry shown in Fig 5B. The entire surface is allowed to actively contract through increasing the surface contractile modulus up to an equilibrium value. The final (deformed) state is defined when the surface contractile moduli γ reaches its prescribed value, and no external loads are applied. Experimentally, this corresponds to the equilibrium state of the mucus. The final configurations represent the equilibrium states.

6. Parameter calibration

The parameters of the model are the bulk and shear moduli, K and G, and surface contractile modulus, γ, penalty parameter, P and penalty surface ratio, . There is a unique relationship between K, G, and Poisson’s ratio ν, allowing to interchangeably use ν in place of K for the calibration procedure. We selected a set of parameters: G = 1.0 Pa, ν = 0.1, γ = 0.03 nN/μm and P = 1.0 nN/μm. The corresponding bulk and elastic moduli were K = 0.9 Pa and E = 2.2 Pa within the range of reported values for mucus [87,88].