Dynamic single cell analysis in a proximal-tubule-on-chip reveals heterogeneous epithelial colonization strategies of uropathogenic Escherichia coli under shear stress

Abstract The urinary tract is a hydrodynamically challenging microenvironment and uropathogenic Escherichia coli (UPEC) must overcome several physiological challenges in order to adhere and establish a urinary tract infection. Our previous work in vivo revealed a synergy between different UPEC adhesion organelles, which facilitated effective colonization of the renal proximal tubule. To allow high-resolution real-time analysis of this colonization behavior, we established a biomimetic proximal-tubule-on-chip (PToC). The PToC allowed for single-cell resolution analysis of the first stages of bacterial interaction with host epithelial cells, under physiological flow. Time-lapse microscopy and single-cell trajectory analysis in the PToC revealed that while the majority of UPEC moved directly through the system, a minority population initiated heterogeneous adhesion, identified as either rolling or bound. Adhesion was predominantly transient and mediated by P pili at the earliest time-points. These bound bacteria initiated a founder population which rapidly divided, leading to 3D microcolonies. Within the first hours, the microcolonies did not express extracellular curli matrix, but rather were dependent on Type 1 fimbriae as the key element in the microcolony structure. Collectively, our results show the application of Organ-on-chip technology to address bacterial adhesion behaviors, demonstrating a well-orchestrated interplay and redundancy between adhesion organelles that enables UPEC to form microcolonies and persist under physiological shear stress.


Introduction
The ability of bacteria to adhere to tissue is essential to establishing infection. Bacteria attachment is heavily influenced by the physiology of the infection site and in many sites shear stress from fluid flow is an important factor. In the urinary tract, shear stress varies in the kidney as the body regulates renal function and changes dramatically upon bladder voiding. Our group has previously used a spatio-temporally controlled in vivo model of kidney infection to study the early stages of kidney colonization by uropathogenic Escherichia coli (UPEC) (Månsson et al. 2007). We identified a synergistic relationship between the important UPEC adhesion organelles P and Type 1 fimbriae (Melican et al. 2011). P fimbriae-mediated colonization of the proximal tubule epithelium during the first hours of infection under high shear stress, whereas Type 1 fimbriae enabled bacteria to colonize the tubule's lumen away from the epithelium. Supported by previous data, this led us to hypothesize a major role for Type 1 fimbriae in mediating interbacterial binding and a "biofilm-like" lifestyle (Justice et al. 2004, Melican et al. 2011, Conover et al. 2016. Neither fim-briae were found to be essential to infection, but in vivo they influenced the kinetics of infection progression (Melican et al. 2011).
The complexity of in vivo models has prevented more mechanistic evaluation of this fimbrial synergy. To address this, we evaluated the usability of in vitro microfluidic technology known as "Organ-on-chip." Organ-on-chip models are designed to recapitulate the physiology and/or the structure at organ and tissue level. Importantly for our work, these models include shear stress, which is often lacking from in vitro assays and is critical to bacterial binding in vivo. This technology has been widely adapted as biomimics of organs such as the liver, kidney, heart, lung, and intestine, but studies on host-pathogen interactions are relatively scarce.
In this work, we establish a "Proximal-tubule-on-chip" (PToC) platform to evaluate the initial binding and colonization characteristics of UPEC to renal epithelial cells under physiological shear stress conditions. We employ single cell tracking to describe distinct initial adhesion behaviors of UPEC in the PToC. We define the role for PapG in mediating initial adhesion behavior and ad-hesion duration. We show how FimH facilitates the formation of extensive 3D microcolonies, which are resistant to flow. These microcolonies do not express extracellular matrix (ECM) in the early hours of infection but rather are dependent on Type 1 fimbriaemediated interbacterial binding. This work uses the PToC to describe the mechanistic and synergistic roles of these two important fimbriae in the initial colonization strategy of UPEC to renal epithelium, and demonstrates how they lead to effective epithelial colonization and microcolony formation during the first hours of infection under flow.

Stationary and rolling adhesion of UPEC under shear stress in a PToC
To study UPEC-host interactions under physiologically relevant conditions, we developed a PToC. This design features six microchannels, seeded with human renal epithelial cells (A498) (Fig. 1A). To mimic the conditions of the renal tubule in the microchannels, we mimicked the shear stress generated by the primary filtrate flow in the S1 proximal tubule, estimated to be 0.104 dyn/cm 2 (Essig and Friedlander 2003). Renal epithelial cells were subjected to a flow of 75 μl/min of CO 2 independent media (CO 2 IM). Under these conditions the cells exhibited healthy morphology, confluence, and proliferation, with a viability of 91.4 ± 6.05%, comparable to a 95.5 ± 2.9% viability under static conditions ( Fig. 1B and C). To investigate how UPEC adheres to renal cells under shear stress, we used the UPEC strain CFT073 (wt) (Table S1, Supporting Information), which we confirmed expressed Type 1 and P fimbriae in CO 2 IM (Table S2, Supporting Information). Under flow, 0.5 ml of CFT073 (≈ 2 × 10 8 cfu) was slowly infused via the inlet, over 1 min. To capture the modes of initial bacterial adhesion to renal cells, we performed time-lapse microscopy for 10 s at a high frame rate. We observed that the majority of bacteria were rapidly expelled by the flow, without interacting with cells, while other bacteria resisted the flow and were displaced at a lower speed (Movie 1).
To analyse the behaviors of these bacteria, which resisted initial clearance, we performed single-cell trajectory analysis (SCTA) on this population ( Fig. 2A and B). The position of individual bacteria was tracked between sequential frames. The majority of bacteria were unbound showing an almost constant displacement with the flow (Fig. 2C and D). Plotting their displacement over time showed a linear relationship with their average velocity exceeding 30 μm/s ( Fig. 2D; Movie 2, bacteria in green circle). Despite the flow, a minority population did manage to interact with the epithelial cells, either rolling or bound. Rolling bacteria exhibited fluctuations in displacement, resulting in a less linear relationship with time ( Fig. 2E and F). Their average velocity did not exceed 30 μm/s ( Fig. 2F; Movie 2, bacteria in blue circle), suggesting that these bacteria established moderate binding to cells, attenuating their displacement. The term rolling was chosen to correlate to previous studies describing a similar mode of adhesion on abiotic surfaces (Thomas et al. 2004). A few bacteria managed to strongly resist displacement by the flow (Fig. 2G and H). These bacteria that exhibited stationary adhesion over time were designated as bound ( Fig. 2G and J; Movie 2, bacterium in red circle). Their relationship between displacement and time was nonlinear, and their average velocity ranged from 0 to 30 μm/s. These phenotypes were not static. We observed unbound bacteria that abruptly attached to a cell (Fig. 2H) and attached bacteria that abruptly detached (Fig. 2I). To obtain a better overview of initial adhesion, we plotted mean velocity against displacement linearity (Fig. 2K). The majority of bacteria were unbound (R 2 ≥ 0.95, mean velocity > 30 μm/s) with the remaining minority either rolling (R 2 > 0.95, mean velocity ≤ 30 μm/s) or bound (R 2 < 0.95, mean velocity ≤ 30 μm/s). This data showed that under shear stress, only a subset of bacteria successfully adhered to renal cells upon initial exposure. Adhesion could be either weak, allowing bacteria to roll on the cell surface, or strong, enabling bacteria to immobilize themselves on the cell.

UPEC's initial adhesion under shear stress is transient, and strengthened by the P fimbriae tip adhesin PapG
With the PToC established, we wanted to understand how rolling and bound bacteria could serve as initiators for colonization. After introduction of bacteria into the PToC, it took ∼15 min for the full inoculum to flow through the microchannel. Over this period, several bacteria initiated adhesion. To quantify binding duration, we selected an area of interest (≈ 2 × 10 4 μm 2 ) and tracked for 1 h all bacteria with a duration of adhesion ≥ 1 min (Movie 3). During the 15 min inoculum infusion, we observed a high number of bacterial adhesion events, followed by numerous detachment events (Fig. 3A). After the inoculation period, adhesion and detachment events stabilized for the remaining 45 min. By calculating the total number of adherent bacteria at any given time point (net bacterial adhesion), we confirmed that a number of bacteria maintained adhesion after the inflow of the inoculum (Fig. 3B). Measuring the binding duration showed that a large proportion of bacteria managed only short-lived adhesion (Fig. 3C), 50.2 ±11.9% of bacteria adhered only for 1-2 min, while 42.8 ± 8.2% and 7 ± 4.8% adhered for 2.5-30 min and > 30 min, respectively (Fig. 3D). This data shows the huge variability in bacterial adhesion and that there is a rapid turnover of adherent bacteria. Despite these fluctuations, a small number of bacteria consolidated strong adhesion to initiate a founder population (Fig. 3E). P fimbriae is known to be a key factor in UPEC binding to renal epithelial cells and therefore we tested the role of this adhesin in the PToC. We constructed an isogenic mutant of CFT073, which lacked the PapG tip adhesin of the P fimbriae, CFT073 papG papG_2, hereafter referred to as papG (Table S1, Supporting Information). The papG strain was inoculated into the PToC and followed for the first hour. During the inoculation phase, adhesion events were followed by an equal number of detachment events (Fig. 3F). Thereafter, net bacterial adhesion demonstrated a low but stable number of adherent bacteria during the first hour of infection even in the absence of PapG (Fig. 3G). The frequency distribution of binding duration showed, however, that most bacteria were transiently adhering for only 1-2 min (Fig. 3H). We found that 85 ± 1.6% of papG adhered for 1-2 min, a significantly higher percentage (P-value = .0347) compared to wt (Fig. 3I). Only 12 ± 4.7% of the bacteria adhered for 2.5-30 min, significantly less than wt (P-value = .0092). Only 3 ± 3.9% of tracked papG bacteria attached for > 30 min. Collectively, this data shows that the presence of PapG contributed to the binding duration of CFT073 in the PToC, and that bacteria unable to express PapG had a reduced ability to mediate long-term adhesion under physiological shear stress (Fig. 3J).

Rapid proliferation and microcolony formation facilitate UPEC establishment of infection under shear stress
During the first hour of infection in the PToC, we observed a rapid turnover of adherent bacteria. Attachment and detachment events decreased after the initial inoculum had passed through the channel, but thereafter remained stable. As there was no more inflow of bacteria, this suggested that new attachment events must originate from bacterial multiplication in the channel. We analysed time-lapse recordings from CFT073 wt and observed active division in adherent bacteria (Movie 3). This is exemplified in Fig. 4(A) and Movie 4, where a bacterium attached at the initiation of infection divided seven times. After each division, one daughter cell remained adhered, while the other was released. To measure mean bacterial generation time, we tracked multiple dividing bacteria and found a mean generation time of ≈ 20 ± 3.1 min (Fig. 4B), demonstrating that adherent bacteria were rapidly dividing in the PToC. This data suggested that attached bacteria serve as seeding points, releasing newly produced daughter cells. More importantly, by dividing every ≈ 20 min, bacteria partly overcome the transient nature of their adhesion. Bacteria bound for > 20 min have sufficient time to divide and shed new bacteria in the microchannel.
Occasionally, both daughter cells remained attached after division. These bacteria rapidly established prominent microcolonies within the first hour of infection ( Fig. 5A; Movie 5). Time-lapse analysis showed that microcolonies expanded threedimensionally; the bottom layer of bacteria was directly attached to the host cell, and the top layers of bacteria extending into the microchannel ( Fig. 5A and Figure S1a-c, Supporting Information; Movie 5). The rapidly proliferating bacteria in microcolonies had a high shedding rate (Fig. 5A, purple arrow), with an increasing number of bacteria detected in the flow through of the PToC upon the development of microcolonies ( Figure S2, Supporting Information). Unexpectedly, bacterial microcolonies exhibited prolonged binding to host cells (> 1 h), compared to single bacteria despite the shear stress ( Fig. 5A and Figure S1a-c, Supporting Information). We observed several microcolonies that alternated between attachment and detachment, as they were displaced by the flow (Figure S1d, Supporting Information; Movie 5). By 3-4 h, the majority of host cells were colonized by multiple microcolonies, while other cells had only a few single bacteria attached, demonstrating a nonuniformity of colonization ( Figure S3a, Supporting Information). Bacteria in mature microcolonies were oriented both horizontally and vertically to the cell surface (Fig. 5B).
Bacterial microcolonies typically remained attached to renal epithelial cells for more than an hour. As we had found that PapG enhances binding strength and prolongs single bacteria's adhesion to renal cells, we investigated how this adhesin would affect microcolony adhesion. Time-lapse videos of infection with the papG strain, unable to express PapG, showed that a minority of bacteria adhered to cells and proliferated with an optimal generation time of ≈ 20 min, similar to wt infection ( Figure S4, Supporting Information). Daughter cells were either released in the microchannel or remained attached, eventually establishing microcolonies ( Fig. 5C). Microcolonies of papG eventually reached a size similar to microcolonies of wt bacteria (Figs 5C and 3H; 30 min). The colonization pattern was also similar to wt infection, with large microcolonies grown on some host cells and scarce colonization on others ( Figure S3b, Supporting Information). Adhesion of the papG microcolonies was significantly longer compared to the brief binding of single papG bacteria, with microcolonies remaining attached for > 1 h. These results show that the lack of PapG did not prevent bacteria from establishing microcolonies. Microcolony formation compensated for the short-lived binding of papG bacteria and enabled them to withstand shear stress and eventually colonize renal cells in a similar pattern to wt bacteria.
Overall, this data shows that the strong initial adhesion combined with rapid proliferation of single bacteria led to UPEC microcolony formation under shear stress. Once part of a microcolony, bacteria exhibited enhanced resistance to shear stress compared with single bacteria and firmly established colonization of renal cells. Simultaneously, these expanding microcolonies shed multiple new bacteria in the microchannel, enhancing the ability of the bacterial population to colonize.

Interbacterial binding mediated by FimH enables UPEC to form microcolonies on renal cells under shear stress
Since P fimbriae were not shown to be critical to microcolony formation, we turned our attention to the Type 1 fimbriae, as our previous in vivo work had shown that the tip adhesion of Type 1 fimbriae, FimH, is involved in interbacterial binding (Melican et al. 2011). We constructed CFT073 fimH, an isogenic mutant lacking the FimH tip adhesin, hereafter referred to as fimH (Table S1, Supporting Information). We infused fimH into the PToC and monitored infection with time-lapse microscopy as previously. The bacteria's distribution of binding duration was comparable to wt and adherent bacteria proliferated with a generation time of ≈ 20 min ( Figures S4 and S5, Supporting Information). Daughter cells either remained attached or were shed into the microchannel (Movie 6). Cellular colonization by fimH lacked, however, microcolony formation but rather occurred as a single layer always directly bound to the cell surface ( Fig. 5D; Movie 6). By 6.5 h, the majority of the cell surface was colonized by a bacterial monolayer ( Figure S3c, Supporting Information). Occasional microcolonies could be observed in fimH but they were rare. Complementation of the CFT073 fimH with a constructed pBAD-fimH restored the ability to form microcolonies ( Figure S6a and Tables S1 and S2, Supporting Information) and addition of Dmannose into the PToC, masking FimH binding, prevented CFT073 wt from forming microcolonies ( Figure S6b, Supporting Information). Collectively, these data further strengthen the role of the FimH adhesin in microcolony formation. Together this indicates that FimH facilitates the formation of microcolonies and the colonization of bacteria in areas not in direct contact with the host cell surface. Although bacterial proliferation is rapid, a difficulty in forming microcolonies delays fimH colonization, leading to a different colonization pattern.

The role of ECM in UPEC microcolony formation
Our earlier in vivo kidney work, and others' work on the bladder had suggested that UPEC adopted a "biofilm-like" lifestyle in vivo (Justice et al. 2004, Melican et al. 2011, Conover et al. 2016, Ho et al. 2020. A biofilm lifestyle is often associated with the production of a bacterial ECM composed of biopolymers such as amyloid curli fibres (ECM-curli) and cellulose, the former with welldocumented roles in promoting adhesion to biotic and abiotic surfaces (Romling et al. 1998, Bokranz et al. 2005, Flemming and Wingender 2010, DeBenedictis et al. 2016, Luna-Pineda et al. 2019. To determine if UPEC microcolonies contained ECM-curli, we used optotracing technology (Choong et al. 2016(Choong et al. , 2021. Following manufacturers instructions, CFT073 biofilms were grown for 4 days in static culture, in the presence of EbbaBiolight 680 (Ebba680), a nontoxic, optically active fluorescent tracer molecule that reports bacterial production of ECM materials (Choong et al. 2016, 2021, Antypas et al. 2018. Visual analysis of UPEC microcolonies showed the presence of ECM-curli detected by Ebba680 (Fig. 6A  and B). This indicates that UPEC microcolonies do express ECMcurli when grown in conditions favorable to biofilm formation.
To determine if ECM-curli fibers were involved in the UPEC microcolony formation in the PToC in the first hours of infection, we added Ebba680 into the flow-through media of the PToC and imaged bacterial colonization over 3.5 h using fluorescence microscopy. While microcolonies formed as observed in previous experiment, no fluorescent signal indicative of ECM-curli was detected over the time course of the experiment (Fig. 6C). The same result was obtained for the fimH strain (Fig. 6D). Collectively, this data indicates that while UPEC can form extensive microcolonies under flow, ECM-curli does not appear to play a role in the 3D microcolony formation at this very early time-frame. Together with our earlier data, this would indicate that the fitness factor enabling microcolony formation at this early stage, was Type 1 fimbriae.

Discussion
By developing the PToC assay for UPEC colonization, we have enabeled a quantitive analysis of the role of the UPEC adhesins at the single cell level under dynamic flow. We have delineated different aspects of UPEC behavior under shear stress, including epithelial adhesion, rapid division, and microcolony formation. Adhesion of bacteria to epithelial cells is essential for infection initiation. P fimbriae-mediated adhesion prolonged binding duration to renal cells and enabled UPEC to resist shear stress for a longer time. Bacteria, which established stationary adhesion became the founders of infection. Rapid proliferation of the founder population led to the formation of microcolonies, whose structure was mediated by Type 1 fimbriae and, at least in the first hours, independent of ECM-curli expression. Rapid division in microcolonies increased the bacterial population locally and increased the shed-ding of bacteria, and thereby enabled a greater spread of infection. While these adhesion factors enhanced initial adhesion as well as microcolony formation, neither were found to be essential to colonization. Overall, these results demonstrated that UPEC possess a repertoire of redundant fitness and virulence determinants, which enhance bacteria's perseverance in the dynamic environment of the proximal tubule. These in vitro findings strongly support our previously reported in vivo findings that P and Type 1 fimbriae act in synergy to mediate effective early colonization of the dynamic microenvironment of the kidney tubule (Melican et al. 2011).
With the methodological advancement of the PToC, we identified the rapid formation of UPEC microcolonies under flow. Microcolony formation has previously been shown to be critical to withstanding shear stress in blood stream infections with, e.g. Neisseria meningitidis (Mairey et al. 2006, Mikaty et al. 2009), but not previously implicated in UPEC kidney infection in this way. Our data shows that bacterial ECM-curli expression does not appear to be the key factor at this very early time frame, but instead demonstrates a clear role of Type 1 fimbriae and interbacterial binding in microcolony formation. Our data suggests that these microcolonies may undergo a transition from a Type 1-mediated microcolony to an ECM-curli rich microcolony over time. It suggest that we have different types of biofilms during UPEC colo- nization. Our results showed that FimH mediates interbacterial interactions enabling bacteria to form microcolonies and expand beyond the host cell surface. Considering that the fim operon is almost omnipresent in UPEC strains (Arthur et al. 1989, Rahdar et al. 2015, bacteria would rarely lack the ability to establish these early microcolonies.
The usability of organ-on-chip methodology is rapidly expanding and offers a high-resolution alternative to in vivo experimentation. The advantages of the model presented here are resolution, speed, and the ability to apply real-time quantative analysis. We can follow the first stages of bacterial-epithelial interaction under physiological flow and over-time, to understand the dynamics and kinetics of early colonization. Kidney infection is an ideal target for these types of approaches as shear stress is a predominant factor. There are several organ-on-chip models mimicking the liver (Sivaraman et al. 2005, Carraro et al. 2008, Novik et al. 2010, kidney (Baudoin et al. 2007, Snouber et al. 2011, Jang et al. 2013, lung (Huh et al. 2010, and intestine (Kim et al. 2012, Kim andIngber 2013). The technology is making its way into infection research with a particular interest for vascular infections (Mairey et al. 2006, Mikaty et al. 2009, Claes et al. 2015.
Our study highlighted that UPEC utilizes multiple adhesion factors to consolidate colonization efficiently under shear stress. The PToC enabled us to monitor the dynamic interplay between host and pathogens with full spatial-temporal resolution in ways that are not feasible in traditional static cell cultures or animal models. We were able to demonstrate and quantify the transient nature of bacterial adhesion, and also investigate UPEC's colonization strategy under shear stress. Monitoring of the infection at this level of detail is currently unattainable in animal models of kidney infection. We envision that future development of PToC with the addition of other types of host cells, both proximal and distal to the infection site, as well as different media compositions mimicking different kidney microenvironments, will help us understand the full complexity of pyelonephritis.

PToC setup
A slide containing six microchannels (μ-slide VI 0.4 , ibidi) was used with dimensions of 17 mm × 3.8 mm × 0.4 mm. 2 × 10 4 A498 cells in 30 μl RPMI 1640 medium were loaded in each microchannel 1 day before an experiment and incubated for 1 h at 37 • C. After microscopic confirmation of cell adhesion to the bottom of each microchannel, 60 μl medium was added in each microchannel reservoir and the slides were incubated overnight at 37 • C. The following day, PToC was assembled. A 50-ml aliquot of prewarmed CO 2 -independent medium (Gibco), supplemented with 10% FBS and 1% Glutamax (CO 2 IM), was prepared. Media were additionally supplemented with 0.1% arabinose and 100 μg/ml ampicillin when strains CFT073 pBAD and CFT073 pBAD-fimH were tested. d-Mannose (3%) (Sigma) was added when masking FimH binding of CFT073 wt. The media aliquot was connected via a 1-mm internal diameter silicone tubing (VWR) to: (i) a tubing pump (Model ISM 80606, Ismatec), (ii) a bubble trap (IBI Scientific), (iii) an inlet (in-line luer injection port, ibidi), (iv) a microchannel of the slide seeded with cells, and (v) a collection tube. All these components and the tubing were interconnected via luer adaptors (IDEX). The pump was set at a flow rate of 75 μl/min and the flow of medium was initiated. Then, the slide was placed inside a portable 37 • C incubator (ibidi) designed to fit on the stage of a phase-contrast inverted microscope (Nikon TS 100). The microscope was connected to a camera (Hamamatsu) to monitor the microchannels with time-lapse microscopy controlled by the NIS-Elements Basic Research software (version 4, Nikon).

Flow rate calculation in PToC
The linear flow velocity of the primary renal filtrate is constant along the proximal tubule (Bonvalet and de Rouffignac 1981). Therefore, we could simulate the renal proximal tubule environment if we achieved the same linear flow velocity in the microchannel of the PToC. To calculate linear flow velocity in the proximal tubule, we used formula (i): where V = linear flow velocity, Q = flow rate, and A = crosssectional area of the channel. To first calculate the cross-sectional area (A) of the proximal tubule, we applied formula (ii): where r = radius. The inner diameter of the proximal tubule, however, decreases along this nephron segment. Thus, we can only estimate the flow rate in the first segment of the proximal tubule, which has an average internal diameter of 41.5 ± 6.2 μm in humans (Møller and Skriver 1985). Therefore, the cross-sectional area of the proximal tubule is ≈ 0.00135 mm 2 . Since the S1 segment is adjacent to the Bowman's capsule, the flow rate of the primary filtrate in the S1 segment of the proximal tubule corresponds approximately to the Single Nephron Glomerular Filtrate Rate (SNGFR), which has been estimated between 60 and 71 nl/min in the human kidney (Oken 1982, Lazzara andDeen 2007). By inserting the SNGFR and the average cross-sectional area of the S1 proximal tubule in formula (ii), we found that the velocity of the primary filtrate in the S1 segment is ≈ 48.15 mm/min. To achieve the same velocity in a microchannel, we calculated its cross-sectional area (1.52 mm 2 ), using formula (ii). The flow rate needed to maintain the same velocity in a microchannel as in the proximal tubule is approximately 73.19 mm 3 /min as calculated with formula (i) or 73.19 μl/min. Due to fine-tuning limitations of the pump, we applied a flow rate of 75 μl/min for all experiments.
According to the slide's manufacturer, shear stress in a microchannel is calculated as where T = shear stress and η = viscosity. The viscosity of cell culture medium is ≈ 0.0079 dyn × s/cm 2 (Rinker et al. 2001). Thus, shear stress in the microchannel is 0.104 dyn/cm 2 , which is comparable to the shear stress value in the human proximal tubule estimated in another study (Essig and Friedlander 2003).

LIVE/DEAD viability assay
A498 cells were incubated in the PToC under 75 μl/min or static conditions in CO 2 IM for 8 h. After incubation, microchannels were disconnected from the tubing and a mix of 2.5 μM ethidium homodimer-1 (dead stain) and 0.5 μM Calcein AM (live stain) from the LIVE/DEAD™ Viability/Cytotoxicity Kit (Invitrogen) was added. Cells were incubated for 20 min at 37 • C and 10 images from one microchannel per condition were captured under fluorescence microscopy, using a GFP and Cy3 filter. Live and dead cells were counted in each image and the average viability (%) was then calculated.

Bacterial mutant and plasmid construction
Bacterial strains, plasmids, and oligonucleotides used in this study are listed in Table S1 (Supporting Information). Isogenic mutants of CFT073 were constructed with one-step inactivation of chromosomal genes (Datsenko and Wanner 2000) using the pSIM6 Red expression plasmid (Datta et al. 2006). Briefly, to construct CFT073 fimH, the complete reading frame of fimH was replaced by a kanamycin resistance cassette, which was PCRamplified from pKD4 with fimH_KOF and fimH_KOR primers. The kanamycin resistance cassette was subsequently eliminated by FLP recombination using pCP20 (Cherepanov and Wackernagel 1995). CFT073 papG papG_2 was constructed by first replacing the complete reading frame of papG with a kanamycin resistance cassette, which was PCR-amplified from pKD4 with papG_KOF and papG_KOR primers. Then, the complete reading frame of papG_2 was replaced by a chloramphenicol resistance cassette that was PCR-amplified from pKD3 with papG_KOF and papG_KOR primers. Both kanamycin and chloramphenicol cassettes were subsequently eliminated with FLP recombination. All deletions were confirmed by PCR, using fimH_F, fimH_R, papG_F, papG_R, papG2_F, papG2_R primers, and by sequencing. PCR reactions to amplify antibiotic resistance cassettes were performed with Phusion DNA polymerase (NEB) at 55 • C annealing temperature. PCR confirmation reactions were performed using illustra Taq PCR beads (GE Healthcare) at 58 • C annealing temperature. To construct complementation vector pBAD-fimH, the fimH ORF was cloned into pBAD under the control of an arabinose-inducible promoter. Briefly, fimH ORF was PCR-amplified with SacI_fimH_FW and HindIII_fimH_RV primers (Table S1, Supporting Information) using gDNA from CFT073 as a template and ligated into pBAD that was previously linearized by double digestion with SacI and HindIII restriction enzymes (NEB). Electrocompetent CFT073 fimH were transformed with pBAD and pBAD-fimH. Construction of pBAD-fimH was confirmed by sequencing. Control experiments were performed with an empty pBAD vector.

Bacterial growth curves
Growth curves were prepared by diluting LB overnight cultures 1:100 in fresh LB, pelleting bacteria at 8000 rpm for 5 min, and resuspending in either CO 2 IM or LB. A volume of 150 μl of each resuspension was transferred in a 96-well plate in triplicates, incubated in a plate reader (Biotek) at 37 • C, measuring OD 600 every 10 min for 5 h.

Testing fimbrial expression in bacterial mutants
Exponential phase bacterial cultures in CO 2 IM were pelleted and concentrated 100 times in saline or saline + 0.05 M methylmannopyranoside (Sigma). For yeast agglutination, 25 μl of the saline or saline + 0.05 M methyl-mannopyranoside resuspension were transferred on a microscopy glass slide and mixed with either an equal volume of fresh baker's yeast suspension in saline (15 mg/ml) or saline only (negative control) for 30 s. For blood agglutination, 25 μl of the saline resuspension was transferred on a microscopy slide and mixed with an equal volume of a 2% suspension of isolated human red blood cells in saline or with saline only (negative control). Results for both yeast and blood agglutination were observed macro-and microscopically.

Infection in PToC
One microchannel seeded with A498 cells was used per condition tested. Each microchannel was exposed to 75 μl/min of CO 2 IM as described above. To prepare the bacterial inoculum an LB overnight culture was diluted 1:100 in 10 ml prewarmed CO 2 IM and incubated at 37 • C under shaking conditions to an OD 600nm ≈ 0.4-0.5. Then, 0.5 ml (≈ 2 × 10 8 cfu) of the exponential phase culture was slowly introduced over the course of 1 min in the microchannel with a syringe via the inlet (t = 0 of the infection). This corresponds to a multiplicity of infection (MOI) of 10 000:1 (bacteria: renal cells). For a detailed description on appropriate MOI selection, see Supplementary Note 2. A random microscopic field of view was chosen along the central lane of the microchannel and infection was then monitored with time-lapse phase-contrast microscopy at 40x magnification. Alternatively, infection was monitored by manually capturing images at selected time points. Samples were collected at the desired time points by placing a 1.5-ml microcentrifuge at the outlet and collecting medium exiting the microchannel for 2 min (≈ 175 μl). Collected samples were immediately transferred to a 96-well plate, and OD 600 was measured in a plate reader (TECAN).

Time-lapse microscopic analysis of bacterial colonization
All time-lapse microscopy videos were recorded from three biological replicates per strain at either 10 frames per second (fps), or 1 or 2 frames per minute (fpm). Time-lapse videos were analyzed in ImageJ v1.46r (Schneider et al. 2012). To perform SCTA of bacteria entering the microchannel, 2 × 10 s time-lapse videos per biological replicate were analyzed. A starting area (78.98 μm × 209.93 μm) was selected on the leftmost part of the field of view (Fig. 2B). Using the ImageJ plugin MTrackJ (Meijering et al. 2012), all single bacteria found within this area on the first frame (t = 0 s) were followed either until the last frame of the video (t = 10 s) or until they exited the field of view. Based on bacteria's trajectories, their displacement per 0.1 s and mean velocity (V) were calculated. R 2 of displacement vs. time was calculated to assess linearity.
To measure the duration of bacterial adhesion to renal cells, we randomly selected two areas of 1 × 10 4 μm 2 each, in the microscopic field of view. All bacteria that landed on renal cells in this area between 0 and 60 min of the infection were tracked using MTrackJ. Out of all tracked bacteria, we further analyzed only the ones attaching for at least ≥ 1 min. Adhesion events for each frame were calculated based on the number of new bacteria found attached and detachment events based on the number of bacteria detaching per frame. The smoothness of the attachment and detachment curve was adjusted with the method of Savitzky and Golay (1964) to reduce noise and improve understanding of the graphs. Net adhesion corresponds to the total number of bacteria found attached per frame. Binding duration for each bacterium was calculated based on the number of frames that it remained attached to a cell. The absolute frequency was calculated with the center of the first bin at 1 min, and a bin width = 0.25. Relative frequency for binding duration was calculated by arbitrarily categorizing bacteria in 3 binding duration ranges: (i) 1-2 min, (ii) 2.5-30 min, and (iii) > 30 min. Based on the total number of bacteria attaching for at least ≥ 1 min, the % of bacteria in each binding duration range was calculated.
To measure the generation time of dividing bacteria, time-lapse recordings at 1 or 2 fpm were analyzed to identify bacteria that remained attached in our field of view for at least two sequential divisions. The elapsed time between the two frames that the two sequential divisions occurred was defined as the generation time.
For microcolony analysis, the full microscopic field of view (473 μm × 473 μm) from time-lapse videos at 1 or 2 fpm was examined. After representative microcolonies were selected, the field of view was cropped around them for clarity.
The rounding onset of individual host cells was determined using MTrackJ on time-lapse videos. We calculated % healthy cells by dividing the number of cells with healthy morphology at each time point with the total number of cells in the field of view. Healthy cell curves (%) were constructed for three biological replicates per strain. The average curve was then calculated for each strain.

Surface biofilm assay and ECM staining by EbbaBiolight
The surface biofilm assay was performed as described by the manufacturer (https://www.ebbabiotech.com/collections/ebbabi olight). Briefly, cover glasses (Menzel Gläser) were sterilized with 70% ethanol and modified to fit inclined in 12 well plates (Corning). Overnight culture of CFT073 was diluted 1:100 in fresh LB without salt medium in the 12 well plates with the inclined coverslips. EbbaBiolight 680 (Ebba Biotech, Sweden) (0.5 μl/ml) was added to experimental wells but not control wells. Plates were incubated for 4 days at 37 • C. After 4 days, the coverslips were washed twice in PBS and fixed with 4% PFA for 1 h at room temperature. The coverslips were washed twice in PBS, then one side of the coverslip was cleaned with 70% ethanol drenched wipes (Kimtech) before the other side was mounted against microscope slides (VWR) with mounting medium (Dako). The slides were imaged in Lionheart™ FX Automated Microscope (Ramcon, Sweden) with fluorescence from EbbaBiolight 680 detected using the propidium iodide channel (Excitation 531 nm, Emission 647 nm).

ECM detection in PToC
PToC was assembled as described above, with 0.5 μl/ml EbbaBi-olight680 (Ebba Biotech) added to the flow-through medium. The Lionheart FX microscope (BioTek) was used to monitor the infection. Montage images of the middle section of the PToC were taken at infection time 0.5, 1.5, 2.5, and 3.5 h with brightfield and propidium iodide channel (excitation 531 nm, emission 647 nm).

Movie editing
Representative frames or full-length time-lapse recordings of the full microscopic field of view (473 μm × 473 μm) were exported as 16-bit Tiff files using NIS-Elements software. Using ImageJ, Tiff files were imported as a stack, cropped around the infection area of interest, and trimmed in length. Brightness and contrast were adjusted linearly. Bacteria and microcolonies were annotated with circles and an ID number using MTrackJ.

Statistics
All statistical analysis was performed using Prism 6. The mean relative frequencies of binding duration between strains were compared with t-test and healthy cells (%) curves were compared with the log-rank test. from F.X.C. H.A., T.Z., F.X.C., K.M., and A.R.D. analyzed the data. H.A., K.M., and A.R.D. wrote the manuscript with input from T.Z. and F.X.C.