Inverted air-liquid interface cultures support live imaging of SARS-CoV-2 infection and mucociliary clearance.
To model the physiology of the human airway in a format amenable to live imaging, we used an inverted primary human bronchial epithelium air-liquid interface (ALI) culture system following the method of Zaderer et al. (Fig. 1a)23. Briefly, primary HBEs were seeded on the underside of permeable transwell supports to enable high-resolution live imaging of the apical surface of the cells through an inverted microscope. After differentiation, these cultures developed a characteristic pseudostratified morphology resembling human bronchial epithelium with multiciliated cells and goblet cells as confirmed by immunostaining for markers of cell-cell junctions (actin), cilia (acetyl a-tubulin), and mucus (MUC5AC) (Fig. 1b). While these largely formed a monolayer on the structured support, some ALI cultures contained furrows and intraepithelial lumens of varying connectivity henceforth referred to as crypts (Supp. Figure 1). Donors differ in their propensity to form these structures, as previously observed24.
To track the motion of mucus on the apical surface, we stained live cultures with either CellMask Orange Plasma Membrane (CMO) or NucView 530 (NV). CMO labels the plasma membrane upon application to the apical surface, but after 1–2 hours accumulates in dead cells and presumptive membrane debris in the mucus (Supp. Video 1). NV, a fluorogenic caspase 3/7 substrate, labels apoptotic cells, which are shed into the mucus allowing persistent monitoring of MCC (Supp. Video 2). As illustrated by CMO staining, the apical mucus disc in inverted ALI cultures often rotated over time, indicative of coordinated MCC (Fig. 1c, Supp. Video 2). Where rotary MCC was not present, the mucus flowed in a disorganized manner consistent with uncoordinated patches of locally coordinated multiciliated cells (Supp. Video 1)25. To observe multiciliated cells, we stained live cultures with SPY650-tubulin (Supp. Video 3). Imaging cultures with this dye revealed that the epithelium is sometimes quite dynamic, with cell migration leading to formation & dissolution of densely ciliated furrows (Supp. Video 4). The brighter structures visible in SPY650-tubulin correspond to the ciliated furrow and crypt structures visible in sections. 4/114 cultures had patchy ciliation with a hypermobile phenotype (Supp. Video 5). SPY650-tubulin also labels cilia with sufficient intensity to acquire high framerate videos of ciliary motion and determine ciliary beat frequency (Supp. Video 6). Taken together, these data confirm that the inverted ALI culture model resembles human bronchial epithelium and supports MCC for live imaging.
Immunostaining of inverted ALI cultures for ACE2 (the SARS-CoV-2 receptor) and TMPRSS2 (the major protease cofactor), confirmed that these cells express both factors required for SARS-CoV-2 entry at the apical surface of the epithelium and among the cilia (Fig. 1d, e). To assess the permissivity of these cultures to SARS-CoV-2 infection, we inoculated cultures with SARS-CoV-2 USA/WA1-2020 at a multiplicity of infection (MOI) of 1 via the apical surface, then quantified viral nucleocapsid (N) RNA in apical rinsates (mucus eluted in PBS) over time by quantitative PCR (qPCR) (Fig. 1f). To avoid disturbing the replication kinetics, each infected culture was only used for one rinsate timepoint. The concentration of N RNA in the mucus increased over time, consistent with productive viral replication and release from the apical surface. Maximum N concentration was observed at 120 hours post infection (HPI) at 1.9 x 108 N copies per square millimeter of culture surface area.
To confirm permissivity, we infected differentiated ALI cultures with SARS-CoV-2 as above and stained for viral replication compartments with anti-double-stranded RNA (dsRNA) antibodies or for viral antigens with anti-N or anti-Spike (S) antibodies at 72 HPI (Fig. 1g). Each marker of infection was detectable above background and displayed distinct localization patterns within the epithelium. S primarily localized in both a perinuclear compartment as well as at the apical surface of the cells below the cilia. N localized more broadly within the cytoplasm with occasional staining in the cilia. dsRNA occurred in a perinuclear compartment, consistent with the formation of membrane-associated replication complexes. Puncta of co-localized S and N occur within the cilia & mucus layer (Fig. 1g, white arrowheads). Punctate protein antigen staining generally did not co-localize with dsRNA, suggesting that at least some of the puncta represent actual virions rather than autofluorescent or noninfectious debris. If virions are released into mucus, that suggests that MCC could be relevant for viral particle spread. Overall, these results indicate that inverted ALI cultures are a suitable model for studying early SARS-CoV-2 spread in the context of mucociliary clearance.
Live imaging of inverted ALI cultures reveals kinetics and spatial aspects of SARS-CoV-2 spread.
To characterize SARS-CoV-2 spread within the bronchial epithelium, we performed live imaging of inverted ALI cultures after infection with a SARS-CoV-2 USA/WA1-2020 derived fluorescent reporter virus that has eGFP inserted into the ORF7a locus (icSARS-CoV-2/eGFP)1. Differentiated ALI cultures from 9 independent donors were infected in technical triplicate and imaged longitudinally for up to 9 days post-infection. Using CMO to simultaneously track MCC, we observed that most cultures had comet-like foci of infection that closely followed the tracks of the mucosal discs (Fig. 2a, Supp. Video 7). When visualized by temporal projection (Fig. 2a, right panel), the motion of the mucus (yellow and green) was observed to precede the development of infection foci (cyan). Similar results were obtained when using NV to track MCC and SPY650-tubulin to track ciliary motion (Supp. Video 8, Supp. Figure 2). The epithelium often appeared to rupture and heal in infected cultures, with 18/23 surveyed infected cultures having ruptures. (Fig. 2b, Supp. Video 9). Ruptures occurred both within foci and among bystander cells. No ruptures were observed in untreated uninfected cultures (0/19 surveyed).
To quantify infection kinetics, we measured the number of GFP + spots at each frame, subtracting spots present at baseline (generally attributable to background around the rim of the transwell) (Fig. 2c). The peak number of GFP + spots and the peak GFP + fractional area were positively correlated with the number of copies of N RNA in apical rinsate at 120 HPI, validating the metric (Fig. 2d). Isolated GFP + cells first became apparent around 20 HPI, with a median of 14 initial infected cells per culture (Fig. 2e). The viral inoculum contained 5 x 105 plaque-forming units (PFU, titered on Vero E6 cells) per well, corresponding to an MOI between 0.5 and 1, so the relative scarcity of early infected cells suggests there are formidable barriers to initial infection in these cultures. After the initial stage of infection, there was considerable heterogeneity in the infection kinetics both within and between donors. For most cultures, the number of GFP + spots increased rapidly around 48–72 HPI, peaking at a median of 103 HPI (Fig. 2f). Infected cells generally maintained viability and persisted within the epithelium for multiple days (Fig. 2g, Supp. Video 10). Nevertheless, overall infection remained relatively low, with peak infection in the most infected culture only ever reaching 3.3 x 104 GFP + spots or 20% of the total surface area. While the density of GFP + cells could be much higher in infected foci, large swathes of the surface remained uninfected in most infected cultures.
Notably, infection kinetics in different cultures correlated with the morphology of the infection foci and with MCC patterns (Fig. 2h, i). In some cultures, infection was restricted to small, circular plaque-like foci with a high density of infected cells in small, defined areas. These cultures generally had disorganized MCC and a low peak infection rate. Most cultures had comet-like foci, with a single infected cell initially appearing around 16–20 HPI and multiple infected cells appearing subsequently downstream of MCC tracks around 36–48 HPI. These cultures generally had larger areas of coordinated MCC, though it could be more disorganized or more rotary. A third subset of cultures had infection appear diffusely over the entire surface of the culture around 36–48 HPI. These cultures reached the highest peak infection and all displayed efficient MCC in a rotary pattern (Fig. 2i). Note that a fourth focus type was occasionally observed that appeared restricted to subepithelial crypts or furrows in the culture. These foci were identified by the presence of horizontally oriented cells in patterns matching the whorls of the cilia marker, which becomes quite bright in the crevices and crypts of the epithelium. Such foci grew rapidly but were restricted to the crypt to the extent that the crypt had limited access to the broader apical surface. Taken together, these data suggest that MCC can impact the spatiotemporal dynamics of SARS-CoV-2 spread.
MCC facilitates SARS-CoV-2 spread.
Since MCC patterns in individual cultures correlated with the morphology of SARS-CoV-2 foci, we hypothesized that restriction of MCC would limit SARS-CoV-2 spread. To mechanically restrict MCC, we applied a layer of low melting temperature agarose to the apical surface of the cultures immediately after mucus removal, CMO staining, and infection (Fig. 3a, Supp. Video 11). As seen in the temporal projection of the CMO channel (Fig. 3b), overlaying the cultures with agarose restricted mucus movement as reflected in the larger areas of static gray (right) as opposed to more saturated colors showing movement over time (right). Imaging of icSARS-CoV-2/eGFP spread following agarose overlay in multiple donors showed a clear restriction of infection to small, plaque-like foci (Fig. 3c, Supp. Video 11) similar to those observed in individual cultures with disorganized MCC above (Fig. 2g, h). Quantification of this effect in multiple donors confirmed lower peak infection percentage and slower spread upon agarose overlay (Fig. 3d, e), except for one culture which supported infection in an intraepithelial crypt (Fig. 3f, Supp. Video 12). These crypts are not exposed to the apical surface and thus not accessible to the agarose overlay. Rapid spread within the crypt further suggests that agarose overlay specifically inhibits infection by MCC restriction as opposed to a nonspecific mechanism, such as hypoxia, osmotic changes, or thermal damage to cells during overlay application.
To address this question in a different way, we knocked out axonemal dynein genes in primary HBEs prior to differentiation of inverted ALI cultures. Axonemal dyneins drive ciliary motion by generating the sliding motion of microtubule doublets within cilia26. DNAH5 and DNAI1 were targeted here on the basis of their common occurrence in human primary ciliary dyskinesia and the successful production of DNAH5 and DNAI1 KO mice, suggesting that these mutations are well tolerated27,28. After initial expansion, undifferentiated HBEs were electroporated with in vitro synthesized CRISPR-Cas9 ribonucleoproteins (crRNPs) targeting DNAH5, DNAI1, or CYPA as a control. Cells were seeded at high density on the undersides of collagen coated transwells (Fig. 4a). Multiciliated cells still differentiated and the apical surface of the cultures were ciliated as usual (Fig. 4b).
To determine the impact of these perturbations on ciliary motion, cultures were stained with SPY650-tubulin and visualized by high-speed video microscopy. Beat frequency was determined using a Fast Fourier Transform (FFT)-based power spectral density analysis. For each pixel, the power of the signal at each possible frequency (given the sampling rate, i.e. the framerate of the video) was calculated, and the frequency with maximal power was taken to be the dominant beat frequency (Fig. 4c)29. Most pixels with power greater than 40 corresponded to beating cilia, consistent with the SPY650-tubulin staining and the characteristic patchy appearance of the ciliary beat frequency images. Compared to the CYPA KO cultures, the DNAH5 and DNAI1 KO cultures had showed substantially diminished beat frequencies (Fig. 4c, d). Given the polyclonal nature of the KO cells, motile cilia were still evident in both axonemal dynein KO cultures, with the DNAH5 KO culture having an intermediate phenotype and the DNAI1 KO culture having very few motile cilia (Fig. 4c, d). MCC in each culture was additionally assessed by imaging with NV, which likewise showed movement of apoptotic cell debris in the apical space of the CYPA KO culture, but not in that of the axonemal dynein KOs (Fig. 4e).
Each culture was subsequently challenged with SARS-CoV-2/eGFP and imaged over 120 hours. The CYPA KO culture had comet-like foci that spread over time (Fig. 4f, Supp. Video 13). In contrast, infection of the DNAI1 KO culture resulted in small plaque-like foci, similar to the agarose overlay, while the DNAH5 KO culture had an intermediate phenotype consistent with its intermediate effect on ciliary motion (Fig. 4f, Supp. Video 13). This correlated with peak number of infected GFP + cells and levels of SARS-CoV-2 RNA in the apical rinsate (i.e., the mucus) of each culture (Fig. 4g, h). In sum, these data suggest that mechanical and/or genetic perturbation of ciliary motion and MCC inhibits SARS-CoV-2 spread after infection.
SARS-CoV-2 infection only modestly inhibits ciliary motion.
MCC facilitates the spread of SARS-CoV-2 early in infection, but this effect is limited. Comet-shaped foci do not grow indefinitely, suggesting that either downstream cells develop resistance to infection or that MCC becomes dysfunctional at later time points. In uninfected cultures, mucus movement most often continues for the duration of any given experiment (Fig. 5a). However, in SARS-CoV-2 infected cultures, mucosal movement typically stops at an average of 100 HPI (Fig. 5a). Previous studies have suggested that MCC dysfunction during SARS-CoV-2 infection may be driven by the loss of multiciliated cells, which are the major target of infection in airway epithelium, or by the enhanced secretion of MUC5AC-containing mucus by goblet cells10,12.
To differentiate between these possibilities, we first monitored ciliary motion of both infected and bystander cells in inverted ALI cultures from three donors at 24, 48, and 72 HPI. Cultures were stained with SPY650-tubulin to track ciliary motion. Five infected cells were identified at 24 HPI, when GFP signal first became detectable, and tracked over the next 48 hours (Fig. 5b). All five cells tracked over this time period continued beating at a regular beat frequency. A cell that is beating and ensconced within the monolayer must be alive and performing metabolic activities to support ciliary motion, so these findings suggest that multiciliated cells retain viability and function for days following SARS-CoV-2 infection. Furthermore, two of the five cells initiated foci during tracking, indicating that productive infection does not necessarily inhibit ciliary motion. Likewise, of the newly observed GFP + cells at 48 HPI, most continued to beat at 72 HPI.
To better survey the population beat frequency of the infected & bystander cells over time, we performed per pixel quantification of the above infections again using an FFT-based power spectral density analysis. The fraction of pixels with maximum signal power greater than 40 was used to approximate the culture area covered with beating cilia. The fraction of beating pixels in a field of view (FOV) increased over time in both mock and infected cultures, likely as a result of improved signal detection as the SPY650-tubulin incubation period increased (Fig. 5c). Between mock and infected cultures, the fraction of beating pixels was comparable at 24 HPI, but differed at 48 and 72 HPI with infected cultures showing fewer beating cilia. Despite the decrease in beating cilia coverage, many FOVs in infected cultures were dense with beating cilia and the beat frequency distribution of the cilia was qualitatively similar between mock and infected cultures (Fig. 5d), consistent with our cell-level observations described above. The shape of the ciliary beat frequency distributions was bimodal, with one population of pixels beating around 7 Hz and another approaching 0 Hz. The higher frequency population reflects beating cilia while the lower frequency population appeared to arise from pixels that contain debris moving through the field of view. This suggests that while individual multiciliated cells may retain viability and functionality after infection, there are fewer beating cilia in infected cultures over time.
To understand whether these differences between mock and infected cultures are driven by infected cells or bystander cells, we compared the fraction of beating area and beat frequency between neighboring GFP + and GFP- pixels in the same fields of view (infected & near bystanders), GFP- pixels in fields of view that contained no GFP + pixels in infected cultures (distant bystanders), and all pixels from mock cultures (Fig. 5e, f). The fraction of beating cilia (i.e., pixels with a power > 40) significantly differed between neighboring GFP + and GFP- pixels only at 24 hpi, and differences between the groups within infected cultures were dwarfed by the overall mock vs. infected difference (Fig. 5e). Furthermore, the beat frequency distribution was qualitatively similar between neighboring GFP+ & GFP- pixels (Fig. 5f). One caveat to these analyses is that motile cilia tend to present as two patches of pixels with a defined beat frequency to either side of the cell body, corresponding to the start and end positions of the ciliary beat. As such, the frequency value and the signal intensity of a given pixel may not necessarily correspond to the same cell, potentially diluting differences between infected cells and their neighbors. Nevertheless, these data suggest that decreases in ciliary motion during infection are due to impacts on both infected and bystander cells, rather than specifically infected cells.
Besides ciliary motion, the other major variable contributing to MCC is mucus secretion. To test whether the mucus influenced the spatial restriction of infection at later timepoints, we rinsed cultures at 120 HPI, shortly after the peak of infection. After rinsing, we observed apoptotic cells moving in mucus over foci of infected cells, consistent with active MCC (Fig. 6a). Interestingly, in some fields of view apoptotic cells appeared to move more quickly over infected foci than over adjacent bystanders (Fig. 6a, asterisks & arrowheads). Cultures that displayed rotary MCC early in infection resumed rotary MCC after rinsing, and those with comet-like foci saw the extension of the comets (Fig. 6b, Supp. Videos 5 & 14). The number of GFP + spots also increased after rinse in each culture (Fig. 6c). This suggests that stalled infection and cessation of MCC in infected cultures may be due to secretion of a soluble factor, likely mucus, and not solely depletion of infected multiciliated cells. Taken together, these data support a complicated role for MCC in SARS-CoV-2 infection, acting as an innate barrier for initial infection while serving to facilitate spread after infection is established.