Phenotyping ciliary dynamics and coordination in response to CFTR-modulators and Thymosin-α1 in Cystic Fibrosis respiratory epithelial cells

The diagnosis and treatment of respiratory disorders are challenging and would benefit from new approaches to systematically assess ciliary beating dynamics and to test new drugs. A novel approach based on multiscale differential dynamic microscopy (multi-DDM) is shown to quantitatively assess collective beating of cilia in a non-biased automated manner, in human airway epithelial cells (HAECs) derived from subjects with cystic fibrosis (CF) and grown in 2D air-liquid interface culture. Multi-DDM can readily detect changes in both ciliary beat frequency (CBF) and cilia coordination that result from perturbations to the mucosal layer. The efficacy of three CFTR-modulating treatments is investigated: ivacaftor (VX-770) with lumacaftor (VX-809), VX-809 alone and Thymosin alpha 1 (Tα1) alone. All three treatments restore coordination of cilia beating in the CF cells, albeit to varying degrees. We argue cilia are affected by these treatments through the physical properties of the mucus. Phenotyping cilia dynamics through multi-DDM provides novel insight into the response of ciliary beating following treatment with drugs, and has application in the broader context of respiratory disease and for drug screening. One sentence summary A semi-automated and unbiased assay based on multiscale differential dynamic microscopy (multi-DDM) detects changes in the coordination and frequency of ciliary beating in F508del/F508del primary human airway cells under different conditions and in response to CFTR-modulating compounds.


Introduction
Respiratory disorders affect millions of people worldwide and can result from both genetic and environmental causes (1,2). Many respiratory disorders are characterised by abnormal ciliary beating, be this a causal or derivative behaviour. Current approaches to systematically analysing collective cilia beating are unfortunately limited, as is our understanding of the physical properties required to sustain healthy transport of mucus out of the airways, known as mucociliary clearance (MCC) (3). Most approaches to phenotyping ciliary motion and coordination are time-consuming and difficult to standardise across labs (4). Videomicroscopy examination of airway biopsies is often 3 used to estimate ciliary beat frequency (CBF) and to inspect waveforms on individual cilia, but this is usually performed manually and requires experienced personnel. Semiautomated approaches to measure CBF have been developed recently, however even these assess only a subset of the sample and thus cannot detect the broad distribution of CBF that can occur within a given biopsy (10, 14 -17). Furthermore, while CBF is a first and readily accessible phenotype, it is not sufficient to diagnose pathologies: it is clearance as a whole that has to function.
A key parameter of healthy mucociliary clearance relates to how the ciliary beating is coordinated across large (many cell) distances. Despite its importance, the characterisation of cilia coordination in the context of human respiratory disease has not been well explored. Approaches probing both CBF and cilia coordination have been reported, however these either require manual selection of the area for analysis (4), or are not suitable for samples grown in 2D air-liquid interphase (ALI) culture, which is the standard method for culturing clinical HAEC samples (10)(11)(12). In ALI culture, the ciliated cells typically exist in patches, and coordination across the entire sample is highly variable (13,14), precluding for example assays based around bead-clearance. As such, there is very little data to describe how collective and coordinated ciliary beating arises in healthy human airway cells, and how this goes awry in disease. We recently reported a video analysis algorithm based on differential dynamic microscopy (DDM), which we called multiscale DDM (multi-DDM) (15). This allows the characterization of collective ciliary beating in human airway epithelial cells in a fast and fully-automated manner. The input required to run the DDM or multi-DDM algorithms are typically 10 second bright-field optical microscopy videos of the ALI cultured cells in situ, taken at 40x magnification and moderately high frame rate (~150 frames per second). By considering the frame differences at various time intervals, transformed in Fourier space, the method extracts temporal and spatial coherence of any dynamics in the video. In particular, for videos with oscillating features such as motile cilia, the CBF is obtained without the need to neither segment nor select regions. In multi-DDM (15) we showed that the spatial scale of coordinated cilia dynamics can be measured.
We focus this study on cystic fibrosis (CF): a life-threatening genetic disorder caused by loss-of-function mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) gene (16). Defective CFTR activity in airway epithelial cells leads to loss of airway surface liquid and incompletely hydrated mucin, resulting in a thick layer of mucus that obstructs the airways and promotes chronic bacterial infections and inflammatory lung damage. In individuals with CF, cilia beating is greatly compromised to the extent that the mucus cannot be properly cleared, and the range of cilia movement is severely restricted (16,17). Of the nearly 2000 CFTR mutations that have been identified (www.genet.sickkids.on.ca), only the most common mutations expressed by large groups of subjects have been targeted for drug screening due to the high cost and time-consuming nature of clinical trials. However, a recent study showed that selected CFTR-modulating drugs used to treat patients with the common F508del-CFTR mutation could in some cases be effective in treating patients with rare, uncharacterized CFTR mutations that are currently not registered for treatment, if only cost-effective assays could be used to screen such samples (18).
In the present study, we performed multi-DDM analysis showing that primary HAECs obtained from subjects with the F508del/F508del mutation in CFTR exhibit unique cilia coordination and CBF dynamics compared to cells from healthy subjects. multi-DDM data quantify in a very direct way the loss of cilia coordination with distance, and how this phenotype is affected by mucus properties in both direct perturbations and pharmacological intervention, providing new important information about coordination of cilia dynamics in the context of CF. As a proof-of-concept, we used this approach to demonstrate the efficacy of the CFTR-modulating drug combination of VX-770 (ivacaftor/ KALYDECO ® ) and VX-809 (lumacaftor, together termed ORKAMBI ® ) on HAECs derived from subjects homozygous for F508del CFTR. We further investigated the effect of VX-809 alone on these cells, as well as a recently-reported CFTR-modulating compound Ta1 (19). The development of a rapid, quantitative assay for characterising collective cilia beating dynamics in a standard cell culture model has important applications in diagnostics and drug screening. To our knowledge, these data represent the first in-depth, quantitative assessment of cilia coordination in HAECs derived from subjects with CF and how this changes in response to CFTR-modulating drugs. Although this study applies multi-DDM analysis to investigate ciliary beating dynamics in the context of CF, our approach to phenotyping cilia dynamics could be applied to the other respiratory diseases in which ciliary beating is affected.

Multi-DDM analysis of CBF and cilia coordination in healthy HAECs
One of the first parameters normally probed in clinical samples of HAECs, and one which is a general indication of ciliary function, is the CBF. Unlike standard clinical practice whereby a single CBF value is calculated from a user-selected region-of-interest (ROI) (20,21), multi-DDM provides the user with a distribution of CBF values measured across the imaged fields of view (FOV). This is important, since CBF can vary across a single sample ( Figure 1A -B), and thus a single point measurement is not an accurate 6 representation of the entire sample. Dividing the entire FOV into square subsets of a predefined size (called tiles -64 x 64 pxl, 9.3 µm per side; Fig. 1C) and running the DDM algorithm on each of these tiles generates a distribution of measured CBF across the entire FOV (Fig. 1D, E). In the measurement of CBF, DDM simply extends in a systematic and user-free fashion the concept of probing intensity modulation over time, in a pixel or group of pixels.
The standard clinical practice for analysing CBF in ALI cultures does not capture any information regarding the spatial coordination of cilia beating. However it is clear that cilia across the airway epithelium must beat in a coordinated fashion in order to efficiently propel the mucus layer upwards, and uncoordinated ciliary beating is a hallmark of some ciliopathies, for example PCD (22)(23)(24). To assess the degree of spatial coordination we deployed multi-DDM, where the size of the tiles on which the algorithm is run is systematically changed (Fig. 1F): as the tile probed is decreased in size, cilia beating within the tile becomes more coordinated, which is seen as an increase in a characteristic decay time of the resulting signal (Fig. 1G). Plotting the inverse of these decay times against their tile size produces a sigmoidal dataset that functions as a quantitative measure of cilia coordination within a given sample (Fig. 1H). This data can be fitted empirically to extract a parameter (essentially the position of the sigmoidal curve) that pinpoints the spatial scale of cilia dynamics coordination. The "shoulders" in the sigmoids are defined as the points where the line that best approximates the central slope meets the respective

Multi-DDM analysis of CBF and cilia coordination in CF HAECs
We used multi-DDM to compare CBF and cilia coordination in HAECs from healthy donors and from subjects with CF. In CF cells defective CFTR activity leads to loss of airway surface liquid and incompletely hydrated mucin, resulting in a thick mucus layer that greatly restricts cilia beating. The distribution of CBFs for a typical sample with the F508del/F508del CFTR mutation is shown in Fig. 2A  We hypothesise that the differences in ciliary beating between cells from healthy subjects versus subjects with CF is attributable to the physical properties of the mucus, since the CFTR mutation affects mucosal properties and not the intrinsic structure or function of cilia (25,26). To test this, we analysed F508del/F508del HAECs both in the presence of mucus and immediately after a wash treatment to remove mucus. Average CBFs obtained 8 following a wash treatment are typically higher than when mucus is present, and this is true for cells obtained from both healthy and CF subjects ( Fig. 2B and Supplementary Table 1). We next assessed the degree of ciliary coordination in the CF samples in the presence of mucus and immediately following a wash. The curves generated using the multi-DDM algorithm show the typical sigmoidal shape for both conditions, however we observe that the position of the left shoulder of the sigmoid (λ 2 ) differs depending on the presence or absence of mucus (Fig. 2C). In the absence of mucus (i.e., immediately following a wash) the ciliary coordination length scale decreases towards values obtained from healthy samples. Direct comparison of λ 2 between the two conditions reveals a statistically significant (p<0.01) shift to smaller values when the mucus is removed in the HAECs obtained from CF subjects. This is also observed to a lesser extent when mucus is removed from HAECs obtained from healthy subjects (p<0.05; Fig. 2D). The shift to smaller values represents a decrease in the length-scale of ciliary coordination, meaning decreased coordination in the absence of mucus. This shift implies that the mucus, acting like an elastic gel raft, helps cilia to synchronise their dynamics. The difference between the shoulder values obtained from two treatments of a given sample can be used to quantitatively differentiate between ciliary beating dynamics.

Multi-DDM charts the evolution of ciliary beating phenotypes over time
The experiments with multi-DDM presented above analysed and compared different HAEC samples maintained under constant cell culture conditions, which is useful to obtain a snapshot comparison between two samples but does not convey information about how the cilia dynamics phenotype evolves over time in response to various conditions. This is particularly relevant for understanding cellular responses to pathogenic infections and/or pharmacological responses. Having identified that the presence of mucus has a dramatic effect on both CBF and cilia coordination, we tested the effect of successive cycles of washing and allowing the mucus to regenerate. HAEC samples obtained from three different subjects with the F508del/F508del mutation were subjected to successive rounds of mucus removal and regeneration for two days. Briefly, DDM data were collected in the presence of mucus (0 hrs) and then immediately after mucus was removed (+2hrs), then again once the mucus had regenerated (+24hrs). This cycle was repeated at 26 and 48hrs (Fig. 3A). We

Multi-DDM detects partially restored ciliary beating following treatment with CFTRmodulating drugs and Thymosin-a1
Having Thymosin-a-1 (Ta1; ZADAXIN) is a naturally occurring polypeptide that is used as an immunomodulator in viral infections, immunodeficiency and malignancies (31,32).
Recently, Ta1 4A, B). Analysis of the sigmoidal curves reveals a shift to smaller length-scale for coordination in all subjects, but most notably in subject 1 upon treatment with Ta1 ( Fig. 4C -E). Consistent with this, the average difference between the left shoulder point before and after the treatment differs significantly from the control at both 24 and 48 hours, and is greater than either VX-770/VX-809 or VX-809 alone (Fig. 4E). The markedly decreased coordination length scale observed in response to all three drug treatments is indicative of partial restoration of normal ciliary beating dynamics, which is consistent with the previously reported roles of these drugs in modulating CFTR functionality in CF cells (26,(32)(33)(34).

Discussion
Probing the collective cilia beating dynamics and mucociliary clearance across cells at reported an organoid-based screen to assess the efficacy of CFTR-modulating drugs and showed its effectiveness in predicting drug responsiveness for subjects with rare, uncharacterized CFTR genotypes (18). The study is an excellent example of how patients that fall into this category could benefit from such a screen, however the methodology of the assay, which involves isolation of colonic crypts, cell selection and expansion, organoid culture and seeding in Matrigel droplets, is costly, requires considerable expertise and is open to variation due to the use of Matrigel and other undefined animalbased products. For these reasons, we believe the assay is unlikely to be widely adopted by practising clinicians. In contrast, our assay is based on primary HAECs grown in ALI culture, which is a simpler cell culture already widely used amongst clinicians.
Furthermore, and unlike current approaches for measuring ciliary beating parameters, our approach is unbiased, automated and considers not only CBF but also the spatio-temporal dynamics of cilia coordination across the entire FOV, the latter of which conveys important information regarding collective cilia beating and fluid transport in a clinicallyrelevant setting. The analysis of multiple cilia beating parameters, a "deep phenotyping" of cilia dynamics, is critical in order gain a more complete picture of how different candidate drug treatments affect mucociliary clearance. This was clearly the case with regard to Ta1 treatment, where despite only a modest improvement in CBF that was not statistically significant, we observed a marked improvement in cilia coordination, even greater than that seen in response to VX-770/VX-809.
We envision a number of applications for the multi-DDM approach presented here.
Firstly, it should now be possible to quantitatively assess a wide range of putative drugs for the treatment of CF based on changes in ciliary beating and coordination. In vivo, MCC is mediated by the coordinated action of thousands of beating cilia, and so a quantitative approach to measure ciliary beating dynamics is highly suited to the analysis of therapeutic modulators of MCC. This is particularly relevant for drugs such as ENaC inhibitors, mucus hydrators, and DNases/mucus disruptors, which do not modulate CFTR function, and so cannot be tested using Ussing chamber analysis nor the Forskolininduced swelling assay (34). A second application of our approach is in the analysis of CFTR-modulating drugs, where it may be useful as a complementary assay alongside Ussing chamber analysis. Theoretically, restoration of CFTR ion channel function as measured by Ussing chamber analysis should lead to mucus rehydration and increased MCC. In practise, however, the degree to which CFTR activity is restored varies among patients, and the relationship between the Ussing chamber results and the observed clinical response is not always clear. In the future, it would be interesting to investigate whether there is any relationship between data obtained via Ussing chamber, multi-DDM and in vivo efficacy to see if there is a threshold of ion transport that is required in order to improve MCC. Application of multi-DDM to assess putative drug treatments would require the culture of HAECs isolated from the subject and application of multi-DDM analysis before and after the treatment, which is quite straightforward. An additional application of multi-DDM is to investigate ciliary beating dynamics in diseases other than 18 CF, such as in ciliopathies, where over 187 genes have been associated with a defect in cilia form or function (1,35). Characterising ciliary beating in different classes of ciliopathies and in the range of different tissues in which ciliary beating is affected will help to better understand this broad class of disease. This in turn may lead to better diagnostic tools, particularly for diseases in which the underlying genetic mutation remains completely uncharacterised.
In conclusion, we have developed a straightforward, quantitative assay based on multi-DDM to characterise and detect changes in ciliary beating of HAECs grown in ALI culture in both healthy and disease contexts. Our approach improves on previous methods of ciliary beat analysis in that it is unbiased and automated, and because it captures the spatio-temporal coordination of collective cilia beating which is crucial for MCC, and not simply the CBF. We provide a proof-of-principle by applying multi-DDM to HAECs obtained from subjects with the F508del/F508del mutation in CFTR to assess their responsiveness to known CFTR-modulators VX-770, VX-809 and Ta1. Our data indicate that multi-DDM can quickly and efficiently detect compounds that result in restored ciliary beating dynamics -in this case VX-770/VX-809 and Ta1, and to a lesser extent VX-809 alone. Our approach provides a straightforward, quantitative assay to assess patient responses to putative and existing drug treatments for CF, which may be applied to CFTR-modulating and non-CFTR-modulating drugs alike. Finally, since multi-DDM is ultimately a means to assess cilia beating dynamics, its use is not limited to CF but may be applied to other diseases in which ciliary beating is affected.

Study design
The overall objective of this study was to use the recently-developed multi-DDM algorithm to characterise ciliary beating in cells from subjects with CF, and to test whether the approach was sensitive enough to detect changes in ciliary beating over time and in response to various drug treatments. As a first-pass, we applied multi-DDM to primary focus on this mutation as it is the most common cause of CF and one for which at least one therapeutic drugs has been approved (ORKAMBI ® ). At least sixteen -often morevideos were taken of each of the samples at each data collection point, avoiding the perimeter of the cell culture insert due to the accumulation of mucus in that region (an experimental artefact caused by the geometry of the vessel). Multi-DDM analyses were performed as previously described (15) and in a such a manner as to ensure the entire field-of-view was represented in an unbiased manner. This study was not blinded.

Human material and cell culture
HAECs obtained through the clinic were isolated from nasal brushings of the inferior turbinate performed using a cytology brush on CF subjects attending the Adult Clinic at National Jewish Health (Denver, CO, USA). Samples were collected using a protocol gassed with 5% CO 2 /95% O 2 , and maintained at 37°C and pH 7.4. Inserts were voltage clamped and short-circuit currents were measured. Sodium transport through the epithelial sodium channel was inhibited using 100 µM amiloride, and after currents became stable, CFTR function was measured by activation using a cocktail of 10 µM forskolin/100 µM 3-isobutyl-1-methylxanthine, followed by potentiation using 1 µM VX-770, then inhibition using 10 µM CFTR(inh)-172.

Video acquisition
At least 20 videos were acquired from each samples at time 0h, 24, 48 and 72h on a Nikon

22
Eclipse Ti-E inverted microscope (Nikon Instruments, Japan) with a 40x objective (Plan Apo λ 40x, N.A. 0.95, Nikon). Digital high-speed videos were recorded under bright field illumination at a sampling frequency of 150fps using a Grasshopper ® 3 GS3-U3-23S6M-C CMOS camera (FLIR Integrated Imaging Solutions GmbH, Germany). Samples of epithelial cells were imaged in a custom made chamber, where temperature, CO 2 and humidity were continuously monitored and maintained at values of 37°C, 5% and 90% respectively. Any videos that showed drifting or duplicated areas of analysis were not included in the analyses.

Data processing
Videos were uploaded to our custom multi-DDM algorithm pipeline (coded in MATLAB, the MathWORKS) and processed as previously described (15). Briefly, the DDM algorithm prescribes to take the algebraic difference of several couples of frames separated by a lag time τ. These differences are then Fourier transformed in space, and the results averaged, to yield an averaged 2D power spectrum. If the anisotropy of the sample's dynamics is not of interest for the analysis, an azimuthal average of the 2D power spectrum is taken. By repeating the process for different values of the time lag τ we build the Image Structure Function I(q,τ) (39,40). As detailed in (15)

Statistical analyses
CBF results are presented as mean ± SD, unless specified otherwise in the figure caption.
λ 2 measurements on a single sample are presented as measurement ± 68% confidence interval, while averages across subjects are presented as geometric mean ± geometric SD.
Statistical analysis were performed by paired or unpaired two-tailed Student's t-test using MATLAB, and by weighted one-tailed t-test using R. The 95% confidence level was considered significant. Details about P values and which test was used to calculate them are in the figure captions.

List of Supplementary Materials
Supplementary Table 1: CBF of samples featured in Fig. 2 Supplementary Table 2: CBF of samples in Fig. 3 Supplementary Table 3: CBF of samples in Fig. 4 Supplementary Figure 1: distribution of CBF measured before and after washing