Research ResourceImaging

A quantitative super-resolution imaging toolbox for diagnosis of motile ciliopathies

See allHide authors and affiliations

Science Translational Medicine  18 Mar 2020:
Vol. 12, Issue 535, eaay0071
DOI: 10.1126/scitranslmed.aay0071

Ciliopathy insights

Primary ciliary dyskinesia (PCD) results from genetic mutations and structural defects that impair the motility of cilia, the cellular protrusions that sweep mucus along the surface of the airway. Here, Liu and colleagues developed a quantitative imaging workflow to improve the diagnosis of PCD. Using super-resolution microscopy on nasal airway cells isolated from patients with PCD, the authors detected mislocalized PCD-related proteins, cellular structural defects, and impairments in ciliary beating. These imaging and analysis methods could help complement standard methods of diagnosis, such as genetic testing, and could provide insight into pathology caused by variants of uncertain significance.


Airway clearance of pathogens and particulates relies on motile cilia. Impaired cilia motility can lead to reduction in lung function, lung transplant, or death in some cases. More than 50 proteins regulating cilia motility are linked to primary ciliary dyskinesia (PCD), a heterogeneous, mainly recessive genetic lung disease. Accurate PCD molecular diagnosis is essential for identifying therapeutic targets and for initiating therapies that can stabilize lung function, thereby reducing socioeconomic impact of the disease. To date, PCD diagnosis has mainly relied on nonquantitative methods that have limited sensitivity or require a priori knowledge of the genes involved. Here, we developed a quantitative super-resolution microscopy workflow: (i) to increase sensitivity and throughput, (ii) to detect structural defects in PCD patients’ cells, and (iii) to quantify motility defects caused by yet to be found PCD genes. Toward these goals, we built a localization map of PCD proteins by three-dimensional structured illumination microscopy and implemented quantitative image analysis and machine learning to detect protein mislocalization, we analyzed axonemal structure by stochastic optical reconstruction microscopy, and we developed a high-throughput method for detecting motile cilia uncoordination by rotational polarity. Together, our data show that super-resolution methods are powerful tools for improving diagnosis of motile ciliopathies.


Cilia are cellular protrusions built on a microtubule scaffold that are critical for signaling, sensing, and motility (1). In the airways, motile cilia sweep mucus-trapped pathogens and particulates introduced through breathing, by beating coordinately on the surface of epithelial cells (1, 2). Defective motile cilia lead to impaired airway mucociliary clearance and chronic infections, resulting in oto-sino-pulmonary disease (3). Because motile cilia and their associated proteins also function outside of the airway, defective motile cilia can lead to heart malformations (4) and infertility (5).

Primary ciliary dyskinesia (PCD) is currently the only recognized human disease caused by defects in motile cilia. PCD is a recessive genetic disease affecting 1:10,000 to 1:15,000 individuals (6) with frequencies up to ~1:2000 in specific ethnic populations (7). Without early diagnosis and treatment, PCD can progress to irreversible respiratory failure, leading to lung transplantation or death by early to mid-adulthood (8). Most patients with PCD present with clinical manifestations of the disease from birth (9); however, their diagnosis is often delayed to late childhood or adulthood (10, 11). The reason for underdiagnosis is multifactorial, including inherent limitations of diagnostic tests, new genes yet to be found, and lack of awareness of the disease (12, 13). Available epidemiological data suggest that early diagnosis and treatment are important strategies for disease management because lung function tends to stabilize when therapy to address the lung disease is initiated (12, 14, 15).

PCD is a genetically heterogeneous disease: To date, mutations in more than 50 genes have been linked to PCD, and more are awaiting validation. Most PCD proteins are components of multiprotein complexes critical for cilia motility associated with the axoneme—the microtubule-based scaffold of motile cilia (Fig. 1, A and B)—such as the outer dynein arm heavy, intermediate, and light chains (DNAH5, DNAH11, DNAH9, DNAI1, DNAI2, DNAL1, and NME8), which propel the relative sliding of microtubules; the nexin-dynein regulatory complex (N-DRC) proteins (GAS8, CCDC65, and CCDC164), which link pairs of peripheral microtubule doublets; radial spoke (RS) complexes (RSPH1, RSPH3, RSPH4A, RSPH9, and DNAJB13), which link the peripheral microtubule doublets with the central pair of microtubules; molecular ruler proteins (CCDC39 and CCDC40), which are critical for the accurate recruitment and spacing of RS, N-DRC, and axonemal dyneins; central pair complex proteins (HYDIN, STK36, and SPEF2), which organize the central pair; and cytoplasmic axonemal dynein assembly and docking factors (DNAAF1 to DNAAF6, LRRC6, SPAG1, CCDC103, ZMYND10, C12ORF59c, TTC12, ARMC4, CCDC114, CCDC151, and TTC25) (16, 17). A few PCD proteins are associated with structures at the base of motile cilia (RPGR, OFD1, and GAS2L2). In recent years, next-generation sequencing has catalyzed the discovery of a new class of PCD genes encoding transcription factors and regulators (CCNO, MCIDAS, and FOXJ1) that control cilia number by regulating the multiciliogenesis transcriptional program.

Fig. 1 A super-resolution molecular map of the PCD proteome.

(A) A bright-field image showing a section of airway pseudostratified epithelium from a healthy volunteer. Note the multiciliated cells and motile cilia. Scale bar, 10 μm. (B) Illustration of the motile cilia structure. TZ, transition zone; DA, distal appendage; BF, basal foot. (C) 3D-SIM micrographs (maximum intensity projection) of airway multiciliated cells labeled with antibodies recognizing 21 PCD proteins. Scale bar, 10 μm. α-Tub, α-tubulin.

Mutations in specific genes have been linked to PCD in many cases because their resulting defects are detectable either by transmission electron microscopy (TEM), through qualitative observation of differences in axoneme cross sections, or by high-speed video microscopy (HSVM), through qualitative observation of altered cilia motility (13, 17). In most diagnostic centers around the world, the gold standard for confirming a diagnosis of PCD remains the demonstration of either a hallmark ciliary ultrastructural defect by TEM (outer dynein arm defect, outer and inner dynein arm defect, or microtubular disarrangement with inner dynein arm defect) and/or the presence of biallelic pathogenic genetic mutations in a known disease-causing gene by sequencing of a gene panel (3, 11). However, both tests have pitfalls: TEM detects differences in ultrastructure due to lack of ciliary components, but this analysis remains difficult to standardize, and it is semiquantitative. Some PCD proteins still show regular axonemal ultrastructure by conventional TEM despite being severely truncated or missing because of early termination mutation (DNAH11, DNAH9, RSPH4A, RSPH9, RSPH1, RSPH3, GAS8, CCDC65, CCDC164, RPGR, OFD1, GAS2L2, HYDIN, and SPEF2) (1821). Current clinical genetic test panels include targeted sequencing of upward of 32 genes, but this analysis is expected to have only ~70% sensitivity (3, 22) due to mutations outside gene panel coverage or in noncoding regions. In cases where a candidate genetic mutation is a variant of uncertain significance (VUS), validation by other diagnostic methods is essential (23). Whole-genome sequencing remains an option that is both expensive and challenging: Sequence analysis can take several months or years depending on the mutation due to locus and allele heterogeneity (24). In addition, sequencing is not always a viable option for diagnosis, especially in the European population, where PCD genetic testing is not always funded and some bias remains against genetic testing as a diagnostic tool (25). HSVM is frequently used, but it is not yet reliably quantitative; it is low throughput due to limited cilia sampling and requires expert analysis and confirmation of motility defects after air-liquid interface (ALI) epithelial cell culture (13, 17), decreasing its applicability in most clinical centers (26).

Despite recent progress, it is estimated that ~30% of patients with clinically suspected PCD remain without a confirmed diagnosis (13, 22). It is therefore necessary to develop quantitative, efficient, and comprehensive methods to increase sensitivity of PCD diagnosis and to assess the effect of genetic variants of PCD genes such as VUSs at the protein level.

Conventional immunofluorescence (IF) microscopy has previously been used to detect PCD proteins, but its sensitivity has been limited by availability of validated commercial antibodies, lack of quantitative analysis, and imaging resolution (27, 28). The emergence of super-resolution fluorescence microscopy and quantitative image analysis, together with systematic efforts to develop antibodies against the human proteome, suggests that its application for diagnostic purposes is now possible. Super-resolution microscopy methods increase the in-plane resolution to tens of nanometers from ~250 nm or more of conventional fluorescence microscopes, making imaging of protein assemblies within cilia of patients’ cells a viable possibility (29, 30).

Here, we present a toolbox of imaging methods based on three-dimensional (3D) structured illumination microscopy (3D-SIM) and stochastic optical reconstruction microscopy (STORM) super-resolution imaging that quantitatively detects mislocalization of PCD proteins, nanoscale structural defects, and lack of beating coordination in multiciliated cells from patients with PCD. A panel of 21 commercially available antibodies directed against PCD proteins was validated to detect molecular defects in multiciliated airway cells from patients with PCD and, using the super-resolution imaging toolbox, applied to 31 clinical PCD cases, including ones where TEM was nondiagnostic and/or VUSs were identified in known PCD genes by genetic testing. Together, data reveal both a global and detailed view of multiciliated cells’ protein organization and demonstrate how super-resolution imaging toolbox can be used as a complementary or stand-alone diagnostic tool for confirming or refuting PCD, thereby showing translational potential in clinical diagnosis.


PCD proteome mapping by super-resolution microscopy

To generate metrics enabling the quantitative comparison of airway multiciliated cells from healthy volunteers and patients with PCD, we built a super-resolution imaging map using a panel of 21 antibodies recognizing proteins linked to PCD (Fig. 1C and tables S1 to S3). We refer to this antibody set as the “PCD proteome” panel because it allows direct or indirect detection of mutations leading to lack of expression or targeting of a large number of known PCD proteins (41 of 45), including 10 proteins that cannot be detected by TEM (DNAH11, DNAH9, GAS8, CCDC65, RSPH4A, RSPH9, RPGR, OFD1, GAS2L2, and SPEF2; Fig. 1C and table S3). Most antibodies in the PCD proteome panel were validated by three or more methods recently suggested by international guidelines (31) including (i) labeling of cells from patients with loss-of-function mutations, (ii) cross-validation with antibodies generated against different antigen moieties, (iii) correlation of mRNA in cells with protein detected by IF, and (iv) Western blotting (fig. S1 and table S2).

To generate the PCD proteome map, we used airway multiciliated cells collected from the nostrils of healthy volunteers because nasal tissue is easily accessible with a cytology brush, without an operation and it is functionally similar to the lower airways of the lung (32). Isolated multiciliated cells were then immunolabeled with an optimized protocol and imaged by 3D-SIM, which features straightforward sample preparation, multicolor imaging, and high-throughput image acquisition (29, 33).

The PCD proteome map (Fig. 1C) shows three main localization patterns on motile cilia: 15 PCD proteins are distributed along the entire length of the axoneme, 1 is enriched in the proximal region (DNAH11), and 1 in the distal region (DNAH9). In addition to axonemal distribution, localization at the base of motile cilia was observed for eight PCD proteins (Fig. 1C and fig. S2A). The map is consistent with reported localization of 20 PCD proteins for which, in some cases, antibodies were not commercially available (table S3) and includes the distribution of a recently found PCD protein PIH1D3 (34, 35) (Fig. 1C and fig. S2B), which forms cytoplasmic puncta and has been reported to form a liquid-like organelle in multiciliated cells of Xenopus (36).

Molecular diagnosis of PCD by super-resolution imaging and quantitative image analysis

Once the reference-imaging map of the PCD proteome was established in cells from healthy volunteers, it was then compared to the map generated from cells of patients with PCD. Cells from 31 individuals with suspected or diagnosed PCD (data file S1) were analyzed after screening in the PCD clinic at the Hospital for Sick Children by TEM and exome sequencing by a commercial gene panel.

As discussed, a major limitation of the current PCD diagnostic tests is the lack of reliable quantitative analysis of patients’ samples. We therefore developed a diagnostic workflow including an assay to quantify differences in targeting of PCD proteins. Nasal cells from patients were collected and processed for IF, imaged by 3D-SIM using a commercial turnkey system, and quantified by automated data analysis. Image volumes were acquired either manually or in automated mode, corrected for chromatic aberration, and then fed into a custom-made MATLAB script to generate a maximum intensity projection used to determine the cell boundary and a mean intensity projection to determine a threshold separating the signal from background fluorescence in healthy cells. The script was then instructed to measure fluorescence intensity of PCD proteins in each z plane to quantify PCD protein amount in the whole cell, in the cilia region, and colocalization between PCD proteins and cilia, using α-tubulin as a marker to label the ciliary axonemal region (Fig. 2A). Fluorescence was quantified and compared between cells from patients with PCD and healthy volunteers stained side by side in three or more independent experiments. Images collected by three imaging techniques (wide field, deconvolution, and 3D-SIM) were compared, and 3D-SIM was selected for its resolving power and accuracy in quantification (fig. S3).

Fig. 2 PCD diagnosis by quantification of 3D-SIM images.

(A) Illustration of 3D-SIM quantification acquisition and script workflow. (B) Three examples of PCD cases used for 3D-SIM quantification. Left: Airway multiciliated cells from a patient (PCD17; data file S1) carrying DNAH5 mutations leading to complete loss of targeting to motile cilia. Middle: Airway multiciliated cells from a patient (PCD20; data file S1) carrying DNAH5 mutations leading to DNAH5 targeting to proximal region of motile cilia. Right: Airway multiciliated cells from a patient (PCD11; data file S1) carrying DNAH11 mutations leading to DNAH11 loss of targeting to motile cilia. Scale bar, 10 μm. (C) Results of 3D-SIM images quantification corresponding to three cases in (B). Student’s t test, **P < 0.01, ****P < 0.0001. n = 8, 13, 10, 10, 9, and 9 from N = 3 replicates.

The validity of the method was first tested on cells from patients with PCD with well-characterized loss-of-function mutations using a subset of the antibody panel including antibodies recognizing the protein of interest and at least two other antibodies recognizing proteins not affected by the potential loss of function; experimenters were left blind to the genetic data. For example, cells were analyzed from a patient (PCD17; data file S1) carrying mutations in DNAH5 [heterozygous, 1.6-kb deletion encompassing exon 2 and nonsense mutation c.4237C>T (p.Gln1413stop)], which led to lack of targeting of DNAH5 from ciliary axonemes (Fig. 2B, left). As expected, our quantitative analysis showed that DNAH5 fluorescence in the whole cell and in cilia was significantly lower in the patient’s cells relative to cells from healthy volunteers (P < 0.0001; Fig. 2C, left, and fig. S4A). Next, we analyzed more complex cases such as patient PCD20 (data file S1) carrying a previously uncharacterized VUS [DNAH5 heterozygous, nonsense mutation 9095C>G (p.Ser3032stop) and missense mutation c.11437C>T (p.Arg3813Trp)]. In this case, DNAH5 was not targeting to the entire length of the ciliary axoneme but rather only to the proximal region (Fig. 2B, middle). Quantitative analysis of 3D-SIM images showed that the average DNAH5 fluorescence in this patient’s cells was about half relative to cells of healthy volunteers (0.43 and 0.53, in cilia and in the whole cell, respectively; Fig. 2C, middle, and fig. S4B).

A third, previously undiagnosed PCD clinical case (PCD11; data file S1) presented by genome sequencing three VUSs and one deletion in DNAH11 [heterozygous, c.1328G>A, c7807C>T, c.11479G>A and ~500–base pair deletion encompassing part of exon 82], whose protein product mistargeting is not detectable by conventional TEM. In this case, DNAH11 appeared to be still expressed, as evidenced by cytoplasmic labeling, but did not target to motile cilia [Fig. 2, B and C (right), and fig. S4C]. Together, our data show that fluorescence microscopy and image quantification methods can detect differences in expression and localization patterns not discernable by TEM and, therefore, are useful to aid PCD diagnosis of complex cases.

Molecular diagnosis by PCD antibody panel and machine learning

To increase the sensitivity and automation of our method, it was further improved in two directions: (i) testing as a stand-alone method by analyzing the entire antibody panel on patients’ cells and (ii) implementation with machine learning algorithms to automate diagnosis. To test our method as stand-alone, we used a complex PCD case (PCD22; data file S1) that did not receive a conclusive molecular diagnosis for more than 10 years (Fig. 3A). The case presented with several mutations identified through a PCD gene panel [DNAH5 c13486C>T and 4.4-kb deletion (encompassing exons 40 to 42), DNAH11 c.100G>T (VUS), and DNAH11 c.101A>T (VUS)] (Fig. 3A). When the distribution of the PCD proteome map was observed in cells from this patient, mislocalization of three outer dynein arm proteins (DNAH5, DNAH11, and DNAI1) was detected. DNAH5, usually distributed along the entire length of the ciliary axoneme, localized only to the proximal end; this result explains the previously inconclusive diagnosis obtained by TEM, which showed a mixture of presence and absence of outer dynein arms. DNAH11 and DNAI1, usually distributed along the proximal end and the entire length of the ciliary axoneme, respectively, localized mainly to the cytoplasm (Fig. 3B). All other PCD proteins recognized in the PCD proteome panel localized as in cells from healthy controls (fig. S5).

Fig. 3 PCD diagnosis by antibody panel labeling and machine learning.

(A) Diagnosis history for a complex PCD case (PCD22; data file S1). (B) Diagnosis of a complex PCD case (PCD22; data file S1) by imaging of the PCD proteome: DNAH5 localizes to proximal cilia, and DNAH11 and DNAI1 mainly localize to the cytoplasm. Scale bar, 10 μm. (C) Illustration of machine learning workflow. SVM, support vector machine. (D) Machine learning distinguishes seven patients with DNAH5 mutations (PCD17, PCD20, PCD35, PCD36, PCD38, PCD40, and PCD42; data file S1) from four healthy controls by the percentage of cells classified as PCD. n = 302, 251, 120, 93, 580, 145, 136, 114, 88, and 83. Error bar represents 95% confidence interval (means ± 1.96 SD). WT, wild type; CNV, copy number variants; ODA, outer dynein arm.

To test the possibility of automating diagnosis, we trained a classifier to recognize cells from patients with mutations in DNAH5, the most common PCD-causing gene in the North American population (table S2). After 3D-SIM–automated data collection, images of cells labeled with anti-DNAH5 antibody (green channel) and cilia marker α-tubulin (red channel) were extracted, projected, and analyzed for green/red pixel cross-correlation (fig. S6A). Data obtained from cells of two healthy volunteers and two patients carrying mutations in DNAH5 (PCD23, c.1432C>T and c.11571-1G>A; PCD43, c.7643T>C) (data file S1) leading to lack of targeting to motile cilia were fed to a support vector machine–based (37, 38) classifier to train a model (Fig. 3C). Next, new data obtained from cells of three healthy volunteers and seven patients with DNAH5 mutations (PCD17, PCD35, PCD36, PCD38, PCD40, PCD42, and PCD20; data file S1) were analyzed to test classifier performance (Fig. 3D). The results showed that the algorithm reliably distinguished cells from patients with PCD with DNAH5 loss-of-function mutations and with DNAH5 partially localized to motile cilia from the cells of healthy individuals (Fig. 3D and table S4). To further increase the applicability of the machine learning algorithm to other prevalent PCD genes, we tested cells from patients with PCD with loss-of-function mutations in CCDC39 and DNAH11. As shown in fig. S6 (B and C), the classifier distinguished cells from patients with CCDC39 and DNAH11 mutations from healthy controls. Together, these results show that fluorescence image analysis by machine learning can be applied to diagnose PCD.

Structural analysis of cells from patients with PCD by STORM imaging

As shown above, IF 3D-SIM combined with the PCD proteome panel allows specific, accurate, and straightforward molecular diagnosis of PCD and, therefore, can be potentially standardized for clinical diagnosis. However, 3D-SIM only features a twofold increase in resolution; therefore, its capacity to detect fine structural changes is limited. To improve our understanding of how specific PCD mutations affect the axonemal structure in patients’ cells, we used STORM, a technique achieving ~25-nm in-plane and ~50-nm axial resolution, which allows detailed visualization of distributions of specific PCD proteins (fig. S7) (39, 40). Here, our analysis focused on the molecular ruler, a structure below the diffraction limit located on the doublets of eukaryotic cilia and flagella (41), and specifically on components of N-DRC and RS, which are distributed along the axoneme longitudinal axis in distinct 96-nm intervals (42). As expected, when multiciliated cells from healthy volunteers were labeled with antibodies against N-DRC protein GAS8 and imaged by conventional fluorescence imaging, the ciliary axonemes showed uniform labeling of GAS8 (Fig. 4A). However, when STORM imaging was used, a structure with periodical GAS8 clusters became visible (Fig. 4A). To better characterize this organization, first, we generated fluorescence intensity distributions along cilium fragments to show the periodic pattern (Fig. 4B, top and middle). Subsequently, Fourier transform of the intensity distribution was performed, which identified a main peak around 0.01 nm−1 in the frequency domain corresponding to 98.5-nm periodicity (Fig. 4B, bottom). RS head protein RSPH4A was also imaged by STORM (Fig. 4C) to detect a second independent protein complex periodically distributed along the axoneme. Measurements of GAS8 cluster-to-cluster distances along motile cilia show that the mean spacing is 95.7 ± 10.2 nm (n = 223; Fig. 4D, left), a distance consistent with cryo–electron tomography (cryo-ET) measurements (42). Similar to GAS8, a periodic spacing of 98.3 ± 8.67 nm (n = 81) was also detected for RSPH4A (Fig. 4D, right). The periodic patterns of both GAS8 and RSPH4A appeared as two main clusters in the lateral direction, a distribution that likely results from a combined effect of intrinsic low number of molecules in the complex and reduced labeling efficiency as shown by STORM simulations and antibody titration experiments (Fig. 5).

Fig. 4 STORM detects 96-nm periodicity on motile cilia axonemes.

(A) STORM imaging of GAS8. Left: Comparison between STORM imaging and conventional fluorescence imaging. Insets in white boxes are high-magnification images of three ciliary regions showing periodic distribution. Right: High-magnification image of an intact cilium from the STORM image on the left. (B) Top: STORM image of a short cilium in (A). Middle: Fluorescence intensity distribution along the cilium. Bottom: Fourier transform of the intensity distribution showing 98.5-nm periodicity. a.u., arbitrary units. (C) STORM imaging of RSPH4A showing 98.5-nm periodicity. (D) Left: Histogram of the GAS8 96-nm repeat distribution in (A), with a mean of 95.7 nm, a median of 95.8 nm, and an SD of 10.2 nm (n = 223). Right: Histogram of the RSPH4A 96-nm spacing distribution in (C), with a mean of 98.3 nm, a median of 96.5 nm, and an SD of 8.67 nm (n = 81).

Fig. 5 STORM image is a result of low intrinsic molecule number of GAS8 or RSPH4A and signal detection efficiency.

(A) Illustration of signal efficiency. Each GAS8/RSPH4A complex (single dot in the nine-dot ring structure, end-on view of the axoneme) consists of N GAS8/RSPH4A molecules. Because of the intrinsic affinity of the antibody to the antigen, antibody concentration and incubation time, intrinsic blinking properties of the fluorophores, and sampling frequency, not all GAS8/RSPH4A molecules in the complex are detected during STORM imaging. Here, we define the percentage of detected GAS8/RSPH4A molecules to all GAS8/RSPH4A molecules in the complex as signal efficiency. Suppose that there are N GAS8/RSPH4A molecules in one complex and the signal efficiency is Pse, the percentage of the complexes detected with 0, 1, 2, …, N GAS8/RSPH4A molecules could then be calculated on the basis of the combinatory logic. The actual molecules detected in each complex obey the Gaussian distribution with a mean N*Pse and SD N*Pse*(1-Pse) (three for instance here). (B) Illustration of the STORM simulation process: The plot on the left is 2D projection of GAS8 or RSPH4A along cilium with color representing the relative z position (1-μm length along cilia and 160-nm diameter). A Gaussian random number generator is used to output the number of the molecules detected in one dot inputting N and Pse. The position of each molecule is then determined by a second Gaussian random number generator by inputting the assigned coordinate based on the supposed theoretical structure and localization precision. The position is duplicated certain times to simulate the blinking of the dyes (blinking times also obey Gaussian distribution). Last, coordinates are rendered into a 2D image. The parameters used here are the following: N = 3, Pse = 0.125, blinking times = 30, localization precision = 12 nm (lateral) and 36 nm (axial), and bin size = 5 nm. (C) STORM simulation showing that signal efficiency influences detection of GAS8/RSPH4A structure. (D) STORM simulation showing that the number of GAS8/RSPH4A molecules in the complex influences the STORM image. (E) Representative images showing that increasing antibody concentration and incubation time improve labeling efficiency while compromising image quality through increase in background. Boxed insets are shown at higher magnification.

Next, structural changes in the axoneme of cells from patients with PCD were examined. We focused on motile cilia from patients with PCD with mutations in CCDC39 because its Chlamydomonas homolog, fap59, is responsible for recruitment of N-DRC complex and inner dynein arm proteins. It was previously shown that in Chlamydomonas mutants lacking fap59, the 96-nm axonemal repeat of RS proteins is shortened to 32 nm (41). TEM studies on cross sections of mammalian axonemes, however, had suggested that 96-nm repeat of RS proteins is detected in the absence of CCDC39 (43). This raised the question of whether CCDC39 is required for RS 96-nm repeat spacing in human motile cilia axonemes.

To answer this question, we first examined GAS8 distribution in cells from three patients (PCD14, PCD19, and PCD33; data file S1) with CCDC39 mutations leading to lack of protein targeting to motile cilia (fig. S8A). As expected, GAS8 was not targeted to the ciliary axoneme; however, it was found at the cilia base (fig. S8B), together with N-DRC protein CCDC65 and inner dynein arm protein DNALI1, consistently with previous reports (43, 44) (fig. S8C). To characterize the precise location, we performed dual labeling with antibodies targeting transition zone protein RPGRIP1L/CEP290 and distal appendage protein CEP164 to show that GAS8 is located in the transition fiber region rather than transition zone in patients with CCDC39 mutations (Fig. 6A and fig. S8D). STORM combined with 3D averaging (45) further demonstrated that GAS8 is distributed in a ninefold symmetrical organization, which is consistent with localization to transition fibers because basal feet do not present a ninefold symmetrical organization in motile cilia (Fig. 6, B and C, and fig. S8E) (46, 47).

Fig. 6 RSPH4A remains distributed in 96-nm intervals in CCDC39 mutant cells.

(A) 3D-SIM imaging shows GAS8 is trapped at the transition fibers/distal appendages in cells from patients with CCDC39 mutations. (B) STORM imaging shows GAS8 forms nine–cluster ring structure at the transition fiber. (C) Averaging of STORM images shows nine–cluster ring structure. (D) Airway multiciliated cells from patients with CCDC39 mutations illustrating the 96-nm repeats for RSPH4A shown by STORM imaging. (E) Top: STORM imaging shows RSPH4A distribution along a cilium from a patient with CCDC39 mutations. Middle: RSPH4A intensity distribution along the cilium. Bottom: Fourier transform of the intensity distribution in the middle. (F) Histogram of the RSPH4A 96-nm spacing distribution from cells from three patients with a mean of 97.0 nm, a median of 96.3 nm, and an SD of 11.5 nm (n = 515). (G) A model describing the difference between cells from healthy controls and patients with CCDC39 mutations: RSPH4A forms 96-nm periodicity along microtubule doublets. GAS8 connects nine microtubule doublets thereby favoring the lateral alignment of the 96-nm periodicity. In airway multiciliated cells from patients, GAS8 fails to be transported to the cilia; as a result, the microtubule doublets are not well aligned and show disorganization although RSPH4A still forms 96-nm repeats along each of the microtubule doublets.

Next, RSPH4A distribution was examined in cells from patients with CCDC39 mutations. RSPH4A is considered a main component of RS in human cilia required for RSPH1 and RSPH9 recruitment (41). STORM showed that the 96-nm repeat of RSPH4A was present and showed expected periodicity [97.0 ± 11.5 nm (n = 500)]; however, there was lack of relative alignment of RSPH4A between microtubules in CCDC39 patients’ cells compared to healthy volunteers (Fig. 6, D to F), consistent with the microtubule disorganization phenotype observed by TEM. Together, our data show that CCDC39 is required for targeting of GAS8 from the cilia base to the axonemal region. In the absence of CCDC39 (and consequently GAS8), specific RS proteins such as RSPH4A are still distributed with a 96-nm repeat interval, albeit appearing less ordered (Fig. 6G), suggesting potential differences in the organization of axonemes between Chlamydomonas flagella and human airway cells (42, 48, 49).

Rotational polarity analysis distinguishes cells from patients with PCD, cystic fibrosis, and healthy controls

The PCD diagnostic methods presented above rely on a priori knowledge of the PCD gene/protein and availability of specific antibodies. To circumvent these limitations, we developed an imaging tool that leverages motile cilia coordination as a readout for defects in motile cilia beating (50). When motile cilia beat coordinately, the basal feet—conical structures located at the base of motile cilia—are pointing direction of beating, a process termed rotational polarity (Fig. 7A) (5052). To ensure coordination, basal feet link hundreds of motile cilia in a network on multiciliated cell surface through microtubules (51, 5355). When basal feet are lost by deletion of basal foot protein ODF2, uncoordinated ciliary beating and impaired mucociliary clearance are observed (51). Odf2−/− knockout mice show PCD respiratory symptoms such as coughing, sneezing, sinusitis, and otitis media (51).

Fig. 7 Rotational polarity analysis reveals the misalignment of basal body/basal foot pairs in airway multiciliated cells from patients with PCD.

(A) Schematic showing rotational polarity in cells from healthy individuals and patients with PCD. (B) 3D-SIM twofold increase in resolving power is required for image analysis of rotational polarity in airway multiciliated cells grown in ALI. Boxed region is shown at higher magnification. Image on the right shows the identified basal body/basal foot pairs in the cell shown in the middle. (C) Top: A representative cell from a healthy volunteer. Bottom: A representative cell from a patient with DNAH5 mutations leading to loss of targeting. (D) Alignment vector length distribution for different patient genotypes (PCD and CF) compared to controls. Healthy control, N = 4 samples; DNAH5, N = 4 samples; DNAH11, N = 3 samples; HYDIN, n = 667 cells from three differential trials; CF, N = 2 samples. Means = 0.54, 0.55, 0.59, 0.52, 0.19, 0.32, 0.25, 0.23, 0.18, 0.29, 0.16, 0.18, and 0.56. n = 116, 193, 57, 788, 199, 221, 106, 192, 466, 205, 224, 667, and 348. Data were analyzed with Student’s t test.

Rotational polarity has been previously examined by TEM (50, 5659). This analysis is mainly performed by examining position of neighboring central pairs, which do not distinguish the forward or back orientation of beating, or by examining cross sections at the ciliary base at the requisite focal plane where the basal feet are located: It remains, therefore, a time-consuming, low-throughput, and mainly qualitative method (5760). To overcome these limitations, we devised a strategy using super-resolution fluorescence microscopy—wide-field and deconvolution microscopy fail to resolve basal body/basal foot pairs (fig. S10)—and automated image analysis to quantify rotational polarity in PCD cells (Fig. 7B). To measure rotational polarity, we used patients’ cells differentiated from basal stem cells in ALI. On ALI supports, cells generate a pseudostratified ciliated epithelium oriented top-down, which allows resolution of basal body/basal foot pairs by super-resolution imaging (Fig. 7B). To detect alignment, we first labeled multiciliated cells with markers of the basal body (POC1B) and the basal foot (Centriolin; Fig. 7, A and B). A custom MATLAB script was developed to automatically identify basal body and basal foot pairs, to measure the orientation of each pair relative to the image axis, and to calculate the norm (vector length) of the average direction for all pairs of the entire cell to determine the overall cilia alignment score (1, full alignment; 0, no alignment) (Fig. 7, B and C).

To assess the reliability of the method, we first analyzed cells obtained from multiple healthy volunteers to establish the alignment baseline of ciliary rotational polarity [0.54 ± 0.03 (N = 4 healthy controls, n = 1154 cells)] (Fig. 7C, top). Subsequently, cells from patients with PCD (PCD23 and PCD38; data file S1) with known mutations (DNAH5) leading to immotile cilia were analyzed as positive controls to establish rotational polarity lower bound (Fig. 7C, bottom, and fig. S10]. As expected, in cells from patients with PCD with immotile cilia, rotational polarity dropped to 0.25 ± 0.09 [N = 2 patients, n = 420 cells; Fig. 7, C (bottom) and D].

Rotational polarity analysis was then tested in cells from patients with mutations leading to severe motile cilia beating defects such as DNAH11 (61) (PCD25, PCD39, and PCD40; data file S1) or with subtle defects such as HYDIN mutants (62) (PCD24; data file S1) and DNAH5 leading to partially localization (PCD20 and PCD36; data file S1). We found that rotational polarity was reduced similarly in other PCD patients’ cells [DNAH11, 0.21 ± 0.07 (N = 3, n = 895 cells); HYDIN, 0.18 ± 0.09 (N = 3 differentiation trials, n = 667 cells); and DNAH5 with partial localization, 0.24 ± 0.01 (N = 2, n = 298 cells)] (Fig. 7D, fig. S10, and table S5).

A potential feature of this method is to distinguish specific defects in motile cilia beating from secondary defects in other molecular pathways linked to respiratory diseases. To test this possibility, we analyzed cells derived from two patients with cystic fibrosis (CF; F508/w1282x and F/2183AA>G mutations in CFTR) for rotational polarity. CF is a rare lung disease caused by mutations in CFTR that ultimately causes impaired mucociliary clearance leading to bronchiectasis similar to PCD (63, 64); however, CF affects periciliary fluid hydration rather than motile cilia beating (65, 66). Rotational polarity in cells from patients with CF was similar [0.56 ± 0.04 (N = 2, n = 348 cells)] to healthy volunteers’ cells where mucus was removed from cultures (Fig. 7D and fig. S10), suggesting that rotational polarity disruption is a distinct characteristic of PCD and, therefore, that this method can be used to distinguish primary ciliary motility defects from indirect defects.


Here, we describe the development and application of a super-resolution imaging workflow (fig. S11) to diagnose PCD. In this study, a discovery cohort of 31 patients with highly diverse ethnic background was recruited, including 12 TEM normal/unclear cases and 15 cases with VUSs. Most PCD cases (21 of 31) could be diagnosed by 3D-SIM with the corresponding validated antibody. The remaining 10 cases were diagnosed indirectly with antibodies from the panel, including a patient with HYDIN mutation, which was solved with the recently described anti-SPEF2 antibody (67).

Our 3D-SIM and 3D volume image analysis method features several advantages to diagnose PCD relative to TEM analysis: (i) It is quantitative; (ii) it provides a 3D view of hundreds of motile cilia along their entire length per cell, rather than a few cross sections; and (iii) it allows detection of 14 PCD proteins whose lack of targeting is undetectable by TEM. When applied to cells from our patient cohort, this method was accurate and specific and provided insightful information about how specific genetic mutations affected targeting of PCD proteins in situ. Quantitative image analysis/machine learning coupled with super-resolution imaging further showed strong potential to validate VUSs on known PCD genes. In addition, this method is easily amenable to automation, from image acquisition and analysis to diagnosis by machine learning algorithms. Together, it shows promise of integrating or replacing some of the current clinical diagnostic methods because it is sensitive, simple, and fast and it links specific genotypic mutations to protein mistargeting. It is worth noting that the IF 3D-SIM quantitative method can also be applied to images obtained by conventional wide-field microscopy but only for analysis of images from cells with complete lack of targeting of PCD proteins. This could be particularly useful in countries with limited resources for a PCD diagnostic center. For cases where partial localization of PCD proteins in cilia is observed (which accounted for 27% of all DNAH5 cases in our cohort), deconvolution or 3D-SIM microscopy is critical. To facilitate the translation of this method for clinical diagnosis, the antibody panel could be restricted to 10 antibodies that recognize proteins nondetectable by TEM or that detect other PCD proteins mislocalization indirectly (DNAH5, DNAH11, DNALI1, GAS8, CCDC65, RSPH4A, RSPH9, RPGR, OFD1, and SPEF2; fig. S12).

To resolve the 96-nm axonemal repeat within cilia, STORM imaging was required. High-resolution EM methods such as cryo-ET have been previously used to study the ultrastructure of the axoneme in cells from healthy individuals and from one patient with PCD (42); however, the complexity and costs associated with this methodology are high due to the necessary specialized equipment, computationally intensive subtomogram averaging, and special protocols for cilia purification. Therefore, this methodology is technically challenging and not yet accessible (42). STORM, however, might provide a time- and resource-effective alternative to cryo-ET because it uses small amounts of cells directly isolated from patients’ nasal biopsies without the need of axoneme purification, allows for assessment of the specific distribution of PCD proteins through antibody labeling, and uses microscopes widely available in many imaging facilities worldwide that cost a fraction to purchase and maintain relative to cryo-ET.

Rotational polarity analysis–based PCD diagnosis is versatile, useful in cases where a genetic cause has not yet been established or when other methods fail. It requires super-resolution microscopy because conventional fluorescence microscopy (wide field and deconvolution) fails to resolve basal body/basal foot pairs, a critical step required to measure rotational polarity. Note that this method is both robust and time effective because it allows analysis of hundreds of cells in a few hours, a task that would require days or weeks by TEM, which is burdened by the complexity of analyzing consecutive sections. On the basis of our data, a vector length threshold between 0.49 (lowest value of controls) and 0.37 (highest value of patients) is suggested as a cutoff to distinguish patients with PCD from healthy controls. Multiciliated cells displaying a lack of basal body alignment but no detectable defect in axonemal ultrastructure by EM have been previously described in medical case reports of patients with PCD and were recently linked to a new PCD gene GAS2L2 (5659). This suggests that this method might help identify new PCD genes regulating basal body–cytoskeleton interaction (50, 55, 60).

Among the methods described here, quantitative IF 3D-SIM and rotational polarity analysis are the most promising for translation into clinical diagnostic tools. IF 3D-SIM could be further improved by implementing high-throughput staining and data collection methods such as multichannel microfluidic chips and automatic cell recognition/data analysis algorithms. Rotational polarity can be used as a quantitative alternative in cases that cannot be diagnosed by other methods because assay time is longer due to requirement for ALI culturing that ensures the proper cell orientation for imaging. For all methods discussed, a larger pool of PCD cases and healthy controls with different genetic background needs to be assayed before translation into clinical diagnostic pipeline.

In summary, quantitative 3D-SIM and STORM can be used as part of a diagnostic workflow (fig. S11) to fill the diagnostic gap in PCD, either by complementing the diagnosis of PCD by TEM or genetic testing or by diagnosing PCD cases directly through antibody multiplexing. Development of validated antibodies against every protein in the proteome is under way; therefore, we anticipate that similar imaging strategies will be applied to other ciliopathies and multigenic inherited disease where relevant cells are easily accessible from patients.


Study design

The purpose of this project was to develop a super-resolution imaging–based toolbox to aid the diagnosis of airway disease PCD. To achieve this goal, 31 PCD cases (confirmed diagnosis with biallelic pathogenic mutations in known PCD genes or hallmark ultrastructural ciliary defects on TEM) and 6 healthy volunteers (no diagnosis of chronic lung disease) were recruited. Nasal airway cells were obtained from individuals recruited in the study. All patients were free from acute infection for at least 3 weeks and were nonsmokers. Patients were recruited consecutively during routine visits to the PCD clinic at the Hospital for Sick Children. The researchers were blind to the genotype of PCD cases. To improve diagnosis by super-resolution imaging techniques, three different methods were developed. The details of study design and sample sizes are provided in corresponding figures, legends, and Materials and Methods. No data were excluded from analyses. All data shown were results of experiments repeated at least three times to verify reproducibility. The protocol used in this study was approved by the Hospital for Sick Children Ethics Committee [Research Ethics Board (REB) #1000054690]; all patients consented to participate in this study.

Statistical analysis

Statistics in this study were performed using Student’s t test. For the machine learning analysis (Fig. 3D), error bar represents 95% confidence interval. Details are specified in the main text, figures, and figure legends.


Materials and Methods

Fig. S1. Antibody validation.

Fig. S2. Proteins localizing at the base of cilia and dual color imaging of PIH1D3 puncta and stress granule marker.

Fig. S3. Comparison of 3D-SIM, wide-field, and deconvolution microscopy for IF quantification.

Fig. S4. Quantification of PCD protein in the whole cell, α-tubulin in cilia, and colocalization between PCD protein and cilia for the three PCD cases in Fig. 2B.

Fig. S5. PCD diagnosis by antibody panel labeling.

Fig. S6. Machine learning analysis.

Fig. S7. STORM imaging of PCD protein OFD1.

Fig. S8. Super-resolution microscopy reveals the pattern of GAS8 and RSPH4A for patients with CCDC39 mutations.

Fig. S9. Comparison of 3D-SIM, wide-field, and deconvolution microscopy for rotational polarity analysis.

Fig. S10. Rotational polarity analysis.

Fig. S11. Proposed workflow of PCD diagnosis.

Fig. S12. PCD diagnosis by 10 antibody panel labeling.

Table S1. All antibodies used in this study.

Table S2. PCD gene percentage and antibody validation.

Table S3. Antibody coverage and protein localization summary.

Table S4. Summary for machine learning results.

Table S5. Rotational polarity analysis result.

Data file S1. All PCD cases used in this study.

Data file S2. Raw data for Fig. 2C and fig. S4.

References (6871)


Acknowledgments: We thank all of the patients and volunteers who participated in this study, M. Miki and M. Sawras for coordinating the patients, and J. Avolio for help with nasal cell scraping. Funding: This project is funded by the Ontario Lung Association and start-up funds to V.M. from the Hospital for Sick Children, Canada and Biomedical Research Center, National Health Research Institute, University of Southampton, UK; Z.L. was supported by the SickKids Restracomp Fellowship Program; R.G.H. was supported by HSBC Catalyst Grant. Author contributions: V.M. and S.D.D. conceived the project. V.M. and Z.L. designed experiments. S.D.D. recruited and phenotyped patients. Q.P.H.N. and A.A. did airway cell culture. Z.L. and Q.G. performed immunolabeling and collected and analyzed the data. Q.G. did statistical analysis. Z.L. wrote MATLAB scripts for rotational polarity analysis and STORM simulation and performed STORM experiments. J.K. collected data for machine learning. Y.M., L.E., and A.G. did the machine learning data analysis. R.G.H. collected the TEM data. H.O. and T.M. cultured samples from patients with cystic fibrosis. S.D.D. edited the paper. V.M. and Z.L. wrote the paper. Competing interests: The authors declare that they have no competing financial interests. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. Custom MATLAB scripts and detailed protocols are available at

Stay Connected to Science Translational Medicine

Navigate This Article