If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Address for reprints: Jian Tajbakhsh, PhD, 3rd St Diagnostics, Faculty, Samuel Oschin Comprehensive Cancer Institute, Cedars-Sinai Medical Center, 8700 Beverly Blvd, Los Angeles, CA 90048.
Sputum is a source of exfoliated respiratory epithelial cells transformed early in lung carcinogenesis. Malignant cells are hypomethylated and contain less genomic 5-methylcytosine (5mC). Validating a test that recognizes and quantifies aberrantly hypomethylated cells in sputum, we assessed its potential as a screening tool for detecting early-stage non–small cell lung cancer.
Methods
Cells extracted from sputum were immunofluorescence labeled with an anti-5-methylcytosine antibody and counterstained with 4′,6-diamidino-2-phenylindole (DAPI) delineating global nuclear DNA (gDNA). Via confocal scanning and 3-dimensional image analysis, fluorescence 5mC and DAPI signals were measured in segmented cell nuclei, and a 5mC/DAPI co-distribution map was generated for each imaged cell. Cells were classified as hypomethylated based on 5mC load and 5mC/DAPI co-distribution. The proportion of hypomethylated epithelial cells in the sputum determines whether a patient has lung cancer.
Results
A total of 88 subjects were enrolled: 12 healthy subjects; 34 high-risk subjects with benign chronic lung disorders (10 with chronic obstructive pulmonary disease, 24 with idiopathic pulmonary fibrosis), and 43 subjects with non–small cell lung cancer (27 with stage I-II and 16 with stage III-IV). The test identified early-stage non–small cell lung cancer and distinguished it from the high-risk group with 95.8% (95% confidence interval, 78.9-99.9) sensitivity and 41.2% (95% confidence interval, 24.6-59.3) specificity applying only 5mC, 95.8% (95% confidence interval, 78.9-99.9) sensitivity and 26.5% (95% confidence interval, 12.9-44.4) specificity using solely 5mC/DAPI index, and 100% (95% confidence interval, 98.7-100) sensitivity and 26.1% (95% confidence interval, 26.2-27.8) specificity with the combined parameters.
Conclusions
We tested and validated a novel, noninvasive, highly sensitive screening test for non–small cell lung cancer. With the use of sputum, our test may impact lung cancer screening, evaluation of pulmonary nodules, and cancer surveillance algorithms.
The fluorescence microscopic image shows epithelial cells that have been extracted from human sputum and labeled for global DNA methylation (5mC, false-colored green), and global DNA (DAPI, false-colored blue); both markers are localized in the cell nucleus. Also, the cell bodies have been delineated by a cytoplasmic marker (false-colored red). The normal sputum contains a majority of methylated cells (cell 1 type in the fluorescence image), whereas the cancer sputum comprises a bulk of hypomethylated cells (cell 2 type). Methylated cells are characterized by a high 5mC/DAPI colocalization index (∂) and hypomethylated cells have a low index (< 15°). DAPI, 4′,6-diamidino-2-phenylindole; 5mC, 5-methylcytosine; gDNA, global nuclear DNA.
With the use of sputum, we have developed a novel, highly sensitive screening test for NSCLC. This may impact lung cancer screening, evaluation of pulmonary nodules, and cancer surveillance algorithms.
Currently, there are no reliable noninvasive, nonradiologic screening tests for NSCLC despite being the most common cause of cancer mortality. By using induced sputum specimens to quantify the proportion of hypomethylated cells, our novel test may be able to impact lung screening and surveillance by providing a high-sensitivity adjunct to the traditional algorithm.
When detected early, patients with early-stage non–small cell lung cancer (NSCLC) can undergo complete surgical resection with a relatively high 5-year survival.
The IASLC lung cancer staging project: proposals for revision of the TNM Stage Groupings in the forthcoming (eighth) edition of the TNM classification for lung cancer.
High-risk groups have been identified and include smokers and patients with chronic pulmonary diseases, such as idiopathic pulmonary fibrosis (IPF) and chronic obstructive pulmonary disease (COPD). Additionally, patients who have undergone resection for lung cancer have a high risk (up to 50%) of recurrence within the first 2 years after surgery and require close surveillance. Current approaches in diagnosing lung cancer include bronchoscopy, chest x-ray, computed tomography (CT), and positron emission tomography (PET); each is limited in detecting early-stage lung cancers. CT scans have a 20% to 30% risk of false-positive results.
In addition, CT imaging, even at lower doses (low-dose CT), may increase the risk of developing secondary malignancies due to repetitive exposure of ionizing radiation, as used for serial and frequent monitoring.
Thus, there is opportunity for alternative or adjunctive technology to augment our current approach of diagnosing lung cancer.
Sputum is a valuable tissue surrogate and a proliferative source of upper respiratory cells that undergo malignant transformation. Earlier trials using sputum cytology for noninvasive early detection of lung cancer were dependent on morphological changes of exfoliated epithelial respiratory cells; detection used staining methods such as Papanicolaou, May-Grunwald, Ziehl-Neelsen and Gomori-Methenamine, or noncommercial antibody-based colorimetric staining. These trials were not consistent in detecting lung cancer and provided no clinical advantage.
This approach has been limited as a general detection test because it is dependent on identifying specific abnormalities. The heterogeneity of lung cancer makes this initial approach challenging as applied methods average gene methylation across a large number of cells, thereby obscuring significant subtle information specific to a subgroup of cancerous cells in sputum.
To address these limitations, we propose a novel approach that analyzes the global DNA methylation status of exfoliated respiratory epithelial cells in the sputum on a cell-by-cell basis (Figure 1 and Video Abstract). Global genome-wide DNA methylation in epithelial cells occurs early in tumor development and is ubiquitous across all major types of cancer.
Global hypomethylation affects largely repetitive DNA elements (comprising ∼50% of the human genome), results in genome instability, and parallels closely the degree of stage-defined malignancy.
The availability of more sensitive technologies, combining immunofluorescence labeling, high-resolution confocal scanning, and single-cell image cytometry for quantifying epigenetic biomarkers in extracted respiratory cells, allows the identification of aberrant cells, leading to the detection of lung cancer. The process involves inducing sputum, followed by staining isolated cells with fluorescent reporters that produce a specific pattern in the nuclei of labeled cells visualized by light microscopy and analyzed with a high-content imaging platform. The specific pattern produced is associated with the nuclear distribution of methylated DNA, which is altered in lung cells during early cancer development, differentiating it from normal respiratory cells.
Figure 1The study protocol and different histograms for 5mC load and 5mc/DAPI colocalization index for each of the study groups are described. There is a significantly increase in hypomethylated cells from the normal population to the high-risk benign to early-stage lung cancer to late-stage lung cancer. This similar phenomenon is illustrated in the scatter plots. 3D, 3-Dimensional; 5mC, 5-methylcytosine; DAPI, 4′,6-diamidino-2-phenylindole.
DNA methylation topology differentiates between normal and malignant in cell models, resected human tissues, and exfoliated sputum cells of lung epithelium.
who previously used this cell-phenotyping assay, identified DNA methylation patterns that are specific to cancerous cells of the lung epithelium. The matching sputum of patients with cancer contained cells with aberrant DNA methylation patterns that resembled the signature of cancer cells in the malignant tissue. The fundamentals of our high-content single-cell analysis—based on the 3-dimensional quantitative DNA methylation imaging (3D-qDMI) platform to expose and enumerate aberrantly methylated cells—have been well reported and the approach validated in basic studies using a variety of cells and tissues.
DNA methylation topology differentiates between normal and malignant in cell models, resected human tissues, and exfoliated sputum cells of lung epithelium.
Nuclear DNA methylation and chromatin condensation phenotypes are distinct between normally proliferating/aging, rapidly growing/immortal, and senescent cells.
Following up on the recent encouraging data with cells of the lung origin, we hypothesized that the detection and enumeration of aberrantly methylated cells in sputum may serve as an indication for early tumor development. We report on the first clinical study to assess the feasibility of using DNA methylation topology toward early lung cancer detection.
Materials and Methods
Study Design and Patients
The study protocol was approved by the Cedars-Sinai Internal Review Board (#Pro00039831 and Pro00046876, approval date May 1, 2016, and January 23, 2018). The Institutional Review Board or equivalent ethics committee of Cedars-Sinai Medical Center approved the study protocol and publication of data. The patients provided informed written consent for the publication of the study data. This prospective study enrolled 88 subjects, divided into 3 groups: 11 (12%) in the healthy/normal group with no history of cancer or pulmonary diseases; 34 (39%) in the high-risk/benign group with benign chronic lung disorders (10 COPD, 24 IPF) without cancer; and 43 (49%) in the cancer group with patients who have NSCLC and have not received treatment (27 with stage I-II, 16 with stage III-IV). Subjects were recruited based on International Classification of Diseases 9th Revision diagnosis code. All diagnoses were confirmed via review.
Sample Collection and Processing
Figure 2 depicts the workflow and process. Sputum induction was performed by inhalation of hypertonic saline. The sputum collection process was standardized for all subjects. Samples were collected and kept at 4 °C until processing. A sputum of 1 to 5 mL of volume will yield thousands of epithelial cells and is sufficient for performing the test. Samples were diluted with Sputolysin (Calbiochem) Worksolution, a phosphate-buffered saline (PBS) solution containing 10 mmol/L dithiothreitol. After every incubation, the sample was centrifuged at 400g for 10 minutes at 4 °C to separate cellular and fluid phases. The cell pellet was resuspended in PBS and then transferred onto a microscope slide using a cytospin apparatus for cell attachment to the glass support. Samples were transferred onto glass slides and immediately fixed with 4% paraformaldehyde and kept in PBS at 2 °C to 8 °C until used in downstream processing. Fixed slides were subjected to an immunofluorescence assay.
Figure 2Schematic of workflow process. TIFFs, Tag image transfer files; 5mC, 5-methylcytosine; DAPI, 4′,6-diamidino-2-phenylindole; 2D, 2-dimensional.
Nuclear DNA methylation and chromatin condensation phenotypes are distinct between normally proliferating/aging, rapidly growing/immortal, and senescent cells.
Primary and secondary antibody sets included unconjugated monoclonal mouse anti–5-methylcytosine (5mC) monoclonal antibody (Cat. No. AMM99021, Aviva Systems Biology) at 1 μg/mL concentration and Alexa488-linked donkey anti-mouse immunoglobulin-G (Cat. No. A-21202, Thermo Fisher Scientific) at 5 μg/mL concentration. The specificity/dynamic range of the antibodies was tested as previously reported.
Specimens were counterstained with 4′,6-diamidino-2-phenylindole (DAPI) (Cat. No. D1306, Thermo Fisher Scientific) for detection of global nuclear DNA (gDNA). Cell Mask Red (Cat. No. H32712, Thermo Fisher Scientific) was used to delineate cell cytoplasm for identification of flat epithelial cells. By staining of the large skirt-like cytoplasm, epithelial cells can be easily identified and distinguished from immune cells with a significantly smaller cell body under a fluorescent microscope. The staining lends itself to convenient size exclusion of cells other than epithelial cells in an automated mode. Figure 3 shows impressions of labeled cells isolated from different sputum samples. For each sputum sample, we analyzed an average of 150 cells.
Figure 3Fluorescent images of selected sputum samples. A, High-resolution area on the slide with few cells showing a less crowded sample compared with cell sample in (B) with significantly more immune cells (monocytes), due to a higher degree of lung inflammation. 5mC, 5-methylcytosine; DAPI, 4′,6-diamidino-2-phenylindole.
Subcellular-resolution images were acquired with a TCS SP5 X Supercontinuum confocal scanning microscope (Leica Microsystems). Serial optical sections were collected at increments of 500 nm. The output file format is a series of TIFF images that can be used for 3-dimensional (3D) image analysis. We used an in-house devised algorithm for pattern recognition and multi-parametric high-content analysis, which has been previously described.
This image analysis tool operates in 3 steps: (1) Cells (within imaged cell populations) are processed for 3D segmentation; (2) fluorescence signal in 5mC and DAPI channels residing within nuclei of (only) epithelial cells are recorded; and (3) the system calculates the 5mC and gDNA content of the entire nucleus, and generates co-distribution maps (2-dimensional scatter plots) of 5mC signals and gDNA (represented by DAPI), discerned as 2-dimensional scatter plots for each individual cell in which each pixel on the plot has a 5mC-intensity value (y-axis) and a DAPI-intensity value (x-axis) representing 1 voxel in a cell nucleus (Figure 4). 5mC/DAPI signal colocalization index was calculated as the angle ∂ under the regression line and is considered a parameter for the 5mC/gDNA co-distribution.
Figure 4Global DNA methylation phenotyping of human sputum-derived cells with 3D-qDMI. Confocal images of fluorescently labeled cells. The cytoplasm is delineated by Cell Mask Red (red), the cell nucleus (global DNA) is delineated by DAPI (blue), and gDNA methylation is visualized by an antibody specific to 5mC (green). Left: The sputum of healthy individuals (normal) contains a majority of highly methylated cells (cell type 1) and sporadically a few hypomethylated cells (cell type 2). Right: In contrast, the sputum of a patient with lung cancer contains a large number of hypomethylated cells. The respective nuclear 5mC/DAPI co-distribution patterns, presented as scatter plots, show that normally methylated (type 1) cells in both sputum-donor groups display a steep regression line (defined by ∂) synonymous with a higher 5mC/DAPI colocalization index, whereas hypomethylated cells (type 2) produce a flatter regression line. Moreover, a frequent signature of the lung cancer–specific hypomethylated cells is the less dispersed and narrow co-distribution of 5mC and DAPI. 3D-qDMI was able to successfully distinguish between normal and aberrant cell types based on 2 5mC-biomarker parameters: global nuclear 5mC intensity and 5mC/gDNA co-distribution (index). 5mC, 5-methylcytosine; DAPI, 4′,6-diamidino-2-phenylindole.
Sputum was characterized by its content of abnormal cells, and tests (binary classifiers) were established on the basis of the 2 parameters, 5mC and 5mC/DAPI colocalization (continuous measurements). For each parameter, thresholds for abnormal cells in sputum at the cell level and the percentage of abnormal cells at the patient level were derived. Patients were classified as “positive” or “negative” for lung cancer as follows: On the cell level, cells were considered abnormal if the cell value was less than a threshold where all individual cell values were considered candidate thresholds and normal if the cell value was equal to or above the threshold. On the patient level, the percentage of abnormal cells was calculated for each candidate threshold value of abnormal cells. Patients were classified as positive for lung cancer if the percentage of abnormal cells was greater or equal to a threshold where all individual values of the percentage of abnormal cells were considered candidate thresholds and negative for lung cancer if the percentage of abnormal cells was less than the established threshold. For each candidate threshold value of abnormal cells, a logistic regression model was fit at the patient level with the corresponding percentage of abnormal cells as an independent variable and 2 groups of interest as a dependent variable (eg, early-stage lung cancer vs high-risk patients with benign lung disease (COPD/IPF)), and then sensitivity and specificity were calculated at each value of the percentage of abnormal cells to discriminate between the 2 groups of interest. At each combination of candidate threshold values of abnormal cells and the percentage of abnormal cells, sensitivity and specificity were calculated and the optimal combination of threshold values that most differentiate 2 groups of interest were chosen on the basis of a minimum value for sensitivity of 90% or 95% while maximizing specificity across all possible combinations of thresholds.
We chose the optimal combination of threshold values with the largest value for abnormal cells and smallest value for the percentage of abnormal cells when multiple combinations of threshold values produced the same sensitivity or specificity. For each parameter, sensitivity and specificity and corresponding 95% confidence intervals (CIs) were estimated at the optimal combination of thresholds, and patients were classified as “positive” or “negative.” Test results from the 2 parameters were combined in the following way: If a patient was classified as positive based on either parameter, then the patient was classified as positive. Otherwise, the patient was classified as negative. The performance of the combined test was examined with sensitivity and specificity, and corresponding 95% CIs were estimated. Internal validation for the combined test was conducted by estimating and correcting possible optimism in sensitivity and specificity (optimism-corrected sensitivity and specificity) using the bootstrap method with 100 replicates as described by Harrell and colleagues,
Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. Efficiency of Some Procedures for Logistic Regression Analysis.
Data were resampled with replacement 100 times, each with the same sample size as the original data (bootstrap data). An optimal combination of 2 thresholds was chosen for each bootstrap data. Each optimal combination of thresholds was then applied to both the bootstrap data and original data, and corresponding sensitivity and specificity for the combined test were estimated for both. The difference between these 2 was calculated for sensitivity and specificity and then averaged over 100 bootstrap data, which is called an “optimism.” The optimism-corrected sensitivity and specificity were calculated as the original sensitivity and specificity minus the estimated optimism for sensitivity and specificity, respectively. Statistical analyses were performed using R version 4.0.5 and the R packages (OptimalCutpoints, epiR, tidyverse, boot, and parallel) with 2-sided tests at a significance level of .05.
For comparative cell phenotyping between our 3D approach and conventional cytology, each selected sputum sample was split into 2 equal aliquots (before sample processing). One aliquot was processed for 3D analysis as mentioned earlier. The other aliquot was transferred to the Cedars-Sinai Medical Center Cytopathology Lab for immediate processing. Each sample was fixed in 50 mL of Saccomanno's fixative (Polysciences) for 20 minutes. The specimen was blended at high speed for 2 to 7 seconds in the Waring blender (Cardinal Health), depending on the thickness of mucoid specimens. The blended sample was centrifuged 10 minutes in a benchtop centrifuge (Hettich) at 1866 rpm. Two smears were prepared from the resuspended sediment after decanting the supernatant and vortexing the sediment for 8 to 10 seconds in the vortex mixer. The smears were air dried, rinsed in 95% alcohol for 10 minutes, and stained with the Papanicolaou stain.
The slides were screened by a cytotechnologist and reviewed by a cytopathologist. Cases were recorded as negative, atypical, suspicious, or malignant.
Results
Our study enrolled 11 healthy subjects with no history of cancer or pulmonary issues before testing, 34 high-risk patients with benign chronic pulmonary ailments (COPD/IPF), and 43 patients with pathologically confirmed lung cancer (Table 1). High-risk benign patients and patients with lung cancer were equally divided by gender with an average age of 68 years. The breakdown of subject demographics, lung cancer types, specific lobe, location within each lobe, and pathological stage is shown in Table 2.
Measurements of global 5mC-intensity values and calculated 5mC/DAPI colocalization indices revealed that the sputum of patients with lung cancer contain a majority of aberrantly hypomethylated cells; sputum of normal cells had a majority of normally methylated cells. Hypomethylated cells are characterized by reduced 5mC intensity that ensues in flatter regression lines and ultimately smaller ∂ (on average <15°). It should be noted that hypomethylated epithelial cells were also found in normal sputum, although at a significantly smaller rate than the sputum of patients with lung cancer, as exemplified in Figure 4.
By looking at the generated data on the cell level, we empirically determined cutoffs for the 2 parameters: 150 relative light units for 5mC-intensity and a colocalization index ∂ = 15°. The cutoffs were chosen to yield more than 50% hypomethylated cells for cancer samples and less than 50% hypomethylated cells for noncancer samples. Cells were characterized as hypomethylated if their measured parameter was less than the determined thresholds. Using these cutoffs, we calculated that healthy sputum contains approximately 18% hypomethylated cells (Figure 5, A). This fraction is increased to approximately 42% in the sputum of high-risk benign patients with chronic pulmonary disorders. In early-stage lung cancer, the number of hypomethylated cells increased to approximately 64%, and in advanced-stage lung cancer, this increased to 80%. The same trend in the proportions of hypomethylated cells can be seen for the 5mC/DAPI colocalization index: 26% in healthy donors, an increase to 66% in high-risk benign patients, 86% in early-stage lung cancer, and 93% in advanced lung cancer (Figure 5, B). However, we did not observe any correlation between the number of hypomethylated cells and the location of the tumor (central vs peripheral) or tumor size. Tumor characteristics are provided in Table E1. Of note, we found an average of 20% difference, for both parameters, between high-risk benign patients and patients diagnosed with early-stage lung cancer.
Figure 5Histograms showing mean values for the 2 parameters, 5mC load (global intensity) (A) and 5mC/DAPI colocalization index (B) for the different subject groups. Both parameters indicate that the number of hypomethylated cells proportionally increases from the healthy state over symptomatic (high-risk benign pulmonary diseases) to lung cancer and even progresses during cancer advancement. In particular, early lung cancer stages exhibiting at average approximately 25% more globally hypomethylated cells than high-risk benign conditions. This margin is significant in differentiating the 2 most important groups in early lung cancer detection. 5mC, 5-methylcytosine; DAPI, 4′,6-diamidino-2-phenylindole; LCa, lung cancer.
In a second analysis, we applied a statistical approach to identify optimal thresholds for the 2 parameters to define a hypomethylated (aberrant) cell and the percent of aberrant cells that would differentiate each specific patient study group (healthy, high-risk benign, and patient with cancer). The results are reported in Table E2, Table E3, Table E4, Table E5, Table E6, Table E7, Table E8, Table E9. Table 3 summarizes the optimal thresholds for the 2 parameters that most differentiate 2 groups of interest and the test performance for the 2 parameters and the combined test for the most important 6 differentiations. In all but 1 differentiation, the 5mC/DAPI colocalization index (parameter 2) has a better performance over 5mC intensity (parameter 1) and the combination of both parameters. Parameter 2 differentiates healthy subjects from high-risk benign cases with a 97.1% sensitivity and 63.6% specificity. The test differentiates healthy subjects from patients with stage I and II lung cancer with 95.8% sensitivity and 90.9% specificity. For the differentiation of the high-risk benign group versus the early-stage lung cancer group, 5mC-intensity demonstrates 95.8% sensitivity and 41.2% specificity increased over the 5mC/DAPI colocalization index. The angle under the regression line (5mC/DAPI colocalization index) associated with the definition of a hypomethylated cell was 6.41 or less, and the optimal threshold value for the percentage of abnormal cells associated with a cancer diagnosis was 30.30% for 5mC and 38.4% for 5mC/DAPI index.
Table 3Summary of best performances for each differentiation and parameter
By applying standard sputum cytology, a total of 68% (n = 23 of 34) high-risk benign cases were examined by cytology, consisting of 16 IPF and 7 COPD cases. A total of 74% (n = 32 of 43) lung cancer cases were examined by cytology, including 23 early-stage and 9 advanced-stage cases. From the total of 34 patients with COPD/IPF, only samples of 23 of them were forwarded for sputum cytology; 17 (65%) were correctly identified as negative, and 6 (35%) were declared unsatisfactory due to scant cellularity. Furthermore, cytology found few atypical cells in only 1 stage IV adenocarcinoma case and misdiagnosed 27 lung cancer cases (19 stage I-II and 8 stage III-IV) as being negative for abnormal cells. Four early-stage lung cancer cases were declared unsatisfactory because of scant cellularity.
Discussion
Although DNA methylation imaging was introduced at the end of the 1990s, it did not gain popularity compared with other contemporaneously developed molecular methods, such as PCR-based, array-based, sequencing, high-pressure liquid chromatography, and mass spectrometry. The reasons may be as follows: DNA methylation imaging was applied in combination with radio-labeled or enzymatic reporters for detection, which lack sensitivity and multiplexing capability or affect repeatability/consistency of the assay, and did not provide enough significance in differential results because of low-image resolution. Unlike current molecular approaches and previous low-resolution imaging-based attempts that average 5mC measurements across large population of cells or only measure mean 5mC intensity values in cell nuclei, 3D-qDMI leverages the extraction of differential 5mC-relevant information by accounting for secondary effects of DNA methylation imbalances occurring throughout cellular transformation, specifically hypomethylation of global DNA. The latter mechanism elicits reorganization of the genome within cell nuclei, thus shifting 5mC/DAPI colocalization patterns.
Nuclear DNA methylation and chromatin condensation phenotypes are distinct between normally proliferating/aging, rapidly growing/immortal, and senescent cells.
This phenomenon is well described in cell biological research but has not been used in cancer pathology. The image analysis we developed addresses this gap and displays the relevant changes as intensity distributions of the 2 types of signals that reflect the proposed mechanism: 5mC signals and gDNA represented by DAPI signals, because DAPI intercalates into adenine thymine-rich DNA, the main component of highly repetitive and compact heterochromatic sequences.
By using the 3D-qDMI platform, we explored 5mC/DAPI codistribution patterns that could differentiate between normal and malignant in respective cell models, resected human tissues, and exfoliated sputum cells of the lung epithelium.
DNA methylation topology differentiates between normal and malignant in cell models, resected human tissues, and exfoliated sputum cells of lung epithelium.
Because DNA methylation correlates with early events in carcinogenesis and tumor progression, this differential property of cells can serve as a signature marker in diagnostics and monitoring. Our approach quantifies the proportion of hypomethylated cells, based on the load and distribution of global 5mC (5mC + gDNA represented by DAPI) in the nuclei of epithelial cells extracted from sputum. By using 3D-qDMI, we demonstrate the ability of sputum to be tested for the early detection of lung cancer. 3D-qDMI allows for the rapid, parallel, single-cell resolution characterization of thousands of cells within heterogeneous sputum samples. This capability is highly favorable because it allows for the exclusion of nonepithelial cells from analysis, thus preventing data skewing caused by sample impurity from infiltrating hematopoietic cells and microorganisms. Another huge advantage of our approach is the bench to bedside clinical applicability. The test is a slide-based cytometric approach that is amenable to automation and scaling; it can seamlessly integrate into the workflow for lung cancer screening and serve as an aid in the diagnosis of lung cancer.
We demonstrate the advantages of using 3D-qDMI over traditional sputum cytology analysis. By applying high-resolution, high-content 3D imaging analysis, our novel, noninvasive sputum-based screening test for NSCLC detects malignancy with a sensitivity greater than 95%. The detection of malignancy was obtained in all comparative group analyses. In addition, the test differentiated the high-risk benign group versus the early-stage lung cancer with high sensitivity group, which is the most critical distinction among all groups because it eliminates the outliers at both extremes (healthy population and patients with stage III-IV).
Our results confirm that a sputum-based test is suitable for lung cancer detection. When original sputum samples were simultaneously processed by both methods, cytologic analysis was limited in cancer detection due to scant cellularity in 30% (n = 7) of high-risk benign cases and 12.5% (n = 4) of lung cancer cases. Sputum cytology using Papanicolaou staining demonstrated worse detection, because it only detected abnormal cells in 1 patient with stage IV lung cancer and had an overall false-negative rate of 84.4% (n = 27) in the lung cancer group, with 19 false-negatives in stage I and II and 8 false-negatives in stage III and IV.
Study Limitations
One limitation of the study includes using the same measurement at the cell level and the patient level without multilevel modeling. Therefore, no correlation can be made between the data of the 2 levels. Rather, abnormal versus normal cells were classified at the cell level, and for each possible threshold value of abnormal cells, a corresponding percentage of abnormal cells was calculated for each patient. Then, we examined its association with 2 groups of interest at the patient level, using a logistic regression model.
Conclusions
Given the ease of obtaining sputum samples from patients and the benefits of no radiation exposure, our novel test may impact lung cancer screening, evaluation of pulmonary nodules, detection of disease recurrence in patients postresection, and cancer surveillance algorithms. Using global DNA methylation topology in the cell nucleus discerned by the innovative 3D-qDMI platform may help leverage the potential of sputum in noninvasive early lung cancer detection. Additionally, our testing methodology may be a feasible option for samples showing scant cellularity, such as those obtained from induced sputum and CT-guided or navigational bronchoscopy fine-needle aspiration biopsies.
H.J.S. is a speaker for Intuitive, Medtronic/Covidien, and Pinnacle Biologics. J.T. has submitted a patent application relevant to the methodology applied in this study. All other authors reported no conflicts of interest.
The Journal policy requires editors and reviewers to disclose conflicts of interest and to decline handling or reviewing manuscripts for which they may have a conflict of interest. The editors and reviewers of this article have no conflicts of interest.
The authors thank our many collaborators at Cedars-Sinai who were instrumental in helping this study come to fruition. The authors thank Paul Noble, MD, Isabel Pedraza, MD, Jeremy Falk, MD, Rodney Mardirosian, Claudia Huiza, Charles Yoo, and Carissa Huynh and Yi Zhang MD (both at Biobank and Pathology, Cedars-Sinai Cancer Center). This project was supported by Cedars-Sinai Technology Ventures.
By using sputum as a source for cancer cells in combination with advanced high-resolution fluorescence imaging and 3D image analysis, we have validated a novel noninvasive approach for the early detection of lung cancer. This new test utilizes the level of global DNA methylation as well as the topology of DNA methylation in the nuclei of cells (extracted from sputum) in order to detect lung cancer based on the number of hypomethylated cells in a sputum sample. Video available at: https://www.jtcvs.org/article/S2666-2736(22)00415-6/fulltext. Video available at: https://www.jtcvs.org/article/S2666-2736(22)00415-6/fulltext.
Table E2Optimal threshold values for 5-methylcytosine and 5-methylcytosine 4′,6-diamidino-2-phenylindol colocalization index and estimated sensitivity and specificity for 5-methylcytosine, 5-methylcytosine 4′,6-diamidino-2-phenylindol, and the combined parameters at the optimal thresholds for healthy versus high-risk benign (chronic obstructive pulmonary disease + idiopathic pulmonary fibrosis)
Criterion for identifying optimal thresholds for parameters
Parameter
Optimal threshold value for abnormal cells at the cell level
Table E3Optimal threshold values for 5-methylcytosine and 5-methylcytosine 4′,6-diamidino-2-phenylindol colocalization index and estimated sensitivity and specificity for 5-methylcytosine, 5-methylcytosine 4′,6-diamidino-2-phenylindol, and the combined parameters at the optimal thresholds for healthy versus early-stage lung cancer
Criterion for identifying optimal thresholds for parameters
Parameter
Optimal threshold value for abnormal cells at the cell level
Table E4Optimal threshold values for 5-methylcytosine and 5-methylcytosine 4′,6-diamidino-2-phenylindol colocalization index and estimated sensitivity and specificity for 5-methylcytosine, 5-methylcytosine 4′,6-diamidino-2-phenylindol, and the combined parameters at the optimal thresholds for high-risk benign (chronic obstructive pulmonary disease + idiopathic pulmonary fibrosis) versus all stages of lung cancer
Criterion for identifying optimal thresholds for parameters
Parameter
Optimal threshold value for abnormal cells at the cell level
Table E5Optimal threshold values for 5-methylcytosine and 5-methylcytosine 4′,6-diamidino-2-phenylindol colocalization index and estimated sensitivity and specificity for 5-methylcytosine, 5-methylcytosine 4′,6-diamidino-2-phenylindol, and the combined parameters at the optimal thresholds for high-risk benign versus early-stage lung cancer
Criterion for identifying optimal thresholds for parameters
Parameter
Optimal threshold value for abnormal cells at the cell level
Table E6Optimal threshold values for 5-methylcytosine and 5-methylcytosine 4′,6-diamidino-2-phenylindol colocalization index and estimated sensitivity and specificity for 5-methylcytosine, 5-methylcytosine 4′,6-diamidino-2-phenylindol, and the combined parameters at the optimal thresholds for high-risk benign versus advanced lung cancer
Criterion for identifying optimal thresholds for parameters
Parameter
Optimal threshold value for abnormal cells at the cell level
Table E7Optimal threshold values for 5-methylcytosine and 5-methylcytosine 4′,6-diamidino-2-phenylindol colocalization index and estimated sensitivity and specificity for 5-methylcytosine, 5-methylcytosine 4′,6-diamidino-2-phenylindol, and the combined parameters at the optimal thresholds for controls (healthy + high-risk benign) versus all stages of lung cancer
Criterion for identifying optimal thresholds for parameters
Parameter
Optimal threshold value for abnormal cells at the cell level
Table E8Optimal threshold values for 5-methylcytosine and 5-methylcytosine 4′,6-diamidino-2-phenylindol colocalization index and estimated sensitivity and specificity for 5-methylcytosine, 5-methylcytosine 4′,6-diamidino-2-phenylindol, and the combined parameters at the optimal thresholds for controls versus early-stage lung cancer
Criterion for identifying optimal thresholds for parameters
Parameter
Optimal threshold value for abnormal cells at the cell level
Table E9Optimal threshold values for 5-methylcytosine and 5-methylcytosine 4′,6-diamidino-2-phenylindol colocalization index and estimated sensitivity and specificity for 5-methylcytosine, 5-methylcytosine 4′,6-diamidino-2-phenylindol, and the combined parameters at the optimal thresholds for controls versus advanced lung cancer
Criterion for identifying optimal thresholds for parameters
Parameter
Optimal threshold value for abnormal cells at the cell level
The IASLC lung cancer staging project: proposals for revision of the TNM Stage Groupings in the forthcoming (eighth) edition of the TNM classification for lung cancer.
DNA methylation topology differentiates between normal and malignant in cell models, resected human tissues, and exfoliated sputum cells of lung epithelium.
Nuclear DNA methylation and chromatin condensation phenotypes are distinct between normally proliferating/aging, rapidly growing/immortal, and senescent cells.
Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. Efficiency of Some Procedures for Logistic Regression Analysis.
The study protocol was approved by the Cedars-Sinai Internal Review Board (#Pro00039831 and Pro00046876, approval date May 1, 2016, and January 23, 2018). All patients provided informed written consent for the publication before their study date.
00415-6&id=fx1.jpg){kind=link}
00415-6&id=fx2.jpg){kind=link}
00415-6&id=gr1.jpg){kind=link}
00415-6&id=gr2.jpg){kind=link}
00415-6&id=gr3.jpg){kind=link}
00415-6&id=gr4.jpg){kind=link}
00415-6&id=gr5.jpg){kind=link}
00415-6&id=fx3.jpg){kind=link}