The current trial (NCT02237183) was initially designed as a randomized, double-blind, dose de-escalation trial of inhaled iloprost compared to placebo with two cohorts. The standard use of inhaled iloprost for pulmonary hypertension requires eight or more daily treatments; we considered this to be impractical for a prevention application and chose QID and BID treatments as more realistic. Ampules containing 5 µg of iloprost and placebo were made available by the manufacturer, Bayer. Cohort A (n = 16) was to self-administer 5 µg of iloprost QID and Cohort B (n = 16) was to self-administer 5 µg of iloprost BID, both for 60 days. Placebo was to be self-administered to 8 participants each in the same schedule as in Cohorts A and B, resulting in a total of 48 participants planned. The final trial schema is summarized in Figure 1.

Inclusion criteria included a smoking history of at least 20 pack-years with abstinence for >12 months and either sputum cytologic atypia (≥ mild) or a history of endobronchial dysplasia (≥ mild) within the past 12 months; ages 18 to 85 years old; ECOG performance status of ≤1; normal organ function; the ability to safely undergo bronchoscopy; and the ability and willingness to give informed consent. Exclusion criteria included current use of thiazolidinediones; treatment with iloprost at any time; current or prior malignancy within the past 6 months (other than curatively treated non-melanoma skin cancer, cervical carcinoma in situ, and bladder carcinoma in situ); receipt of chemotherapy or radiotherapy within the previous 6 months; pregnancy or breastfeeding; and the use of anticoagulants other than aspirin or non-steroidal anti-inflammatory drugs. Participants were recruited from the Pulmonary clinics at the Rocky Mountain Regional Veterans Administration Medical Center and by referral from University of Colorado Hospital clinics.

After a pre-screening visit, two sequential 3-day sputum samples were collected, one in Saccomanno’s fluid and the other in a similar but proprietary fixative provided by VisionGate, Inc (10). Individuals who produced sputum with ≥ mild atypia were invited to participate in the trial and underwent screening tests, including physical exam, labs, baseline bronchoscopy, full pulmonary function testing (PFTs) including diffusing capacity for carbon monoxide (DLCO), EKG, exhaled carbon monoxide (CO), 6-minute walk test (6MWT), St. George Respiratory Questionnaire (SGRQ), and Chronic Obstructive Pulmonary Disease Assessment Test (CAT) questionnaire administration.

Iloprost or matched placebo was delivered by the FDA-approved I-neb Adaptive Aerosol Delivery (AAD) device (Accredo). Each participant was taught to self-administer the experimental agent. Treatments last 4–10 min and the device measures compliance and delivered dose. During the initial training visit, a 2.5-µg test dose was administered and participants delivered the full 5-µg doses subsequently. Intolerance was defined as coughing lasting more than 10 min, systolic blood pressure decrease of more than 20 mm Hg, or flushing lasting more than 10 min.

White light bronchoscopy was performed at trial initiation and after 60 days of treatment while participants were still taking drug or placebo. Six predetermined sites, namely, the carinae of the right upper lobe, right middle lobe, superior segment of the right lower lobe, left upper lobe, left upper division/lingula, and superior segment of the left lower lobe, were biopsied, along with any additional sites that appeared abnormal. Additional biopsies were taken for RNA and progenitor cell culture.

Blood and urine were obtained prior to each bronchoscopy, and sputum was collected at trial conclusion. Interval visits were carried out on days 15 (telephone) and 30. Pre-study testing was repeated at the 60-day visit, including PFTs, 6MWT, the SGRQ, and CAT. On day 90, a post-intervention physical exam, history, and toxicity assessment was conducted.

H & E endobronchial biopsies were graded according to the World Health Organization (WHO) grading system (11) and assigned a score according to the following scale: 1 = normal bronchial epithelium; 2 = reserve cell hyperplasia; 3 = squamous dysplasia; 4 = mild dysplasia; 5 = moderate dysplasia; 6 = severe dysplasia; 7 = carcinoma in situ; and 8 = invasive carcinoma. All biopsies were graded in a blinded fashion by the study pathologist (DTM). Biopsies were also assigned a visual inflammation score based on estimated cellularity attributed to inflammatory cells, with 0 = no significant inflammation, <5%; 1 = mild, 5%–25%; 2 = moderate, 25%–75%; 3 = severe, >75%.

Bronchial biopsies frozen in OCT were used for bulk mRNAseq studies. Total RNA was purified using an RNA extraction microkit (Qiagen), followed by mRNA library prep (Illumina) and sequencing (NovaSEQ next-gen sequencer). The data analysis pipeline included trimming paired-RNA fastq files and mapping to human genome hg38 using STAR (12). The fragments per exon kilobase per million mapped reads (FPKM) values were calculated using RNAseq by expectation maximization (RSEM) (13). eBayes (limma R package) was used to fit the model that considers participant heterogeneity, batch effect, and histology. Significant differences between groups (pre-iloprost, post-iloprost, and placebo) were calculated using moderated t-statistics ( 14, 15). The FPKM tables of pre-iloprost vs. post-iloprost and pre- and post-placebo were used as input to identify differentially expressed pathways by GSEA (gene set enrichment analysis) (16) considering Hallmark as reference database (17). The effective gene sets (normalized p < 0.01) of each pathway were collected and visualized using custom R codes.

For single-cell RNAseq (scRNAseq), biopsies were proteolytically digested to prepare single-cell suspensions and used in the Chromium Single-cell 3’ solution capture system from the 10x Genomics platform. Cell ranger 5.0 was used to create cell by gene count tables, preprocessed using Seurat R package (18) and batch-corrected by harmony in R (19). The Uniform Manifold Approximation and Projection (UMAP) and violin plots were generated by the Seurat R package, and heatmap plots of average expression were produced by using the ComplexHeatmap R package (20).

An air–liquid interface (ALI) culture method for bronchial basal progenitor cells following previously described methods was developed (21). Briefly, biopsies were proteolytically digested and single-cell suspensions were used to grow basal progenitor cells following the methods described before (21, 22). Clonal basal cells were then plated on collagen-coated transwell membrane inserts and used for in vitro ALI differentiation. Cells were treated with either vehicle or 1 µM iloprost during the entire culture period (6 weeks) with drug replenishment every 48 h. Membranes were fixed with 10% neutral buffered formalin, embedded in paraffin, sectioned, and H & E stained to determine epithelial histology.

The primary objective of the trial was to evaluate the toxicity of inhalational iloprost administered QID to participants for 2 months using CTCAE v4.0 and compliance. Exploratory endpoints for which the trial was not powered to assess included effect on endobronchial histology, expectorated sputum cytology, and pulmonary function. Participant characteristics and adverse events are summarized using descriptive statistics. Histology was assessed on the WHO 1–8 scale at each biopsy site at baseline and post-treatment, and changes in worst-grade histology were analyzed using the Wilcoxon signed-rank test due to the discrete nature of the outcome. Changes in PFT, 6MWTs, and SGRQ and CAT scores were also analyzed using the Wilcoxon signed rank test.

Because of the expiration of the placebo and the inability to procure more, the study design was modified to eliminate the placebo group after three individuals were enrolled. Owing to the resultant delays to study completion and given that the QID treatment schedule was well tolerated, the protocol was amended to eliminate the BID cohort (cohort B) after five individuals were enrolled.

All the de-identified genomic data will be available at the NCBI db-GAP database.

Screening for potential participants occurred in the Rocky Mountain Regional VAMC pulmonary clinics from 15 November 2015 to 27 November 2019. Of 69 consented individuals, 21 withdrew from the study or were lost to follow-up prior to sputum collection, 14 had screen failures (insufficient sample or lack of cytologic atypia), and 31 met the inclusion criterion of at least mild atypia (18 = mild; 11 = moderate; 2 = severe), as shown in Figure 1. All enrolled in the trial and underwent at least one bronchoscopy (Figure 1). Three additional individuals were enrolled on the basis of known mild or worse endobronchial dysplasia on prior bronchoscopy. Of the 34 enrolled participants, 26 individuals received iloprost QID, 5 received iloprost BID, and 3 received placebo. The baseline characteristics of the study participants are summarized in Table 1. Participants were predominantly male (88%), Caucasian (88%), and elderly (mean age = 64.6; range = 32–84 years old). Median (range) for pack-years smoked was 36 (21.5–88), and years since quit smoking was 19.4 (1–44). Average lung function FEV1% predicted was 86.3% (range = 29%–127%), and FEV1/FVC was 0.72 (0.35–0.92). Of 34 participants, 8 (25%) had COPD by the standard definition of FEV1/FVC < 0.70.

Inhaled iloprost was well tolerated during both treatment schedules. Four participants, two from Cohort A and two from Cohort B, dropped out due to grade 2 toxicities (cough, headache and nausea, dyspnea, and wheezing). Two other individuals from Cohort A were removed for reasons unrelated to toxicity: pulmonary embolism requiring anticoagulation and hesitancy to take an additional medication. No unexpected toxicities were noted; there were no adverse events higher than grade 2 in either iloprost or placebo groups (Supplementary Table 1).

The I-neb AAD device records inhalation treatments to determine compliance. Treatment time ranged from 6 to 20 min with an average of 11 min. Total compliance, measured as mean percent of scheduled doses actually administered, across groups was Cohort A = 67.2% (Min–Max, 19%–98.5%); Cohort B = 70.2% (Min–Max, 13.4%–101%); placebo = 93.3% (Min–Max, 87.9%–99.2%). The predetermined definition of fully compliant was 80%; 46.2% of Cohort A, 60.0% of Cohort B, and 100% of placebo participants had ≥80% of scheduled doses actually administered and thus met the definition of fully compliant.

Baseline bronchoscopy was performed in 34 individuals. A total of 19 had at least one biopsy graded as dysplastic (0 = mild; 14 = moderate; 5 = severe). Sputum cytology prior to bronchoscopy was completed on 33 individuals, including 2 of the 3 who qualified for the study based on a previous dysplastic biopsy. Of the 33 participants with mild or worse sputum atypia, 18 had one or more dysplastic biopsies. The fractions of dysplastic biopsies for varying grades of sputum atypia were as follows: normal, 1/1; mild, 9/18; moderate, 8/13; and severe, 1/2.

Although the trial was not powered to assess histologic response, the predetermined endpoint for treatment effect on histology was the change in maximum histologic score between baseline and final bronchoscopy (Table 2). Additional predetermined endpoints were change in mean histologic score and the dysplasia index, defined as fraction of biopsy sites scored as dysplastic. The early termination of enrollment into the placebo group and Cohort B limited size to three evaluable participants each, making comparison to Cohort A unlikely to be meaningful. Cohort A (22 evaluable participants) showed small and insignificant worsening of maximum (0.130, p = 0.79) and mean (0.08, p = 0.66) dysplasia scores. The dysplasia index showed a small, but not significant, improvement (−7.7%, p = 0.51). An alternative scoring system for grading response was assessed after study completion and results are shown in Supplementary Table 2 (23). Among 22 individuals in Cohort A who had both bronchoscopies, there were n = 138 locations that were biopsied at both time points. There was no significant change in the worst grade histology in this group (Figure 2A). At the lesion level, 68 (49%) had no change in histologic grade (stable), whereas progression occurred in 40 (29%, progressive) and regression occurred in 30 (22% regressive) (Figure 2B), with no statistically significant change (p = 0.513, Stuart–Maxwell test of marginal homogeneity).

One individual had a dramatic worsening of a biopsy site from normal to carcinoma in situ. A subsequent clinically indicated bronchoscopy 6 weeks later demonstrated regression to normal. This participant had an additional follow-up bronchoscopy 13 months later where the biopsy site showed reserve cell hyperplasia. A second participant had a site (carina between anterior and apico-posterior segments of the left upper lobe) that was not sampled on the baseline bronchoscopy that appeared visually abnormal on the final bronchoscopy and was read as carcinoma in situ. This site regressed to severe dysplasia at 3 months and to normal at 15-month follow-up bronchoscopies. In both situations, no treatment for the carcinoma in situ lesions occurred.

The number of participants with pre- and post-treatment sputum cytology (Supplementary Table 3) was too small to discern any meaningful difference caused by iloprost treatment.

Endobronchial biopsies were scored for inflammation on a score from 0 (none) to 3 (severe). Changes in maximum and mean inflammation scores are shown for Cohort A, Cohort B, and placebo groups in Supplementary Table 4. For Cohort A, there was a small, statistically insignificant increase in both maximum (0.3) and mean (0.2) inflammation scores between baseline and final bronchoscopies.

Results of PFTs, 6MWT, and SGRQ and CAT scores at baseline and 60-day visits are summarized in Supplementary Tables 5–7, and 8, respectively. There were no statistically significant or clinically meaningful changes in pre-bronchodilator FEV1 or FVC, post-bronchodilator FEV1 or FVC, and 6MWT-associated measures including distance walked and dyspnea levels associated with the 6MWT, SGRQ, or CAT scores.

Bulk mRNAseq and scRNAseq analysis were used to investigate iloprost-mediated changes in bronchial tissues. Considering the small size of the trial and to avoid molecular differences associated with varying histology grades, we restricted our analysis to histologically normal biopsies (grade 1.0) both at pre-treatment and 2 months post-iloprost (Figure 3A, group A). Four individuals from Cohort A were used in this analysis and were compared to two placebo participants (Figure 3A, group B). For each placebo participant, two biopsies with histology grade 1.0 at baseline and at 2 months bronchoscopy were included in the analysis. Heatmaps of genes significantly altered after iloprost or in placebo are shown in Figure 3B. A detailed list of these genes is provided in Supplementary Tables 9A, B. GSEA was used to identify pathways that are (i) upregulated both in placebo and in iloprost, (ii) downregulated both in iloprost and in placebo, (iii) upregulated only in iloprost, and (iv) downregulated only in iloprost (Figure 3C). This analysis plan was devised to recognize pathways involved in normal repair after bronchoscopy and biopsy (i and ii) or involved in iloprost exposure (iii and iv). DNA repair and oxidative phosphorylation were upregulated in both placebo and iloprost groups, while Wnt-β catenin, Notch signaling, IL2-Stat5, and EMT pathways were downregulated in both groups. Pathways that were upregulated by iloprost only included Kras signaling, TGFβ, and interferon α and γ response pathways, while pathways involved in cell proliferation such as G2M checkpoint, E2F targets, and spermatogenesis were downregulated in iloprost compared to placebo (normalized p < 0.01).

scRNAseq was performed using five paired biopsies (pre- and post-treatment from the same sites) from two participants, all with histology grade 1.0 both pre-treatment and post-iloprost. No placebo samples were available for scRNAseq. A total of 74,519 cells from the 10 biopsies were profiled. UMAP for dimensionality reduction of single-cell data and unsupervised clustering showed small differences in overall cellular distributions in pre- and post-treatment samples (Figures 3D, E; Supplementary Figure 1A). Known cell-type-specific markers for epithelial and non-epithelial (including immune) cells identified cell-type-specific clusters and their relative proportions in pre- and post-biopsies (Figure 3F). The main difference observed was an expansion of CD8-positive T cells (cluster 2) and a decrease in a subset of basal cells (cluster 5) post-iloprost. More in-depth analysis of cluster 5 identified only five genes (MALAT1, HBB, DDX17, SYEN2, and F3) that were exclusive to this cluster and were absent in the other basal cell clusters 3 and 13 (Supplementary Figure 1B). In contrast, T-cell clusters (clusters 0 and 2) in pre- and post-iloprost treatment identified a list of genes that are upregulated post-iloprost (Figure 3G). Significantly, key genes involved in interferon <υ>γ</υ> signaling including CD69, RUNX3, CCL5, Granzyme A (GZMA), and CXCR4 were upregulated by iloprost specifically in T cells compared to non-T cells (Figure 3H).Correlation between in vitro and in vivo responses to iloprost

Eleven dysplastic biopsies (three severe and eight moderate dysplasia) from 10 individuals collected at pre-treatment bronchoscopy were successfully cultured at ALI with vehicle or iloprost and evaluated for response. Cells from 5 (2 severe and 3 moderate dysplasia) of the 11 biopsies responded to iloprost but not vehicle by re-establishment of a normal mucociliary epithelium, while 6 (1 severe and 5 moderate dysplasia) were unresponsive (Figure 4A; Supplementary Figure 2). Of the five biopsy sites that yielded cell cultures that responded to iloprost but not vehicle in vitro, four responded to inhaled iloprost by improving histology to a non-dysplastic grade. Of the six biopsy sites that yielded cell cultures that did not respond in vitro, none responded to inhaled iloprost by improving histology [agreement between response in vivo and in vitro, kappa coefficient 0.81 (95% CI 0.47, 1, Figure 4B)]. All of these participants were on the QID dosing and administered most (87.5% ± 9% for responders and 63.7% ± 29% for non-responders) of their scheduled iloprost doses.