Tricine, Tris, ammonium persulphate (APS), tetramethylethylenediamine (TEMED), sodium dodecyl sulphate (SDS), acrylamide, and N,N′-methylenebisacrylamide were purchased from SERVA (Heidelberg, Germany). The protease inhibitor cocktail (#539134), the exosome release inhibitor GW4869 (#D1692), the horseradish peroxidase (HRP)-conjugated goat anti-immunoglobulin G (IgG) secondary antibody, dithiothreitol, iodoacetamide, ammonium bicarbonate, trifluoroacetic acid, and formic acid were from Sigma-Aldrich (Merck, Darmstadt, Germany). The matrix metalloproteinase (MMP) inhibitor GM6001 (galardin; #364210) was from Calbiochem (Merck, Darmstadt, Germany). The primary antibodies (Abs) to CD44 (#ab9524), alpha-smooth muscle actin (#ab5694), and vimentin (#ab92547) were from Abcam (Cambridge, UK). The Ab to CD63 was from Invitrogen (Ts63; Thermo Fisher Scientific, Waltham, MA, USA), and that to cytochrome c oxidase subunit 4 (COX IV) was from Cell Signaling Technology (3E11; Danvers, MA, USA). The streptavidin–HRP conjugate was supplied by GE HealthCare (Uppsala, Sweden). Sera-Mag SpeedBead Carboxylate-Modified [E7] Magnetic Particles were obtained from Cytiva (Danaher, Washington, DC, USA), and the sequencing-grade modified trypsin was from Promega Corporation (Madison, WI, USA). Acetonitrile and water were purchased from Honeywell (Charlotte, NC, USA), and the ethanol was from Supelco (Merck, Darmstadt, Germany).

The study group consisted of 257 DPLD subjects. Based on their diagnoses, the representative patients were classified into the five cohorts: 46 subjects with IPF, 58 patients with HP, 123 patients with SRC, 14 patients with OP, and 16 patients with CTD-ILD. The diagnoses were established in compliance with current guidelines published as official American Thoracic Society (ATS)/European Respiratory Society (ERS)/Japanese Respiratory Society (JRS)/Latin American Thoracic Association (ALAT) clinical practice guidelines on IPF and HP (18–21) or ATS/ERS/World Association of Sarcoidosis and Other Granulomatous Disorders (WASOG) guidelines on SRC (22) or according to currently used practical diagnostic approaches for CTD-ILD (23, 24) and OP (25), respectively. The diagnoses were established as the result of multidisciplinary team consensus (pneumologists, radiologists, and pathologists) in tertiary healthcare centres specialising in pulmonary medicine, the National Institute for Tuberculosis, Lung Diseases and Thoracic Surgery, Vysne Hagy, Slovakia, and Department of Pneumology and Phthisiology, Faculty of Medicine, Comenius University and University Hospital, Bratislava, Slovakia. The major characteristics together with DLCO of cohorts are presented in Table 1 . Based on CT findings, DPLDs were classified into two categories, fibrotic phenotype with reticular changes and traction bronchiectasis with or without the presence of honeycombing, and non-fibrotic phenotype with ground-glass opacity (GGO), consolidation, and diffuse nodules or cysts.

BALFs were collected by instillation of 120 mL (in three successive 40-mL aliquots) of sterile normal saline mainly into the right middle lobe or into the most affected lobe and aspirated by gentle suction using a flexible fibreoptic bronchoscope. BALF was first filtered through a double layer of sterile gauze and centrifuged at 300 g for 15 min at 10°C, and supernatants were collected, and either analysed directly or frozen for later analyses in a deep frozen box to −80°C. BALF cell differential counts are presented in Table 2 .

For the preparation of cytocentrifuge slides, 1 mL of BALF was collected and processed using a StatSpin Cytofuge 2 cytocentrifuge at 8,500 rpm for 4 min. The slides were then stained with Hemacolor Rapid Staining of Blood (Sigma-Aldrich). Following staining, microscopy and differential cell counts (macrophages, lymphocytes, eosinophiles, and neutrophils) were performed using a Zeiss Axiolab 5 microscope. To determine the absolute cell count, the BALF was filtered through gauze, and the filtered BALF was stained with CD45PC7 (Beckman Coulter). Subsequently, the BALF was centrifuged at 300 g for 15 min at 10°C. Lymphocytes and lymphocyte subsets were discriminated by a NAVIOS flow cytometer (Beckman Coulter France S.A.S.) using tetraCHROME CD45-FITC/CD4-PE/CD8-ECD/CD3-PC5 Antibody Cocktail (Beckman Coulter France S.A.S.). Data were analysed using the KALUZA software (Beckman Coulter France S.A.S.). The CD3, CD4, and CD8 expressions are presented as a percentage and total number of cells. Data are presented in Table 2 .

CT scans were acquired with a clinical CT system (PHILIPS Brilliance iCT SP, Philips Healthcare), with a 64-slice detector and 0.625-mm collimation; the tube potential was 120 kV with automatic tube current modulation. Images were reconstructed with 1-mm slice thickness, with an increment of 0.5 mm and a 768 × 768 graphic matrix for achieving isotropic voxels. A sharper kernel that is used for high-resolution CT reconstructions was applied as per the institutional standard. Patients were in supine position, and scans were performed during deep inspiration. Commercially available software (Contextflow GmbH, Vienna, Austria) was utilised to quantify HRCT disease patterns associated with DPLD (including the percentage of lung anomalies, GGOs, honeycombing, and reticulation) in a cohort of 30 subjects diagnosed with IPF and HP. Subsequently, the obtained data were correlated with CD44 concentration levels in BALF.

The primary human lung fibroblasts MRC-5 cells, from the American Type Culture Collection (ATCC), were cultured in RPMI 1640 medium (Invitrogen) supplemented with 100 U/mL penicillin, 100 μg/mL streptomycin, 2 mmol/L L-glutamine, and 10% heat-inactivated foetal calf serum (FCS) (all from Sigma-Aldrich). In our experimental model, the MRC-5 fibroblasts were standardly cultivated to subconfluency on 24-well cultivation plates (5 × 10⁵ cells/well), washed with the medium, and then incubated 24 h either with the selected BALF samples (IPF; BALF diluted 1:3 with the medium) or with the control mixture [CTR; phosphate-buffered saline (PBS) diluted 1:3 with the medium]. Optionally, the cells were in the course of the experiment co-treated with GM6001 (galardin, MMP inhibitor) or GW4869 (exosome release inhibitor). Afterwards, the cells were rigorously washed (3 times) with the medium and incubated for the next 24 h with the medium only. Afterwards, conditioned media were collected and centrifuged for 5 min at 2,000 g and the supernatants were analysed directly or frozen in a deep frozen box for later analyses by mass spectrometry, Western blotting, or enzyme-linked immunosorbent assay (ELISA). The adherent cells remaining on the wells of the plates were washed and lysed, and the cell lysates were analysed directly or frozen for later analyses by Western blotting. The morphology of the cultivated cells was visualised by using light microscopy phase-contrast imaging.

The BALF samples were used for ELISA analysis by using a commercially available ELISA kit for human CD44 (FineTest, #EH0654). All assays were performed according to the instruction manual recommended by the manufacturer.

To separate exosomes from soluble proteins in BALF, we used the Izon qEV kit (Izon Science, Christchurch, New Zealand) based on size-exclusion chromatography separation. First, a qEV column was cleaned and equilibrated by filtered PBS. Second, on the top of the column, a BALF sample or a cell supernatant sample was applied. Next, fractions were eluted by PBS. After the elution of the first seven fractions (3 mL, void volume), fractions 8–16 (500 μL each) were collected. Then, the isolated fractions were used for evaluation by Western blotting. In addition, the size and concentration of exosomes were measured in BALF and cell supernatants by using an Exoid instrument (Izon) based on tunable resistive pulse sensing (TRPS). TRPS is designed preferentially to measure the size of particles in the range of 40 nm to 10 µm. In our experimental setup, NP150 nanopores were applied, allowing the evaluation of exosomes.

Immunoblotting was performed as described previously (26). Briefly, various samples, including cell supernatants, cell lysates, and the fractions from the size-exclusion chromatography separation, were analysed by SDS polyacrylamide gel electrophoresis (SDS-PAGE) on polyacrylamide gels followed by a transfer at a constant voltage (15 V) to an Immobilon polyvinylidene difluoride membrane (Millipore, Merck, Darmstadt, Germany). The membranes were blocked using 4% non-fat milk and immunostained with the specific primary Ab followed by a secondary HRP conjugate. For visualisation, the chemiluminescence image analyser Azure 280 (Azure Biosystems, Dublin, CA) was used. Densitometric quantifications of corresponding bands were done by means of the AzureSpot software; the bands corresponding to BALF-treated samples (IPF) and control samples (CTR) were normalised to the bands corresponding to the COX IV levels in cell lysates.

For reverse transcription quantitative PCR (RT-qPCR) analysis, the MRC-5 cells, both control and IPF-BALF stimulated as described above, were lysed in TRIzol reagent (Invitrogen Life Technologies), and RNA was extracted according to the manufacturer’s instructions. Complementary DNA (cDNA) was synthesised from 400 ng of total RNA using M-MuLV Reverse Transcriptase (#M0253L, New England Biolabs) and random heptamers. Gene expression was measured via quantitative real-time PCR using Luna Universal qPCR Master Mix (#M3003L, New England Biolabs) with the following primers for human CD44 (CD44f: CTGGGGACTCTGCCTCGT; CD44r: CCGTCCGAGAGATGCTGTAG) and EEF1A1 (EEF1A1f: GTGCTAACATGCCTTGGTTC; EEF1A1r: AGAACACCAGTCTCCACTCG) as an endogenous control. Data were recorded on a CFX96 Touch Real-Time PCR detection system (Bio-Rad) and analysed by the 2^−ΔΔCT method (27).

The activated and control MRC-5 cell supernatants from conditioned media (150 µL) were reduced with 5 mM dithiothreitol and alkylated with 15 mM iodoacetamide. Samples were cleaned and digested using a single-pot, solid-phase-enhanced sample preparation method (28). Briefly, proteins were bound to 170 µg of Sera-Mag SpeedBead Carboxylate-Modified Magnetic Particles (Cytiva), washed with 80% ethanol, resuspended in 100 mM ammonium bicarbonate and digested with 0.6 µg of trypsin (Promega) on a mixing platform for 16 h at 37°C. Samples were then acidified with trifluoroacetic acid (0.5% final concentration), and peptides were eluted.

For liquid chromatography-coupled mass spectrometry, peptides were loaded onto a PepMap Neo C18 trap column (300 μm × 5 mm, 5-μm particle size, Thermo Scientific, Thermo Fisher Scientific, Waltham, MA, USA) and separated with an EASY-Spray PepMap RSLC C18 analytical column with an integrated nanospray emitter (75 μm × 500 mm, 2-μm particle size, Thermo Scientific) on a Vanquish Neo system (Thermo Scientific). Two consecutive linear gradients were applied at a flow rate of 250 nL/min: 2%–24% solution B for 100 min and 24%–40% solution B for 20 min. The two mobile phases used were 0.1% formic acid (v/v) (A) and 80% acetonitrile (v/v) with 0.1% formic acid (B). Eluted peptides were sprayed directly into an Orbitrap Exploris 480 mass spectrometer (Thermo Scientific). Precursors were measured in the mass range 350–1,700 m/z with a resolution of 120,000 and selected for fragmentation in a data-dependent mode using the cycle time strategy (2 s) with a dynamic exclusion of 60 s. Higher-energy collisional dissociation fragmentation was performed with a normalised collision energy of 30%, and tandem mass spectrometry (MS/MS) scans were conducted with an isolation window of (m/z) 2 and a resolution of 30,000.

Obtained datasets were processed by MaxQuant (version 2.4.2.0) (29) with the built-in Andromeda search engine. Carbamidomethylation (C) was set as a permanent modification and acetylation (protein N-terminus) and oxidation (M) as variable modifications. The search was performed against the Homo sapiens protein database (UniProt, downloaded 30.08.2023). The relative quantities of individual proteins were determined by the built-in label-free quantification (LFQ) algorithm MaxLFQ, which provides normalised LFQ intensities for identified proteins (30). The statistical analysis was performed using Perseus v1.6.15.0 (31). Only proteins with two and more valid values in at least one experimental group were retained. Consequently, the missing values were imputed from the normal distribution creating the list of quantified proteins. Principal component analysis was used to evaluate sources of variability among samples and replicates. Next, Student’s t-test was applied with permutation-based false discovery rate correction for multiple testing with a q-value threshold at 0.01.

Both fibroblast-specific expression and exosomal origin were assigned to the quantified proteins using the list of fibroblast markers in the PanglaoDB database [https://panglaodb.se/; (32)] and the list of exosomal proteins in the ExoCarta database [http://exocarta.org/; (33)] and the Vesiclepedia database [http://www.microvesicles.org/; (34)], respectively.

Complete data can be found in Supplementary Table S1 .

The one-sample Kolmogorov–Smirnov test was used to determine whether the investigated population followed a normal distribution. Non-parametric analysis of variance (Kruskal–Wallis) with Dunn’s post-test was used to determine the differences and statistical significance. The results were expressed as the median and interquartile range (IQR). Correlation analysis was performed by Spearman’s test. A P-value <0.05 was considered to indicate statistical significance. The area under the receiver operating characteristic curve was calculated to assess the ability of CD44 to distinguish between fibrotic and non-fibrotic phenotypes of DPLDs. Statistical analysis was performed using the SAS software.

The study was approved by the Ethical Committee of the Faculty of Medicine of Comenius University in Bratislava and the Ethical Committee of the National Institute for Tuberculosis, Lung Diseases and Thoracic Surgery, Vysne Hagy. All investigations were carried out in accordance with the International Ethical Guidelines and the Declaration of Helsinki. Written informed consent for enrolling in the study, personal data management, and study was obtained from all patients and control subjects.

The major hallmark of lung fibrosis is the activation of lung fibroblasts to myofibroblasts. We applied this feature to identify specific fibroblast-derived biomarkers of pulmonary fibrosis in BALF. Figure 1 shows the basic workflow of our experimental approach: briefly, human fibroblasts were activated with the selected BALF from the IPF cohort; the activated fibroblasts were then washed rigorously and further incubated to allow secretion; the secretomes were then proteomically analysed by mass spectrometry; and, finally, the BALFs of all DPLD patients were tested for the presence of the identified candidate by ELISA.

In particular, nine BALF samples were randomly selected from the IPF cohort. Next, we applied MRC-5 cells, i.e., primary human lung fibroblasts, which had been well-characterised for their ability to be activated to myofibroblasts (35). The subconfluent MRC-5 fibroblasts were incubated 24 h either with the selected BALF samples (IPF; BALF diluted 1:3 with the medium) or with the control mixture (CTR; PBS diluted 1:3 with the medium). Afterwards, the cells were rigorously washed (three times) and incubated for the next 24 h with the medium to allow secretion. Afterwards, the cultivated cells were visualised by light microscopy. The phase-contrast images of the MRC-5 cells incubated with IPF-BALF displayed characteristic morphological changes (36) attributed to their activation from fibroblasts to myofibroblasts when compared with the control cells: namely, the IPF-BALF-treated fibroblasts appeared to be more flattened with evident nuclei, they apparently lost the typical stretched shape, and they were seemingly in a growth-arrested state ( Figure 2A ). In response to the IPF-BALF treatment, the MRC-5 fibroblasts increased the expression of vimentin and alpha smooth muscle actin (α-SMA), both markers of fibroblast activation; their expression levels in the cell lysates were normalised to the expression of COX IV, which was used as a housekeeping control protein ( Figure 2B ).

In parallel, the secretomes of stimulated and control MRC-5 fibroblasts were proteomically analysed by mass spectrometry with fibroblast-specific expression assigned to the identified proteins by using the PanglaoDB database. In this way, several fibroblast-specific proteins were found to be significantly enriched followed treatment with IPF-BALF ( Table 3 ), which further confirmed the activation of fibroblasts to myofibroblasts. Some of them [e.g., interleukin (IL)-6 and IL-8] are markers of general inflammation. Recent research highlights a role for CD44 in fibrotic processes (37–40): the role of CD44 in mesenchymal progenitor cells and their differentiation into fibroblasts in IPF, as well as its involvement in the acquisition of a motile phenotype by IPF fibroblasts (in patients fulfilling diagnostic criteria for IPF) and their invasive capabilities, has already been discussed in previous studies. In mice, CD44 expression increases following fibrosis induction with bleomycin. CD44 is involved in enhancing fibroblast motility and invasiveness. Therefore, we hypothesised that CD44 levels would show a more significant increase in fibrotic processes compared with inflammatory diagnoses within DPLD. Based on this, the CD44 protein was chosen for further study.

First, we confirmed the finding from mass spectrometry by Western blotting and ELISA. By means of both methods, we detected significantly higher levels of CD44 in the conditioned media from the IPF-BALF-treated MRC-5 cells when compared with those of control cells. The levels were normalised to COX IV expression in the corresponding lysates, and then the obtained normalised optical densities (ODs) were expressed as a fold change of IPF versus CTR ( Figures 2B, C ). In addition, control donors’ BALFs (four donors with SRC, one donor with inflammatory HP, and one donor without DPLD) were included in the experiments, showing results comparable with those of the PBS controls ( Supplementary Figure S1A ).

Next, we sought for the origin of CD44 secreted from the activated MRC-5 fibroblasts. The CD44 protein is known either to be proteolytically shed from the cell surface by various metalloproteases yielding a soluble ectodomain (41–43) or to be released from cells as a full-length membrane-embedded component of exosomes (44–48). To discriminate between these two possibilities, we performed the fibroblast secretion phase in the presence of the following inhibitors: GM6001 (galardin, MMP inhibitor) and GW4869 (exosome release inhibitor). As shown in Figure 2D , the co-incubation with GW4869 led to a reduction in CD44 secretion by the activated MRC-5 cells. Furthermore, CD63, an exosomal marker, displayed a similar expression profile in cell supernatants ( Figure 2D ; Supplementary Figure S1B ). Notably, the MMP inhibitor GM6001 caused a significant decrease in CD63 ( Figure 2D ; Supplementary Figure S1B ). Interestingly, it was shown that the inhibition of the shedding of desmosomal cadherin desmoglein 2 (Dsg2) with the MMP inhibitor GM6001 resulted in reduced exosomes’ release (49).

Moreover, the majority of the proteins identified by mass spectrometry were assigned to be of potential exosomal origin by using the ExoCarta and Vesiclepedia databases ( Table 3 ). Notably, analysis of CD44 messenger RNA (mRNA) levels in control and IPF-BALF-stimulated MRC-5 cells revealed no significant increase in CD44 expression upon stimulation ( Figure 2E ). This indicates the regulation of CD44 via subcellular distribution and not via gene expression.

These results altogether suggest that the activation of lung fibroblasts by IPF-BALF induces the secretion of CD44.

Based on these results, we tested the levels of CD44 in the BALFs of all DPLD patients with various diagnoses. As shown in Figure 3A , we detected significantly increased concentrations of CD44 in the BALF from the IPF cohort and also in the subgroups with fibrotic phenotype forms of HP and CTD-ILD cohorts. We did not detect increased concentrations of CD44 in BALF in both SRC and OP cohorts. When we separated the selected IPF-BALF by means of a size-exclusion chromatography on an Izon qEV column, which allowed the isolation of exosomes, we detected CD44 in the CD63-positive fractions corresponding to exosomes. In contrast, immunoglobulin was present in the fractions corresponding to soluble proteins ( Figure 3B ). In addition, we analysed BALF from IPF cohorts and from conditioned supernatants of the BALF-activated MRC-5 cells by means of the Exoid instrument measuring the size and concentration of exosomes in solution by the principle of TRPS. In both, BALF and supernatants, we detected vesicles of similar diameters in a range of approximately 150 nm ( Figures 3C, D ; Table 4 ) indicating similar characteristics of exosomes derived in vitro from fibroblasts and collected from BALF in vivo.

These findings implicate that CD44 might be present in BALF from the cohorts with pulmonary fibrosis in the form of an exosomal membrane-anchored receptor.

To evaluate the reliability of BALF-CD44 as a potential marker for pulmonary fibrosis, we conducted logistic regression models with a receiver operating characteristic (ROC). In the frame of our study, we categorised all subjects with DPLD into two groups: fibrotic (including IPF, fibrotic HP, and fibrotic CTD-ILD) and inflammatory ones (including SRC, inflammatory HP, inflammatory CTD-ILD, and OP). Logit models of the CD44 effect on the fibrotic process showed statistically significant differences even after adjusting for confounders (age and smoking) ( Table 5 ). The obtained area-under-the-ROC-curve (AUC) score, 0.8048, showed that CD44, as a biomarker, has a good predictive ability to discriminate fibrotic lung processes from other non-fibrotic DPLD diagnoses ( Figure 4A ). This suggests that measuring the CD44 concentration in BALF effectively distinguishes cases with and without fibrosis.

Notably, in the SRC patient group, only a very small proportion (4%) exhibited fibrotic involvement. SRC has a relatively low tendency to cause fibrosis, and patients in stage IV usually already have a confirmed diagnosis, making lavage testing unnecessary. This explains the limited number of stage IV patients in the study. The graph in Supplementary Figure S1C compares fibrotic SRC fibrosis (stage IV, N = 5) with inflammatory SRC phenotypes (stages I–III, N = 118).

Finally, we performed correlation analyses of the BALF-CD44 levels with the measures obtained independently by other diagnostic methods. First, the lungs of selected cases were examined by HRCT to gain more detailed characteristics, such as lung consolidation, emphysema, GGO, honeycombing, or reticular pattern ( Figure 4B ; Table 6 ). In this respect, the CD44 concentrations in BALF positively correlated with GGOs and reticular patterns ( Table 7 ). Furthermore, BALF-CD44 negatively correlated with DLCO and positively correlated with the total number of macrophages ( Table 8 ).

Taken together, these results suggest that BALF-CD44 is an appropriate marker to discriminate the fibrosing phenotypes of DPLDs.