RNA-sequencing analysis of lung primary fibroblast response to eosinophil-degranulation products predicts downstream effects on inflammation, tissue remodeling and lipid metabolism

Eosinophil isolation was performed under a study protocol approved by the University of Wisconsin-Madison Health Sciences Institutional Review Board. Informed written consent was obtained from subjects prior to participation. Peripheral blood eosinophils were obtained from allergic subjects with (n = 2) and without mild asthma (n = 1). Subjects with prescriptions for low doses of inhaled corticosteroids did not use their corticosteroids the day of the blood draw.

Eosinophils were purified by negative selection as previously described [29] from three female donors with 0.12, 0.2 and 0.47 × 10⁶ eosinophils per ml of whole blood. Briefly, 200 ml of heparinized blood was diluted 1:1 in HBSS and was overlaid above Percoll (1.090 g/ml). After centrifugation at 700 xg for 20 min, at room temperature, the mononuclear cells were removed from the plasma/Percoll interface and erythrocytes were eliminated from the cell pellet by hypotonic lysis. The remaining pellet was resuspended in 2% new calf serum in HBSS. Cells were then incubated with anti-CD16, anti-CD3, anti-CD14 and anti-Glycophorin-A beads from Miltenyi (San Diego, CA), and run through an AutoMACS (Miltenyi). Eosinophil preparations with purity > 99% and viability ~98% were used the same day, ~5 h after the blood draw.

Human lung fibroblasts (HLF) were isolated as described previously [30, 31] using de-identified tissue samples from thoracic surgical resection specimens. These samples were collected and characterized in collaboration with the biobanking services run by the Carbone Cancer Center Translational Science BioCore at the University of Wisconsin-Madison, under Institutional Review Board approval. To obtain non-fibrotic fibroblasts, we utilized adjacent (uninvolved) lung from lobectomy or biopsy specimens from patients undergoing lung resection for pulmonary nodules and who did not have any identifiable lung disease by history or histologic assessment. All specimens used for fibroblast isolation were examined by a pathologist to ensure that no underlying lung disease (emphysema, idiopathic lung diseases, etc.) was present on histology. Briefly, to isolate fibroblasts, tissue specimens were placed in DMEM with 100 units/ml streptomycin, 250 ng/ml amphotericin B, 100 units/ml penicillin, and 10 μg/ml ciprofloxacin. Alveolated lung tissue was minced and plated onto 10-cm plates in growth medium containing DMEM supplemented with 10% FBS, 2 mM L-glutamine, and antibiotics, as above. Expanded populations of fibroblasts were subsequently subcultured after 4–5 days, resulting in the development of a homogenous fibroblast population. All primary cultures were used from passages 5–10 and maintained on tissue culture plastic until the time of experiments. Two different HLF lines from different donors were used in this study.

Peripheral blood eosinophils were cultured at 1 × 10⁶/ml in medium (RPMI 1640 plus 10% fetal bovine serum (FBS)) with IL-3 (2 ng/ml) or IL-5 (4 ng/ml) for 20 h. After 20 h stimulation with cytokines, eosinophils were washed and suspended at 1 × 10⁶/ml in fresh medium (no cytokine), and 1 ml was added to a 24-well plate, which had been coated overnight with HA-IgG (10 μg/ml; 350 μl/well) and saturated with 0.1% gelatin for 30 min at 37 °C. Heat aggregation of human IgG (HA-IgG) was performed in PBS for 30 min at 63 °C, as previously described [29]. IL-3-preactivated eosinophils were also seeded on uncoated (no HA-IgG) wells, as negative control for eosinophil degranulation. After a 6 h incubation, the three types of supernatant fluids (conditioned media: IL-3, IL-3 on HA-IgG (IL3IgG) and IL-5 on HA-IgG (IL5IgG)) were harvested and stored at −80 °C, before analysis by ELISA for EDN release (degranulation), and before addition on cultured HLF. Human EDN ELISA (MBL, Woburn, MA) has a minimum detection limit of 0.62 ng/ml.

HLF were treated in 1:1 dilution of maintenance media (DMEM/10% FBS, 2 mM L-glutamine, and antibiotics as above) and eosinophil-conditioned media. The three types of eosinophil-conditioned media described above (IL-3, IL-3 on HA-IgG (IL3IgG) and IL-5 on HA-IgG (IL5IgG)) were obtained from three different blood donors. Each conditioned medium was tested individually on the HLF from both lung biopsy donors. Controls included HLF cultured in medium only (RPMI with 5% FBS) and HLF cultured with 1 ng/ml plus 0.5 μg/ml of HA-IgG (rhIL3IgG). This latter control was used to account for 1) potential release of intracellular IL-3 during IL-3-activated eosinophil degranulation, and 2) any release of coated HA-IgG during the 6 h degranulation period. After 24 h, HLF were harvest with RLT extraction buffer from Qiagen for total RNA extraction.

Total RNA was extracted from HLF using the RNeasy Mini Kit (Qiagen, Valencia, CA) and treated with DNase (RNase-free DNase kit, Qiagen). Fifteen μl of total RNA were recovered with concentrations ranging from 70 to 180 ng/μl and with ratios 260/280 nm ranging from 1.98 to 2.04. RNA samples were submitted to the University of Wisconsin-Madison Biotechnology Center (Madison, WI) for RNA quality and integrity evaluation via Agilent 2100 Bioanalyzer platform (Agilent Technologies) and whole-transcriptome sequencing. The sequencing library from mRNA was prepared using TruSeq Stranded mRNA Library Preparation Kit (Illumina; San Diego, CA, USA) and RNA-seq (1 × 100 bases) was carried out using Illumina HiSeq 2500 platform. SeqMan Ngen (v.13) and ArrayStar with Qseq module (v.13) software packages (DNAStar, Madison, WI) were used to map sequence reads to human reference genome (GRCh38), apply statistical analyses and identify global gene expression changes for both fibroblast donors. Of note, the two fibroblasts (L20 and L21) were analyzed separately, and only genes up- or down-regulated by more than 1.5 fold for both L20 and L21 were considered in our analyses. Raw data and normalized RPKM (Reads Per Kilobase per Million mapped reads) values have been deposited to the NCBI Gene Expression Omnibus (GEO) database and assigned accession number (GSE102860). Differentially expressed genes were analyzed using DAVID Bioinformatics Resources 6.8(Beta) (National Institute of Allergy and Infectious Diseases (NIAID), NIH) [32] and the Ingenuity Pathway Analysis (IPA) web-based software (Qiagen) to identify the relevant biological mechanisms and pathways.

The reverse transcription reaction was performed using the Superscript III system (Invitrogen/Life Technologies, Grand Island, NY, USA). Gene expression levels were determined by qPCR using SYBR Green Master Mix (SABiosciences, Frederick, MD, USA) or Taqman. For PCR using SYBR Green, forward and reverse specific primers (see Additional file 1: Table E1 for primer sequences) were designed using Primer Express 3.0 (Applied Biosystems, Carlsbad, CA, USA) and blasted against the human genome to determine specificity using http://​www.​ncbi.​nlm.​nih.​gov/​tools/​primer-blast. The reference gene primers, ß-glucuronidase ((GUSB), forward: caggacctgcgcacaagag, reverse: agcgtgtcgacccattc), was used to normalize the samples. Standard curves were performed and efficiencies were determined for each set of primers. Efficiencies ranged between 91 and 96%. Data are expressed as fold change using the comparative cycle threshold (∆∆CT) method as described previously [29].

Eosinophils activated for 20 h with IL-3 or IL-5 were added on coated heat-aggregated (HA)-human serum IgG for 6 h (IL3IgG and IL5IgG). The morphology of the eosinophils pre-activated with IL-3 or IL-5 and seeded on coated HA-IgG is shown on Additional file 2: Figure E1, where pictures of cultures were taken at 2 h, 4 h and 6 h. Additional file 2: Figure E1 suggests that in the earlier time-points, eosinophils adhered to HA-IgG. However, the adhesion was reversible for IL-5-activated eosinophils while IL-3 activation led to cell breaks and release of free granules (Additional file 1: Figure E1). As a negative control for eosinophil degranulation, IL-3-activated eosinophils were seeded in uncoated wells (IL3). In this condition, (no IgG), eosinophils do not adhere to the bottom of the wells (not shown). Conditioned media were harvested 6 h after eosinophils from three donors were seeded on IgG. As shown on Fig. 1, IL3IgG eosinophils released ~15-fold more EDN than the control IL3 (seeded on uncoated wells), and ~2-fold more than IL5IgG. These conditioned media samples were diluted (2-fold) in serum free medium, and added to human lung primary fibroblasts (HLF) for 24 h before RNA expression analyses by next-generation RNA sequencing (RNA-Seq) and RT-qPCR.

In a first analysis, we wished to examine gene expression levels changed in HLF by the degranulation products from IL-3-activated eosinophils. For this purpose, gene expression by fibroblasts incubated with IL3IgG conditioned media (mean values of two donors) was compared to both control fibroblast cultures (medium and rhIL3IgG). Genes whose expression level is changed in the same direction (up or down) by more than 1.5 fold in both HLF lines (L20 and L21) were included in this first dataset. Among these genes, 169 genes were upregulated (Additional file 3: Table E2). Analysis using DAVID Bioinformatics Resources identified 24 genes coding for secreted proteins (including 10 cytokines) and 47 genes coding for membrane proteins (Table 1). Additionally, 131 genes were downregulated (Additional file 4: Table E3), among them, 44 were genes coding for membrane proteins, 23 coded for nuclear proteins, while only 10 coded for secreted proteins.

The differentially expressed genes (up- (Additional file 3: Table E2) and down-regulated (Additional file 4: Table E3)) were analyzed using the Ingenuity Pathway Analysis (IPA) web-based software. The Predictive Causal Analytic tool identified three major and highly overlapping Regulator Effect Networks. These three networks contained part of the 300 genes present in our dataset (target molecules), potential upstream regulators of these genes, and the association between these genes with potential downstream biological functions (Table 2). More than 15% (27) of the 169 genes upregulated in our dataset were part of the target genes grouped into the three networks. There was 85% similarity among the target genes between the three networks. Around 80% of the potential identified upstream regulators (Table 2) were common among the three networks, and more than 65% of the projected functional downstream effects of these genes were similar between the three networks as well. Therefore, the aggregate of all three networks is shown in Table 2. The main predicted downstream functions of the fibroblast genes were related to the immune response, tissue remodeling and lipid metabolism (Table 2).

The upstream regulators are molecules that are predicted to activate or inhibit the expression of the genes in our dataset, according to the literature compiled in the Ingenuity® database. IPA determined the predicted overlap between an upstream regulator and the genes listed in our dataset. Additionally, IPA uses a z-score algorithm to make statistical prediction for activation or inhibition of a specific upstream regulator to genes present in our dataset. As shown in Table 3, the main significant groups of upstream regulators are composed of cytokines, kinases, intracellular signaling molecules and transcription regulators. The upstream regulators with the four highest z-scores for each group is shown in Table 3. Furthermore, other potential significant upstream regulators are listed on Additional file 5: Table E4. Of note and as expected, kinase inhibitors for MAPK, NF-kB, PI3K display negative z-scores predicting inhibitory effects on the target genes (Additional file 4: Table E3). In addition to these potential upstream regulators shown in Table 3, Additional file 4: Table E3 and Additional file 5: Table E4, miR-223 (z-score = −2.0; p = 8.51 × 10⁻³) was also found to be a potential upstream down-regulator of ABCA13, AK8 and EGR3 (Additional file 6: Table E6). This is particularly interesting in the light of a recent report by Zangari et al. where it is shown that neutrophils release miR-223-3p via extracellular vesicles [33]. Notably also, miR-223 expression was increased in bronchoalveolar lavage cells enriched in airway eosinophils 48 h after a segmental allergen challenge [34]. Subsequently, we tested the validity of these upstream regulators by analyzing the expression of the upstream regulator receptors on HLF. For this purpose, we extracted and listed the expression level of receptors for the upstream regulators predicted by IPA from our RNA-seq database. This is shown in Table 2, Table 3 and Additional file 4: Table E3. As shown in Additional file 7: Table E5, HLF expressed relatively high levels of the receptors for TNF-α, IL-1, OSM, IL-6, IFN-γ, IL-17A, IL-17F, IL-27, IFN-α, TNFSF12, C5 and F2. Of note, while IL2RG (common γ-chain receptor) is not expressed, IL-4 can signal via its type II receptor, whose subunits IL4R and IL13RA1 are both expressed by HLF (Additional file 7: Table E5). In addition to IL2RG, the receptors for GM-CSF and IL-18 were not expressed in HLF, in any of our conditions (Additional file 7: Table E5). Notably also, while IL-3-activated eosinophils may be the source of these upstream regulators, HLF might also release some of these regulators during the 24 h culture with eosinophil conditioned media. Therefore, the potential upstream regulators that are extracellularly localized and expressed by HLF were listed in Additional file 7: Table E5. HLF express IL6, TNFSF12, C5, IL32, FN1, APP, HMGB1 and TLR4 (Additional file 7: Table E5), which could all participate in HLF activation in an autocrine way. Eosinophils can produce and/or store a variety of cytokines and other ligands [35]. While we have not yet performed transcriptome and proteome of IL-3-activated eosinophils, we have recently reported RNA-seq data using resting human blood eosinophils [36]. In that study, among the potential upstream regulators identified by IPA, transcripts coding for CAMP (LL37), CHUK, HMGB1, HMOX1, IL-1, IL-18, IL-32, OSM, SELPG and TNFSF12, were detected in eosinophils. Therefore, altogether, our data indicate that IL-1ß, OSM (oncostatin-M), IL-32 and TNFSF12 (Tweak) may be candidates to explain the direct effect of conditioned medium from degranulated eosinophils on HLF. Interestingly, IL32 was upregulated in fibroblasts by IL3IgG conditioned media (Additional file 7: Table E5), suggesting that IL-32 could function endogenously on fibroblasts.

The downstream analysis performed by IPA shows a variety of downstream functions affected by the proteins coded by the genes composing our dataset. As shown in Table 2, downstream functions of 32 genes from our dataset as defined by the Regulator Effect Networks have been regrouped in three major categories: immune response, tissue remodeling and lipid metabolism. Identical to the Regulator Effect Networks (Table 2), another IPA tool called, Downstream Analysis as Networks, identified two major downstream networks, which are linked to the immune response (Fig. 2). These two networks (immune response of cells and homing of cells) are predicted to be regulated by 27 of the 300 genes present in our dataset (Additional file 1: Table E1 and Additional file 3: Table E2), with relatively high positive z-scores (3.180 and 3.407; Fig. 2). In addition to these two networks, angiogenesis, development of vasculature and development of connective tissue cells, are three major mechanistic networks characterizing tissue remodeling (Fig. 3). These three mechanistic networks include as many as 29 genes from our dataset (~10%). Finally, a third mechanistic network, named synthesis of lipid, is affected by 19 genes of our dataset and is illustrated on Fig. 4.

IPA network algorithm also generates networks showing interactions between the genes of our dataset in relation with other neighboring genes present in a Global Molecular Network. Thirty-five total genes, or group of genes, are present in each network. Fifteen networks were formed using > 10 genes from our dataset. The networks with the two highest scored (p-value < 10⁻²³) and including > 15 genes/groups from our dataset are shown on Additional file 8: Figure E2 and Additional file 9: Figure E3. These 15 genes or gene products are directly or indirectly associated with the protein kinase B (PKB or Akt) (Additional file 8: Figure E2) or the c-jun N-terminal protein kinase (JNK) and TGF-ß (Additional file 9: Figure E3), and mostly characterized by the inflammatory response, cell movement and cell/tissue development. Of note, the functional networks created by IPA analysis (Additional file 8: Figs. E2 and Additional file 9: Figure E3) suggest that there is no known tight functional connection either between the 27 genes characterizing “inflammation” (Fig. 2), or between the 29 genes characterizing “tissue remodeling” (Fig. 3) or between the 19 genes characterizing “lipid metabolism” (Fig. 4). Although it is worth noticing that the network 1 (Additional file 8: Figure E2) includes IL12RB1, mir-221, mir-29 and NODAL that are all part of the 27 genes related to inflammation. In addition, the network 2 (Additional file 9: Figure E3) includes APOA1, C3, COL4A3, CYP1B1 and FGF10 that are part of tissue remodeling, and ADH1C, APOA1, C3 and CY1B1 that are all part of the genes characterizing lipid metabolism. This indicates that there are some known functional relationships between genes characterizing “Inflammation”, “tissue remodeling” or “lipid metabolism” (Figs. 2, 3 and 4).

In a second analysis of our RNA-seq database, we restricted the list of genes obtained in the first dataset by eliminating the genes which expression levels were not changed in HLF cultured with IL-3- compared to IL-5-activated eosinophils (IL3IgG versus IL5IgG). For this purpose, gene expression by HLF incubated with IL3IgG conditioned media (average of two donors) for 24 h was compared to fibroblast incubated with IL5IgG conditioned media (average of two donors), and to both controls HLF (medium and rhIL3IgG). Genes whose expression level changed by more than 1.5-fold in the same direction (up or down) in both HLF lines are included in this second restricted dataset. In these comparisons, 35 genes coding for known proteins were upregulated (Table 4), and analysis using DAVID Bioinformatics Resources 6.8 identified eight genes coding for secreted proteins (also shown in Table 1). In addition, in this dataset #2, 37 genes were downregulated (Additional file 6: Table E6), among them, three were genes coding for secreted proteins. Table 4 and Additional file 6: Table E6 give a brief description of the 72 (35 up +37 down) genes and their relationship with the networks created by IPA using the 300 genes in our dataset #1 (Tables 2 and 3; Figs. 2, 3 and 4, Additional file 2: Figure E1 and Additional file 8: Figure E2). Furthermore, Additional file 10: Table E7 shows the genes of our dataset #2 associated with the six most significant global molecular networks formed by IPA (Additional file 8: Figure E2, Additional file 9: Figure E3 and not shown). The genes of our dataset #2, which are part of the three downstream functional categories (immunity, remodeling and lipid metabolism, Figs. 2, 3 and 4), are listed in Table 5. The eight genes shown in Table 5 are involved in at least one of these three categories, and four genes (C3, CXCL1, CXCL8, IL6) are part of all three categories. From Table 5, CXCL1, CXCL8, ICAM1 and IL6 were arbitrarily selected to validate our RNA-seq analyses by RT-qPCR.

For RT-qPCR analysis, conditioned media (IL3IgG, IL5IgG and IL3 (no IgG)) from all three eosinophil donors were used on the two HFL lines. Controls included fibroblasts cultured in medium only and fibroblasts cultured with 1 ng/ml of IL-3 plus 0.5 μg/ml of HA-IgG (rhIL3IgG). In agreement with the RNA-seq data, qPCR data (Fig. 5) confirms that CXCL1, CXCL8, ICAM1 and IL6 are all significantly upregulated by IL3IgG conditioned media compared to IL5IgG and controls.