Surgical adhesions in mice are derived from mesothelial cells and can be targeted by antibodies against mesothelial markers

See allHide authors and affiliations

Science Translational Medicine  28 Nov 2018:
Vol. 10, Issue 469, eaan6735
DOI: 10.1126/scitranslmed.aan6735

Getting adhesions unstuck

Peritoneal adhesions are ectopic fibrotic tissues induced by surgical perturbations that result in postoperative morbidities such as small bowel obstruction. The cellular origin of adhesions remains unclear. Now, Tsai et al. show that mesothelial cells overlying organs and the abdominal wall give rise to adhesions after surgery in mice. The injured mesothelium up-regulated mesothelium-specific genes that were known to be highly expressed during fetal development. Targeting adhesions with antibodies against the mesothelial marker mesothelin eliminated adhesions that had formed after surgery. Injured mesothelium responded to hypoxia, and this was mediated by the HIF1α pathway. Blocking HIF1α with small-molecule inhibitors prevented adhesion formation in mice after surgery.


Peritoneal adhesions are fibrous tissues that tether organs to one another or to the peritoneal wall and are a major cause of postsurgical and infectious morbidity. The primary molecular chain of events leading to the initiation of adhesions has been elusive, chiefly due to the lack of an identifiable cell of origin. Using clonal analysis and lineage tracing, we have identified injured surface mesothelium expressing podoplanin (PDPN) and mesothelin (MSLN) as a primary instigator of peritoneal adhesions after surgery in mice. We demonstrate that an anti-MSLN antibody diminished adhesion formation in a mouse model where adhesions were induced by surgical ligation to form ischemic buttons and subsequent surgical abrasion of the peritoneum. RNA sequencing and bioinformatics analyses of mouse mesothelial cells from injured mesothelium revealed aspects of the pathological mechanism of adhesion development and yielded several potential regulators of this process. Specifically, we show that PDPN+MSLN+ mesothelium responded to hypoxia by early up-regulation of hypoxia-inducible factor 1 alpha (HIF1α) that preceded adhesion development. Inhibition of HIF1α with small molecules ameliorated the injury program in damaged mesothelium and was sufficient to diminish adhesion severity in a mouse model. Analyses of human adhesion tissue suggested that similar surface markers and signaling pathways may contribute to surgical adhesions in human patients.


Adhesions are fibrous tissues that develop after trauma to serosal membranes and surrounding cavities, which are likely to share a common pathology. Adhesions arise in response to instigations such as surgery, infection, or dialysis. Their primary sequelae include small bowel obstruction, chronic pain, female infertility, poor quality of life, and death (14). The National Institutes of Health estimates that ~93% of abdominal surgeries lead to adhesions (5), with a ~20% rate of rehospitalization for adhesion-related complications. The annual cost of surgery-related adhesions is over $1 billion in the United States (1, 3, 58). Although adhesions are a significant health care burden, few definitive strategies exist to prevent or treat adhesion formation.

Adhesions occur through an insult resulting in hypoxia and reactive oxygen species damage leading to inflammation and activation of the coagulation cascade, which results in the formation of a fibrin bridge between adjacent surfaces (4). Much is known about later stages of adhesion formation involving fibrin deposition and fibrinolysis (9); however, the cellular and molecular details of the initial stages require further elucidation. It has been proposed that the frictionless surface of the mesothelium, the epithelial monolayer lining the peritoneal cavity and visceral organs (912), plays a protective role against adhesion formation (13), suggesting that adhesion formation requires removal of the mesothelium. Exposed basement membrane is suspected to be the substrate for fibrin attachments between denuded surfaces, followed by fibroblast accumulation from subparenchymal sources (9, 12). Other studies propose that possible sources of adhesions are myofibroblast metaplasia (14), mesothelial-to-mesenchymal transition (MMT) (15), or extracellular matrix (ECM) deposition by subperitoneal fibroblasts (11, 1620). Because these observations are largely dependent on histology, the early mechanisms continue to be debated. A cell of origin has not yet been definitively identified through lineage-tracing analyses. A few transcriptomic studies have characterized the gene expression changes leading to adhesion initiation (21, 22), although it is unknown which cells sense the initial insults to initiate downstream pathology. Consequently, few effective treatments exist.

Here, we demonstrate by lineage tracing that an activated subset of mesothelial cells could be a cell of origin for adhesions. We identify mesothelin (MSLN) as a specific surface marker that is up-regulated by mesothelial cells participating in adhesion formation and show that these cells are a necessary component of adhesion tissue in mouse and human. We found that administration of anti-MSLN antibodies diminished adhesions in a mouse model and that this treatment was boosted by coadministration of antibodies against CD47 (23, 24). We isolated and purified injured peritoneal mesothelium from mice using fluorescence-activated cell sorting (FACS) and performed RNA sequencing to identify molecular candidates for adhesion initiation. We show that peritoneal injury in a mouse model using surgical ligation and abrasion induced up-regulation of the hypoxia-inducible factor 1 alpha (HIF1α) signaling pathway and that this pathway activated reprogramming of the mesothelium. Disruption of the HIF1α pathway using small-molecule inhibitors was sufficient to curtail adhesion formation in vivo.


Mesothelial cell proliferation after peritoneal injury in mice

To identify and prospectively isolate cells that could be responsible for adhesion development, we analyzed various established rodent injury models for reproducibility in adhesion severity and location. These animal models included those where adhesions were induced by intraperitoneal injections of chemicals, rough abrasion of the cecum and peritoneum (25, 26), cauterization of the peritoneum and abdominal organs (12, 27), and placement of ischemic buttons in the peritoneal wall (15, 2831). We found that the use of chemicals, cauterization, and general or even localized abrasion were unpredictable in the reproducibility of adhesion formation. Instead, adhesion formation was systemic in these models, likely due to generalized inflammation.

We found that inducing formation of a small, single ischemic “button” in the peritoneal wall of mice by pulling up the peritoneal wall with forceps and ligating its base resulted in adhesions between the button and adjacent tissue, with little morbidity (Fig. 1A). This method was reproducible: Adhesion sites were consistently in the same place, making it easier to pinpoint the locations of cells likely to give rise to adhesions. This method was physiologically relevant as the button simulated micro-pockets of ischemia that often form in response to surgical ties. We found that an optional, gentle abrasion of the ischemic button and intestinal surfaces could controllably increase adhesion severity without increasing morbidity in the mice.

Fig. 1 Mouse surface mesothelium proliferates in response to the induction of adhesions.

(A) Adhesions were induced in mice by surgical placement of ischemic buttons in the mouse peritoneum at t = 0 days. Ischemic buttons were analyzed for mesothelial proliferation 3 days after button induction. Adhesions were analyzed 7 days after button induction. (B) Images show H&E staining of ischemic buttons at 30 min and 1, 2, and 4 hours after button induction, followed by abrasion in the peritoneum (black arrowheads) (n = 2 mice per time point). (C) Immunofluorescence staining of ischemic buttons for PDPN and cytokeratin 19 (K19) at 0, 6, 12, and 24 hours after button induction and abrasion of the mouse peritoneum (n = 2 mice per time point). (D) Immunofluorescence staining of string adhesions between the liver (L) and peritoneum (P) for MSLN and PDPN 7 days after adhesion induction in mice. (E) Numbers of mesothelial cells (MSLN+) (top) and double-positive EdU+MSLN+ mesothelial cells (bottom) were counted in n = 20 high-powered fields (0.75 mm × 1 mm) in normal mouse peritoneal tissue versus peritoneal tissue with adhesions. (F) Immunofluorescence staining for MSLN and EdU (MSLN+EdU+ in white arrowheads) in normal mouse peritoneum (P) (top) and adhesions (ADH) between the peritoneum and large intestine (I). Scale bars, 100 μm. Values are presented as means ± SEM; ***P < 0.0005, ****P < 0.0001. Analyses were done with an unpaired t test.

After surgery, mice were allowed to recover and analyzed at 0.5, 1, 2, 4, 24, and 72 hours and 7 days after induction of the ischemic button. Buttons were removed and stained with hematoxylin and eosin (H&E) at 30 min and at 1, 2, and 4 hours after adhesion induction (Fig. 1B). To confirm that mesothelial cells were present after button induction, buttons were stained with antibodies against the mesothelium-specific markers podoplanin (PDPN) and keratin 19 (K19). Although other studies have reported PDPN expression on injured fibroblasts (32, 33), we only observed PDPN expression in cells colabeled with K19 on the surface monolayer of the buttons at 0, 6, 12, and 24 hours after button induction (Fig. 1C and figs. S1 and S2) and on adjacent tissues, suggesting that PDPN+ cells were mesothelial. Confocal microscopy of peritoneal wall tissue confirmed that PDPN expression remained specific to mesothelium during injury, as compared to smooth muscle actin (SMA), the expression of which was characteristic of fibroblasts (fig. S3, A and B).

We found that MSLN, a marker highly expressed on fetal peritoneal mesothelium, with low expression in adult mesothelium (21), was highly up-regulated by the mesothelium at all time points after injury in the mice (figs. S3, C and D, and S4). Intact cellular layers were visible on ischemic buttons with and without abrasion at postoperative time points (Fig. 1C), indicating that the mesothelium was not denuded, nor had it retracted after this injury. At some time points, e.g., at 12 hours after injury, mesothelial cells took on an “activated” phenotype, characterized by loosening of intercellular connections (fig. S2). The mesothelium proliferated at 4 hours after injury as evidenced by its transition from a single-cell layer to a multicellular layer (fig. S5A). By 24 hours after injury, the mesothelium showed further signs of thickening (Fig. 1C and fig. S5B), forming full adhesions after 7 days (fig. S5C). Trichrome staining and additional immunohistochemistry using antibodies against PDPN and MSLN were performed and showed that adhesions contained cells that stained with either PDPN or MSLN (Fig. 1D and fig. S6), suggesting that they may be locally derived from peritoneal mesothelium.

To investigate whether mouse adhesion tissue was derived from local surface mesothelium or circulating cells, we subjected C57BL/6J wild-type mice and C57BL/Ka Rosa26mRFP1 mice (34) that constitutively expressed red fluorescent protein in all cells to parabiosis where both mouse circulatory systems were connected (35, 36). By 14 days, the blood of both mice was chimeric (fig. S7). Adhesions induced in the nonfluorescent C57BL/6J wild-type mouse that underwent parabiosis showed little evidence of contributions from circulating cells. Most of the PDPN+ cells in the adhesions were not fluorescent, although infiltrating blood or tissue myelomonocytic cells were observed (fig. S8) as evidenced by their coexpression of F4/80 stain.

Mice were injected intraperitoneally with 0.025 mg of 5-ethynyl-2′-deoxyuridine (EdU) after adhesion induction and allowed to recover for 7 days; then, their adhesions were stained for EdU, PDPN, and MSLN. Mesothelial cells (PDPN+ or MSLN+) were counted in multiple (n = 20) high-power fields (0.75 mm × 1 mm) in control mice and in mice after adhesion induction (n = 16). Normal undamaged mesothelium showed rare mesothelial cell proliferation. However, there were marked foci of EdU+MSLN+ mesothelial cells in areas of the mouse peritoneum where adhesions were present and around regions of thickening (fig. S9). This suggested that mesothelial cell proliferation contributed to adhesion development in mice receiving ischemic buttons (Fig. 1, E and F).

To study the ultrastructural changes in the mesothelial cells after button induction, we performed conventional and large block-face scanning electron microscopy on uninjured peritoneum and peritoneum on ischemic buttons 30 min and 2, 4, 6, 12, and 24 hours after button induction (figs. S10 to S12). These images confirmed the presence of mesothelial cells in the adhesion tissue and the basement membrane underneath the mesothelium and suggested that, within 1 hour after injury, some mesothelial cells detached from their basement membranes, separated from neighboring cells, and extended ciliary protrusions into the peritoneal space.

In vivo and in vitro lineage tracing of activated mesothelial cells from mouse adhesions

We developed a lineage-tracing assay in which a cell-permeable dye, 5(6)-carboxyfluorescein N-hydroxysuccinimidyl ester (CFSE), was pipetted directly onto the surgically induced buttons to label mesothelial cells in mice. Buttons from mice treated with 10 μM CFSE became fluorescent within 30 min (Fig. 2A). Staining for PDPN and K19 confirmed that no CFSE labeling occurred in underlying tissue (Fig. 2B and fig. S13), indicating that the CFSE dye specifically labeled surface mesothelial cells and did not cross the mesothelial basement membrane. Mice undergoing ischemic button placement and treated with CFSE were allowed to recover for 4 to 7 days. Fluorescence imaging confirmed CFSE staining within adhesions (Fig. 2, C to E) that colocalized with PDPN staining (Fig. 2, F and G, and fig. S14), indicating that labeled mesothelial cells contributed to the adhesions formed after button induction.

Fig. 2 Lineage tracing and clonal analysis of mesothelial cells involved in adhesion formation in mice.

(A) Whole-mount imaging of an ischemic button (IB) imaged in vivo 30 min after adhesion induction and treatment with the stain CFSE. (B) Immunofluorescence staining of PDPN, K19, and CFSE on the surface mesothelium surrounding a single ischemic button (single staining for PDPN, K19, and CFSE is shown in fig. S13). (C to E) Whole-mount imaging of an ischemic button 4 days after adhesion induction and treatment with CFSE stain. Adhesions (ADH) are indicated by white boxes and white arrowheads (n = 3). (F and G) Immunofluorescence staining of adhesions for PDPN and CFSE 7 days after adhesion induction (n = 5). (H) Immunofluorescence staining of adhesion sites for PDPN and ActinCreER; R26VT2/GK3 7 days after adhesion induction and CFSE injection. White arrowheads indicate cells originating from the same precursor. (I) Immunofluorescence staining for PDPN and ActinCreER; R26VT2/GK3 at adhesion sites 7 days after adhesion formation (n = 3). White arrowheads indicate a spindle-like phenotype. Scale bars, 100 μm unless otherwise indicated.

We then crossed ActinCreER mice with R26VT2/GK3 Rainbow mice (35, 37, 38) carrying a multicolored fluorescent reporter. After low-dose tamoxifen administration and Cre induction, the Rainbow reporter was activated in a very low number of cells, with each cell randomly and permanently expressing one color. We and others have found that tracing many cells simultaneously makes it difficult to determine clonality and morphological outcomes (37, 38). Therefore, low-dose tamoxifen studies were chosen to elucidate lineage outcomes of single cells. Adhesion induction was performed on ActinCreER; R26VT2/GK3 offspring, and lineage tracing was conducted on adhesions in vivo for 7 days. Clusters (clones) of PDPN+ cells of the same color were found within adhesion sites (Fig. 2H), suggesting that adjacent cells were derived from single precursor cells. Some PDPN+ mesothelial cells exhibited spindle-like morphology, suggesting a potential transition to a fibroblast phenotype (Fig. 2I and fig. S15).

Mesothelial explants from the kidney capsule and intestines were excised from uninjured control mice and cultured for 14 days. Cells expanded outward from the explant and stained positive for PDPN and K19 (fig. S16), and many PDPN+K19+ cells showed spindle-like phenotypes that were different from their original cobblestone appearance. These in vitro experiments suggested that adhesions may result from a transition or differentiation of epithelial-like mesothelial cells into spindle-like fibrocytic cells (fig. S16).

Targeting MSLN in peritoneal adhesions in mice

To determine whether PDPN+MSLN+ mesothelial cells were a necessary component of adhesions, we depleted these cells from preformed adhesions in mice. Mouse monoclonal anti-MSLN antibodies were injected intraperitoneally immediately after adhesion induction (Fig. 3A) and were found to bind only to damaged areas. Light sheet microscopy of injured mouse peritoneum revealed emerging foci of MSLN expression surrounding injury sites (Fig. 3, A and B; fig. S17; and movies S1 and S2). MSLN expression was specific to injured areas surrounding the ischemic button, leaving adjacent areas unlabeled (Fig. 3B and figs. S17 and S18, controls); antibody binding was absent in deeper tissues.

Fig. 3 Mesothelium is a necessary component of adhesions in mice.

(A) Light sheet microscopy images after injection of anti-MSLN antibodies into the mouse peritoneum after adhesion induction in vivo. Autofluorescence (green) indicates the suture (S) and muscle fibers of the peritoneum (P). (B) Further analysis by light sheet microscopy shows longitudinal expansion of MSLN+ staining that is parallel to the muscle fibers. Volume rendering of 2D images (a virtual z-stack) as a 3D model was done using the MIP or blend mode. 2D representations are MIPs of 36-μm-thick sections from the 3D model. (C) CD47 expression in FPKM (fragments per kilobase of transcript per million mapped reads) at 6, 12, and 24 hours after adhesion formation in response to button induction in mice. (D) Phagocytosis of mesothelial cells treated with anti-CD47 antibody by macrophages. (E) Immunofluorescence staining of macrophages (red) and mesothelial cells (green) showing phagocytosis (denoted by white arrowheads) of mesothelial cells by macrophages. (F) Appearance of adhesions 2 weeks after button induction in mice. (G) Appearance of adhesions after treatment with anti-MSLN antibodies injected at 7, 10, or 13 days after adhesion formation. (H) Adhesion score after treatment of formed adhesions in mice with anti-CD47 antibody alone (n = 8), anti-MSLN antibody alone (n = 3), or a combination of anti-CD47 and anti-MSLN antibodies (n = 5) compared to control untreated mice injected with phosphate-buffered saline (PBS; n = 10). Antibodies were injected at 7, 10, and 13 days after induction of adhesions. Values are presented as means ± SEM; **P < 0.005, ****P < 0.0001. Statistical analyses were done with an unpaired t test. (I) Ischemic buttons (IB) from mice after antibody treatment showing immunofluorescence staining for collagen, fibronectin, CD31, F4/80, and MSLN.

Damaged mesothelium also expressed CD47 (Fig. 3C), a surface molecule highly up-regulated on tumors that blocks macrophage-mediated phagocytosis (24). Blocking CD47 with anti-CD47 antibody has been shown to eliminate many human tumors in immunodeficient mice by enhancing phagocytosis of tumor cells (23, 24, 39, 40). A human mesothelial cell line (MeT5A) was treated with anti-CD47 antibody in vitro and showed increased phagocytosis by macrophages (Fig. 3, D and E) compared to the MeT5A human mesothelial cell line treated with control immunoglobulin G (IgG) antibody.

To investigate the potential therapeutic effect of antibody treatments in vivo, we required a method to evaluate adhesion severity. Previous studies characterized adhesions by focusing on strength (mechanical difficulty in separating adhered tissue) or the number of organs to which they adhered to score adhesion severity (27, 41, 42). We developed a method to evaluate adhesion severity based on histological criteria and adhered surface area (fig. S19).

Anti-CD47 antibody (200 μg) alone, anti-MSLN antibody (200 μg) alone, or a combined injection of both antibodies was administered 7, 10, and 13 days after adhesion induction in mice. A decrease in adhesion burden was observed using anti-MSLN antibody alone and a greater reduction was observed using a combination of anti-MSLN antibody and anti-CD47 antibody (Fig. 3, G and H) compared to vehicle controls (Fig. 3, F to H). Immunohistological analysis of the ischemic buttons resected from the antibody-treated mice confirmed the absence of MSLN+ cells (Fig. 3I), suggesting that mesothelial cells had been removed. This MSLN+ subset may be a necessary component of adhesion tissue and potentially could contribute to adhesion formation. Collagen and fibronectin were still detected on the resected buttons, suggesting that MSLN+ cell removal was sufficient to diminish adhesion burden, despite residual ECM components. Immunofluorescence staining for K19 and PDPN (fig. S20) and H&E staining (fig. S21) showed the persistence of mesothelium on the buttons.

RNA sequencing of mouse mesothelial cells in the early stages of adhesion formation

To determine changes in gene expression in PDPN+MSLN+ cells during adhesion formation, we isolated PDPN+MSLN+ cells using FACS from sites of injury at 6, 12, and 24 hours after adhesion formation after button induction. Our lineage gating scheme included mesothelial markers and excluded blood cells, endothelial cells, and lymphatic cells. The cells selected by FACS were PDPN+LYVE1CD31CD45 (Fig. 4A), and we performed RNA sequencing on this population. Our purification protocol included ischemic buttons to enrich for activated mesothelium. Peritoneal mesothelial cells were similarly isolated from mice that did not undergo surgery (t = 0). Differentially expressed genes were clustered on the basis of expression patterns across the 24-hour time course and then were analyzed for gene set activity (4345) to elucidate early transcript changes occurring in the first 24 hours after adhesion formation (Fig. 4, B to E).

Fig. 4 Mesothelial cells show a distinct transcriptional profile after adhesion induction in mice.

(A) Surface mesothelium was isolated from ischemic buttons, and cells with a PDPN+LYVE1CD31CD45 surface phenotype were obtained by flow cytometry. DAPI, 4′,6-diamidino-2-phenylindole. (B) Heatmap of RNA expression in purified surface mesothelium immediately after button induction in mouse peritoneum (t = 0) and 6, 12, and 24 hours after button induction. Representative genes are shown above gene clusters. (C) Log fold changes in transcript expression were calculated 24 hours after induction of adhesions compared to controls and were plotted against gene identity. (D) Number of up-regulated and down-regulated genes from uninjured and injured mesothelium at 24 hours after button induction is shown versus the number of up-regulated or down-regulated genes in HSCs and their progeny after differentiation. Differentiated hematopoietic cells included multipotent progenitors (MPPs), common myeloid progenitors (CMPs), monocytes, natural killer (NK) cells, CD4+ and CD8+ T cells ± CD69 expression, AML stem cells (LSCs), AML progenitor cells (LPCs), and AML blasts. (E) Heatmap of gene sets up-regulated or down-regulated in mouse surface mesothelium 6, 12, and 24 hours after adhesion formation. NF-κB, nuclear factor κB; BMP, bone morphogenetic protein.

Differentially expressed genes broadly clustered into six expression patterns, peaking or dipping at 6, 12, and 24 hours after button induction (Fig. 4B). Fold changes in transcript expression after 6, 12, and 24 hours were plotted against total gene expression (Fig. 4C). Differentially expressed genes were calculated on the basis of a q < 0.05 threshold and ordered by fold change. At 24 hours after injury, ~8000 genes were significantly (q < 0.05) differentially expressed compared to control mice that did not undergo surgery (Fig. 4C).

We compared the number of differentially expressed genes in mesothelial cells between injured and control mice 24 hours after button induction to genes expressed during hematopoietic stem cell (HSC) differentiation. Transcriptional changes in activated mesothelial cells were greater than those found among HSCs, macrophages, and B cells. Transcriptional changes in activated mesothelial cells were greater than those found in HSCs undergoing differentiation to multipotent progenitors, in activated T cells compared to resting CD69 T cells, and in acute myeloid leukemia (AML) cells compared to AML blast cells (Fig. 4D). Together, this suggested extensive reprogramming in the mesothelium after button induction and injury.

Gene set activity analysis (Fig. 4E) showed early up-regulation of genes involved in angiogenesis and hypoxia, followed by expression of genes involved in the inflammatory response encoding chemokines, chemotactic factors, cytokines, and nuclear factor κB pathway components (Fig. 4E). Genes encoding proliferation factors were up-regulated 24 hours after button induction, corroborating data from our pulse-chase experiments. Gene sets associated with ECM formation were down-regulated within 24 hours after injury, including fibronectin 1 and collagens (COL1A1, COL1A2, and COL3A1; fig. S22). The HIF1A gene was highly up-regulated within 6 hours of adhesion formation, indicating that sensing of hypoxia in the mesothelium may contribute to adhesion initiation. The tumor growth factor (TGF) signaling pathway, which is thought to contribute to adhesion formation, was down-regulated in the mesothelium 24 hours after surgery (Fig. 4E).

We observed expression of the mesothelium-specific markers PDPN and K19 at all time points after adhesion initiation in mice (Fig. 5A). Many markers, which are highly expressed by peritoneal mesothelium during mouse fetal development with low expression in the uninjured adult, were up-regulated after injury. These included MSLN, which was up-regulated 80-fold, uroplakin1B (UPK1B), and Wilms’ tumor antigen 1 (WT1), which peaked at 24 hours after adhesion initiation (Fig. 5A). Genes associated with fibroblasts, specifically S100A4 (FSP1), E-cadherin (CDH1), and SMA (ACTA2), were not expressed in uninjured mesothelium (t = 0) but were up-regulated after injury (Fig. 5A). Adhesion molecules were also up-regulated, indicating that cell-to-cell contact and cellular migration were important to adhesion initiation (fig. S22). These RNA sequencing data show that injured mesothelial cells undergo a massive genomewide transcriptional rearrangement involving a mesothelial program with an increased fibrogenic composition.

Fig. 5 Mouse mesothelial gene expression compared to fibroblast gene expression.

(A) Expression of genes (indicated above graphs) in mouse mesothelial cells by RNA sequencing (FPKM) over 24 hours. (B) Immunofluorescence staining of adhesions 7 days after button induction for PDPN, MSLN, CD44, S100A4, and K19. Scale bars, 100 μm. ADH, adhesions; L, liver; P, peritoneal wall; I, intestine.

We validated our RNA sequencing screen using immunohistochemical staining of proteins encoded by gene targets in mouse adhesions (Fig. 5B and fig. S23B). There was strong protein staining detected by immunofluorescence and RNA expression detected by in situ hybridization for the target genes, demonstrating that the RNA sequencing data were recapitulated in vivo. Adhesions did not stain positive for HIF1A expression at later stages after injury initiation; this transcription factor was up-regulated within 6 hours of injury and down-regulated soon thereafter. Injured mesothelium broadly expressed S100A4, MSLN, and K19, whereas WT1 was restricted to a subset of injured mesothelium, suggesting heterogeneity within adhesions.

Effect of HIF1α blockade on adhesion formation in mice

Hypoxia has been shown to be an instigator of adhesion formation (9), although which cells sense hypoxia and their contribution to adhesion formation have not been well characterized. We found that mesothelial HIF1A gene expression was up-regulated early during adhesion formation before decreasing shortly thereafter (Fig. 5A). It is established that the HIF1α transcription factor is regulated through stabilization and posttranslational modification (46). We investigated protein partners of HIF1α known to increase HIF1α activity and found them to be up-regulated at similar time points during adhesion development (fig. S24).

We analyzed the potential effects that HIF1α could exert on surface mesothelium in vitro. Primary mouse mesothelial cells that were cultured under low oxygen conditions (5% O2 or with 100 μM CoCl2) for 3 days showed little morphological difference from mesothelial cells cultured under normoxic conditions (fig. S25). Mesothelium cocultured with macrophages, which are constitutively present in the peritoneum, developed dense fibrotic foci that stained positive for PDPN and HIF1α (Fig. 6, A and B), suggesting that macrophages may contribute to mesothelial transformation.

Fig. 6 Inhibition of HIF1α is sufficient to prevent adhesion formation in mice.

(A) Mesothelial macrophage cocultures in vitro under normal oxygen conditions (normoxia) immunostained for PDPN or HIF1A expression. (B) Mesothelial macrophage cocultures in vitro under hypoxia conditions (5% O2 incubator) stained for PDPN or HIF1A expression. Scale bars, 250 μm. (C) Gross anatomy, immunofluorescence staining, and H&E staining of mouse adhesions 7 days after button induction and treatment with echinomycin (20 μg/kg) daily for 7 days. (D) Adhesion score after treatment of mouse adhesions with echinomycin or PX12, small-molecule inhibitors of HIF1α. (E) Numbers of double-positive mesothelial cells (MSLN+PDPN+) and triple-positive EdU+MSLN+ PDPN+ mesothelial cells were counted per high-powered field (0.75 mm × 1 mm) after treatment of mice with the HIF1α inhibitor echinomycin (n = 16 fields) and were compared to untreated control mice (n = 5 fields). (F) Expression measured by RNA sequencing (FPKM) of selected target genes during a 24-hour time course after adhesion induction and treatment of mice with the HIF1α inhibitor echinomycin (n = 2). Heatmap showing RNA sequencing of surface mesothelium immediately after echinomycin treatment and button induction and 6, 12, and 24 hours after echinomycin treatment and button induction. Scale bars, 100 μm unless otherwise noted. Values are presented as means ± SEM; **P < 0.005, ****P < 0.0001. Statistical analyses were done by an unpaired t test.

In vivo analysis of the HIF1α signaling pathway was performed using small-molecule inhibitors of HIF1α, specifically cryptotanshinone (inhibits HIF1α activation) (47), FM19G11 (represses α subunit targets) (48), echinomycin (blocks DNA binding of HIF1α) (49), and PX-12 (blocks transcriptional activity of HIF1α) (50). Mice underwent adhesion induction and were treated with cryptotanshinone (200 mg/kg), FM19G11 (2 mg/kg), echinomycin (10 or 20 μg/kg), or PX12 (25 mg/kg) immediately after injury, 4 hours after injury, and every 24 hours for 7 days. Mice were euthanized after 7 days and adhesions were imaged and scored. H&E staining showed an intact layer of MSLN+ mesothelium (Fig. 6C). Adhesion burden was significantly (P < 0.0001) reduced (Fig. 6D) in mice treated with PX12 (n = 7) or echinomycin (n = 17), with many mice showing no observable adhesions (a score of 0), suggesting that HIF1α blockade was sufficient to prevent adhesion formation. Treatment with HIF1α inhibitors did not affect wound healing; all mice appeared healthy, and abdominal wall closure was complete after 7 days, similar to the situation observed for untreated control animals that underwent injury.

We injected EdU, as previously described, into mice that had undergone button induction and adhesion initiation and had been treated with echinomycin (20 μg/kg; n = 5). Mice were analyzed 7 days after EdU injection and were found to have reduced EdU incorporation into PDPN+MSLN+ cells (Fig. 6E) compared to untreated injured control animals pulsed with EdU, suggesting that HIF1α inhibition decreased mesothelial cell proliferation. Mesothelial cells of echinomycin-treated mice exhibited both activated (fig. S26A) and normal uninjured phenotypes (fig. S26B), suggesting that HIF1α reduced cell proliferation but not cell activation and adhesion formation.

Transcriptional changes in mesothelial cells treated with HIF1α inhibitors

To determine the transcriptional changes downstream of HIF1α, 10 mice were treated with echinomycin (20 μg/kg) before and immediately after induction of four ischemic buttons in the peritoneum of each mouse. Damaged mesothelium at 24 hours after button induction was isolated from the buttons and RNA was extracted for sequencing. Known HIF1α targets VEGFβ and transferrin were down-regulated after echinomycin treatment (fig. S27). Expression of proliferating cell nuclear antigen and the origin recognition complex family of genes was analyzed (fig. S28, A and B) (5153) to confirm that echinomycin did not inhibit DNA replication in our purified mesothelial cell subset.

RNA sequencing analysis revealed that ~200 genes up-regulated 24 hours after injury showed decreased expression after HIF1α inhibition. More than 600 genes with decreased expression 24 hours after injury were up-regulated in response to HIF1α inhibition (Fig. 6F). Some genes associated with adhesion formation were unaffected (ITGAM, ITGB1, and ITGB2; fig. S26). Expression of PDPN, MSLN, and S100A4 increased after treatment with echinomycin and injury, whereas expression of UPK1B and WT1 decreased (Fig. 6F), suggesting that WT1 and UPK1B may be downstream targets of HIF1α and could play a role in adhesion formation.

Up-regulation of MSLN in human peritoneal adhesions

To determine whether the potential mechanism of adhesion formation and target gene expression observed in mice was similar in human tissue, human adhesion samples were obtained from patients requiring abdominal surgeries (n = 6). Samples were fixed, sectioned, and stained with H&E and trichrome. The human adhesions were filled with cells and showed collagen deposition (Fig. 7A). Human adhesion samples stained positive for expression of MSLN, PDPN, CD47, and S100A4 by immunofluorescence and UPK1b expression by in situ hybridization (Fig. 7B), demonstrating that many target genes expressed in mouse adhesions were similarly expressed in human adhesions.

Fig. 7 Gene expression in human peritoneal adhesions.

(A) H&E staining (left) of representative human adhesion tissue from n = 6 patients undergoing surgery showing mesothelial cells present within the adhesion. Trichrome staining (right) of representative human adhesion tissue from n = 6 patients undergoing surgery showing mesothelial cells within areas of fibrosis. (B) Immunofluorescence staining and in situ hybridization for PDPN, MSLN, S100A4, WT1, CD47, and UPK1B in human adhesion tissue isolated from patients undergoing surgery (n = 6). Scale bars, 100 μm unless otherwise noted.


The cellular origins of peritoneal adhesions and early molecular events are unclear, largely because the cell of origin has not been identified. Studies have suggested that adhesions originate from stromal, submesothelial, or mesothelial cells through an MMT (14, 15). Here, we use multiple lineage-tracing approaches and propose that the mesothelium contributes to and is incorporated into adhesions because of its proliferation outward into the peritoneal cavity. We show that preventive or postadhesion treatment targeting these mesothelial cells results in reduction of adhesion severity in mice.

Despite our finding that the mesothelium is a contributor to peritoneal adhesions, it is likely that other cell types play a role. Although CFSE stained only surface mesothelial cells, there was marked heterogeneity within this surface cell population that was difficult to elucidate using only immunostaining. Future studies will need to focus on specifically labeled subsets of these cell populations within the surface mesothelium to carefully trace individual cellular contributions to adhesions. At the same time, the scope of our study was limited only to the mesothelium. Our lineage tracing and subsequent FACS-based RNA sequencing analyses were designed to gate out all other cell populations that may play a role in adhesion formation. Therefore, our study cannot exclude the fact that other cells such as submesothelial fibroblasts may also contribute to adhesions. Further studies need to be done to specifically label these cells genetically or chemically and then to trace these cells after adhesion induction surgery to document their contributions to adhesion formation.

We propose a model of early adhesion formation (fig. S29) where, after injury, the peritoneal mesothelium responds by induction of genes responsible for cell proliferation and differentiation. This enables separation of the mesothelium from its basement membrane, likely through the down-regulation of collagens and other ECM proteins, and enables the mesothelium to move into the peritoneal space. For simplicity, our model omits interactions between mesothelial cells and other cell types (fibroblasts and hematopoietic cells) that likely also contribute to adhesions. Specifically, many fibroblasts often share similar mesothelial markers when activated (32, 33). It is also likely that activated mesothelium recruits leukocytes, as evidenced by an increase in chemokines and cytokines (Fig. 4E), and initiates coagulation events. The interplay between inflammatory and mesothelial cells will be important to elucidate in future studies.

The RNA sequencing studies presented here reveal marked early gene expression changes occurring within the first 24 hours after adhesion formation. Injured mesothelium up-regulated expression of fibroblast genes S100A4, CDH2, and ACTA2, which, together with in vivo and in vitro staining, suggest that the mesothelial cells become fibroblast-like cells as previously described (15) but migrate into the peritoneal cavity rather than into the abdominal organs (21). Unexpectedly, our results show TGFβ down-regulation in PDPN+MSLN+ cells, suggesting that TGFβ plays a role later in adhesion formation or is expressed by other cell types.

The MSLN+ cells in the postinjury surface mesothelium involved in adhesion formation were heterogeneous by several gene expression criteria. It is plausible that there was a further adhesion-prone subset of cells within the MSLN+PDPN+ mesothelium. Whether different mesothelial cell subsets share common embryonic origins or represent separate mesothelial cell types, as recently revealed in dermal mesenchyme (22), remains to be studied.

The mesothelium likely senses hypoxia and responds through HIF1α up-regulation. Disruption of the HIF1α pathway by small-molecule inhibitors was sufficient to induce a decrease in adhesion formation in mice, implicating a role for HIF1α in the pathogenesis of adhesion formation. Our findings implicate WT1 as a downstream target of HIF1α and a likely candidate in further regulating adhesion formation.

Many mesothelial genes were highly up-regulated during peritoneal injury in mice and had low or no expression in most other tissues, making them potential targets for treatments to reverse adhesions. We identified adhesion-specific targets and then showed that an antibody against these targets could be used to treat adhesions in a mouse model. Furthermore, many malignancies, such as mesotheliomas, pancreatic ductal adenocarcinoma, ovarian cancer, and some lung cancers, express MSLN (5457). Our finding that targeting MSLN enhanced phagocytosis of MSLN-positive mesothelium further suggests that certain cancers may be susceptible to a combined anti-MSLN/anti-CD47 immunotherapy, although this remains to be tested.


Study design

The objective of this study was to determine whether the mesothelium contributed to the formation of adhesion tissue in a mouse model. Further objectives assessed whether small-molecule inhibitors of HIF1α could prevent adhesion formation and whether antibodies against two markers expressed by mesothelial cells, MSLN and CD47, could treat preformed adhesions in mice. In all treatment studies, each experiment used five mice that received treatment and five untreated control mice. All in vivo experiments were repeated at least twice; many were repeated up to six times to ensure statistical power and adequate sample sizes (see individual experiments for n). Mice were excluded only due to morbidity within 5 days of adhesion induction. Blinding was not performed. Early endpoints were taken if mice were found to be morbid. All outliers were reported and included in statistical analyses.

Adhesion induction

All animal experiments were carried out in strict accordance with the guidelines set forth by the Association for Assessment and Accreditation of Laboratory Animal Care International and Stanford University’s Administrative Panel on Laboratory Animal Care (protocol number 10266) in the United States or the European Animal Welfare Act, Directive 2010/63/EU. The protocol was approved by the Committee on the Ethics of Animal Experiments of the Government of Bavaria, Germany, and received permission no. 55.2-1-54-2532-150-2015. All surgery was performed under anesthesia, and all efforts were made to minimize suffering. Adhesion induction surgeries were done on wild-type B6 (C57BL/6J, the Jackson Laboratory) mice aged 6 to 10 weeks. Mice were anesthetized by inhaled isoflurane until they were unconscious as confirmed by toe-pinch test. The abdomen was disinfected with betadine and PBS. A left mid-clavicular incision was made in the skin running down the length of the mouse. A similar left mid-clavicular incision was made in the peritoneum running down the length of the peritoneum. The peritoneum was gently folded to the right and held down by a hemostat. A single, ischemic button was placed on the right half of the peritoneal wall by clamping a small (~5 mm diameter) piece of peritoneum with a hemostat and ligating the base with a 4-0 silk suture (683G, Ethicon) twice before the clamp was released. Light abrasion on the button (20 times) and on the adjacent liver, cecum, and small and large bowels (7 times) was optionally performed (depending on the desired adhesion severity) with a surgical brush. Light brushing with fewer repetitions was performed to avoid pinpoint bleeding. The peritoneum was closed using 4-0 silk sutures, and the skin was stapled closed (EZ Clips, 9 mm; Braintree Scientific Inc.). Mice were allowed to recover on a heating pad and injected with buprenorphine (0.05 to 0.1 mg/kg). Mice were followed closely and monitored daily for signs of morbidity for 7 days until euthanasia. Adhesed tissues were dissected, scored, and fixed in 2% paraformaldehyde (PFA) overnight at 4°C.

Adhesion scoring

We assigned a single score for an adhesion, taking into consideration both surface area contact and molecular makeup as described in the text. In our scheme, an adhesion with a score of 0 has no adhesion between two areas, with limited mesothelial thickening on the button (fig. S18A). The area stains positive for MSLN and fibronectin (fig. S18A). Mice with an adhesion score of 0 had no signs of morbidity and survived the 7-day recovery period (n = 25).

An adhesion score of 1 indicated a “string” adhesion, connecting the two adhesed areas with a light fibrous bridge (fig. S18B). The string and surrounding areas were immunopositive for MSLN and fibronectin (fig. S18B). Most mice with an adhesion score of 1 had no signs of morbidity and survived the 7-day recovery period (n = 11).

All adhesions with scores of 2 or above involve a single direct contact between two tissues. The adhesion contact itself was light and usually involved contact between the peritoneum and an abdominal organ (fig. S18C). The adhesion between the two adhesed areas stained positive for MSLN, fibronectin, and F4/80, suggesting that some macrophage infiltration has occurred (fig. S18C). Most mice with an adhesion score of 2 had no signs of morbidity and survived the 7-day recovery period (n = 12).

An adhesion score of 3 was characterized by direct, continuous contact of three areas, usually between the peritoneum and two abdominal organs or two distinct, noncontinuous areas of a single organ (fig. S18D). The adhesed area stained positive for MSLN, fibronectin, F4/80, pan collagen, and CD31 (fig. S18D). Most mice with an adhesion score of 3 had little signs of morbidity and survived the 7-day recovery period (n = 9).

An adhesion score of 4 included direct and continuous contact of four or more areas, usually between the peritoneum and three abdominal organs, or multiple separate areas of one or two organs (fig. S18E). The adhesed area stained positive for MSLN, fibronectin, F4/80, pan collagen, and CD31 (fig. S18E). Most mice with an adhesion score of 4 showed little signs of morbidity and survived the 7-day recovery period (n = 14).

An adhesion score of 5 was characterized as full compaction/encapsulation of the abdominal organs. Most organs were adhesed to the peritoneum as well as to each other as a single, rigid mass (fig. S18F). The adhesed areas stained positive for MSLN, fibronectin, F4/80, collagen, and CD31 (fig. S18F). Most mice with an adhesion score of 5 were visibly morbid with low survival rates at 7-day postinjury (n = 5).

The amount of surface area contact does not always dictate molecular phenotype and vice versa, as the severity of the adhesion likely exists on a continuum represented by our discrete criteria described above. We have observed adhesion areas with high surface area contact (a score of 3 or 4) with little or no collagen or macrophage involvement (F4/80). Conversely, we have also observed adhesion areas with low surface area contact (a score of 2) with high collagen and macrophage infiltration. In these cases, we scored adhesions based on surface area contact or number of organs involved, as we predict that this is a more significant indicator of clinical outcome.


Parabiosis surgeries were done on age-matched (4 to 6 weeks old) female wild-type B6 (C57BL/6J, the Jackson Laboratory) and C57BL/Ka Rosa26 mRFP1 mice. Mice that would undergo parabiosis were housed together in a single cage for 10 days before surgery. Mice were anesthetized by inhaled isoflurane until they were unconscious as confirmed by toe-pinch test. The sides of the mice were shaved and cleaned with 70% ethanol and betadine. Mice were laid next to each other, and incisions from elbow to knee were made on adjacent sides. The elbow and knee joints were ligated using 4-0 sutures (Ethicon), and the loose skin from adjacent mice was stapled together. Mice were allowed to recover on a heating pad and injected with buprenorphine (0.05 to 0.1 mg/kg). Mice were followed closely and monitored daily for signs of morbidity for 14 days. Staples were removed after 14 days, and mice were bled retro-orbitally to assay for chimerism.


Tissues were fixed in 2% PFA overnight at 4°C and were embedded and frozen in optimal cutting temperature compound (OCT; Sakura) or embedded in paraffin. Frozen sections were cut at 10 to 12 μm throughout the adhesed organs and saved for immunofluorescence. Paraffin sections were cut at 5 μm, and H&E and Masson’s trichome stains were performed via standard protocols.


Immunofluorescence studies were performed on frozen sections. Frozen sections were thawed at room temperature for 10 min and washed twice in PBS. Slides were blocked in 5% serum for 30 min at room temperature. Sections were subsequently stained with primary antibodies against PDPN (1:100; hamster monoclonal, Abcam), MSLN (1:200; rabbit monoclonal, Abbiotec), fibronectin (1:100; rabbit monoclonal, Abcam), F4/80 (1:100; rat monoclonal, Abcam), CD31 (1:100; rabbit monoclonal, Abcam), pan-collagen (1:100; rabbit polyclonal, Abcam), WT1 (1:100; rabbit polyclonal, Abcam), HIF1α (1:100; mouse monoclonal, Abcam), S100A4 (1:100; rabbit polyclonal, Abcam), and CK19 (1:100; rabbit monoclonal, Abcam, EP1580Y and 1:100; rabbit polyclonal, Abcam) overnight at 4°C, and washed three times in PBS. Slides were stained and were incubated with secondary antibodies conjugated to Alexa Fluor 488, Alexa Fluor 594, or Alexa Fluor 647 for 1 to 2 hours at room temperature. Stains were washed once with phosphate-buffered saline with Tween-20 (PBST) and three times with PBS before nuclear staining with Hoechst 33342 (Life Technologies) for 2 min and mounted with Fluoromount G (Southern Biotech).

In situ hybridization

Tissues were fixed in 10% (v/v) neutral-buffered formalin at room temperature for 24 to 32 hours, dehydrated, and embedded in paraffin. Tissue sections cut at a thickness of 5 μm were processed for RNA in situ detection using the RNAscope 2.5 HD Assay–RED according to the manufacturer’s instructions (Advanced Cell Diagnostics). RNAscope probes used were Mm-Upk1b (NM_178924.4, 46-966) and Hs-UPK1B (NM_006952.3).

EdU pulse chase

After adhesion induction, mice were immediately injected subcutaneously with 0.025 mg of EdU (Life Technologies) in 90% PBS and 10% ethanol. Mice were traced for 7 days and euthanized. Adhesed tissues were dissected and fixed with 2% PFA overnight, frozen in OCT (Sakura), and sectioned at 12 μm. EdU-positive cells were visualized with the Click-iT EdU Imaging Kit (Life Technologies).

Lineage tracing

Adhesion induction surgeries were performed on wild-type mice as previously described. Six hundred to 1000 μl of CFSE (Thermo Fisher Scientific) was pipetted on top of the button and into the peritoneal cavity. Mice were closed and allowed to recover for 4 to 7 days.

Clonal analysis

Adhesion induction surgeries were performed on ActinCreER; R26VT2/GK3 mice as previously described (34). 4-Hydroxy tamoxifen (1 mg/kg) was added on top of the button. Mice were closed and allowed to recover for 7 days. In vitro clonal analyses were performed by explanting mesothelium from the renal capsule of ActinCreER; R26VT2/GK3 mice and culturing them as previously described (21). Explants were treated with 8 mM 4-hydroxyl tamoxifen for 14 days and imaged for clonality.

Flow cytometry

A modified adhesion induction surgery was performed according to previously described procedures on wild-type C57BL/6J (the Jackson Laboratory). To obtain adequate amounts of mesothelial cells, four buttons were established per mouse (two on each side of the peritoneal wall). Twenty buttons were taken for a single triplicate (n = 5 mice per triplicate), and three cohorts of littermate mice were euthanized at 6, 12, or 24 hours after injury. No abrasion of the button or abdominal organs was done. Mice were allowed to recover for 6, 12, or 24 hours and euthanized. Ischemic buttons were obtained by cutting the base and placed in dissociation medium [Dulbecco’s modified Eagle’s medium (DMEM; 10565-042, Life Technologies), collagenase IV (50 mg/ml; Worthington Biochemical), and 20 μM CaCl2] and dissected further using a single edge razor and incubated in dissociation media for 30 min at 37°C. The subsequent cell suspension was filtered through a 100-μm filter and spun and washed with 2% fetal bovine serum (FBS) in PBS. Cells were treated with 1 ml of ACK (ammonium-chloride-potassium) lysis buffer (Life Technologies) for 5 min at 4°C and spun and washed. Cells were blocked with 1% goat serum (Life Technologies) for 10 min and stained with anti-PDPN (1:100; 8.1.1, BioLegend), anti–LYVE-1 (lymphatic vessel endothelial hyaluronan receptor 1; 1:100; ALY-7, eBioscience), anti-CD31 (1:100; 390, eBioscience), and anti-CD45 (1:100; 30-F11, BioLegend) for 30 min at 4°C. Cells were spun down, filtered, and resuspended in 200 μl of 2% FBS in PBS. Cells were sorted using a FACSAria (BD Biosciences); PDPN+LYVE1CD31CD45 cells were sorted directly into 750 μl of TRIzol LS (Life Technologies).

RNA sequencing and analysis

Total RNA from sorted mesothelial population was isolated using TRIzol (Thermo Fisher Scientific) as per the manufacturer’s recommendation and was facilitated by the addition of linear polyacrylamide (Sigma-Aldrich) as a carrier during RNA precipitation. Purified total RNA was treated with 4 U of RQ1 RNase-free deoxyribonuclease (DNase; Promega) at 37°C for 1 hour to remove trace amounts of genomic DNA. The DNase-treated total RNA was cleaned up using an RNeasy micro kit (Qiagen). Total RNA (10 to 50 ng) was used as input for complementary DNA (cDNA) preparation and amplification using Ovation RNA-Seq System V2 (NuGEN). Amplified cDNA was sheared using Covaris S2 (Covaris) using the following settings: total volume, 120 ml; duty cycle, 10%; intensity, 5; cycle/burst, 100; total time, 2 min. The sheared cDNA was cleaned up using Agencourt Ampure XP (Beckman Coulter) to obtain cDNA fragments ≥ 400 base pairs (bp). Five hundred nanograms of sheared and size-selected cDNA was used as input for library preparation using the NEBNext Ultra DNA Library Prep Kit for Illumina (New England BioLabs) as per the manufacturer’s recommendations. Resulting libraries (fragment distribution, 300 to 700 bp; peak, 500 to 550 bp) were sequenced using HiSeq 4000 (Illumina) to obtain 2× 150-bp paired-end reads. The reads obtained were trimmed for base call quality and the presence of adapter sequences using Skewer (58). High-quality reads thus obtained were aligned to mouse genome using OLego (59), and the quantity of expressed mRNAs was estimated using Cuffdiff2 (60) and represented as FPKM.

Gene expression intensities of each data point and results of pairwise comparison by Cuffdiff are further analyzed and visualized by Gene Expression Commons ( (43). To find similar expression pattern genes, gene expression intensities in FPKM at four time points are first standardized into z values for each gene, and then correlations to a gene of interest (e.g., HIF1A) were computed against entire genes by Pearson product-moment correlation coefficient. Genes with a P value of less than 0.05 are listed as similar expression pattern genes.

Differentially regulated genes by Cuffdiff pairwise comparison were subjected to gene set activity analysis. For each gene set defined by Gene Ontology (44), differentially regulated genes were divided into up-regulated genes and down-regulated genes on the one hand and those that are in the gene set and not in the gene set on the other hand, and then Fisher’s exact test was performed. Next, q values (estimated false discovery rate) were computed using the qvalue R package provided by the Storey lab (45). Gene set activity is presented by color scale based on odds ratio and P value. In pairwise setting, gene set with an odds ratio of more than 1 is considered as “activated.”

Antibody treatments

Adhesions were induced in wild-type B6 (C57BL/6J, the Jackson Laboratory) and allowed to recover for 7 days. Two hundred micrograms of monoclonal anti-MSLN (B35) antibody was administered via intraperitoneal injections at 7, 10, and 13 days after surgery. Two hundred micrograms of monoclonal anti-CD47 (MIAP301) (BioXCell) was coadministered via intraperitoneal injections at the same frequency. Mice were euthanized 17 days after initial surgery and scored for adhesion severity. B35 anti-MSLN antibody was a gift from A. Miyajima.

Whole-mount immunostaining and tissue clearing

Whole-mount samples were stained and cleared with a modified 3DISCO protocol (61, 62). In short, samples stored in PBS-GT [DPBS (14190094, Gibco), 0.2% Gelatin solution (G1393, Sigma-Aldrich), 0.5% Triton X-100 (X100, Sigma-Aldrich), and 0.01% thimerosal (T8784, Sigma-Aldrich)] were incubated with primary antibodies [anti-PDPN (ab11936, Abcam), anti-MSLN (250519, Abbiotec), and anti–α-SMA (ab5694, Abcam)] in PBS-GT, with shaking, for 36 hours at room temperature. Excessive antibody was removed by thorough washing in PBS-GT for 6 to 12 hours and refreshing the solution every 1 to 2 hours. Incubation with fluorophore-coupled secondary antibodies (A-21451 and A-21207, Molecular Probes) in PBS-GT for 36 hours was followed by thorough washing in PBS-GT as described above. Samples were then dehydrated in an ascending series of tetrahydrofuran (186562, Sigma-Aldrich) (50, 70, 3× 100%; 60 min each) and subsequently cleared in dichloromethane (270997, Sigma-Aldrich) for 30 min and eventually immersed in benzyl ether (108014, Sigma-Aldrich). Cleared samples were imaged in 35-mm glass bottom dishes (81218, Ibidi) using a laser scanning confocal microscope (LSM710, Zeiss).

Multiphoton microscopy

Immunostained and noncleared samples were imaged using a Leica SP8 MP (Leica Microsystems) at the Imaging Core Facility of the Biomedical Center (LMU, Martinsried). Briefly, samples were embedded in 2% agarose (Lonza, 859081) in a 35-mm dish (351008, Corning) and submersed in PBS (14190094, Gibco). A 25× water dipping objective (HC IRAPO L 25×/1.00 W) was used in combination with a tunable laser (InSight DS+ Single, Spectra-Physics), set to 1080 and 1225 nm excitation, respectively. Data from confocal and multiphoton microscopy were analyzed using Imaris 9 (Bitplane, UK). Brightness and contrast were adjusted for better visualization.


Peritoneum was imaged with a LaVision BioTech UltraMicroscope (LaVision BioTech GmbH, Germany) equipped with a 2× objective lens and additional zoom optics (0.63× to 6.3×). Antibodies against MSLN (B35) were labeled with the IRDye 800CW fluorophore (LI-COR, Lincoln, NE) according to the manufacturer’s instructions. One milligram of the antibody conjugate in PBS was injected intraperitoneally immediately after surgery and left for 30 min. Fluorescence signal was collected at 800 nm using 750 nm excitation, and autofluorescence emission was collected at 690 nm using 650 nm excitation. The sample was scanned in 4-μm steps in the z-direction. Fluorescence signal was collected at 800 nm using 750 nm excitation, and autofluorescence emission was collected at 690 nm using 650 nm excitation. Data were further processed using Imaris 8.2.1 (Bitplane, UK). Three-dimensional (3D) rendering was done using the maximum intensity projection (MIP) or blend mode. 2D representations are MIPs of 36-μm-thick sections from the 3D model.

In vitro mesothelial culture

Wild-type B6 (C57BL/6J, the Jackson laboratory) were euthanized, and the mesothelium was excised from the renal capsule and intestine. Excised mesothelium was cut into smaller fractions and placed into culture dishes pretreated with EmbryoMax 0.1% Gelatin solution (EMD Millipore) for 30 min and cultured in DMEM (Life Technologies) with 10% FBS, 1% penicillin/streptomycin, and 1% nonessential amino acids at 37°C for 7 days. Mesothelial cells were cocultured with macrophages (added to confluency) under normal or hypoxic conditions (5% O2 incubator or 100 μM CoCl2).

Preparation of primary mouse BMDMs

To prepare primary bone marrow–derived macrophages (BMDMs), BALB/c mice were humanely euthanized and disinfected with 70% ethanol. An incision was made along the legs, and the muscle was removed from the bones. The femur and tibia were removed from the body and rinsed in PBS. The bones were flushed with a 6-ml syringe and a 23-gauge needle, and the marrow was resuspended in 10 ml RPMI 1640. The suspension was centrifuged for 5 min at 1200 rpm, and the pellet was resuspended in 5 ml of ACK lysis buffer (Invitrogen) for 5 min to remove blood cells. The suspension was filtered over a 70-μm Falcon cell strainer and centrifuged again. The pellet was resuspended in 40 ml of macrophage media [RPMI 1640 + 10% FBS + 10% penicillin/streptomycin + MCSF (macrophage colony-stimulating factor) (10 ng/ml)] and plated on four 10-cm petri dishes. On day 4, the macrophage medium was replaced. On day 7, the macrophages were lifted from the dish and used.

Preparation of human macrophages

Leukocyte reduction system chambers were obtained from the Stanford Blood Center from anonymous donors, and peripheral blood mononuclear cells (PBMCs) were enriched by density gradient centrifugation over Ficoll-Paque Premium (GE Healthcare). Four hundred million PBMCs were plated per 15-cm dish (Corning), washed vigorously three times after a 30-min period of adherence, and then left to differentiate into macrophages by culture for 7 to 10 days in IMDM + GlutaMAX (Invitrogen) supplemented with 10% AB-Human Serum (Gemini), penicillin (100 U/ml), and streptomycin (100 μg/ml; Invitrogen).

Phagocytosis assay

Phagocytosis assays were performed by coculture of 50,000 macrophages with 100,000 human mesothelial cells (MeT5A; American Type Culture Collection) labeled with Calcein AM (Thermo Fisher Scientific) and antibodies for 2 hours in serum-free medium and then analyzed using a Cytoflex cell analyzer with a high-throughput sampler (Beckman Coulter). Antibodies used for treatment included human IgG4 isotype control (Eureka Therapeutics) and anti-CD47 clone 5F9. Primary human macrophages were identified by flow cytometry using anti-CD45 antibodies (BioLegend). Dead cells were excluded from the analysis by staining with DAPI (Sigma-Aldrich). Phagocytosis was evaluated as the percentage of green macrophages using FlowJo v9.4.10 (Tree Star) and was normalized over the isotype control. Statistical significance was determined by two-tailed paired Student’s t test.

Small-molecule inhibitor treatments

Adhesion induction surgeries were performed on 4- to 8-week-old wild-type B6 (C57BL/6J, the Jackson laboratory) as previously described. Mice were treated with cryptotanshinone (200 mg/kg; Sigma-Aldrich), FM19G11 (2 mg/kg; Sigma-Aldrich), echinomycin (10 or 20 μg/kg; Sigma-Aldrich), or PX12 (25 mg/kg; Sigma-Aldrich) immediately after injury, 4 hours after injury, and every 24 hours subsequently for 7 days.

Scanning electron microscopy

Conventional scanning electron microscopy. Tissue specimens (mesothelium buttons) were fixed overnight in 4% PFA with 2% glutaraldehyde in 0.1 M Na-cacodylate buffer (pH 7.3), postfixed in 1% OsO4 (1 hour), and dehydrated in a graded ethanol series (50, 70, 90, and 100%; 10 min each) before critical point drying with liquid CO2 in a Tousimis Autosamdri-815B apparatus (Tousimis, Rockville, MD), mounting on Aluminum SEM stubs (Electron Microscopy Sciences, Hatfield, PA), and sputter coating with 50 Å of gold palladium using a Denton Desk II Sputter Coater (Denton Vacuum, Moorestown, NJ). Specimens were visualized with a Zeiss Sigma FESEM (Zeiss Microscopy, Thornwood, NY) operated at 2 to 3 kV using InLens Secondary Electron (SE) and SE2 detection. Images were captured in Tagged Image File Format (TIFF) using a store resolution of 2048 pixels × 1536 pixels and a line averaging noise algorithm.

Large block-face (thin section) scanning electron microscopy to visualize internal ultrastructure. Tissue specimens (mesothelium buttons) were similarly fixed in 4% PFA with 2% glutaraldehyde in 0.1 M Na-cacodylate buffer (pH 7.3) before en bloc staining with a series of heavy metals as described by Tapia et al. and Deerinck et al. (63, 64). Typically, specimens were treated consecutively with ferrocyanide-reduced OsO4 (1% OsO4 with 1.5% tetrapotassium ferricyanide; 1 hour), freshly prepared and syringe-filtered, 1% thiocarbohydrazide (40 min), 2% OsO4 (1 hour), and 1% uranyl acetate (overnight), before dehydration in a graded ethanol series (50, 70, 90, 100%; 10 min each), followed by 2× in 100% acetonitrile (10 min each). Repeated washing with H2O (3 × 5 min) was included between steps of staining and before dehydration. Tissue was then infiltrated with 25, 50, and 75% EMBed 812 embedding resin (Electron Microscopy Sciences, Hatfield, PA) in acetonitrile, followed by 100% resin (2 × 3 hours) and finally embedding in pure resin with polymerization for 48 hours at 60°C. Ultrathin sections (200 nm each) were collected on conductive silicon wafer substrates and visualized with a Zeiss Sigma FESEM (Zeiss Microscopy, Thornwood, NY) operated at 5 to 7 kV using Backscattered Electron detection. Signal inversion resulted in micrographs with gray levels similar to transmission electron microscopy images (65). Images were captured in TIFF using a store resolution of 2048 pixels × 1536 pixels and a line averaging noise algorithm.

Statistical analysis

All statistical analyses were performed using GraphPad Prism 6. All results were expressed as means ± SEM. Unpaired t test was used to compare two groups. P < 0.05 was considered significant.


Fig. S1. The surface mesothelium remains after gentle abrasion injury.

Fig. S2. The surface mesothelium remains on buttons after placement.

Fig. S3. PDPN and MSLN are exclusive to surface mesothelium.

Fig. S4. The surface mesothelium buttons are MSLN+ after placement.

Fig. S5. The surface mesothelium thickens in response to injury.

Fig. S6. The surface mesothelium thickens in response to injury.

Fig. S7. Chimerism in parabiotic mice.

Fig. S8. Circulating cells do not contribute significantly to adhesions.

Fig. S9. EdU+ cells are found in areas of tissue thickening.

Fig. S10. Scanning electron microscopy of peritoneal buttons.

Fig. S11. Scanning electron microscopy of peritoneal buttons.

Fig. S12. Large block-face scanning electron microscopy of peritoneal buttons.

Fig. S13. CFSE treatment labels only surface mesothelial cells.

Fig. S14. CFSE-labeled mesothelial cells contribute to adhesions.

Fig. S15. PDPN+ ActinCreER; R26VT2/GK3 cells give rise to adhesions.

Fig. S16. In vitro mesothelial transition.

Fig. S17. Anti-MSLN antibody specifically binds the surface mesothelium.

Fig. S18. Antibodies against host IgG do not bind mesothelial cells.

Fig. S19. Adhesion scoring characterized by surface area contact and molecular markers.

Fig. S20. K19+PDPN+ mesothelium persists after antibody treatment.

Fig. S21. Mesothelium persists after antibody treatment.

Fig. S22. ECM genes are down-regulated after adhesion induction.

Fig. S23. Validation of enriched genes via immunofluorescence.

Fig. S24. Transcript levels of HIF1A stabilization proteins.

Fig. S25. Mesothelial in vitro culture assay.

Fig. S26. HIF1A and mesothelial gene targets after HIF1α inhibition.

Fig. S27. HIF1A and mesothelial gene targets after HIF1α inhibition.

Fig. S28. Echinomycin does not affect DNA replication.

Fig. S29. A proposed mesothelial centric model of adhesion formation.

Date file S1. Source data.

Movie S1. 3D rendering of light sheet microscopy imaging of injured surface mesothelium.

Movie S2. 3D rendering of light sheet microscopy imaging of injured surface mesothelium.