Excess shed mesothelin disrupts pancreatic cancer cell clustering to impair peritoneal colonization

Abstract

Peritoneum is the second most common site of metastasis in patients with pancreatic ductal adenocarcinoma (PDAC). Peritoneal colonization is impaired in PDAC cells with knockout (KO) of the cancer surface antigen mesothelin (MSLN) or by introducing Y318A mutation in MSLN to prevent binding to mucin‐16 (MUC‐16). MSLN has a membrane‐bound form but is also shed to release soluble MSLN (sMSLN). Their individual roles in peritoneal metastasis are unknown. Here, a C‐terminal truncated MSLN mutant (∆591) incapable of cell membrane insertion but proficient in secretion was engineered. Expression of ∆591 MSLN failed to rescue peritoneal metastasis in MSLN KO cells and inhibited peritoneal colonization when overexpressed in WT PDAC cells. Exposing PDAC cells to conditioned medium (CM) containing excess sMSLN impaired cancer cell clustering in vitro and in peritoneal fluid in vivo, while CM containing only Y318A sMSLN did not. These data demonstrate that interaction of membrane‐bound MSLN with MUC‐16 promotes cell clustering that is critical for efficient peritoneal metastasis. However, peritoneal colonization by MSLN KO cells was rescued by expression of ∆591 mutant MSLN bearing Y318A mutation, suggesting that sMSLN also has a MUC‐16‐independent role in peritoneal spread. Alterations in inflammatory signaling pathways occurred following KO cell exposure to CM containing sMSLN, and CM from cancer cells with intact peritoneal metastasis provoked increased KO cell secretion of IL‐1α. While excess sMSLN inhibits cell clustering and peritoneal colonization, sMSLN may also promote PDAC peritoneal metastasis independent of MUC‐16.

MSLN is a PDAC cancer surface antigen that is typically membrane‐bound, but is subsequently shed through the action of cell surface proteases. PDAC cells expressing MSLN and MUC‐16 on the cell surface have enhanced homotypic cell clustering that is associated with increased peritoneal metastasis. In the presence of excess shed MSLN (sMSLN) clustering is disrupted and peritoneal metastasis is impaired. IL‐1α production is increased when cancer cells are stimulated with endogenous levels of sMSLN or a mutant sMSLN product that lacks ability to bind MUC‐16. Increased IL‐1α production is also associated with increased peritoneal metastasis.

Abbreviations

CM

conditioned medium

FDR

false discovery rate

GLPD1

GPI‐specific phospholipase D1

GPI

glycosylphosphatidylinositol

GSEA

gene set enrichment analysis

IP

intraperitoneal

KO

knockout

MPF

megakaryocyte potentiating factor

MSLN

mesothelin

MUC‐16

MUCIN‐16

NES

normalized enrichment score

ns

not significant

PCA

principal component analysis

PDAC

pancreatic ductal adenocarcinoma

sMSLN

soluble MSLN

TME

tumor microenvironment

WT

wild type

1. INTRODUCTION

Pancreatic ductal adenocarcinoma (PDAC) is a deadly disease that kills 87% of patients within 5 years of diagnosis. ¹ Approximately 80% of patients already have detectable metastases when their disease is discovered. PDAC most commonly metastasizes to the liver; however, 50% of patients have peritoneal metastasis at the time of death making peritoneal disease the second most common site of PDAC metastasis. The biology of peritoneal metastasis in PDAC has been understudied in part because the most utilized pre‐clinical models of PDAC rarely develop peritoneal disease. Most information about peritoneal metastases in PDAC is extrapolated from research on gastric and ovarian cancers. ²

Mesothelin (MSLN) is a cell surface glycoprotein normally expressed on peritoneal, pleural, and pericardial surfaces but also on the surface of many cancer cells, including ~85% of human PDAC. ³ High expression of MSLN is correlated with earlier cancer‐specific mortality in patients who undergo tumor resection. ⁴ , ⁵ Our pre‐clinical studies using murine PDAC models have demonstrated that deletion of MSLN from cancer cells impairs peritoneal colonization without affecting cell growth in culture or subcutaneous tumor growth in mice. Cancer cells lacking MSLN have impaired angiogenesis at nascent sites of peritoneal metastasis due to a paracrine signaling mechanism between cancer cells and macrophages. ⁶ , ⁷ Moreover, MSLN alters the polarization of macrophages to allow for a more tumor‐permissive tumor microenvironment (TME). ⁷

The MSLN gene encodes a 71‐kDa precursor protein that is cleaved intracellularly to release a 31‐kDa N‐terminal fragment known as megakaryocyte‐potentiating factor (MPF). The remaining 40 kDa mature MSLN contains a glycosylphosphatidylinositol (GPI) linkage site near the C‐terminus which allows for cell surface attachment. Mature membrane‐bound MSLN is cleaved from the cell surface by extracellular proteases, including members of the ADAM, MMP, and BASE families, resulting in release of soluble MSLN (sMSLN) with varying truncations of the C‐terminus. ⁸ Shed sMSLN can be detected in the intratumoral fluid within the TME ⁹ and in patient malignant ascites. ¹⁰ In patients with mesothelioma and ovarian cancer, sMSLN subsequently migrates from the TME into the blood so efficiently that it can serve as a serum tumor biomarker to monitor progression and therapeutic regression of disease. ¹¹ However, in PDAC, transit from the extracellular fluid to the circulation appears to be impaired. ⁹

MSLN has no intracellular domain and is thought to primarily function through interaction with Mucin‐16 (MUC‐16), ¹² its only known binding partner in PDAC. MUC‐16 is a heavily glycosylated megaprotein (~4000 kDa) which associates with MSLN through N‐glycans in the Ser/Thr/Pro‐rich tandem repeat region although a specific binding site has not been identified. ¹³ Tyrosine 318 (Y318) is a crucial residue within the MUC‐16‐interacting regions of MSLN. ¹⁴ Point mutation to Y318A ablates MSLN‐MUC‐16 binding, and complementation of MSLN knockout (KO) PDAC cells with Y318A mutant MSLN failed to restore efficient peritoneal colonization even though complementation with wild‐type (WT) MSLN did. ⁶ Expression of MUC‐16 on the surface of ovarian cancer cell lines has been shown to increase heterotypic cell–cell interactions with peritoneal cell culture monolayer and with ex vivo peritoneum. ¹² , ¹⁵ Membrane‐bound MSLN on serosal surfaces is thought to play a role in this process. However, it is currently unclear whether other products of the MSLN gene (precursor, sMSLN or MPF) made by cancer cells may also play a role in enhancing peritoneal metastasis. In addition, in ovarian cancer cell line models, MSLN KO reduced proliferation in suspension, increased anoikis, and impaired peritoneal dissemination, and did so even in cell lines lacking detectable MUC‐16, raising the question of whether some pro‐tumorigenic activities of MSLN may occur independent of MUC‐16. ¹⁶ The answers to these questions may have important therapeutic implications given that MSLN is highly targetable and existing MSLN‐directed therapeutics have not been tested for their ability to disrupt pro‐tumorigenic effects of MSLN. Here, we examine the role of various MSLN products on peritoneal metastasis and identify membrane‐bound MSLN and sMSLN as important mediators of peritoneal spread.

2. MATERIALS AND METHODS

2.1. Cell culture

Human pancreatic cancer cell lines KLM1 and T3M4 were the gifts of Udo Rudloff and Mitchell Ho (NCI, Bethesda, MD), respectively. Identity of the cell lines was confirmed by STR testing. Cells were cultured in RPMI 1640 medium (Gibco, Thermo Scientific) supplemented with l‐Glutamine (2 mmol/L), penicillin (100 U), streptomycin (100 μg), and 10% FBS (Hyclone, Thermo Scientific). All cells were maintained in a 5% CO₂ adjusted incubator at 37°C and tested free of mycoplasma. Synthesis of MSLN KO KLM1 and T3M4 cell lines and engineering of KLM1 KO cells over‐expressing full‐length WT and Y318A mutant MSLN were described previously. ⁶ , ⁹ Characterization of T3M4 parent and KO cell line for expression of MSLN and MPF was described previously. ⁹

2.2. Engineering of +MPFf, +MSLNf, and ∆591 MSLN cell lines

MPF and MSLN sequence fragments were cloned into the 19 416‐M06‐369 Empty Puro Vector by CCR Protein and Nucleic Acids Production Center (Frederick, MD). KLM1 KO cells were stably nucleofected with MPFf or MSLNf expression constructs or empty vector using a Nucleofector 2b device (Amaxa, Lonza). Cells were selected in 0.5 μg/mL puromycin for 2 weeks, then screened for MSLN expression by flow cytometry and ELISA of CM. Lentiviral expression plasmids for Δ591 MSLN mutants were produced by GenScript USA Inc. and lentiviruses were synthesized and packaged by NINDS Viral Production Core Facility (Bethesda, MD). KLM1 KO cells were transduced as described previously, ⁶ and single cell clones were isolated under puromycin selection following serial dilution. MSLN expression in the clones was assessed by ELISA of CM. Sequence files for vector and MSLN inserts are included in the Supporting Data.

2.3. Flow cytometry

Cells were harvested by trypsinization then resuspended in PBS + 3% FBS. For assessment of MSLN expression, cells were incubated with mouse anti‐human MSLN MN‐1 (BioXcell: 612916A2) for 30 min on ice. For assessment of MUC‐16 expression, the primary antibody was purified anti‐CA125 (MUC16) Purified anti‐CA125 Antibody (MUC16) (Biolegend: 666904). For both, this was followed by incubation in PE‐conjugated goat anti‐mouse IgG F(ab′)2 Fragment Goat Anti‐Mouse IgG (H+L) (Jackson Immunoresearch: 115‐116‐146) staining for 30 min. Alternatively, Alexa647‐conjugated MN (labeled by Biolegend; Cat.: 98713; Clone: MN; Lot #: B298023; 0.5 mg/mL) was used for single‐step labeling for MSLN. Following labeling, cells were re‐suspended in FACS buffer (3% FBS in PBS) prior to analysis on a FACS Sony SA3800 or Canto II.

2.4. ELISA assays

All serum/media concentrations were measured using Quantikine ELISA Kits (R&D Systems, for MSLN #DMSLN0, IL‐1α #DLA50, LIF #DLF00B). For assessment of MPF, streptavidin assay plates (Meso‐Scale Diagnostics, MSD, Rockville, MD) were incubated with 25 μL biotinylated MPF capture antibody (MPF49 Ab) solution for 1 h. The plates were then washed before 50 μL of recombinant MPF calibrators, 10‐fold diluted samples, or reference samples were added. Samples were incubated 1 h, washed, and then 25 μL of Sulfo‐Tag MPF detection antibody (MPF25 Ab) solution was added. After another 1 h incubation, 2× read buffer was added to the plates and signal was detected using a QuickPlex instrument (Meso‐Scale Diagnostics) within 15 min. These data were analyzed using WorkBench 4.0 software (Meso‐Scale Diagnostics), and concentration of the analyte was extrapolated from the calibrator standard curves.

2.5. Cell growth assays in adherent culture

Cells were plated at 1 or 5 × 10⁵ cells/well in 6‐well plates (Corning). At appropriate time, cells were detached with 0.05% Trypsin–EDTA (ThermoFisher: 253001200), resuspended, and counted using Cellometer Auto T4 (Nexcelom Bioscience).

2.6. Colony formation in soft agar

1 × 10⁴ cells were suspended in 0.3% LMP (low melting point) agarose (Invitrogen) in RPMI supplemented with 20% FBS and then seeded on top of 0.6% agarose in 6‐well plates in triplicates. Complete medium (500 μL) was added on top of the agarose layer once a week. After 28 days in culture, the colonies were stained using 0.2% crystal violet. The mean number of colonies were counted with ImageJ in 10 non‐overlapping fields per well.

2.7. Immunoblots

Cells were detached with 0.05% Trypsin–EDTA (ThermoFisher: 253001200), washed in PBS, and lysed in RIPA buffer (ThermoFisher: 89901) containing protease/ phosphatase inhibitors (Thermo Scientific: 78444). Protein concentrations were determined by BCA assay (Pierce: 23227). Proteins were separated by SDS‐PAGE, transferred to PVDF membrane, and blocked in 5% milk solution. The following primary and secondary antibodies were used for MSLN: Mouse‐human MSLN (BioXCell) and Goat Anti‐Mouse IgG (H+L)‐HRP Conjugate (Bio‐Rad, #1706516); for GAPDH: GAPDH (Cell Signaling, Cat#:5147) and Goat Anti‐Rabbit IgG (H+L)‐HRP Conjugate (Bio‐Rad; #1706515).

2.8. Mouse experiments

All animal experiments were performed in accordance with NIH guidelines and approved by the NCI Animal Care and Use Committee. Female 6‐ to 8‐week‐old athymic nude mice (Charles River, Frederick, MD) were inoculated into the peritoneum (IP) with 1 × 10⁶ cells in RPMI 1640 (Gibco, MD) with no additives. Mice were euthanized after ~6 weeks, all visible tumor within the abdominal cavity was dissected out and weighed to assess tumor burden. Blood was obtained through submandibular puncture into Serum Gel Z/1.1 micro tubes (SARSTEDT), then centrifuged at 10 000 g for 5 min to separate serum.

2.9. Cell clustering assays

For in vitro assays, cells (7.5 × 10⁵ in 6‐well, or 1.5 × 10⁵ in 12‐well format) were tagged with CellTracker™ Green CMFDA Dye (ThermoFisher) then dispersed into ultra‐low attachment plates (Corning) for 24 h. For in vivo cell clustering imaging, cells were inoculated into the mouse abdomen as described above, then 4 h post‐injection, the mice were euthanized and the abdomen was lavaged with 2 mL of RPMI 1640. Peritoneal lavage fluid containing the labeled cells was transferred into ultra‐low attachment dishes (Corning) and fluorescent cells were immediately imaged (in vitro: eight fields/6‐well size, or four fields/for 12‐well size; in vivo: three to five fields) using ZOE Fluorescent Cell Imager (Bio‐Rad).

2.10. Conditioned media (CM) generation and treatment

KLM1 parent and derivative cells were plated at equal density, then old medium was replaced with equivolume of fresh RPMI after 48 h. CM was collected 24 h later and stored in aliquots at −80°C for later use. KLM1 or T3M4 KO cells were grown for 48 h after plating, before old growth medium was replaced with designated CM and treated for 4 h.

2.11. Scratch assay to assess cell motility

Cells were cultured for at least two passages, then seeded (1.5 × 10⁴–6.0 × 10⁴) onto a 96‐well plate in at least triplicate wells in complete growth medium. Once the cells reached 100% confluency, cells were washed in RPMI1640, then fresh RPMI1640 was added. Scratches in the monolayer were made with Sarturios Wound Maker, then the wells were washed twice with 1X PBS. RPMI1640 (100 μL) was added to each well and the plate was loaded onto Incucyte Live‐Cell Analysis System. Images were captured every 6 h from 0 to 36 h. The wound width, wound confluence and relative wound density were measured by Incucyte® Scratch Wound Analysis Software Module.

2.12. RNA harvest and deep sequencing

CM‐treated cells were lysed and RNA extracted using RNeasy Kit (Qiagen). Total RNA from quintuple samples was sequenced by the Center for Cancer Research Sequencing Facility (Bethesda, MD) using an Illumina HiSeq 2500 instrument. Sequencing quality was first examined by FastQC (version 0.11.8), Preseq (version 3.1.2), Picard tools (version 2.23.7), and RSeQC (version 4.0.4). The Illumina reads in FASTQ format were then trimmed by Cutadapt (version 4.0) to remove the adaptor sequences and low‐quality reads. STAR (version 2.7.0f) two‐pass mode was used to map reads to human reference genome hg38 and RSEM (version 1.3.3) was used to quantify the gene and transcript expression levels. Differentially expressed genes were identified by DESeq2 (version 1.42.0). Gene Set Enrichment Analysis (GSEA) was performed using the GSEA software (version 4.3.0) with 1000 permutations. ¹⁷ , ¹⁸ Normalized enrichment score (NES)| ≥ 2 and false discovery rate (FDR) < 0.05 were used as the cut‐off criteria.

2.13. Statistics and schemas

GraphPad Prism 7 software and Microsoft Excel were used for all statistical analysis and graphing. Data are presented as averages with error bars representing standard deviations unless stated otherwise. One‐way ANOVA for multiple comparisons with post‐hoc Tukey test and two‐way ANOVA with Sidak's multiple comparison test were used to determine statistical significance. All experiments were confirmed by repeat. Statistical significance is indicated as follows: ns = not significant, * for p < .05, ** for p < .01, *** for p < .001. Schemas were generated using BioRender.

3. RESULTS

3.1. Effect of MPF and MSLN fragments on IP tumor growth

To determine which form(s) of MSLN contribute to acceleration of PDAC peritoneal metastasis, we repeatedly utilized the experimental paradigm outlined in Figure 1A. Specifically, we took MSLN KO human PDAC cells that were known to be deficient in peritoneal colonization and forced them to express various engineered MSLN fragments (Figure 1B) through stable transduction or transfection. KO cells and those complemented with engineered MSLN fragments were then inoculated IP into athymic nude mice to assess for the ability of each construct to rescue peritoneal growth.

FIGURE 1.

Experimental design. (A) Schema depicting experimental plan for testing pro‐tumorigenicity of protein products of MSLN gene. Red bar shows MPF protein, which is expected to be secreted. Blue bar shows MSLN protein. Lighter blue indicates residues C‐terminal to the GPI anchor which is required for membrane‐bound MSLN. Constructs missing the GPI‐anchor are expected to be secreted‐only. Darker blue residues indicate those which may contribute to sMSLN. Y318A point mutation ablates MSLN interaction with MUC‐16. (B) Schematic showing each MSLN variant complemented into KO cells.

First, we synthesized expression constructs designed to produce MPF alone or the mature MSLN protein (MPFf and MSLNf, respectively, Figure 1B). These constructs were stably transduced into KLM1 MSLN KO pancreatic cancer cells to produce KO+MPFf and KO+MSLNf cell lines. As expected, MSLNf was expressed on the cell surface while MPFf was not (Figure 2A). In culture, each of these cell lines grew at a similar rate as the control cell lines that lacked CRISPR deletion of MSLN (Mock) or KLM1 MSLN KO cells that were stably transfected with empty vector (KO+vec) (Figure 2B). The cell lines were each injected IP into athymic nude mice and tumors were allowed to grow for ~6 weeks. Mice were euthanized, all tumors were harvested from the abdominal cavity and total weight of tumor burden was determined for each mouse. As seen previously, ⁶ tumors grew robustly in mice injected with mock cells. However, tumors struggled to grow in mice injected with KO+vec, KO+MPFf, or KO+MSLNf (Figure 2C). Although complementation of MSLNf or MPFf failed to rescue tumor growth, ELISA analysis of tumor lysates showed that MSLNf and MPFf were strongly expressed in the KO+MSLNf and KO+MPFf tumors (Figure 2D,E). We considered that both MPF and MSLN together may be necessary to stimulate IP growth of tumor. To test this, we repeated the mouse experiment, but co‐injected a 1:1 ratio of +MSLNf and +MPFf cells into the mice. The combination was still unable rescue tumor growth (Figure 2F). These data suggest that neither MPF nor mature MSLN alone can mediate enhancement of IP tumor growth. Given that a single point mutation (Y318A) in mature MSLN ablated phenotype rescue by full‐length MSLN in our previous experiments, ⁶ the lack of MSLNf activity was surprising. The initial MSLNf construct included a pre‐protypsin leader sequence to provide a membrane localization signal in the absence of the native N‐terminus. This leader sequence was fused directly to the first residue of mature MSLN. We hypothesized this could interfere with binding of MSLNf to MUC‐16 or other putative binding partners important for MSLN pro‐tumorigenic activity. Therefore, we designed two additional MSLNf constructs: MSLNf#2 which lacked a leader sequence and MSLNf#3 which included a 9 amino acid flexible GGSx3 linker between the pre‐protrypsin leader sequence and the start of mature MSLN (Figure S1A). MSLNf#2 was poorly expressed upon transduction into KLM1 MSLN KO cells, while MSLNf#3 was easily detectable (Figure S1B,C). KO cells stably transduced with these two constructs were injected IP into nude mice as described above; however, neither MSLNf#2 nor MSLNf#3 rescued the enhanced IP growth phenotype in KLM1 KO tumors (Figure S1D). These data demonstrated that engineered MSLNf constructs could not mediate enhancement of IP tumor growth.

FIGURE 2.

Assessing the activity of MPF versus membrane‐bound mature MSLN. KLM1 MSLN KO cells were stably transfected with empty vector (KO+vec), MSLNf, or MPFf. (A) KLM1 derivative cells were assessed for membrane‐bound MSLN expression by flow cytometry. (B) Growth rate of the cell lines on tissue culture plastic was measured. There was no significant difference between the groups. (C–F) Cell lines were injected IP into nude mice and allowed to grow for ~6 weeks. (C) Total burden of peritoneal tumor was measured, ***p < .001. (D, E) MSLN and MPF concentration in tumor lysate was measured by ELISA assay. (F) Tumor burden of co‐injected +MSLNf and +MPFf cells was assessed, *p < .05; ns, not significant.

3.2. Complementation of KLM1 MSLN KO cells with secreted‐only MSLN variants impacts peritoneal colonization

To determine whether soluble, shed MSLN (sMSLN) could promote PDAC IP metastasis, we synthesized expression constructs with a premature C‐terminal truncation at residue 591 (∆591) that eliminated the canonical GPI anchoring site that attaches mature MSLN on the cell surface. A ∆591 construct with the WT MSLN sequence (∆591WT) and one bearing a Y318A point mutation (∆591Mu) were synthesized (Figures 1B and 3A). Y318A point‐mutation has previously been shown to disrupt binding to MUC‐16/CA125 ¹⁴ and to render the full‐length MSLN unable to restore enhanced peritoneal colonization in MSLN KO cells. ⁶ The constructs were stably transduced into KLM1 MSLN KO pancreatic cancer cells. Anti‐MSLN antibody successfully recognized the ~40 kD mature MSLN band and ~70 kD MSLN pre‐cursor upon immunoblotting of the cell lysates (Figure 3B). ELISA assay of CM from the cells showed that both ∆591WT and ∆591Mu cell line clones secreted WT or Y318A sMSLN into the culture medium (Figure 3C). As expected, no surface expression of MSLN was seen by flow cytometry, indicating that the ∆591 truncation was successful in eliminating the membrane‐bound form of MSLN (Figure S2A). Interestingly, a shallower C‐terminal truncation made within the canonical GPI anchoring sequence was unsuccessful at eliminating surface expression of MSLN (Figure S2B,C). In culture, the MSLN ∆591 cell lines grew more slowly than the parent cell line (Figure 3D). In addition, both KO+∆591WT and KO+∆591Mu cells had consistently impaired colony formation in soft agar (Figure 3E). To understand if complementation of KO cells with sMSLN‐producing ∆591 constructs could restore enhanced peritoneal colonization, the cell lines were injected IP into athymic nude mice. Immunoblot of mouse tumor lysates confirmed continued MSLN expression in the ∆591WT MSLN and the ∆591Mu MSLN tumors and that expression of ∆591WT MSLN was significantly higher than comparators (Figure 3F). Complementation of MSLN KO cells with ∆591WT MSLN failed to rescue IP tumor growth, while the KLM1 parent and MSLN KO‐positive and ‐negative controls behaved as seen previously (Figure 3G). ∆591WT MSLN expression was insufficient to rescue IP tumor growth.

FIGURE 3.

Assessing the activity of secreted‐only MSLN in KO cells. (A) Schema depicting truncation mutant (Δ591) to remove the GPI anchoring site and prevent membrane association of mature MSLN. KLM1 MSLN KO cells were stably transduced with WT and Y318A (Mu) Δ591 expression vectors and single cell clones were isolated. (B) Immunoblot of Δ591 clones to assess for MSLN expression. (C) Conditioned medium of Δ591 clones plated at equal density was assayed for MSLN concentration by ELISA. (D) Representative experiment showing growth rate of the KO+ Δ591 cell lines on tissue culture plastic as measured by counting cell number in triplicate wells. (E) Cells were suspended in soft agar and colonies were counted after ~3 weeks. Figure depicts raw data of triplicate wells (marker) in each experiment (bar). **p < .01 or ***p < .001 indicate statistically significant difference as compared to parent for at least 2 of 3 experiments (in same direction of change). (F, G) Cell lines were injected IP into nude mice and allowed to grow for ~6 weeks. (F) Tumors were lysed and MSLN expression was assayed by immunoblot. (G) Total burden of peritoneal tumor dissected from the mouse abdominal cavity, *p < .05. Each point represents one animal. Results were confirmed by repeat using other Δ591WT and Δ591Mu clones. (H) MUC‐16 surface expression was assessed by flow cytometry. Left—Representative tracing. Right—Summary of geometric means in relation to Parent over multiple experiments.

Interestingly, complementation of the MSLN KO cells with ∆591Mu MSLN did successfully rescue IP tumor growth despite disruption of the MUC‐16‐binding site (Figure 3G). It has previously been reported that MSLN participates in a reciprocal regulatory loop with the tumor suppressing micro‐RNA mIR‐198. ¹⁹ As this pathway could function independently of MUC‐16‐to‐MSLN binding and could be differentially affected by ∆591WT versus ∆591Mu MSLN, mIR‐198 expression was assessed in each cell type. No changes were observed (Figure S2D). Similarly, KO+∆591Mu cells had no motility advantage over KO or parent cell lines (Figure S2E). MUC‐16 is an important mitogen in PDAC. Interestingly, both KO+∆591Mu and KO+∆591WT expressed less MUC‐16 on the cell surface than parent or MSLN KO cells, although the trend did not reach statistical significance in the KO+∆591WT line (Figure 3H). The decreased growth, colony formation and surface MUC‐16 expression seen in KO+∆591Mu cells was difficult to reconcile with the ∆591Mu complementation rescuing IP growth phenotype in KO cells. While the data demonstrating IP tumor rescue using the ∆591Mu MSLN could result from an artifactual gain‐of‐function secondary to over‐engineering, the data also allow for an alternative hypothesis that sMSLN has a secondary pro‐tumorigenic effect unrelated to MUC‐16 binding that is obscured when WT sMSLN is overexpressed.

3.3. ∆591WT MSLN mutant disrupts pancreatic cancer cell–cell interaction and clustering

Our data show that absence of all MSLN (both membrane‐bound and sMSLN, as occurs in MSLN KO cells) slows PDAC peritoneal metastasis and that overexpression of engineered soluble‐only WT MSLN cannot rescue IP metastasis. This led us to hypothesize that interaction of membrane‐bound MSLN with MUC‐16 is the most dominant mechanism for pro‐tumorigenic activity. If so, then overexpression of ∆591WT MSLN would be expected to inhibit IP tumor growth in a model with membrane‐bound MSLN present, by competing with native full‐length MSLN for MUC‐16‐binding sites. By acting as an inhibitor of the MUC‐16/membrane‐bound MSLN interaction, ∆591WT MSLN would have a dominant‐negative effect. By contrast, ∆591Mu MSLN would not be expected to interfere due to the Y318A point mutation which renders the mutant MSLN unable to interact with MUC‐16 (Figure 4A). To test this, ∆591WT and ∆591Mu MSLN expression constructs were transduced into the KLM1 Parent cell line (which expresses native MSLN at endogenous levels) to generate Parent(P)+∆591WT and P+∆591Mu cell lines. ELISA assay of culture medium from the cells showed that both P+∆591WT and P+∆591Mu cell lines secreted ~100× the WT or Y318A sMSLN into CM as the parent cell line (Figure 4B). As expected, the new cell lines retained surface expression of MSLN, indicating that transgene expression does not change localization of native MSLN in these cells (Figure S2A). The new cell lines had similar growth in culture, colony formation in soft agar, motility, and surface MUC‐16 expression as compared to Parent and MSLN KO cells (Figures 3E,H 4C, and Figure S2E). Consistent with our hypothesis, the P+∆591WT cell line grew less tumor than the Parent line when injected IP into nude mice (p < .05), although the inhibition was not as profound as seen with uncomplemented KO cells (Figure 4D). P+∆591Mu grew tumors of similar size as the Parent cell line, as hypothesized. Immunoblot of tumor samples demonstrated continued tumor expression of MSLN in vivo (Figure 4E), and mice bearing tumors from ∆591 constructs had serum concentrations of MSLN ~10‐fold higher than that of mice inoculated with KLM1 parent cells (Figure 4F). These data show that excess WT sMSLN inhibited the tumor‐promoting activity of full‐length MSLN, although this inhibition is slight in comparison to complete deletion of MSLN. Moreover, this inhibition did not occur in the presence of Δ591Mu, indicating it was dependent upon the ability of the excess sMSLN to bind MUC‐16, rather than a MUC‐16 independent effect of sMSLN.

FIGURE 4.

Overexpression of sMSLN inhibits the pro‐tumorigenic activity of cells expressing membrane‐bound MSLN. (A) Proposed model showing how sMSLN might block cell–cell association by interfering with interactions between MUC‐16 and membrane‐bound MSLN. (B–H) KLM1 cells were stably transduced with Δ591 MSLN expression vectors and pooled cells were used for experiments. (B) Expression of MSLN in conditioned medium as detected by ELISA. (C) Representative experiment showing growth rate of the cell lines on tissue culture plastic as measured by counting cell number in triplicate wells for each time point. No significant difference in growth was observed. (D–F) Cell lines were injected IP into nude mice and allowed to grow for ~6 weeks, *p < .05; ***p < .001; ****p < .0001; ns, not significant. (D) Total burden of peritoneal tumor dissected from the mouse abdominal cavity. Each point represents one animal. (E) Tumors were lysed and MSLN expression was assayed by immunoblot. Each lane is lysate from one mouse tumor. (F) Serum MSLN expression was assayed by ELISA. (G, H) Indicated cells were tagged with CellTracker dye and visualized by fluorescence microscopy. Representative images of multiple replicates with four to eight fields imaged per replicate. (G) Labeled cells were plated at equal density onto low adherence plates for 24 h before visualization. (H) Labeled cells were injected IP into nude mice. Mice were euthanized 4 h later and peritoneal lavage fluid was collected for visualization.

Interaction of membrane‐bound MSLN with MUC‐16 is reported to enhance cell–cell interactions between cancer cells and heterotypic cell attachment to mesothelial surfaces. ¹² Cell–cell attachments are required for cells to form clusters and cell clustering increases efficiency of metastasis. ²⁰ We previously showed that both WT and MSLN KO cells fail to undergo anoikis when dispersed onto low adherence plates. ⁶ To test the clustering ability of the P+∆591 MSLN cell lines, live cells were tagged with a fluorescent dye and allowed to disperse in low‐adherence dishes for 24 h. Inhibited clustering of MSLN KO cells as compared to Parent cells was observed (Figure 4G). P+∆591WT MSLN cells exhibited reduced clustering, similar to KO cells, while P+∆591Mu MSLN cells clustered similarly to the Parent cells as predicted by our model. In addition, complementation of KO cells with ∆591 MSLN could not restore clustering in KO cells, unlike what we observed previously upon complementation of KO cells with full‐length MSLN. ⁶ These in vitro data are consistent with our hypothesis that excess sMSLN interferes with homotypic cancer cell attachment. To determine whether sMSLN can similarly disrupt cell clustering in the peritoneal cavity, each dye‐labeled cell line was injected IP into nude mice and peritoneal fluid was collected 4 h later. Fluorescent microscopy demonstrated that Parent and P+∆591Mu cells assembled into large multi‐cellular clusters while KO cells largely failed to do this. Occasional smaller clusters of P+∆591WT cells were visualized, but most fields clustered poorly, similar to KO (Figure 4H). The presence of excess WT sMSLN impaired the ability of cancer cells to cluster within the peritoneal cavity.

3.4. Excesses of native sMSLN disrupt clustering

To determine whether native sMSLN would behave similarly to our engineered ∆591 constructs, we collected CM from KLM1 Parent and MSLN KO cells as well as KO cells stably expressing full‐length WT (KO+WT) or Y318A (KO+Mu) MSLN to use as a source for sMSLN (Figure 5A). Although these four cell lines grow in culture at an equal rate, ⁶ when cells were plated at equal density, the transduced cell lines KO+WT and KO+Mu generated medium with sMSLN concentrations >10‐fold higher than parent cell line (Figure 5B), demonstrating that the transgenic promoter causes overexpression of MSLN. When these CM stocks were used in a drug‐like fashion to “treat” KLM1 and T3M4 cells growing on low adherence plates, only the +WT CM disrupted Parent cell clustering (Figure 5C). These data confirmed that native sMSLN functioned similarly to the ∆591 engineered constructs to inhibit formation of cancer cell clusters.

FIGURE 5.

Excess of sMSLN prevents cell clustering. (A) Schema showing method used to produce MSLN‐containing CM. Mock cells underwent transient transfection with Cas9 vector like MSLN KO cells, but no gRNA was included, so they retain endogenous MSLN expression (MSLN+). Transduction of KO cells with full‐length MSLN WT or Y318A expression construct produces the KO+WT and KO+Mu cell lines which overexpress MSLN (MSLN++). (B) ELISA to assess MSLN concentration of CM from equal numbers of cells plated for each type. (C) Cells were tagged with CellTracker dye then plated at equal density onto low adherence plates for 24 h before visualization by fluorescence microscopy. Representative images of multiple replicates with 4–8 fields imaged per replicate. (D) Schema for treatment of KLM1 and T3M4 KO cells with CM prior to RNA harvest. (E) PCA plots following RNA deep‐sequencing of CM‐treated KO cells. (F) Quantitation and directionality of GSEA pathway changes in KO cells treated with KO CM (negative control) as compared to MSLN‐containing CMs.

3.5. Transcriptional changes invoked by acute exposure of KO cells to CM containing sMSLN

To understand the signaling pathways activated by sMSLN that might promote peritoneal metastasis as seen in the KO+∆591Mu model, KLM1 and T3M4 MSLN KO cells were exposed to the four different CMs described above for 4 h, and RNA deep sequencing was performed on extracted RNA (Figure 5D). Principle component analysis (PCA) revealed that KO cells treated with KO CM clustered separately from the other treatment groups (Figure 5E). GSEA was used to identify pathways that were differentially expressed in KO cells following treatment with MSLN‐containing CM versus KO CM. Interestingly, almost every pathway identified to have a statistically significant change (as defined by FDR < 5%) was downregulated upon 4 h of exposure to MSLN‐containing CMs (Figure 5F). As KO cells lack membrane‐bound MSLN, pathways associated with cell clustering could not be assayed using this experiment. However, MUC‐16‐dependent effects of sMSLN could be determined by identifying unique pathways altered by treatment with KO+WT or Parent CM but not KO+Mu CM. Just one pathway met this criteria in both cell lines: Xenobiotic Metabolism (Figure 6A,B).

FIGURE 6.

Soluble MSLN exposure results in release of IL‐1α. (A, B) Venn diagrams outlining GSEA pathway changes that occur when MSLN KO cells are exposed to sMSLN‐containing CM as compared with exposure to CM from MSLN KO cells. Pathways with statistically significant differential expression are listed. Those in bold were significantly changed in both KLM1 and T3M4. (C, D) KO cells were treated with stock CM isolated from KO, Mock, +WT or +Mu cells for 0 or 4 h, **p < .01; ****p < .0001; ns, not significant. The concentration of IL‐1α (C) or LIF (D) in CM was then measured by ELISA.

To focus on effects anticipated to be MUC‐16‐independent, we examined the pathways (24 for T3M4 and 30 for KLM1, all those within the blue circle in Figure 6A,B) that were altered after treatment with +Mu CM. There were 17 pathways enriched (with NES < −2) in at least one cell line, but all 3 that were significant in both KLM1 and T3M4 were inflammation‐related (Figure S3). To identify individual genes that might be common drivers of these pathway hits, we filtered the pathway gene lists in Figure S3 to identify included genes with statistically significant change (FDR < 0.05) and presence in the gene list for 2 or more differentially expressed pathways for both KLM1 and T3M4. Sixty‐two genes were identified. The top 20 gene hits for each cell line (as sorted by greatest fold change) are reported in Figure S4. Of those common to both KLM1 and T3M4, IL1A, LIF, CXCL1, and PTGS2 (COX‐2) were considered most relevant in the context of the athymic nude mouse model that we used for the described experiments, as these proteins have known functions beyond their effects on T lymphocytes. CXCL1 was not chosen for further validation due to low expression in these cell types. IL1A and LIF were further explored due to their robust expression and previous association with poorer survival in cancer patients. ²¹ , ²² While the RNA‐Seq data detected decreased IL1A and LIF mRNA in both T3M4 and KLM1 following treatment with sMSLN‐containing CM, increased IL‐1α secretion by KO cells was measured following CM treatment, particularly with +Mu and Mock CM (Figure 6C), and no changes in LIF secretion were found (Figure 6D). In summary, CM lacking sMSLN or containing excess of sMSLN provoked less IL‐1α secretion than exposure to CMs of constructs associated with enhanced peritoneal colonization, implicating tumor‐secreted IL‐1α as a potential MUC‐16 independent mediator of peritoneal metastasis.

4. DISCUSSION

In this study, we have shown that sMSLN interferes with homotypic cell adhesion of pancreatic cancer cells, most likely by blocking the binding of membrane‐bound MSLN to MUC‐16. This disruption of cancer cell clustering is associated with decreased peritoneal colonization and metastasis. Moreover, our data suggest that sMSLN has a pro‐metastatic MUC‐16‐independent signaling function associated with cancer cell secretion of IL‐1α. Our experiments were unable to establish a role for MPF, the second protein product of MSLN gene, in pro‐metastatic activity. This is the first time that an autocrine effect of sMSLN has been examined in the setting of peritoneal metastasis.

The MSLN gene encodes a precursor protein that is heavily processed post‐translationally and produces both MPF and MSLN protein products. Here, we used molecular engineering to produce artificial MSLN constructs that separate MPF and MSLN coding regions for the purpose of isolating the independent effects of these proteins. We failed to detect pro‐tumorigenic activity of either protein in isolation or when cells secreting MPFf were co‐injected with KO+MSLNf to provide an MPF source. On the surface, this conflicts with previous work demonstrating that MPF negatively impacts survival of a mouse lung cancer model, ²³ and also numerous studies that established a pro‐tumorigenic role for MSLN in pancreatic cancer and other tumor types. In the case of MSLNf, we expect the lack of activity is due to improper post‐translational processing and offers a cautionary tale against using over‐engineered products as a sole source to generate mechanistic insights. It also leaves open the question of what was improperly processed given that MSLN protein expressed from the MSLNf construct successfully translocated to the cell membrane and had similar cell surface localization as when MSLN was expressed from the endogenous gene. Alternatively, our failed experiments with MSLNf could suggest that sMSLN is responsible for the primary pro‐tumorigenic activity of MSLN, not membrane‐bound MSLN. However, this hypothesis does not fit with our model as expression of WT sMSLN also failed to rescue activity. In the case of MPF, a role for this protein in pancreatic cancer tumorigenesis has not been established and our experiments using MPFf do not support activity of this protein in PDAC. This may be tumor‐type or cell‐line specific but could also be due to issues with post‐translational processing of the protein in the absence of the C‐terminal domains of the MSLN precursor protein. Due to the complex life cycle of MSLN and MPF, it remains a difficult problem to dissect their individual roles.

Inactivity of MPFf and MSLNf complementation suggested that proper processing of these MSLN gene products may not occur when the MSLN coding sequence is separated from the MPF coding sequence. This concern dissuaded us from using these recombinant tools in our subsequent experiments, so that we relied upon MSLN‐precursor‐derived products to assess the role of sMSLN. This caused us to use whole CM. This methodology, while more physiologic, obscures our ability to firmly attribute the activities we have observed to a direct effect of MSLN protein product(s) versus contributions of other factors differentially expressed by MSLN‐producing cell lines, and is an important limitation of our study.

Previously, we have shown that MSLN KO significantly delays peritoneal colonization by reducing angiogenesis in nascent peritoneal tumor deposits by means of a paracrine mechanism that we were unable to elucidate. ⁶ Recent studies by others have completed the paracrine signaling story, demonstrating that pancreatic cancer cell MSLN promotes macrophage secretion of VEGF which feeds back to support tumor cell growth. ⁷ In our current study, we have used MSLN KO cells as the control for poor peritoneal tumor growth and attempted to determine which form(s) of MSLN can rescue this deficiency. Our new studies demonstrate that membrane‐bound cancer cell MSLN promotes cancer cell clustering in addition to angiogenesis at attachment sites to the peritoneum. Previous studies have shown the importance of cancer cell clusters in the context of both hematogenous and peritoneal spread, and preventing cancer cell clustering appears to be a viable therapeutic target to limit metastasis. In hematogenous spread, circulating tumor cells (CTCs) can travel as individual cells or in clusters. CTC clusters are usually homotypic, but heterotypic clusters that have inclusion of tumor stromal or immune cells are also observed. ²⁴ , ²⁵ Although singlet CTCs are more common than CTC clusters, CTC clusters have a higher proliferation rate, estimated at 20‐ to 100‐fold greater metastatic potential than singlet CTCs, and are associated with poor prognosis clinically. ²⁵ , ²⁶ In breast and lung cancer models, disrupting intercellular adhesion of the cells within CTC‐clusters suppresses metastasis. ²⁶ , ²⁷ Similar studies of clustering in hematogenous spread have not been performed in pancreatic cancer. However, recent data using pancreatic (and other) cancer cell lines clearly demonstrated that peritoneal lavage fluid or cleared ascitic fluid promote cancer cell clustering that leads to peritoneal metastatic deposits. ²⁰ Our work clearly implicates MSLN in the formation of peritoneal cancer cell clusters, raising the likelihood that MSLN additionally facilitates peritoneal colonization through this mechanism. Use of excess sMSLN to compete with membrane‐bound MSLN for MUC‐16 binding sites inhibited cell clustering and peritoneal colonization, while an excess of a mutant sMSLN incapable of binding MUC‐16 had no effect. These data provide strong circumstantial evidence that the enhancement of cell clustering and peritoneal metastasis associated with membrane‐bound MSLN expression is MUC‐16‐dependent.

In our modeling, we have used two different means to generate sMSLN: overexpression of a C‐terminal truncated MSLN incapable of GPI‐linkage to the membrane and addition of CM from pancreatic cancer cells that secrete varying amounts of MSLN, along with MPF. These overexpression models are convenient tools for elucidating the mechanism of sMSLN action. Experiments using either method of generating sMSLN resulted in the same conclusion, namely, that sMSLN is likely a competitor for MUC‐16‐binding sites and can impair homotypic cell adhesion critical for the most efficient peritoneal colonization. Importantly, we did not use purified recombinant mature MSLN in our assays out of concern that MSLN generated in an artificial system might be post‐translationally processed differently than that generated by cancer cells; recombinant MSLN could produce an sMSLN product with different activity than native. It is now well understood that shedding of membrane‐bound MSLN from the cancer cell surface occurs due to the overlapping activity of multiple extracellular enzymes and produces sMSLN products with a wide variety of C‐termini. sMSLN produced by the action of ADAM, MMP, or BASE proteases is believed to constitute the majority of sMSLN species and lacks all components of the GPI moiety. It is also missing a short span of the mature MSLN C‐terminus, beyond the cleavage point within the MSLN amino acid sequence. ⁸ Alternatively, GPLD1 (GPI‐specific phospholipase D1) can theoretically cleave within the GPI linkage to release a mannose‐containing sMSLN product. This form of sMSLN has previously been shown to bind macrophage mannose receptor CD206 and alter the polarization of tissue‐resident macrophages, ²⁸ which then promote tumor growth and pulmonary metastasis through secretion of vascular endothelial growth factor and the neutrophil chemotactic factor S100A9. ⁷ Mannose‐containing sMSLN cannot be generated by the Δ591 truncation used in our experiments as the GPI attachment site has been completely ablated; yet Δ591Mu MSLN‐expressing tumor cells had increased IP metastasis compared to MSLN KO cells. Investigations detailing activity differences among different sMSLN products would be an important next step in future experiments.

Our results suggest that sMSLN has a MUC‐16‐independent autocrine function that is difficult to detect in the presence of membrane‐bound MSLN. Examination using bulk RNA sequencing of MSLN KO cells acutely exposed to sMSLN‐containing medium, identified decreased activation of inflammatory pathways, accompanied by increased secretion of IL‐1α. Our experiments showed that stimulation of MSLN KO cells with all four types of sMSLN‐containing CM stimulated IL‐1α secretion to some degree. However, the increase in IL‐1α was highest (reaching statistical significance), only with addition of Mock or Y318A CM, the CM models mimicking our in vivo IP growth experiments with Parent and KO+Δ591Mu, respectively. Notably, these were the cell lines that displayed intact peritoneal dissemination, suggesting that IL‐1α may play a role in peritoneal colonization. Importantly, complementation of Δ591Mu in KO cells successfully rescued peritoneal growth despite these cells having significantly decreased cell surface MUC‐16. Previously, knockdown of MUC‐16 has been shown to decrease proliferation, migration, colony formation, cell adhesion, tumor growth, and metastasis of pancreatic cancer in part through reduction of AKT and MAPK activation. ²⁹ For Δ591Mu complementation to overcome the concomitant reduction in surface MUC‐16 which impairs KO+Δ591Mu cell growth in culture and colony formation in soft agar, insinuates a strong MUC‐16‐independent pro‐tumorigenic effect requiring non‐autocrine elements within the peritoneal milleu. IL‐1α has previously been implicated as a mediator of cross‐talk with cancer‐associated fibroblasts in the PDAC microenvironment, cells that are frequently complicit in supporting tumor cell growth and metastasis. ³⁰ Additional experiments using a fully immune‐competent model system would be ideal for further pursuing these leads but are outside the scope of the present study.

In summary, we have identified MSLN as an important mediator of pancreatic cancer cell clustering that promotes peritoneal colonization. Interventions to disrupt cancer cell clustering might constitute an important future strategy for limiting PDAC metastasis.

AUTHOR CONTRIBUTIONS

T.E., N.P., S.J., L.R.A., M.W.R., N.M., and X.Z. performed experiments, analyzed the data, and prepared figures. C.‐H.T. performed all bioinformatic analyses and prepared these figures. C.A. conceived the idea, designed the experiments, and assisted with figure preparation. T.E., N.P., and C.A. drafted the manuscript. All authors reviewed and approved the final manuscript.

DISCLOSURES

The authors declare no conflicts of interest. Dr. Alewine receives drug support from Astra Zeneca, Minneamrita, ProDa LLC, and HCW Biologics for use in ongoing clinical trials.

Supporting information

Figure S1..

Figure S2..

Figure S3..

Figure S4..

Dataset S1..

Dataset S2..

Dataset S3..

Dataset S4..

Dataset S5..

Ewa T, Panchwagh N, Tai C‐H, et al. Excess shed mesothelin disrupts pancreatic cancer cell clustering to impair peritoneal colonization. The FASEB Journal. 2024;38:e70247. doi: 10.1096/fj.202400446R

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available in the Material and Methods, Results, and/or Supplemental Material of this article. They are also available at GEO GSE260587 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE260587).