Multiple proteases are involved in mesothelin shedding by cancer cells

Mesothelin (MSLN) is a lineage restricted cell surface protein expressed in about 30% of human cancers and high MSLN expression is associated with poor survival in several different cancers. The restricted expression of MSLN in normal tissue and its frequent expression in cancers make MSLN an excellent target for antibody-based therapies. Many clinical trials with agents targeting MSLN have been carried out but to date none of these agents have produced enough responses to obtain FDA approval. MSLN shedding is an important factor that may contribute to the failure of these therapies, because shed MSLN acts as a decoy receptor and allows release of antibodies bound to cell-surface MSLN. We have investigated the mechanism of shedding and show here that members of the ADAM, MMP and BACE families of proteases all participate in shedding, that more than one protease can produce shedding in the same cell, and that inhibition of shedding greatly enhances killing of cells by an immunotoxin targeting MSLN. Our data indicates that controlling MSLN shedding could greatly increase the activity of therapies that target MSLN. Liu et al find that mesothelin (MSLN), a candidate target for antibody-based cancer therapy, is targeted by several proteases, and that inhibition of shedding facilitates killing of cells by an immunotoxin targeting MSLN.

M esothelin (MSLN) is a GPI-anchored membrane protein that is synthesized as a 70 kDa precursor protein and is proteolytically cleaved, generating a 30 kDa polypeptide known as megakaryocyte potentiating factor that is released from the cell and a 40-45 kDa glycoprotein (MSLN) 1-6 that is connected to the cell surface by phosphatidyl inositol 7,8 . MSLN binds to MUC16 and has a role in cell adhesion 9,10 and tumor progression 11 . MSLN is shed from the cell surface generating a pool of free antigen in pleural fluid, ascites, and in blood. Serum MSLN is used as a screening biomarker for cancer patients and to monitor responses 12,13 . Unfortunately, shedding is an impediment to antibody-based therapies, because it provides a sink within tumors that reduces their efficiency and also promote their release from the cell surface before they can kill the cell 6 .
Sheddases are proteases that cleave membrane-bound protein substrates close to or within the cell membrane and release the ectodomain from the cell. The best known sheddases are members of the ADAM family [14][15][16] . Substrates for ADAMs include growth factors, cytokines, chemokines, and adhesion molecules. ADAM17 also known as TACE was shown to mediate MSLN shedding in A431 cells 6 . However, TACE knockdown only reduced MSLN shedding by 50% indicating other sheddases have a role in shedding. To understand MSLN shedding in more depth, we determined the cut sites generated when MSLN is shed in different cancer cell lines. Using RNAseq data to determine sheddase expression, we knocked down RNAs encoding various sheddases to find which are important for shedding. Our results show that MSLN is cleaved at many sites close to the cell membrane, that ADAM10, ADAM17, BACE2, BACE1, and MMP15 all have a role in MSLN shedding, and that more than one sheddase can catalyze MSLN release in the same cell line.

Results and discussion
To determine the sites at which MSLN is cut, we studied five different cancer cell lines: KLM1 (pancreatic), KB31 (cervical), OVCAR8 (ovarian), A431/H9 cells (epidermoid carcinoma transfected with MSLN), and RH16, a primary mesothelioma cell line. Using culture medium, we immunoprecipitated MSLN with an antibody to the amino terminus and identified the product by immuno-blotting and Coomassie Blue staining. Figure 1a shows an SDS gel in which MSLN appears as a 40-45 kDa broad band typical of a glycosylated protein. A431 cells do not express MSLN. Because a single band with a molecular weight close to that of full-length MSLN was detected, we conclude that the major cleavage site is near the cell membrane. Figure 1b lists the most C-terminal MSLN peptide identified by mass spectrometry (for full-length sequence of the peptide see Supplementary Table 1). It was previously reported that MSLN from human ascites, pleural fluid, or the A431/H9 cell line is cut at two locations near the C terminus 6,17 . One cleavage is at residue 586 between Y and L, and the other is at residue 591 between L and S (Fig. 1b, cut sites 3 and 6). We have now identified several other cut sites (Fig. 1b). All the cut sites are located close together and close to the plasma membrane. These results establish that the region close to the cell membrane is the principal location of cutting by sheddases. The multiplicity of sites raises the possibility that different enzymes are responsible for shedding in different cells.
ADAM family. To identify the proteases responsible for shedding, we analyzed several RNAseq databases to determine which sheddases are expressed in our cell lines (https://depmap.org/portal/ccle). In KLM1, KB31, and OVCAR8 cells, several members of the ADAM family are expressed, specifically ADAM9, 10, 15, and 17 (Supplementary Table 2). ADAM17 was previously shown to be important for shedding of MSLN in A431/H9 cells 6 . Figure 2a, b (full image Supplementary Fig. 1) show that lowering ADAM17 expression did not decrease MSLN secretion in KB31 but did decrease shedding in KLM1 cells by 45%. These data identify ADAM17 as a sheddase for KLM1 and indicate that other proteases must be involved in shedding. ADAM9, 10, and 15 also have protease activity 14 and are expressed in these cell lines (Supplementary  Table 2). Figure 2c, d shows that ADAM10 knockdown diminished shedding in KLM1 and OVCAR8 cells while knockdown of both ADAM17 and ADAM10 produced a greater decrease in shedding than each alone Fig. 2e Supplementary Fig. 2).
MMPs. MT-MMPs (membrane-type matrix metalloproteinases) are a subgroup of the MMP family that can also act as canonical sheddases 16 . MMP14, 15, 17, 24, and 25 are all MT-MMPs. We found that MMP14 is highly expressed in KLM1 cells and MMP15 is highly expressed in KB31, KLM1, and OVCAR8 cells (Supplementary Table 2). We observed that knockdown of MMP15 reduced MSLN expression in KB31 cells (Fig. 3a, b). Knockdown of MMP14 or MMP15 did not reduce MSLN shedding in KLM1 cells (Fig. 3c, d) or OVCAR8 cells ( Supplementary  Fig 5), but slightly increased shedding. BACE1 and BACE2. BACE1 and BACE2 can cleave amyloid precursor protein (APP) and have been extensively studied in Alzheimer's disease 18 . They have also been implicated in processing other proteins 16,19,20 . We find that BACE2 and BACE1 RNAs are expressed in many cell lines (Supplementary Table 2). To determine if BACE2 protein is also expressed, we did western blots and observed BACE2 expression in KB31, MS751, T3M4, and AsPC1 cells ( Supplementary Fig. 6). To determine if BACE2 is involved in shedding, two siRNAs targeting human BACE2 were transfected into KB31 cells. Western blots show that when BACE2 protein is knocked down (Fig. 4a, full image Supplemental Fig. 7), total cell-associated MSLN is greatly increased (Fig. 4b), cell surface MSLN is increased as measured by flow cytometry (Fig. 4c), and shed MSLN is substantially decreased (Fig. 4d). These data indicate that in KB31 cells, BACE2 knockdown increases cell-associated MSLN by reducing MSLN shedding.
Since BACE1 RNA is also expressed in KB31 cells, albeit at a lower level than BACE2 (Supplementary Table 2), we knocked them down (Fig. 4e) and found that knockdown of BACE1 or BACE2 lowered shed MSLN levels to the same extent (Fig. 4f). However, knocking down both did not further lower MSLN shedding.
Because BACE2 is expressed more abundantly than BACE1 in our cell line panel (Supplementary Table 2), we examined the role of BACE2 in a cervical cancer line, MS751, and two pancreatic cancer lines, AsPC1 and T3M4. Figure 5a shows that siRNAs for BACE2 lower BACE2 protein levels (full image Supplemental Fig. 8). Figure 5b shows that in the medium of these three lines MSLN is significantly decreased. In MS751 there is a 60% decrease, in T3M4 a 55% decrease, and in AsPC1 a 23% decrease. We also knocked down BACE1 in KLM1 and BACE1  Fig. 9). In summary, these data indicate that BACE2 but probably not BACE1 has a major role MSLN shedding.
Small-molecule inhibitors to BACEs, ADAMs, and MMPs reduce MSLN shedding. Lanabecestat (LY3314814) and LY2886721 are potent inhibitors of BACE1 and BACE2 that have been used in clinical trials 16,21,22 . We tested the inhibitors on cell lines where MSLN shedding was affected by BACE2 knockdown. Both drugs inhibited MSLN shedding by over 50% in KB31 and by 30-40% in MS751, T3M4, and AsPC1 cells (Fig. 6a). Neither agent inhibited MSLN shedding in KLM1 and OVCAR8 cells (Fig. 6a), which have very low BACE2 (Supplementary Table 2). We then measured the activities of the ADAM inhibitor TMI-1 and the MMP inhibitor Marimastat on several cell lines. Figure 6b shows that the TMI-1 greatly decreased MSLN shedding in KLM1, as expected from the knockdown experiments, and it also affected OVCAR8 and T3M4. Marimastat inhibited shedding in KB31, KLM1, T3M4, and OVCAR8 cells. Marimastat is a potent broad-spectrum inhibitor of all major MMPs 23 . We only knocked down membrane-bound MMPs; it is possible other MMPs play a role in MSLN shedding.
Because our knockdown studies showed both MMP15 and BACE inhibited shedding in KB31 cells, we tested inhibitors of these enzymes. Figure 6c shows that Lanabecestat inhibits shedding by 30%, that Marimastat lowers shedding by 50%, and the combination by 80%. This result shows shedding can be mediated by two enzymes in the same cell line. Furthermore, we suspect an additional sheddase is involved, since shedding was not completely inhibited. To determine if this inhibition would result in enhanced immunotoxin action, we treated cells with the inhibitors for 20 h and then added SS1P, an immunotoxin that kills MSLN-expressing cells 1 . We found that the IC 50 changed from 4.6 ng/ml with no inhibitor to 1.1 ng/ml with Marimastat, to 1.3 ng/ml with Lanabecestat and to 0.29 ng/ml with both (Fig. 6d). The inhibitors were also examined in OVCAR8 cells. As shown in Fig. 6e, they stimulated SS1P killing; the IC 50 fell from 7.8 to 1.3 ng/ml with TMI-1, to 2.8 ng/ml with Marimastat and to 0.94 ng/ ml with both. These data show that in a single cell line more than one protease participates in MSLN shedding and inhibition greatly sensitizes the cells to an agent targeting MSLN.
Do BACEs participate in shedding of other proteins. The unexpected finding that BACEs have a major role in MSLN shedding raises the possibility that they would affect the release/ shedding of other proteins. To investigate this, we treated five cell lines with 20 μM Lanabecestat for 20 h and measured the content of several proteins and cytokines in the medium (Supplementary Table 3). As expected MSLN levels were decreased in all the lines, although the decrease in KLM1 was small and not significant.
There was also a small inhibition of shedding of Her1 (EGFR) in four lines and CA125 in one cell line (MS751). There was no effect on the folate receptor, which like MSLN is PI linked or on TGF-β1 and galectin 3. We conclude that the effect of BACEs on MSLN shedding is relatively specific.
MBTPS1 and MEPRIN. MBTPS1 has also been found to have a role in protein processing 16 . We knocked down MBTPS1 in KB31 and KLM1 cells and found no change in MSLN shedding ( Supplementary Fig. 10). Because there is no R residue in the region of MSLN where cutting occurs and cutting by MBTPS1 requires an R residue at P4, this result was expected. We also considered the possibility that meprin may participate in shedding, but CCLE shows that MEP1A and MEP1B RNA levels are very low in our cell line panel.

Conclusion
We show here that MSLN shedding is an unexpectedly complex process, which involves many different enzymes that cleave MSLN at several sites close to the cell membrane. We utilized several inhibitors that greatly decrease MSLN shedding and greatly enhance killing of cells by immunotoxins that target MSLN. Cutting at site 4 was only observed in KB31 cells, in which shedding is inhibited by BACE knockdown and the BACE inhibitor Lanabecestat, which did not inhibit MSLN shedding in KLM1 (Fig. 6a), OVAR8 (Fig. 6a) and RH16 ( Supplementary  Fig. 11). Cutting at site 3 is likely due to ADAM17, since it is found in cell lines sensitive to ADAM17 knockdown and inhibition by the ADAM inhibitor TMI-1. KB31 cells do not have cut site 3 and are not sensitive to TMI-1. We would expect these
Knockdown experiments. All cells were transfected using Lipofectamine RNAi-MAX reagent according to the manufacturer's protocol in 6 well or 96 well. Either 5 × 10e5 cells were plated in each well of six-well plates, or 5000 cells were plated in each well of 96-well plates. After 24 h on plate, siRNAs were transfected at a final concentration of 60 nM in a volume of 2 ml or 100 µl, respectively. siRNA oligo sequences are listed in Supplementary Table 4.
Soluble MSLN assay. Cell culture media was collected after siRNA transfection or inhibitor treatment at indicated times. The culture media were centrifuged at 3000 r.p.m. table-top centrifuge for 10 min, diluted 5-10-fold in diluent #100 and assayed using the R-PLEX human MSLN antibody set following the manufacturer's instructions (Mesoscale Discovery, Rockville, MD). The electrochemiluminescence was measured on a MESO QuickPlex SQ instrument. Shed MSLN levels are highly dependent on cell density and growth conditions. In these studies, the ratio of cells to medium is higher in six-well plate format leading to higher MSLN levels, which varied from 5 to 50 ng/ml under different experimental conditions. The MSLN levels were usually normalized using WST-8 assays to correct for differences in cell growth.
Real time PCR. RNAs were made using the Triazole reagent (Invitrogen). Reverse transcription and cDNA synthesis used the Quantitect Reverse transcription kit following the manufacturer's instructions (Qiagen). Primers are listed in Supplementary  Statistical significance compared to a negative control siRNA is noted by asterisks (*P < 0.05, **P < 0.01, ****P < 0.0001).
dilutions of SS1P were added in triplicate. Cells were incubated for an additional 72 h before running a WST-8 cell viability assay (Dojindo Molecular Technologies   The extracts were lyophilized, desalted using C18 columns (Thermo scientific), and reconstituted in 0.1% TFA.
Mass spectrometry acquisition and data analysis. The samples were injected on to Easy nLC 1000 HPLC (Thermo Scientific) coupled to an Orbitrap Fusion mass spectrometer (Thermo Scientific). The Easy nLC was configured with Acclaim Pepmap100 C18 2 cm trap column followed by 25 cm analytical column (Thermo Scientific) and the samples ran on a 90 min gradient at 250 nl/min. The precursor scans were performed in the Orbitrap at a resolution of 60,000 with a mass range of 375-1500m/z, followed by MS/MS analysis at a resolution of 7500 in the ion trap. Normalized collison energy was 29, and charge state 1 and unassigned charge states were excluded. Acquired MS/MS spectra were searched against human Uniprot protein database or Mesothelin (Uniprot number Q13421 iso #1) using the SEQUEST algorithm in the Proteome Discoverer 2.2 (Thermo Scientific, CA). Enzyme was set to Trypsin (Semi), precursor ion tolerance was set at 10 ppm, and the fragment ions tolerance was set at 0.6 Da along with methionine oxidation included as dynamic modification. Peptide validation was done by either the Percolator or Target Decoy PSM Validator with FDR set at 0.1%.
Statistics and reproducibility. All data are presented as mean ± standard deviation (SD). The error bars represent the SD of triplicates or more as indicated in the figure legends. All experiments were repeated at 2-3 times. Statistical differences between groups were measured by unpaired T-test, two-tailed, using GraphPad Prism 7. A P value of less than 0.05 was considered significant. All data are available upon request.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Original mass spectra has been deposited in MassIVE (https://massive.ucsd.edu) under the identifier MassIVE: MSV000085950. The source data behind graphs can be found in Source Data. All other data generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.