Opposing roles for ADAMTS2 and ADAMTS14 in myofibroblast differentiation and function

  1. Edward P Carter
  2. Kubra K Yoneten
  3. Nuria Gavara
  4. Eleanor J Tyler
  5. Valentine Gauthier
  6. Elizabeth R Murray
  7. Peter ten Dijke
  8. Angus J Cameron
  9. Oliver Pearce
  10. Richard P Grose

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Crosstalk between cancer and stellate cells is pivotal in pancreatic cancer, resulting in differentiation of stellate cells into myofibroblasts that drives tumour progression. To assess cooperative mechanisms in a 3D context, we generated chimeric spheroids using human and mouse cancer and stellate cells. Species‐specific deconvolution of bulk‐RNA sequencing data revealed cell type‐specific transcriptomes underpinning invasion. This dataset highlighted stellate‐specific expression of transcripts encoding the collagen‐processing enzymes ADAMTS2 and ADAMTS14. Strikingly, loss of ADAMTS2 reduced, while loss of ADAMTS14 promoted, myofibroblast differentiation and invasion independently of their primary role in collagen‐processing. Functional and proteomic analysis demonstrated that these two enzymes regulate myofibroblast differentiation through opposing roles in the regulation of transforming growth factor β availability, acting on the protease‐specific substrates, Serpin E2 and fibulin 2, for ADAMTS2 and ADAMTS14, respectively. Showcasing a broader complexity for these enzymes, we uncovered a novel regulatory axis governing malignant behaviour of the pancreatic cancer stroma. © 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Article activity feed

  1. Version published to 10.1002/path.6214
  2. Review Commons

    Note: This rebuttal was posted by the corresponding author to Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Reply to the reviewers

    1. General Statements [optional]

    We are grateful to the reviewers for highlighting the value and power of our 3D chimeric dataset to explore cancer/stellate interactions in pancreatic cancer invasion. We also appreciate their support of our findings identifying divergent roles for the two related enzymes ADAMTS2 and ADAMTS14. We thank the reviewers for their detailed comments, which have allowed us to prepare a significantly stronger and clearer manuscript.

    Following the reviewers comments we have made three major changes to the manuscript, which we will outline here in addition to the point-by-point rebuttal.

    1. i) Revised manuscript structure. We have modified the structure of the manuscript, which we hope improves the clarity and accessibility of the work.

    Figure 1 remains the description of our 3D invasion model and our approach to identify stellate cell and cancer cell transcriptomic information from this context.

    Figure 2 describes our focus on proteases and now includes concordance of our data with clinical data sets. This is also now where we describe the strikingly opposing roles for ADAMTS2 and ADAMTS14 in regulating invasion.

    Figure 3 is now the figure demonstrating that ADAMTS2 and ADAMTS14 have an equal contribution to collagen processing from stellate cells. This is an important experiment given that the main physiological roles for these enzymes are in the processing of collagen, and the importance of collagen for cancer progression. It was therefore reasonable to hypothesise that the effect of these enzymes on invasion could be due to differences in their collagen processing in this context. The finding that both have an equal effect on collagen processing points towards a wider, and more diverse, role for these enzymes in regulating biology.

    Figure 4 describes the divergent roles of these two enzymes on myofibroblast differentiation, and by extension TGFβ bioavailability. In this figure we now include experiments with TGFβ reporter constructs, which demonstrate an increase in active TGFβ following loss of ADAMTS14 and a reduction in TGFβ activity following loss of ADAMTS2.

    Figure 5 is our matrisomic experiment to identify enriched enzyme-specific substrates following knockdown of either ADAMTS2 or ADAMTS14.

    Figure 6 details our investigation into the substrate responsible for the reduction in invasion following loss of ADAMTS2. As the previous matrisomic experiment identified only two enriched ADAMTS2 substrates, we investigated both in our 3D assays, identifying SERPINE2 as the responsible substrate. Further analysis identified a reduction in plasmin activity in ADAMTS2 deficient cells. This was rescued with co-knockdown of SERPINE2, implicating this pathway as being crucial for mediating the effect of ADAMTS2. Additionally, we now include experiments demonstrating that concomitant knockdown of SERPINE2 alongside ADAMTS2 rescues the reduction in TGFβ activity observed with ADAMTS2 loss alone.

    Figure 7 describes our analysis of ADAMTS14 substrates. As the matrisomics identified a large change in proteins following ADAMTS14 knockdown, we performed an siRNA screen of candidates to identify those responsible for ADAMTS14 phenotype. This, followed by further validation in our 3D invasive assay, revealed Fibulin2 as the responsible substrate. Fibulin2 has a well-established role in regulating TGFβ release from the matrix. In accordance with this we present new data using TGFβ reporter constructs, which demonstrate that the increase in active TGFβ following ADAMTS14 knockdown can be reversed with co-knockdown of Fibulin2.

    1. ii) Improvement of the clinical significance of our chimeric data set and ADAMTS proteins. Ideally, we would like to present IHC images of ADAMTS2 and ADAMTS14 expression in PDAC tissue samples to corroborate our in vitro findings. However as these enzymes are secreted, this precludes antibody based imaging, as it would not provide cell type specific information. RNA scope presents an alternative, however we have experienced technical issues with this technique due to RNA degradation in PDAC tissue and unavailability of ADAMTS2/14 specific probes. In place of this we have used a range of publically available resources.

    We have compared our chimeric data set with human clinical data using the resource published by Maurer and colleagues (PMID: 30658994). This paper presents transcriptomic data from PDAC tumour and stromal compartments using laser microdissection of clinical tissue. In accordance with our data set, the majority of metzincins, including ADAMTS2 and ADAMTS14, are expressed in the stromal compartment. These data are presented in updated figure 2.

    We have also examined ADAMTS2 and ADAMTS14 expression in PDAC and CAF subtypes using publically available data sets. Using the TCGA dataset, we identified that ADAMTS2 and ADAMTS14 are highly expressed in PDAC tumours compared to normal counterparts. As the majority of PDAC is comprised of stroma, the bulk transcriptomic data from TCGA, combined with the results from the Maurer publication, lead us to conclude that this expression reflects the stromal origin of these proteases. In addition, using publically available single cell RNA sequencing data published by Luo and colleagues (PMID: 36333338), we identified ADAMTS2 and ADAMTS14 expression in the prominent PDAC CAF subtypes, inflammatory and myofibroblastic CAFs. Together these data demonstrate that these enzymes are enriched in clinical disease, which when combined with our mechanistic 3D studies implies a greater role for these enzymes in disease progression than previously appreciated.

    iii) Improved mechanistic link between ADAMTS2 and ADAMTS14 with TGFβ bioavailability

    To strengthen the association between ADAMTS2 and ADAMTS14 function, their substrates SERPINE2 and Fibulin2, and TGFβ bioavailability, we have performed the following experiments using TGFβ reporter constructs:

    We have taken conditioned media from stellate cells lacking either ADAMTS2 or ADAMTS14, along with co-knockdown of their substrate, and stimulated a recipient cell line expressing a SMAD Luciferase reporter. These cells express luciferase in response to TGFβ stimulation. In accordance with a role for ADAMTS14 and Fibulin2 in regulating TGFβ, we demonstrate that following ADAMTS14 knockdown there is a strong increase in active TGFβ in the media (Figure 4I), which is abrogated with co-knockdown of Fibulin2 (Figure 7F).

    We have also obtained a fluorescent reporter, CAGA-eGFP, which expresses GFP in response to TGFβ stimulation in order to examine TGFβ activity in 3D cultures. Stellate cells expressing this construct were embedded in collagen: Matrigel hydrogels following knockdown of either ADAMTS2 or ADAMTS14 and CAGA fluorescence recorded after 72 hours of culture. In accordance with our data, stellate cells deficient in ADAMTS14 showed increased fluorescence in 3D, indicative of increased TGFβ activity, which was abrogated with co-knockdown of Fibulin2 (Figure 4J, K and 7G, H). Equally, loss of ADAMTS2 reduced TGFβ activity in 3D culture, which was rescued with co-knockdown of SERPINE2 (Figure 4J, K and 6 D, E).

    These experiments confirm a link between the ADAMTS enzyme, its relevant substrate, and TGFβ bioavailability. Together with extensive published work linking SERPINE2 and Fibulin2 with TGFβ release we are confident in our proposed mechanism for the dichotomic relationship of ADAMTS2 and ADAMTS14 in regulating TGFβ and thus myofibroblast action.

    2. Point-by-point description of the revisions

    This section is mandatory. *Please insert a point-by-point reply describing the revisions that were already carried out and included in the transferred manuscript. *

    Reviewer #1 (Evidence, reproducibility and clarity (Required)):

    This study aims to explain the opposing contributions of stromal stellate cells/CAFs to PDAC. By first identifying stroma-specific proteases, followed by a process of candidate selection and elimination, the authors find that two specific metalloproteases that share enzymatic activity against collagen in fact have differential activity on TGFb availability. This could be interpreted as a way of shaping the CAF population and tumor-promotin or -restricting properties of the stroma.

    There are several flaws that the authors could address to improve the manuscript:

    1. In the flow of experiments and analyses, there is a strange mix of fully unbiased discovery phases followed by functional experiments that do not consider all possible candidates to test, and vice versa. For instance, from the mixed-species transcript analysis, ADAMTS2 and -14 are chosen based on their shared collagenase activity based on literature. However, the authors then perform again a proteomics analysis to identify things from the entire matrisome that are cleaved by these enzymes? Then, for ADAMTS2 a co-silencing approach is done on one selected candidate (Serpine2), but for ADAMTS14 an siRNA screen is performed? The problem of this approach is that the rationale for some studied enzymes is very strong, where as for others it is not.

    We thank the reviewer for their comment and trust the revised manuscript provides more clarity for the rationale of our approach. We performed the chimera sequencing as a discovery experiment to reveal the communication between cancer and stellate cells in a 3D, invasive context. We present the chimera experiment and data here as a resource for the community, with our analysis of ADAMTS2 and ADAMTS14 function serving as a first example of the biological insight this data set can reveal. Other insights revealed from this dataset are active avenues of research in our group.

    Our finding that ADAMTS2 and ADAMTS14 have dramatically opposing roles in regulating invasion was especially striking given their equal contribution to collagen processing in this context. This led us to conclude that the divergent nature of these enzymes must be due to enzyme-specific substrates. A substrate repertoire for these enzymes has been previously published (PMID: 26740262) and we reasoned that the responsible substrate would be enriched following knockdown of the relevant enzyme. Thus we preformed matrisomics on cells lacking either of these enzymes, which did indeed reveal enrichment of known, enzyme-specific substrates that we could use for further analysis.

    The matrisome following ADAMTS2 knockdown was minimally changed and only presented enrichment of two ADAMTS2 substrates. As there was only a minimal cellular phenotype in 2D following loss of ADAMTS2, we decided to concentrate our studies on the two identified substrates in our 3D assay. Conversely as the matrisome following ADAMTS14 knockdown was dramatically different from control cells, and ADAMTS14 knockdown presented a clear phenotype in αSMA expression, we decided to perform a screen of all matrisome hits. This highlighted the role of IL-1β in mediating myofibroblast differentiation, which has been reported elsewhere and validated our approach. Further, this refined the number of enriched ADAMTS14 substrates to two, MMP1 and Fibulin2, with Fibulin2 being identified as the responsible candidate in our 3D assays.

    The ECM is more than just collagen. Choosing these two metalloproteases based on their shared collagen substrate is an approach that perhaps oversimplifies the ECM a bit, and again, does not provide the strongest rationale that these metalloproteases are most likely to explain counteracting stromal activities on tumor growth and progression.

    We fully agree with the reviewers comment and feel our work acutely demonstrates this point. Loss of either ADAMTS2 or ADAMTS14 had similar effects on collagen processing; implicating their divergent roles on invasion was independent of their effects on collagen regulation. This work therefore showcases the incredible complexity of ECM regulation in tumour progression. As discussed in the manuscript, collagen along with other elements of the ECM can regulate tumour progression and we believe our work adds an additional facet to this.

    Related to the above: How were the stellate cells used for the matrisome analysis grown? In the suspension setup or adherent? This will have a large impact on the outcome. Is there for instance hyaluronic acid in this matrix?

    The matrisome analysis was conducted on cells cultured in 2D. Vitamin C was added to the media to promote matrix production. We agree that this is not truly reflective of the in vivo situation but as a discovery tool this led us to identify the ADAMTS2 and ADAMTS14 substrates responsible for the function observed in 3D.

    1. Performing the species-specific transcript analysis both ways is a neat approach, but why did the authors ignore the opportunity to formally overlay/compare the two stromal gene sets to define likely candidates based on statistics?

    We primarily used this approach as a discovery tool to identify key differences between cancer and stellate cell compartments. Comparing the two species data sets is problematic as the murine cancer cells express many elements found in the stellate cells, while the human data set presents a cleaner comparison. This is evident from comparing metzincin expression in the two data sets. The human data set (Figure 2A) shows clear separation between cancer and stellate compartments, which is less evident in the murine data set (Supp figure 2A). As noted in supplementary figure 1A, unlike the human cancer cells used in this study, the murine cancer cells are capable of invading without stellate support (although when cultured with stellate cells invasive projections are always stellate led). Nevertheless the murine data set matches the human, although with less clarity.

    Minor comments: The bioinformatics Methods need more details on how reads were mapped to the different genomes. How many mismatches were allowed and was the mapping done separately or using for instance Xenofilter?

    We have improved the methodology section to include more detail for this separation. Using STAR aligner, reads were mapped to host species using a combined human and mouse genome. Ambiguous reads were subsequently discarded from the analysis. While there are bioinformatic packages that seek to match ambiguous reads to parent species we did not use these for our analysis.

    The authors use the knowledge on the activities of both ADAMTS2 and -14 on collagen as a rationale to choose these two. Is there really a need for the paragraph (and associated figures) from line 102 on?

    Given the prominent role collagen has been shown to have in regulating PDAC progression and the primary role for ADAMTS2 and ADAMTS14 being collagen processing, we initially hypothesised that the divergent role for these enzymes on invasion could be due to differences in collagen processing in this context. The fact that both equally contribute to collagen processing is surprising and adds to the novelty of our findings that these enzymes have a more complex role in regulating stromal biology.

    We have altered the structure of the manuscript to emphasise this point. The divergent roles of ADAMTS2 and ADAMTS14 on invasion are now presented in Figure 2, with their equal role in collagen processing now presented after in Figure 3. Figure 4 onwards now details the opposing roles of these enzymes in myofibroblast differentiation and our investigation into the enzyme-specific substrates responsible for this.

    Abstract, line 21; some words are missing?

    We thank the reviewer for bringing this to our attention and have now amended the abstract.

    Were the siRNA screen hits validated?

    Yes, hits relevant for our further investigations, MMP1 and Fibulin2, are presented in the manuscript.

    What is the genotype of the mouse cancer cells? KPC-derived?

    DT6066 are KPC derived while R254 are derived from KPF mice. This has been added to the methods with relevant reference.

    Reviewer #1 (Significance (Required)):

    The trick of dissecting tumor from stromal signals in spheroid cocultures by RNA-Seq is a cool trick, but not new and the authors should probably cite some prior work.

    We have included reference to other work where researchers have used species deconvolution to explore heterocellular interactions (Lines 68-72). However, we believe our work is one of the first to use this approach to explore cellular interactions in an in vitro, 3D, invasive context.

    What this all means for patients (or in vivo tumors even) remains unclear. There is some debate on whether highly activated CAFs (ACTA2/aSMA+ cells, some call them myCAFs) are indeed tumor-restrictive or whether they promote invasion. The authors appear to argue the latter (which I can agree with) but without any translational work to show what the net outcome of this mechanism is, the study remains descriptive and perhaps of limited interest.

    We contend that our 3D invasion model is a powerful tool to understand the role of stellate cells in leading invasion. We have shown the utility of this model in several studies to dissect the biology of this cell type, revealing the importance of the nuclear translocation of FGFR1 in stellate invasion (PMID: 36357571), the role of the kinase PKN2 in regulating stellate heterogeneity (PMID: 35081338) and the influence of cancer cell-derived exosomes on stellate invasion (PMID: 33592190).

    CAFs within PDAC stroma are highly plastic and can adopt multiple functions depending on distinct environmental cues. Thus, identifying how they are regulated is of paramount importance if they are to be therapeutically targeted. We contend that our mechanistic studies using heterocellular 3D models can aid in the dissection of the biology of these cells with more granularity than offered by clinical or in vivo studies, particularly in the context of secreted proteases. To add clinical relevance for our findings we have compared our chimera data set with previously published laser microdissected tumour and stroma PDAC tissue (Figure 2B), and identified ADAMTS2 and ADAMTS14 expression in prominent CAF subtypes (inflammatory and myofibroblastic) from published single cell RNA seq data taken from tumours (Supp figure 2C). As these enzymes are produced in multiple CAF subtypes, genetically targeting them in vivo appears prohibitive. The generation of ADAMTS2 and ADAMTS14 specific inhibitors would be required to assess their roles in vivo.

    Reviewer #2 (Evidence, reproducibility and clarity (Required)):

    The manuscript by Carter and colleagues examines that role of cancer-associated fibroblasts (CAFs) in regulation of invasion in a 3D co-culture assay with epithelial cells. The authors propose that invasive chains of cancer cells are led by fibroblasts. The authors utilise a system of co-culture to create chimeric human-mouse fibroblast-cancer cell spheroids (both directions utilised, to eliminate species bias) to allow for in situ sequencing of the co-operating transcriptional programmes of each cell type during 3D invasion. From this powerful approach, this allowed the authors to identify two key two collagen-processing enzymes, ADAMTS2 and ADAMTS14, as contributing to CAF function in their system. The authors identify that these two enzymes have opposing roles in invasion, and map some of their key substrates in invasion, which extend beyond collagen-processing. The authors propose that one key function is to control the processing of TGFbeta, the latter of which is a regulator of the myofibroblast subpopulation of CAF. Overall, the findings of the manuscript are interesting, but need some further proof-of-principal demonstrations to extend their findings to support the claims within.

    • The authors demonstrate a clear role of ADAMTS2 and ADAMTS14 in stellate function during differentiation and invasion. Is there any evidence of such changes in patient materials? Could the authors query publicly available databases of micro-dissected stroma vs epithelium to validate the translational relevance of their findings?

    We thank the reviewer for their suggestion; we have now explored clinical relevance of ADAMTS2 and ADAMTS14 expression in two ways. We have used previously published work by Maurer and colleagues (PMID: 30658994), which descibes transcriptomic analysis of laser microdissected tumour and stroma from pancreatic cancer tissue. In accordance with our chimeric data set the majority of metzincins, including ADAMTS2 and ADAMTS14, are expressed in the sromal compartment (Figure 2B). We have also used publically available scRNA seq data to examine ADAMTS2 and ADAMTS14 expression in distinct CAF subtypes (Supp Figure 2C). Both ADAMTS2 and ADAMTS14 are expressed in inflammatory and myofibroblastic CAFs, with ADAMTS14 expression lower than that of ADAMTS2. Given the complexity of CAF heterogeneity it is possible that ADAMTS2/14 secretion by one population regulates the resulting phenotype of surrounding CAFs, however this hypothesis if beyond the scope of our current work.

    Major comments:

    • Page no. 4, Line 71, The authors conclude that the invasion in the chimeric spheroids is "led by" stellate cells. This is a key concept in the manuscript. How do the authors define the "led by" phenomena? What is the frequency that this occurs?

    In our experience all invasive projections are stellate led, defined as a stellate-labelled nucleus present at the tip of invasive projections. Indeed the human cancer cells used in this study are incapable of invading in the absence of stellate cells (Supp figure 1 A). We have previously reported this model where we demonstrated FGFR1 activity in the stellate cells is crucial for invasion (PMID: 36357571). Others have demonstrated the general importance for fibroblasts in leading invasion (PMID: 18037882, 28218910). Interestingly in our study, mouse cancer cells were capable of invading in the absence of stellate cells. However, when cultured with stellate cells, projections were predominantly stellate led.

    • For Figure 2A and S2A, the text suggests that the heatmap represents the stellate vs cancer cell expression (as shown in Figure 1B and S1B) in the respective species but the labelling below the heatmap suggests they are all cancer cells (Mia, Pan, R2 and DT). Is this a typo? Could the authors clarify this?

    We use Mia, Pan, R2 and DT to define the sphere combination from which the data originated. We have improved the clarity of the heatmaps by colour coding the different cell types within each sphere, and matching it with the cell type data presented in the heat map. We hope this improved labelling makes the heatmaps more accessible.

    • The text and the figures are lacking information about the cell line names used in the experiment, e.g, Figure 2C, 2D, 2SB, 2SC and 2SD does not indicate what cell line was used in the study. This is the same with other figures as well. Please indicate in all instances exactly which samples are queried.

    We have now included reference to the cell type and stellate cell species used in each experiment in relevant figure legends. Key 3D invasive experiments were conducted with both human and mouse stellate cells.

    • It's mentioned in the text that the authors have used the cancer and stellate cells in a 1:2 ratio but the numbers of stellate cells look different between different spheroids confocal images. e.g. The numbers look very different between the Miapaca2:PS1 vs Miapaca2:mPSC spheroids. Is this simply the representative images, or are their bona fide differences. This, in turn, would impact on claims of cells being 'led' by stellate cells. Can the authors clarify?

    This is a consequence of the method by which the stellate cells were immortalised. Human PS1 stellate cells were immortalised with hTERT, while mouse stellate cells were immortalised with SV40. A consequence of this is that the mouse stellate cells proliferate faster in 3D than the human stellate cells, with both proliferating slower than the cancer cell compartment. So while spheroids start at 1000 cells (666 stellate, 333 cancer) with stellate cells as the prominent component they are quickly overtaken by the cancer cells. Despite this difference in proliferation we find no difference in the invasive capacity of the stellate cells, with invasive projections always stellate led irrespective of whether they are human or mouse.

    • While for most of the experiments the authors generated the chimeric spheroids first and then performed the respective experiments, it appears that for the invasion assay simply co-culture of Cancer cells and stellate cells was done. Is this correct? Have the authors tried performing the assay with the chimeric spheroids to see if the stellate cells still invade?

    The Boyden chamber migration assay was conducted by seeding a co-culture of stellate and cancer cells in the apical compartment then imaging their migration to the basolateral side. This provided a second method to predominantly showcase the enhanced migration of cells lacking ADAMTS14 in a manner that could be quantified over time. We have not tried placing spheroids in the apical compartment and imaging invasion through the pores.

    • The authors claim that ADAMTS2 and ADAMTS14 regulate the bioavailability of TGFB, and this is a key reason that these regulate CAF differentiation. However, there is no direct demonstration of this concept, which is conspicuous by absence. Could the authors either directly demonstrate this, or remove such notions from the results, and explicitly state that this is an untested speculation in discussion? Examples of this are:

    o Line 173, authors state "ADAMTS2 facilitates TGFβ release through degradation of the plasmin inhibitor, SERPINE2 (Figure 5D)"

    o Line 196 authors conclude "Together these data implicate 197 ADAMTS14 as a key regulator of TGFβ bioavailability (Figure 6F)."

    o Line 240 states "This reduces the activation of Plasmin, preventing the release of TGFβ (Figure 5C)." Since this is just a model without detailed experiments, It will be better to propose rather than conclude.

    We appreciate the reviewer’s concern and have now added additional experiments to strengthen the association of ADAMTS enzymes and TGFβ bioavailability.

    Using a TGFβ-responsive luciferase reporter we demonstrate that the media from stellate cells lacking ADAMTS14 has greatly increased amounts of active TGFβ (Figure 4), which is abrogated when Fibulin2 is knocked down alongside (Figure 7). This links ADAMTS14 and Fibulin2 to TGFβ activity. Given the extensive literature detailing a role for Fibulin2 in regulating matrix TGFβ release through interactions with fibrillin (e.g, PMID: 19349279, 12598898, 12429738) we believe this is how ADAMTS14 is regulating myofibroblast differentiation. As we do not directly examine the association of Fibulin2 with fibrillin in this manuscript we have amended the associated statements to reflect this.

    We have also used a TGFβ-responsive fluorescent reporter to examine TGFβ activity of stellate cells in 3D. Consistent with our results, loss of ADAMTS2 reduces, while loss of ADAMTS14 enhances, TGFβ activity (Figure 4), which can be reversed with concomitant knockdown of their respective substrates SERPINE2 (Figure 6) and Fibulin2 (Figure 7).

    • Figure S5C shows a less invasive phenotype in the NTCsi + ADAMTS14si spheroids compared to the NTCsi + NTCsi control. However, there appears no appreciable difference between NTCsi + ADAMTS14si and NTCsi + NTCsi spheroids' brightfield images in Figure 5SD.

    Could the authors comment on this?

    We thank the reviewer for bringing this to our attention and apologise for our mistake. The images were positioned erroneously. This has now been corrected and the images reflect the quantification that demonstrates a clear increase in invasion following loss of ADAMTS14, which is abrogated with co-knockdown of Fibulin2.

    Minor Comments:

    • Page no. 2, Line 20 has an incomplete sentence "Crosstalk between cancer and stellate cells is pivotal in pancreatic cancer, resulting in differentiation 21 of stellate cells into myofibroblasts that drive."

    Apologies for the error. This has been rectified.

    • Figure 2C; Figure S2C and Figure S5E lack quantification for the western blots.

    We have now included densitometry for all western blots, presenting values relative to the respective loading control and normalised to the experimental control. Values are averages taken from all biological repeats with significance indicated where relevant.

    • Why did the authors choose to investigate the Metzincin family? Could the authors provide their reasoning to investigate these proteins, to the exclusion of other candidates?

    We focused on the metzincin family, as they are best known for their involvement in cancer invasion. A goal for this manuscript is to present our chimera data set as a discovery tool for the community. While this initial manuscript focuses on protease activity, we have further projects on-going that have used this data set to identify important elements of cancer/stellate communication.

    • Info about the number of fields imaged per sample for the microscopy data is missing in the figure legends (e.g. Figure 2F and 2I, Figure 5SF).

    We have now included a statement in each relevant figure legend to indicate that quantification was performed on at least five fields of view per biological repeat.

    • Any particular reason why the ADAMTS2 expression was not checked through Western blotting like ADAMTS14 in Figure S2B.

    We attempted to examine ADAMTS2 by western blotting but were unable to find an antibody that produced consistent results with our samples, and corroborated consistent knockdown by PCR.

    • The legends for Figure 3SC and 5SF mention that "Images are representative of at least two biological replicates". How many technical replicates were used? It would be useful if the relative intensity of the images is measured and plotted in a graph.

    We have now moved these images to the main figure alongside quantification of αSMA intensity. Images are collected from two biological repeats with quantification obtained from at least five fields of view per image. Together these data strongly demonstrate that loss of ADAMTS14 increases αSMA fibre intensity, which is blocked by either an inhibitor of TGFβ signalling (Figure 4), or co-knockdown of Fibulin2 (Figure 7).

    Reviewer #2 (Significance (Required)):

    This work provides an examination of the cross talk between fibroblasts and cancer cells in a 3-Dimensional culture model of pancreatic tumour cell invasion. By using chimeric human-mouse spheroids, the authors are able to identify cell-type specific transcripts by bulk RNA sequencing in situ. This advance is not to be underestimated as a number of existing approaches for cell type-specific profiling (eg. single-cell sequencing) relies upon dissociation of cell communities prior to sequencing. It is very likely that transcriptional programmes change during this isolation process. This approach allows the authors to identify transcriptional co-operating programmes in situ. This data provides a resource to understand this key co-operation of these two cell types during tumourigenesis, and will be of interest to the pancreatic cancer field. In addition, the mapping of the key substrate of these enzymes provides further insights that may be useful in understanding the expanded target repertoire of these enzymes beyond collagen processing.

    We thank the reviewer for their strong support of our chimeric spheroid approach and resulting investigation into the dichotomic roles of ADAMTS2 and ADAMTS14.

    Reviewer #3 (Evidence, reproducibility and clarity (Required)):

    AMADTS2 and ADAMTS14 belong to the disintegrin and metalloproteinase with thrombospondin motif protein family, mainly produced by pancreatic stellate cells (PSC) and are related to cancer cell invasion. This study reveals that ADAMTS2 and ADAMTS14 have opposite roles in myofibroblast differentiation based on experiments testing HSC driven cancer cell invasion, variant expression of HSC activation makers and the related downstream targets analysed upon RNA sequencing analyses. The authors (TA) established PSC/cancer cell chimeric spheroids for investigating the crosstalk between these cell types in 3D in vitro. Based on their findings, they claim that ADAMTS2 and ADAMTS14 have different functions regarding PSC and TGF-β activation. However, their conclusions mainly rely on quantitatve data of invasion and mechanistic details are completely lacking.

    Comments: Typos, even in the abstract, e.g. first sentence incomplete

    We apologise for the error in the abstract and have rectified this in the revised manuscript.

    Introduction is rather sparce with one third of the text repeating the results of the study

    Our manuscript details a discovery experiment using chimeric spheroids to identify cancer cell and stellate cell transcriptomes in a 3D invasive context. We then showcase the power of this data set by using it to identify and then describe divergent roles for ADAMTS2 and ADAMTS14 in shaping stellate cell biology. Given this two-tiered approach we incorporated text that would normally be placed in the introduction into the results section (e.g. our description of the importance of collagen processing in PDAC, presented as a prelude to the results from figure 3). We feel this improves the flow of the manuscript, rather than having information that isn’t necessarily relevant to the reader at the outset.

    Some citations do not at all fit with the position where they are placed; needs approval

    We have examined this in detail and are confident in our use of appropriate references throughout.

    In this study, it is said that TS2 (AMADTS2) and TS14 (ADAMTS14) have opposite functions on myofibroblast differentiation, with individual depletion leading to distinct matrisomal phenotypes in PSC. However, both similarly contribute to collagen processing. As we know, collagen is increased in response to TGFβ signaling, since TS2 depletion (knock down, kd) inhibits and TS14 kd is suggested to promote TGFβ activation, it is expected that this has impact on the available collagen levels. How do the authors explain that nevertheless the kd effect on collagen is very similar?

    The primary effects of these enzymes are on the processing of pro-collagen to its mature form, rather than on the production of collagen. This is evidenced in figure 3B where collagen expression in the whole cell lysate is the same following ADAMTS2 knockdown, and slightly reduced with loss of ADAMTS14, but the mature form is lost in the cell culture supernatant.

    While myofibroblast differentiation is associated with increased collagen production, it is possible that this is perturbed in a situation where the cell is surrounded by collagen that is incompletely processed (e.g. through biomechanical feedback). Given that our results clearly indicated that the effect of ADAMTS2 and ADAMTS14 on invasion is independent of their roles in collagen processing, this avenue is beyond the scope of the current manuscript.

    The authors claim that TS2 facilitates TGFβ release and TS14 is a key regulator of TGFβ bioavailability. However, throughout the whole data, there is no experimental evidence for this conclusion. TGFβ activation, LAP concentration and downstream effects should be provided.

    Most of the conclusions in the manuscript are based on effects to invasion and the estimated quantification histograms. "Black boxes in between the treatment, e.g. knockdown and readout, that relate to the signals and mechanisms remain black boxes throughout. For example, the impact of the treatments on stellate cell activation markers, the cancer cells invasion signaling, the SERPINE2- and Fibulin2-dependent myofibroblast differentiation pathways should be mechanistically investigated.

    We disagree with this comment. Our invasive model shows a clear role for ADAMTS2 and ADAMTS14 in regulating invasion, which is mitigated by disrupting their substrates SERPINE2 and Fibulin2.

    ADAMTS2 loss is associated with a reduction in plasmin activity, which again is mitigated with concurrent loss of SERPINE2. Equally, inhibition of plasmin activity with Aprotinin matches the loss of invasion observed with loss of ADAMTS2. Plasmin has a well-established role in mediating TGFβ release from the matrix. We have now included additional experiments using a TGFβ fluorescent reporter in 3D culture. This demonstrates that loss of ADAMTS2 reduces TGFβ activity, which can be rescued with co-knockdown of SERPINE2 (Figure 6). Our data therefore support a mechanism where ADAMTS2 blocks TGFβ release from the matrix, and therefore myofibroblast differentiation, through its regulation of SERPINE2 activity.

    We have strengthened our proposed mechanism for ADAMTS14 regulation of TGFβ through Fibulin2 with the use of both luciferase and fluorescent TGFβ reporter constructs. Using these reporters, we demonstrate that stellate cells lacking ADAMTS14 exhibit increased TGFβ activity (Figure 4), which is mitigated with co-knockdown of Fibulin2 (Figure 7). Combined with the effects on αSMA expression and 3D invasion, our data fit with a model where ADAMTS14 regulates TGFβ bioavailability through Fibulin2.

    The authors investigate one cell line each for their conclusions; we know that different cell lines behave differently; can they confirm that the finding they present is of general validity or a finding that is specific for the tested cancer/PSC cell lines. Can the principle findings also be proven in primary cells. More importantly, the authors should proof their findings in PaCa tissue of patients as follows: Expression of the proteases in the tissue, related variation of matrisome signatures, e.g. by snRNASeq, to confirm relevance of the finding.

    All our key 3D invasive experiments are repeated with both human and mouse stellate cells, adding strength to our proposed association with ADAMTS2 and SERPINE2, and ADAMTS14 and Fibulin2, on the invasive capacity of stellate cells. As detailed above we have explored the clinical relevance of our findings by examining laser dissected tumour and stromal data from PDAC tissue, and scRNA fibroblast data. These data confirm that ADAMTS2 and ADAMTS14 are predominantly expressed in the stromal compartment of the tumour and are associated with key CAF subtypes present in the PDAC environment, inflammatory and myofibroblastic CAFs.

    Details related to the figures: Figure 1: Are the numbers of PSC and PaCa cells integrated in the spheres related to the numbers found in patients?

    The 2:1 ratio of stellate to cancer cells used to produce spheres is a technical requirement and reflects the numbers in patients (PMID: 23359139). Cancer cells will proliferate substantially faster than the stellate cells so at the end of the experiment (day 3) the spheres are predominantly cancer cells. Nevertheless the stellate cells are able to drive invasion of the cancer cells, which can be quantitatively assessed in this model.

    B, it seems that the PSC in the spheroid are not equally distributed but instead are all located in close vicinity to eachother in a cloud; is that the representative situation for the spheres and is this similar in the PaCa cancer tissue? Does this have influence on the results?

    We have replaced this image with a more representative image that shows mouse stellate cells dispersed throughout the sphere.

    Figure 2: It is interesting to hear that BMP1, which is actually a ligand for BMP signaling is a protease for Collagen. How does this work?

    While the BMP family generally belong to the TGFβ superfamily, BMP1 is the exception in that it is a C-terminal collagenase. Please refer to reference 21 in the manuscript (PMID: 33879793), which details the role of BMP1 on collagen processing and the resulting effect on PDAC progression.

    C, Quantification of all blots should be presented.

    We have now included densitometry for all western blots, presenting values relative to the loading control and normalised to the experimental control. Values are averages taken from all biological repeats with significance indicated by stars.

    Figure S2: TS2 kd and TS14 kd should be confirmed and provided by both qrt PCR and WB data.

    We were unable to assess ADAMTS2 knockdown by western blot due to the quality of available antibodies. We are confident that either western or PCR confirmation of knockdown is sufficient, especially given the strong phenotype observed with the resulting knockdown.

    Figure 3: F; this result is arguing against the conclusion that TGFb bioavailability is a function of the ADAMs, since the kd impacts on the treatment result with exogenous TGFb. This suggests an effect downstream of ligand activation by proteasomal cleavage, e.g. receptor activation or signal transduction; this needs clarification. H, I: TGFβR inhibitor reduces TS14kd enhanced αSMA expression. How is unclear and needs clarification, since from F we know that already activated TGFβ needs TS2 to fully induce αSMA expression.

    SupplFig.3: B, C, as above!

    αSMA expression in stellate cells requires continuous exposure to TGFβ over 48 hours. Active TGFβ has an incredibly short half-life (minutes) and so requires positive feedback to maintain signalling. We propose that following ADAMTS2 knockdown the cells are incapable of releasing further TGFβ to maintain the phenotype. Equally following ADAMTS14 knockdown the cells are able to release more TGFβ, which is incapable of initiating signalling when the receptor is blocked.

    Figure 4: TIMP1 is a canonical TGFb signaling target gene in fibrosis. How the authors explain that TIMP1 is upregulated in both knockdowns, when they claim that TS2 and 14 have opposing functions on TGFb activation. This result as well puts their conclusions as regards TGFb and also the myofibroblast phenotype into question. Especially, since TIMP1 signifies stellate cell activation not only in the pancreas, but also in the liver and kidney. C, D, E should be explained in more detail and all details of the results should be presented.

    TIMP1 is a substrate for both ADAMTS2 and ADAMTS14, so its enrichment following knockdown of either is unsurprising, reflective of reduced cleavage of TIMP1. Both our 3D invasive assessment in Figure 6 and αSMA imaging in supplementary figure 5 demonstrate that TIMP1 is not responsible for the effect observed as a consequence from loss of either ADAMTS2 or ADAMTS14.

    This holds also for the different myofibroblast phenotypes. All data should be included. From recent scRNASeq investigations, several myofibroblast populations were described and compared, e.g my-stellate cells vs i-stellate cells. To which of these phenotypes the identified populations belong?

    As mentioned above, we have interrogated publically available data sets and identified ADAMTS2 and ADAMTS14 expression in multiple CAF subtypes. As these proteases are secreted it is probable that one CAF subtype can control the phenotype of surrounding CAFs through ADAMTS2 and ADAMTS14 production. While intriguing, this hypotheses is beyond the scope of the current work.

    Figure 5: C, Only brightfield images are provided, confocal images are suggested for comparison of +/- Aprotinin treatment.

    We do not think the addition of confocal images will add to the comparison. Aprotinin clearly reduces invasion, which coupled with the action of stellate-derived SERPINE2 on invasion, and reduced plasmin activity following ADAMTS2 knockdown, suggests that plasmin is important for regulating the effects of ADAMTS2 on invasion.

    The efficiency of TS2 and Serpine2 kd should be provided by qrt PCR and WB.

    TS2 kd promoted SERPINE2 expression should also be presented by qrt PCR and WB.

    We are confident that either western or PCR confirmation of knockdown is sufficient. Of note is that following ADAMTS2 knockdown, SERPINE2 expression is unchanged (sup figure 4C). This would indicate that the enrichment of SERPINE2 observed in the matrisome following loss of ADAMTS2 is reflective of reduced cleavage, rather than a change in expression.

    Figure 6: A, why ta use aSMA and not invasive activity as a readout here?

    Increased αSMA expression following ADAMTS14 knockdown provides a strong, clear, 2D phenotype to act as a readout for an siRNA screen with high-content imaging. Performing such a screen with our 3D invasive model is currently impractical.

    There are many parameters leading to decreased aSMA expression upon kd; (1) why only MMP1 and Fibulin were selected as candidates?

    From our αSMA screen, MMP1 and Fibulin2 knockdown were the only candidates that were able to both prevent an increase in αSMA seen with ADAMTS14 loss alone, and are known ADAMTS14 substrates. Further validation in our 3D invasive model demonstrated that Fibulin2 and not MMP1 was responsible for the effect of ADAMTS14 loss on invasion.

    (2) the single kd control of the screen candidates is missing!

    We feel this control is not needed, as the goal of the experiment was to establish which candidate was responsible for mediating the effects brought about by ADAMTS14 knockdown. Increased αSMA expression with IL-1β loss validates our approach, as this is a known negative regulator of TGFβ signalling.

    (3) Can it be expected that all these matrisomal proteins are involved in aSMA expression regulation? I have doubts.

    We agree with the reviewers comment, from the siRNA screen (sup figure 5B) it is clear that the majority of the identified matrisome proteins have a minimal effect on αSMA expression following loss of ADAMTS14.

    C, D, E, why MMP1 was not also tested in these assays?

    Our spheroid assay clearly demonstrated that invasion was enhanced following ADAMTS14 knockdown even with co-knockdown of MMP1. Given the strong rescue observed with co-knockdown of Fibulin2 we proceeded to further analyse this candidate over MMP1.

    F, Fibrillin is shown in the figure but not described in the text. It would be quite interesting to see whether Fibrillin kd has the same effect as TS14 kd on LTGF-β activation (which of course need to be shown experimentally).

    The association of fibrillin with TGFβ release is well established as it underpins the biology behind Marfan syndrome. Loss of fibrillin, or mutations to its TGFβ binding sites results in a phenotype consistent with super active TGFβ signalling.

    E, what is the meaning of αSMA intensity quantification? By IF staining of αSMA? PSC αSMA expression should be quantified by qrt PCR and WB.

    We have now incorporated the confocal images analysing αSMA expression into the main figure and labelled the quantification accordingly. We feel this improves the clarity of the figures. Every western blot is now presented with quantification.

    Also here, kd efficiency of TS14 and Fibulin2 should be provided by qrt PCR and WB.

    Figure S5E should be part of figure 6, qrt PCR of Fibulin2 should be added.

    We have moved this western blot to the main figure (Fig 7C). We feel additional PCR validation of Fibulin 2 knockdown is not necessary.

    Figure 5/6 and throughout: It is claimed that ADAMTS2 and ADAMTS14 regulate TGFβ bioavailability through SREPINE2-Plasmin and Fibulin2. As mentioned above, TGFβ activation is only mentioned in the schemes, but no experimental evidence is given. In addition, according to previous studies, ADAMTSs can activate latent TGFβ directly by interaction with the LAP of latent TGFβ. .

    We have now included extra experimental evidence to support an association of ADAMTS proteins with TGFβ bioavailability. Using a TGFβ luciferase reporter construct, we demonstrate that active TGFβ is increased following loss of ADAMTS14, which is abrogated with concomitant loss of Fibulin2. This provides further evidence that ADAMTS14 is mediating its effects on myofibroblast differentiation / invasion through TGFβ release.

    Figure 3B, C, and 6D: We are confused from the migration/invasion assays. Invasion should be based on migration of tumor cells, whereas in the migration assays only stellate cells seem to be active? Can you explain this to us? According to Figure 3B, stellate and cancer cells are cocultured in the chamber. Is this the same condition as for the experiment presented as figure 6D?

    In our migration assay, stellate and cancer cells are co-cultured in the apical chamber and cell migration imaged over time. We pooled data of both cancer and stellate cell migration following stellate specific knockdown of either ADAMTS2 or ADAMTS14, which showed an increase in cell migration following loss of ADAMTS14. In figure 7, we again use this assay to demonstrate that Fibulin2 expression accounts for the phenotype observed from loss of ADAMTS14.

    In summary, this study for the first time found that ADAMTS2 and ADAMTS14 have opposite roles on myofibroblast differentiation, which is shown by using chimeric spheroids of stellate and pancreatic cancer cells. The authors claim a therapeutic potential for pancreatic cancer by regulating ADAMTS2/14-mediated stellate cell activation, which should avoid cancer cell invasion. The approach is interesting and there is preliminary evidence, however the study has many gaps and requires substantive workload.

    We thank the reviewer for their support of our findings. We hope the additional data, combined with the known role for these substrates in the regulation of TGFβ, strengthens the clarity of our manuscript.

  3. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #3

    Evidence, reproducibility and clarity

    AMADTS2 and ADAMTS14 belong to the disintegrin and metalloproteinase with thrombospondin motif protein family, mainly produced by pancreatic stellate cells (PSC) and are related to cancer cell invasion. This study reveals that ADAMTS2 and ADAMTS14 have opposite roles in myofibroblast differentiation based on experiments testing HSC driven cancer cell invasion, variant expression of HSC activation makers and the related downstream targets analysed upon RNA sequencing analyses. The authors (TA) established PSC/cancer cell chimeric spheroids for investigating the crosstalk between these cell types in 3D in vitro. Based on their findings, they claim that ADAMTS2 and ADAMTS14 have different functions regarding PSC and TGF-β activation. However, their conclusions mainly rely on quantitatve data of invasion and mechanistic details are completely lacking.

    Comments:

    Typos, even in the abstract, e.g. first sentence incomplete Introduction is rather sparce with one third of the text repeating the results of the study Some citations do not at all fit with the position where they are placed; needs approval

    In this study, it is said that TS2 (AMADTS2) and TS14 (ADAMTS14) have opposite functions on myofibroblast differentiation, with individual depletion leading to distinct matrisomal phenotypes in PSC. However, both similarly contribute to collagen processing. As we know, collagen is increased in response to TGFβ signaling, since TS2 depletion (knock down, kd) inhibits and TS14 kd is suggested to promote TGFβ activation, it is expected that this has impact on the available collagen levels. How do the authors explain that nevertheless the kd effect on collagen is very similar?

    The authors claim that TS2 facilitates TGFβ release and TS14 is a key regulator of TGFβ bioavailability. However, throughout the whole data, there is no experimental evidence for this conclusion. TGFβ activation, LAP concentration and downstream effects should be provided.

    Most of the conclusions in the manuscript are based on effects to invasion and the estimated quantification histograms. "Black boxes in between the treatment, e.g. knockdown and readout, that relate to the signals and mechanisms remain black boxes throughout. For example, the impact of the treatments on stellate cell activation markers, the cancer cells invasion signaling, the SERPINE2- and Fibulin2-dependent myofibroblast differentiation pathways should be mechanistically investigated.

    The authors investigate one cell line each for their conclusions; we know that different cell lines behave differently; can they confirm that the finding they present is of general validity or a finding that is specific for the tested cancer/PSC cell lines. Can the principle findings also be proven in primary cells. More importantly, the authors should proof their findings in PaCa tissue of patients as follows: Expression of the proteases in the tissue, related variation of matrisome signatures, e.g. by snRNASeq, to confirm relevance of the finding.

    Details related to the figures:

    Figure 1: Are the numbers of PSC and PaCa cells integrated in the spheres related to the numbers found in patients? B, it seems that the PSC in the spheroid are not equally distributed but instead are all located in close vicinity to eachother in a cloud; is that the representative situation for the spheres and is this similar in the PaCa cancer tissue? Does this have influence on the results?

    Figure 2: It is interesting to hear that BMP1, which is actually a ligand for BMP signaling is a protease for Collagen. How does this work? C, Quantification of all blots should be presented.

    Figure S2: TS2 kd and TS14 kd should be confirmed and provided by both qrt PCR and WB data.

    Figure 3: F; this result is arguing against the conclusion that TGFb bioavailability is a function of the ADAMs, since the kd impacts on the treatment result with exogenous TGFb. This suggests an effect downstream of ligand activation by proteasomal cleavage, e.g. receptor activation or signal transduction; this needs clarification. H, I: TGFβR inhibitor reduces TS14kd enhanced αSMA expression. How is unclear and needs clarification, since from F we know that already activated TGFβ needs TS2 to fully induce αSMA expression.

    SupplFig.3: B, C, as above!

    Figure 4: TIMP1 is a canonical TGFb signaling target gene in fibrosis. How the authors explain that TIMP1 is upregulated in both knockdowns, when they claim that TS2 and 14 have opposing functions on TGFb activation. This result as well puts their conclusions as regards TGFb and also the myofibroblast phenotype into question. Especially, since TIMP1 signifies stellate cell activation not only in the pancreas, but also in the liver and kidney. C, D, E should be explained in more detail and all details of the results should be presented. This holds also for the different myofibroblast phenotypes. All data should be included. From recent scRNASeq investigations, several myofibroblast populations were described and compared, e.g my-stellate cells vs i-stellate cells. To which of these phenotypes the identified populations belong?

    Figure 5: C, Only brightfield images are provided, confocal images are suggested for comparison of +/- Aprotinin treatment. The efficiency of TS2 and Serpine2 kd should be provided by qrt PCR and WB. TS2 kd promoted SERPINE2 expression should also be presented by qrt PCR and WB.

    Figure 6: A, why ta use aSMA and not invasive activity as a readout here? There are many parameters leading to decreased aSMA expression upon kd; (1) why only MMP1 and Fibulin were selected as candidates? (2) the single kd control of the screen candidates is missing! (3) Can it be expected that all these matrisomal proteins are involved in aSMA expression regulation? I have doubts. C, D, E, why MMP1 was not also tested in these assays? F, Fibrillin is shown in the figure but not described in the text. It would be quite interesting to see whether Fibrillin kd has the same effect as TS14 kd on LTGF-β activation (which of course need to be shown experimentally). E, what is the meaning of αSMA intensity quantification? By IF staining of αSMA? PSC αSMA expression should be quantified by qrt PCR and WB. Also here, kd efficiency of TS14 and Fibulin2 should be provided by qrt PCR and WB.

    Figure S5E should be part of figure 6, qrt PCR of Fibulin2 should be added.

    Figure 5/6 and throughout: It is claimed that ADAMTS2 and ADAMTS14 regulate TGFβ bioavailability through SREPINE2-Plasmin and Fibulin2. As mentioned above, TGFβ activation is only mentioned in the schemes, but no experimental evidence is given. In addition, according to previous studies, ADAMTSs can activate latent TGFβ directly by interaction with the LAP of latent TGFβ. . Figure 3B, C, and 6D: We are confused from the migration/invasion assays. Invasion should be based on migration of tumor cells, whereas in the migration assays only stellate cells seem to be active? Can you explain this to us? According to Figure 3B, stellate and cancer cells are cocultured in the chamber. Is this the same condition as for the experiment presented as figure 6D?

    In summary, this study for the first time found that ADAMTS2 and ADAMTS14 have opposite roles on myofibroblast differentiation, which is shown by using chimeric spheroids of stellate and pancreatic cancer cells. The authors claim a therapeutic potential for pancreatic cancer by regulating ADAMTS2/14-mediated stellate cell activation, which should avoid cancer cell invasion. The approach is interesting and there is preliminary evidence, however the study has many gaps and requires substantive workload.

    Significance

    AMADTS2 and ADAMTS14 belong to the disintegrin and metalloproteinase with thrombospondin motif protein family, mainly produced by pancreatic stellate cells (PSC) and are related to cancer cell invasion. This study reveals that ADAMTS2 and ADAMTS14 have opposite roles in myofibroblast differentiation based on experiments testing HSC driven cancer cell invasion, variant expression of HSC activation makers and the related downstream targets analysed upon RNA sequencing analyses.

    The authors (TA) established PSC/cancer cell chimeric spheroids for investigating the crosstalk between these cell types in 3D in vitro. Based on their findings, they claim that ADAMTS2 and ADAMTS14 have different functions regarding PSC and TGF-β activation. However, their conclusions mainly rely on quantitatve data of invasion and mechanistic details are completely lacking.

    Comments:

    Typos, even in the abstract, e.g. first sentence incomplete Introduction is rather sparce with one third of the text repeating the results of the study Some citations do not at all fit with the position where they are placed; needs approval

    In this study, it is said that TS2 (AMADTS2) and TS14 (ADAMTS14) have opposite functions on myofibroblast differentiation, with individual depletion leading to distinct matrisomal phenotypes in PSC. However, both similarly contribute to collagen processing. As we know, collagen is increased in response to TGFβ signaling, since TS2 depletion (knock down, kd) inhibits and TS14 kd is suggested to promote TGFβ activation, it is expected that this has impact on the available collagen levels. How do the authors explain that nevertheless the kd effect on collagen is very similar?

    The authors claim that TS2 facilitates TGFβ release and TS14 is a key regulator of TGFβ bioavailability. However, throughout the whole data, there is no experimental evidence for this conclusion. TGFβ activation, LAP concentration and downstream effects should be provided.

    Most of the conclusions in the manuscript are based on effects to invasion and the estimated quantification histograms. "Black boxes in between the treatment, e.g. knockdown and readout, that relate to the signals and mechanisms remain black boxes throughout. For example, the impact of the treatments on stellate cell activation markers, the cancer cells invasion signaling, the SERPINE2- and Fibulin2-dependent myofibroblast differentiation pathways should be mechanistically investigated.

    The authors investigate one cell line each for their conclusions; we know that different cell lines behave differently; can they confirm that the finding they present is of general validity or a finding that is specific for the tested cancer/PSC cell lines. Can the principle findings also be proven in primary cells. More importantly, the authors should proof their findings in PaCa tissue of patients as follows: Expression of the proteases in the tissue, related variation of matrisome signatures, e.g. by snRNASeq, to confirm relevance of the finding.

    Details related to the figures:

    Figure 1: Are the numbers of PSC and PaCa cells integrated in the spheres related to the numbers found in patients? B, it seems that the PSC in the spheroid are not equally distributed but instead are all located in close vicinity to eachother in a cloud; is that the representative situation for the spheres and is this similar in the PaCa cancer tissue? Does this have influence on the results?

    Figure 2: It is interesting to hear that BMP1, which is actually a ligand for BMP signaling is a protease for Collagen. How does this work? C, Quantification of all blots should be presented.

    Figure S2: TS2 kd and TS14 kd should be confirmed and provided by both qrt PCR and WB data.

    Figure 3: F; this result is arguing against the conclusion that TGFb bioavailability is a function of the ADAMs, since the kd impacts on the treatment result with exogenous TGFb. This suggests an effect downstream of ligand activation by proteasomal cleavage, e.g. receptor activation or signal transduction; this needs clarification. H, I: TGFβR inhibitor reduces TS14kd enhanced αSMA expression. How is unclear and needs clarification, since from F we know that already activated TGFβ needs TS2 to fully induce αSMA expression.

    SupplFig.3: B, C, as above!

    Figure 4: TIMP1 is a canonical TGFb signaling target gene in fibrosis. How the authors explain that TIMP1 is upregulated in both knockdowns, when they claim that TS2 and 14 have opposing functions on TGFb activation. This result as well puts their conclusions as regards TGFb and also the myofibroblast phenotype into question. Especially, since TIMP1 signifies stellate cell activation not only in the pancreas, but also in the liver and kidney. C, D, E should be explained in more detail and all details of the results should be presented. This holds also for the different myofibroblast phenotypes. All data should be included. From recent scRNASeq investigations, several myofibroblast populations were described and compared, e.g my-stellate cells vs i-stellate cells. To which of these phenotypes the identified populations belong?

    Figure 5: C, Only brightfield images are provided, confocal images are suggested for comparison of +/- Aprotinin treatment. The efficiency of TS2 and Serpine2 kd should be provided by qrt PCR and WB. TS2 kd promoted SERPINE2 expression should also be presented by qrt PCR and WB.

    Figure 6: A, why ta use aSMA and not invasive activity as a readout here? There are many parameters leading to decreased aSMA expression upon kd; (1) why only MMP1 and Fibulin were selected as candidates? (2) the single kd control of the screen candidates is missing! (3) Can it be expected that all these matrisomal proteins are involved in aSMA expression regulation? I have doubts. C, D, E, why MMP1 was not also tested in these assays? F, Fibrillin is shown in the figure but not described in the text. It would be quite interesting to see whether Fibrillin kd has the same effect as TS14 kd on LTGF-β activation (which of course need to be shown experimentally). E, what is the meaning of αSMA intensity quantification? By IF staining of αSMA? PSC αSMA expression should be quantified by qrt PCR and WB. Also here, kd efficiency of TS14 and Fibulin2 should be provided by qrt PCR and WB.

    Figure S5E should be part of figure 6, qrt PCR of Fibulin2 should be added.

    Figure 5/6 and throughout: It is claimed that ADAMTS2 and ADAMTS14 regulate TGFβ bioavailability through SREPINE2-Plasmin and Fibulin2. As mentioned above, TGFβ activation is only mentioned in the schemes, but no experimental evidence is given. In addition, according to previous studies, ADAMTSs can activate latent TGFβ directly by interaction with the LAP of latent TGFβ. .

    Figure 3B, C, and 6D: We are confused from the migration/invasion assays. Invasion should be based on migration of tumor cells, whereas in the migration assays only stellate cells seem to be active? Can you explain this to us? According to Figure 3B, stellate and cancer cells are cocultured in the chamber. Is this the same condition as for the experiment presented as figure 6D?

    In summary, this study for the first time found that ADAMTS2 and ADAMTS14 have opposite roles on myofibroblast differentiation, which is shown by using chimeric spheroids of stellate and pancreatic cancer cells. The authors claim a therapeutic potential for pancreatic cancer by regulating ADAMTS2/14-mediated stellate cell activation, which should avoid cancer cell invasion. The approach is interesting and there is preliminary evidence, however the study has many gaps and requires substantive workload.

  4. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #2

    Evidence, reproducibility and clarity

    The manuscript by Carter and colleagues examines that role of cancer-associated fibroblasts (CAFs) in regulation of invasion in a 3D co-culture assay with epithelial cells. The authors propose that invasive chains of cancer cells are led by fibroblasts. The authors utilise a system of co-culture to create chimeric human-mouse fibroblast-cancer cell spheroids (both directions utilised, to eliminate species bias) to allow for in situ sequencing of the co-operating transcriptional programmes of each cell type during 3D invasion. From this powerful approach, this allowed the authors to identify two key two collagen-processing enzymes, ADAMTS2 and ADAMTS14, as contributing to CAF function in their system. The authors identify that these two enzymes have opposing roles in invasion, and map some of their key substrates in invasion, which extend beyond collagen-processing. The authors propose that one key function is to control the processing of TGFbeta, the latter of which is a regulator of the myofibroblast subpopulation of CAF. Overall, the findings of the manuscript are interesting, but need some further proof-of-principal demonstrations to extend their findings to support the claims within.

    • The authors demonstrate a clear role of ADAMTS2 and ADAMTS14 in stellate function during differentiation and invasion. Is there any evidence of such changes in patient materials? Could the authors query publicly available databases of micro-dissected stroma vs epithelium to validate the translational relevance of their findings?

    Major comments:

    • Page no. 4, Line 71, The authors conclude that the invasion in the chimeric spheroids is "led by" stellate cells. This is a key concept in the manuscript. How do the authors define the "led by" phenomena? What is the frequency that this occurs?

    • For Figure 2A and S2A, the text suggests that the heatmap represents the stellate vs cancer cell expression (as shown in Figure 1B and S1B) in the respective species but the labelling below the heatmap suggests they are all cancer cells (Mia, Pan, R2 and DT). Is this a typo? Could the authors clarify this?

    • The text and the figures are lacking information about the cell line names used in the experiment, e.g, Figure 2C, 2D, 2SB, 2SC and 2SD does not indicate what cell line was used in the study. This is the same with other figures as well. Please indicate in all instances exactly which samples are queried.

    • It's mentioned in the text that the authors have used the cancer and stellate cells in a 1:2 ratio but the numbers of stellate cells look different between different spheroids confocal images. e.g. The numbers look very different between the Miapaca2:PS1 vs Miapaca2:mPSC spheroids. Is this simply the representative images, or are their bona fide differences. This, in turn, would impact on claims of cells being 'led' by stellate cells. Can the authors clarify?

    • While for most of the experiments the authors generated the chimeric spheroids first and then performed the respective experiments, it appears that for the invasion assay simply co-culture of Cancer cells and stellate cells was done. Is this correct? Have the authors tried performing the assay with the chimeric spheroids to see if the stellate cells still invade?

    • The authors claim that ADAMTS2 and ADAMTS14 regulate the bioavailability of TGFB, and this is a key reason that these regulate CAF differentiation. However, there is no direct demonstration of this concept, which is conspicuous by absence. Could the authors either directly demonstrate this, or remove such notions from the results, and explicitly state that this is an untested speculation in discussion? Examples of this are:

      • Line 173, authors state "ADAMTS2 facilitates TGFβ release through degradation of the plasmin inhibitor, SERPINE2 (Figure 5D)"
      • Line 196 authors conclude "Together these data implicate 197 ADAMTS14 as a key regulator of TGFβ bioavailability (Figure 6F)."
      • Line 240 states "This reduces the activation of Plasmin, preventing the release of TGFβ (Figure 5C)." Since this is just a model without detailed experiments, It will be better to propose rather than conclude.

    • Figure S5C shows a less invasive phenotype in the NTCsi + ADAMTS14si spheroids compared to the NTCsi + NTCsi control. However, there appears no appreciable difference between NTCsi + ADAMTS14si and NTCsi + NTCsi spheroids' brightfield images in Figure 5SD. Could the authors comment on this?

    Minor Comments:

    • Page no. 2, Line 20 has an incomplete sentence "Crosstalk between cancer and stellate cells is pivotal in pancreatic cancer, resulting in differentiation 21 of stellate cells into myofibroblasts that drive."
    • Figure 2C; Figure S2C and Figure S5E lack quantification for the western blots.
    • Why did the authors choose to investigate the Metzincin family? Could the authors provide their reasoning to investigate these proteins, to the exclusion of other candidates?
    • Info about the number of fields imaged per sample for the microscopy data is missing in the figure legends (e.g. Figure 2F and 2I, Figure 5SF).
    • Any particular reason why the ADAMTS2 expression was not checked through Western blotting like ADAMTS14 in Figure S2B.
    • The legends for Figure 3SC and 5SF mention that "Images are representative of at least two biological replicates". How many technical replicates were used? It would be useful if the relative intensity of the images is measured and plotted in a graph.

    Significance

    This work provides an examination of the cross talk between fibroblasts and cancer cells in a 3-Dimensional culture model of pancreatic tumour cell invasion. By using chimeric human-mouse spheroids, the authors are able to identify cell-type specific transcripts by bulk RNA sequencing in situ. This advance is not to be underestimated as a number of existing approaches for cell type-specific profiling (eg. single-cell sequencing) relies upon dissociation of cell communities prior to sequencing. It is very likely that transcriptional programmes change during this isolation process. This approach allows the authors to identify transcriptional co-operating programmes in situ. This data provides a resource to understand this key co-operation of these two cell types during tumourigenesis, and will be of interest to the pancreatic cancer field. In addition, the mapping of the key substrate of these enzymes provides further insights that may be useful in understanding the expanded target repertoire of these enzymes beyond collagen processing.

  5. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    This study aims to explain the opposing contributions of stromal stellate cells/CAFs to PDAC. By first identifying stroma-specific proteases, followed by a process of candidate selection and elimination, the authors find that two specific metalloproteases that share enzymatic activity against collagen in fact have differential activity on TGFb availability. This could be interpreted as a way of shaping the CAF population and tumor-promotin or -restricting properties of the stroma.

    There are several flaws that the authors could address to improve the manuscript:

    1. In the flow of experiments and analyses, there is a strange mix of fully unbiased discovery phases followed by functional experiments that do not consider all possible candidates to test, and vice versa. For instance, from the mixed-species transcript analysis, ADAMTS2 and -14 are chosen based on their shared collagenase activity based on literature. However, the authors then perform again a proteomics analysis to identify things from the entire matrisome that are cleaved by these enzymes? Then, for ADAMTS2 a co-silencing approach is done on one selected candidate (Serpine2), but for ADAMTS14 an siRNA screen is performed? The problem of this approach is that the rationale for some studied enzymes is very strong, where as for others it is not.
    2. The ECM is more than just collagen. Choosing these two metalloproteases based on their shared collagen substrate is an approach that perhaps oversimplifies the ECM a bit, and again, does not provide the strongest rationale that these metalloproteases are most likely to explain counteracting stromal activities on tumor growth and progression.
    3. Related to the above: How were the stellate cells used for the matrisome analysis grown? In the suspension setup or adherent? This will have a large impact on the outcome. Is there for instance hyaluronic acid in this matrix?
    4. Performing the species-specific transcript analysis both ways is a neat approach, but why did the authors ignore the opportunity to formally overlay/compare the two stromal gene sets to define likely candidates based on statistics?

    Minor comments:

    The bioinformatics Methods need more details on how reads were mapped to the different genomes. How many mismatches were allowed and was the mapping done separately or using for instance Xenofilter?

    The authors use the knowledge on the activities of both ADAMTS2 and -14 on collagen as a rationale to choose these two. Is there really a need for the paragraph (and associated figures) from line 102 on?

    Abstract, line 21; some words are missing?

    Were the siRNA screen hits validated?

    What is the genotype of the mouse cancer cells? KPC-derived?

    Significance

    The trick of dissecting tumor from stromal signals in spheroid cocultures by RNA-Seq is a cool trick, but not new and the authors should probably cite some prior work.

    What this all means for patients (or in vivo tumors even) remains unclear. There is some debate on whether highly activated CAFs (ACTA2/aSMA+ cells, some call them myCAFs) are indeed tumor-restrictive or whether they promote invasion. The authors appear to argue the latter (which I can agree with) but without any translational work to show what the net outcome of this mechanism is, the study remains descriptive and perhaps of limited interest.

  6. Version published to 10.1101/2022.09.20.508645v1 on bioRxiv