Group I Paks are essential for epithelial- mesenchymal transition in an Apc-driven model of colorectal cancer

p21-activated kinases (Paks) play an important role in oncogenic signaling pathways and have been considered as potential therapeutic targets in various cancers. Most studies of Pak function employ gene knock-out or knock-down methods, but these approaches result in loss of both enzymatic and scaffolding properties of these proteins, and thus may not reflect the effects of small molecule inhibitors. Here we use a transgenic mouse model in which a specific peptide inhibitor of Group I Paks is conditionally expressed in response to Cre recombinase. Using this model, we show that inhibition of endogenous Paks impedes the transition of adenoma to carcinoma in an Apc-driven mouse model of colorectal cancer. These effects are mediated by inhibition of Wnt signaling through reduced β-catenin activity as well as suppression of an epithelial-mesenchymal transition program mediated by miR-200 and Snai1. These results highlight the potential therapeutic role of Pak1 inhibitors in colorectal cancer.

Available adenoma tissues should be analyzed for in situ: 1) Epithelial and mesenchymal markers to substantiate the conclusions on EMT; 2) GST, to verify GST-PID expression in the tumors; 3) Ki67 and Caspase-3 for the effects on cell growth in vivo 4) miR200; ESPR1; phospho-Snai; and other key signaling events detected in vitro to verify that these signaling events are affected by PID in tumors. 6. Authors claim that an EMT program is mediated by miR200; alternative CD44 splicing; the splicing factor ESPR1; and phosphorylation of Snai. However, the mediating function is only examined for miR200, while claims about other mediating factors need to be supported by experimental evidence testing directly their function as mediators of EMT in the used models. 7. The legend for F1C is missing. 8. F1F lacks blotting for total Pak1 and also lacks quantification. 9. F3A claims to display cell viability, although the actual analysis is of the total number of mitochondria in the cell population (MTT). To measure viability, authors will need to analyze apoptosis and other potential causes of cell death. 10. All immunoblots (and qPCR gels) need to be properly quantified (based on exposures within the quantitative linear range) and each be based on at least three independent experiments. 11. All the tissue stainings need to be quantified, both for the morphology and in particular the markers (signaling, etc.) labelled by antibodies. 12. All the cell stainings (e.g. SF5E) must be also be properly quantified. 12. FS4. A more informative heat map should be provided and identify the DE genes. Also, an Ontology enrichment analysis of DE genes should be presented.
Reviewer #2 (Remarks to the Author): In this manuscript, the authors investigate the effect of inhibiting Pak-1 in a mouse model using an inducible form of a peptide that constitutes the auto-inhibitory domain of Pak-1. They show that expressing this peptide specifically in the distal intestinal and colonic epithelium decreases the severity of the tumour phenotype of Apc mutant mice.
Although the manuscript is overall well written and the data presented clearly, I have a number of reservations/queries. If Pak-1 is required for activation of Wnt signalling (as stated in the introduction) then its inhibition should be deleterious for normal tissue homeostasis in the intestinal crypt. However, that is not what is observed, challenging the idea that Pak-1 is indeed required for Wnt signalling (which is crucial for normal crypt homeostasis). This requires an explanation and also determining if the expression of the PID causes any changes in the normal epithelium in greater detail. For instance, is PID expressed in ALL cells when induced?
The statement that APC-null cells assume stem cell characteristics (line 217/218) is not accurate. They behave more like transit amplifying cells in that they continually divide. Normal stem cells cycle slowly, not rapidly.
Has it been established how scaffolding functions of Pak-1 are affected by the PID?
The nature of the mutant Apc mouse is not explained and only referenced late in the text. This is not helped by the nomenclature changing from APC/+ to APC null, to APCloxP/+. That means the nature of this particular model is difficult to know. I assume it is the same mouse as described in Hinoi et al 2007, but this needs to be made much clearer early on. Particularly, since in most APC mutant mouse models, tumours arise predominantly in the small intestine. In the model used here, tumours seem to arise mostly in the colon. This needs to be explained much better to allow comparison to existing studies.
Invasiveness and motility are mentioned in the results (Line 179/180) but in the methods only 'motility' assays are described.
The conclusion that Pak-1 directly regulates Snai1 is no supported by the data. Lack of active Pak1 correlated with decreased Snai1 phosphorylation, that does not mean Pak-1 is the direct kinase.
What is the explanation for the cohort that completely lacks tumours?
Which animals were used for the analysis shown in Figures 3-5? Those without any tumours? That should be stated explicitly. Would tissue from APC/PID mice that did develop tumours yield different results?
In summary, although the findings are interesting and indicate a key function of Pak-1 in tumour progression in an APC mutant model, there are many details that need to be explained and/or considered to allow better understanding of the results and their implication. We appreciate these comments, but had to take a different approach than that suggested by the reviewer to address this important concern. The mouse model used in our manuscript is not well-suited for preclinical drug studies, because the CDX2P-Cre;Apcf lox / + mice take about 10 months to develop CRC (and they don't all do so synchronously), at which point, one would need to start treating relatively large cohorts for at least several weeks. Even then, I'm not sure there would enough statistical power to prove a point. Investigators have indeed used CRC GEMM models for preclinical studies, but such experiments are most conveniently done in a more synchronized model such as CDX2P-CreER T2 ;Apc fl/fl mice, in which both alleles of Apc are deleted in the colon upon tamoxifen administration. Such mice rapidly develop CRC and have been used to test the antiproliferative effects of drugs such as rapamycin (Hardiman et al., PLoS One, e96023, 2014). The recently published melanoma studies cited by the reviewer did not face such steep logistical hurdles as we would face using our CDX2P-Cre;Apcf lox / + mice.

Reviewer
Instead, in the revised manuscript we used a xenograft approach, which has the virtue of displaying relatively synchronous, observable tumor growth. We realize that the xenograft approach has its own limitations, but believe that they get at the issue of treating established disease with Pak inhibitors, as might be encounted in CRC patients. We hope the reviewer accepts the need for and validity of this approach. We agree with the reviewer and now include the suggested experiments to rule out significant contributions of the GST moiety and also non-Pak related interactions. These data are presented in the Results section and shown in Supplementary Fig. S5A and S5B.

The expression of the Pak inhibitory peptide (PID) is present already at the start of tumor development. A very interesting observation is that stem cell markers were downregulated in isolated cells (SF4B)
. This suggests that the observed colon cancer prevention may be due to depletion of stem cells from the colon (and/or inhibition of stem cell properties), thereby depriving the tissue from tumor originating cells. However, this finding appears to be underemphasized and insufficiently analyzed in vivo. To better clarify this, the authors should stain the colon tissues (normal, adenoma and adenocarcinoma tissues from wt and PID expressing mice) for these stem cell markers.
We now present data regarding stem cell markers in the colon tissues. These IHC data are presented in Supplementary Fig. S7C. They show a reduction in Msi1 (and possibly Bmi1) expression in CRC tissue from mice expressing the PID. . Also, long-term treatment with Wnt pathway inhibitors such as axitinib was reported to block tumor formation in cancer cells, zebrafish, and Apc min/+ mice, but did not affect intestinal homeostasis (Qu et al., PNAS 113:9339, 2016). These points are made explicit in the revised discussion.

The mechanistic efforts rely heavily on cell culture, while tissues remain under analyzed. Available adenoma tissues should be analyzed for in situ: 1) Epithelial and mesenchymal markers to substantiate the conclusions on EMT; 2) GST, to verify GST-PID expression in the tumors; 3) Ki67 and Caspase-3 for the effects on cell growth in vivo 4) miR200; ESPR1; phospho-Snai; and other key signaling events detected in vitro to verify that these signaling events are affected by PID in tumors.
We now present extensive IHC data on these tumors. In some cases, IHC-grade antibodies were not currently available (e.g., anti p-Snail) or did not stain properly (anti-GST), but we were able to obtain data for markers in each category, including Msi, Bmi1, ESRP1, E-Cadherin, and Caspase 3). These data are presented in Supplementary Fig. S7. 6. Authors claim that an EMT program is mediated by miR200; alternative CD44 splicing; the splicing factor ESPR1; and phosphorylation of Snai. However, the mediating function is only examined for miR200, while claims about other mediating factors need to be supported by experimental evidence testing directly their function as mediators of EMT in the used models.
We now show that manipulating ESRP levels with shRNA leads to corresponding changes in CD44 splicing (Fig. 5E). We also tried a similar functional analysis of CD44 by expressing CD44v in PID-cells (using a vector from Toru Hiraga), but this vector failed to alter CD44v levels. We could not source any other CD44v expression vectors.

The legend for F1C is missing.
This issue has been fixed.

F1F lacks blotting for total Pak1 and also lacks quantification.
This issue has been fixed.

F3A claims to display cell viability, although the actual analysis is of the total number of mitochondria in the cell population (MTT). To measure viability, authors will need to analyze apoptosis and other potential causes of cell death.
We now present data regarding apoptosis in these colon cell lines and are presented in Supplementary Fig. S3B.

All immunoblots (and qPCR gels) need to be properly quantified (based on exposures within the quantitative linear range) and each be based on at least three independent experiments.
This issue has been addressed.

All the tissue stainings need to be quantified, both for the morphology and in particular the markers (signaling, etc.) labelled by antibodies.
This issue has been addressed.
11. All the cell stainings (e.g. SF5E) must be also be properly quantified.
This issue has been addressed.

FS4. A more informative heat map should be provided and identify the DE genes. Also, an Ontology enrichment analysis of DE genes should be presented.
The heat map has been updated and the DE genes identified. In addituon, we now present GO enrichment data.
This issue has been addressed.

If Pak-1 is required for activation of Wnt signalling (as stated in the introduction) then its inhibition should be deleterious for normal tissue homeostasis in the intestinal crypt. However, that is not what is observed, challenging the idea that Pak-1 is indeed required for Wnt signalling (which is crucial for normal crypt homeostasis). This requires an explanation and also determining if the expression of the PID causes any changes in the normal epithelium in greater detail. For instance, is PID expressed in ALL cells when induced?
See point #4 above. Unfortunately, we were not able to use IHC to document PID expression on a cell-by-cell basis. We believe that PID expression leaves enough residual Wnt signaling to support normal tissue homeostasis, but not enough for effective tumorigenesis, similar to the argument made by Fearon's group in Feng et al. (PLOS Genet. 11, 2015).

The statement that APC-null cells assume stem cell characteristics (line 217/218) is not accurate. They behave more like transit amplifying cells in that they continually divide. Normal stem cells cycle slowly, not rapidly.
We have edited the text to address this point.

Has it been established how scaffolding functions of Pak-1 are affected by the PID?
That is an interesting point. In truth, the answer is "not really." It is assumed that binding to SH3-domain containing proteins like PIX or other binders such as PP2A and/or POPX are unaffected, but I don't think anyone has really looked into this issue. We added some discussion aroung this point in the revised manuscript. The preparation of the cell lysates was also not described, or at least I could not find it in the relevant section in the methods.
We have added the methodology for invasion along with motility in Materials and Methods and have now included methods for cell lysis.
One thing that is not really discussed, other than a mention in the introduction, is how the effects of PAK-1 on cytoskeletal dynamics may contribute to the observed effects. At least a small paragraph about this should be included in the discussion.
That is true. Some of the effects of Pak blockade may well be due to loss of cytoskeletal signaling. As we did not examine this possibility in detail in our experiments, we didn't discuss it in detail in the paper. However, we now include a few new sentences in the Discussion to address this important possibility. The reviewer raises a good point, and one that we had considered while the experiments were ongoing. The issue here is in obtaining a comparable data set. At the Jackson Lab Gene Expression Data (GXD) site (http://www.informatics.jax.org/expression.shtml), there are data regarding transcript expression from C57BL/6 mouse colonic epithelium, but these data are derived from cells that were obtained directly from the colon and lysed without culturing. In contrast, our PID+ and PID-cells were primary cultures from the adenomas formed in the Apc mouse model. In retrospect, perhaps we ought to have made an additional primary colonic epithelial line from normal mouse colonic epithelia, but those still wouldn't be from adenomas, as none would form in WT mice.