ROBO2 is a stroma suppressor gene in the pancreas and acts via TGF-β signalling

Whereas genomic aberrations in the SLIT-ROBO pathway are frequent in pancreatic ductal adenocarcinoma (PDAC), their function in the pancreas is unclear. Here we report that in pancreatitis and PDAC mouse models, epithelial Robo2 expression is lost while Robo1 expression becomes most prominent in the stroma. Cell cultures of mice with loss of epithelial Robo2 (Pdx1Cre;Robo2F/F) show increased activation of Robo1+ myofibroblasts and induction of TGF-β and Wnt pathways. During pancreatitis, Pdx1Cre;Robo2F/F mice present enhanced myofibroblast activation, collagen crosslinking, T-cell infiltration and tumorigenic immune markers. The TGF-β inhibitor galunisertib suppresses these effects. In PDAC patients, ROBO2 expression is overall low while ROBO1 is variably expressed in epithelium and high in stroma. ROBO2low;ROBO1high patients present the poorest survival. In conclusion, Robo2 acts non-autonomously as a stroma suppressor gene by restraining myofibroblast activation and T-cell infiltration. ROBO1/2 expression in PDAC patients may guide therapy with TGF-β inhibitors or other stroma /immune modulating agents.

#1 This point is difficult to see due to what may be compression artefacts in Figure 1a) ---------They use an ex vivo explant growth model to culture exocrine cell factors (similarly to a method previously described). This model is used with exocrine cells from PDX1C RE-Robo2flox/flox and control PDX1C RE animals. Gene expression changes are shown in Figure 2d,f. Robo2f/f explants show significantly more mesenchymal, stellate cells #2 The VIM color should be made consistent between panels 2c, 2e ---------The authors isolated epithelial (EpC AM+) and mesenchymal (C d140a/Pdgfr-a+) fractions from both types of explants, and show that the mesenchymal cells do not have a floxed Robo2 : from this, they conclude that the mesenchymal fraction is not derived from an epithelial origin. They conclude it must derive from further expansion of existing fibroblastic cells. Since these are not perturbed by Robo2, the Robo2 effect must be non-autonomous.
#3 Their flow cytometry markers should be shown alongside E-C ad, Vim and aSMA (or something subset of these). #4 The point in italics above seems to be a key element of the paper, but is not explained in sufficient clarity. A cartoon might be helpful.
-----------TGF-ß pathway inhibition blocks the effects of epithelial Robo2 loss. The TGF-ß pathway is implicated by correlative evidence in Figure 3. #5 The blot in 3g is not very convincing? (it might be fine given the experimental evidence that follows).
--------In Figure 4, the authors convincingly show that TGB1 receptor inhibition via galunisertib undoes the gene expression and cellular expansion phenotype seen from Robo2 loss.The authors investigate an C aerulein-treatment acute pancreatitis model, using their Robo2f/f mice.They show stromal expansion (increased interacinar space) at D3.
The authors claim that by D8, the tissue returned to baseline? #7 I'm not sure this is apparent in Figure 5a. (a relevant 'normal comparison' is in Supp Fig 2a).
------------They show that, like in vitro, galunisertib treatment could undo the Robo2-associated changes in the context of acute pancreatitis modelling. (Shown in Fig 5). The authors show that Slit1-/animals recapitulate key features of the Robo2f/f model. #8 The RNA in situ results in Supp Figure 9 are once again, difficult to interpret.
----------------Finally, the authors show human PDAC can be subclassified using Robo1 and Robo2 expression and that Robo1 expression, in the context of Robo2 loss has a substantially worse prognosis -as predicted by both IHC or RNA-Seq scoring.
A suggestion for the RNA fish images: Figure 1a-c : label the image with dashed lines: endo/acinar/duct?
• Overall, interesting results -impressive use of in vitro modelling to identify an in vivo intervention.
Reviewer #2: Remarks to the Author: In this manuscript from Pinho et al, the authors describe a function for ROBO2 as a stroma suppressor gene in the pancreas. The authors show that deletion of Robo2 leads to activation of Robo1 positive myofibroblasts and induction of TGF beta and Wnt pathways. They further present data associating the ROBO2low;ROBO1high phenotype as indicative for patients with the poorest survival. The in vitro data cell culture data are somewhat weak (see points below) and mostly pancreatitis related. Similar the animal model used is an acute pancreatitis model. Hence the authors discuss the role of Robo in context with pancreatic cancer. Therefore, a comparison of KC mice and KC -Robo2F/F mice would significantly strengthen the points suggested in this manuscript. Figure 1 a-c needs proper quantifications. How many mice per experimental group were analyzed? A quantification of relative expression between acinar cells and ductal cells in NMP, AP and KPC as well as relative increase in expression between NMP, AP and KPC needs to be included. In Figures 2d, 2f and 3a, 3c the authors compare the pdx1cre with the Robo2F/F culture (similar in 4a-4d). However, the cell types in each culture are quite different (with mainly ADM cells in pdx1cre and mainly fibroblasts in Robo2F/F). The results shown are somewhat trivial, since it is expected that if more fibroblasts are present in one of the culture conditions that the amount of fibroblast markers increases. Is there a difference in the expression of these genes in fibroblasts from either pdx1cre or Robo2F/F (after their separation from other cells in this ADM culture)? In Fig. 3d and 3f it is unclear if this is an IF picture from RoboF/F. if this is the case, then the control needs to be show, too. In addition, it is unclear if the quantifications shown are comparison of fibroblasts from pdx1cre or Robo2F/F (which would be the proper way to compare the role of switch in Robo isoforms in fibroblast expansion), or analyses of a heterogenous mix of cells. Figure 4e needs quantifications. For Figure 6e a table with the fold differences of other genes analyzed needs to be included in the supplemental data. In Figures 1-6 the authors describe a role of Robo1/2 switching for fibroblast expansion. In the tumor samples in Figure 7a none of the stromal areas express robo1 or 2. Instead both seem to only express in tumor cells. In my opinion this figure is completely disconnected to previous figures. Other points: Most of the figure legends are very superficially described. For example in Fig. 3d and 3f it is unclear if this is an IF picture from RoboF/F. Some of the figures are organized in a different order of how they are discussed in the text, which needs to be changed. It did not really become clear why the authors call the assay in Fig. 2a an ADM culture with "exocrine cell fractions with minimal contaminating fibroblasts". This actually does not look like a typical ADM assay, but rather as a typical assay that allows the outgrowth of fibroblasts. In several figures it is unclear what the size bars indicate.
Reviewer #3: Remarks to the Author: Pinho and collaborators have studied the role of Robo1 and Robo2 receptors in pancreaitc cancer using a combination of in vitro assay and in vivo models. Their data suggest that Robo1/2 might be involved in PDAC . However, at this stage the work is not a final product, and rather an heterogeneous collection of data, many of which too preliminary. I also think that the ratio of in vitro and in vivo data is not good and that the in vivo ones should be preponderant. Although I read the papet three times and I am still unsure I understood the message. Slit1, 2 and 3 bind similarly to Robo1 and Robo2, so if Robo2 is down and Robo1 is up why is there a problem? The authors should have provided some schematics summarizing the data and working model? How does this work?
To start with : C ould the authors present clearly where are Robo1 and Robo2, Slit1, Slit2 and Slit3 are expressed in the normal and pathological pancreas ? C ould they confirm Robo1 and Robo2 expression in acinar cells using immunostaing with commercially available anti-Robo antibodies (several have been validated and some were used for their human studies) ? This also applies to caerulein-induced pancreatitis. The in situ data are not very convincing including the colocalization of Robo2 with SMA+ cells (Fig S1b). Double immuno would help. Slit1, 2 and 3 were all previously detected in pancreatic islets (Yang et al PNAS) which does not really fit with the current data. As Slits function redundantly how removing a single Slit (Sli1) can cause defect? Likewise, the Robo1 KO exist and is viable. It should also be studied to support the conclusions.
They should also show Robo2 expression/level (blot or immuno) in pancreata cultures (control and Robo2F/F ; Figure 2) . What they currently show to support the presence of Robo2 in fibroblastic cells is rather indirect. The absence of C re expression should obviously be confirmed using cre immuno of preferentially a reported such as tdTomato.
The authors did not use an inducible line and in Pdx1-cre, and it is known that cre is expressed very early on in development (around E8.5) and therefore some of the observed defects could be related to abnormal pancreas development or cell specification. What supports the normal organization of the pancreas in Pdx1-cre ;Robo2lox mice ? The data supporting normal histology are very weak (only H&E). They should confirm that pancreatic function is normal (RTPC R is not enough) and quantify the histological data. A previous study (that the authors fully ignore) showed that Slit/Robo signaling controls beta cell survival (Yang et al., PNAS 110, pp 16480-16485, 2013) They should then discuss potential discrepancy with this earlier work.
Other points All the evidence supporting abnormal gene expression in Robo2FF animals with acute pancreatitis comes from RTPC R ( Figure S5). Figure S4 : why is Robo1 expression not upregulated after Robo2 silencing ? "Indicative of stromal changes, and an increase of Tgfb1, albeit only a trend (Fig.5b,c). "a tendency for increased C d3+ T lymphocyte infiltration at D3" (S. Fig. 11c,e), "Some typical Wnt target genes showed a tendency for increased expression, albeit not statistically significant" If some data are not significant this should be clearly stated rather than suggesting that there is a difference. And the picture on the figures should also reflect the results (on S11 it seems that there is a huge increase of C D3 cells) C ould they also validate the specificity of the Robo2 antibody (used for human immunostaining) on Robo2F/f mice ? Were they previously validated? the anti-Robo1 reference ab72790 is an anti DLST based on the abcam site. The Robo2 ab75014 is against the mouse protein and should work on mouse pancreatic tissue. Why don't they also showRobo1/2 distribution in normal adult human pancreas? It is otherwise difficult to know what Robo2 low or high mean.

We would like to thank the reviewers for their interest in our work and for their constructive comments and suggestions, as we believe they have led to an improved version of our manuscript.
Below we address the particular issues raised by each reviewer.

Reviewer #1:
This paper addresses the role of the Robo2 gene in pancreatic cancer using a mouse model. The main findings are noted below with comments immediately following.
The cell type expression of Robo1 and Robo2 is examined by RNA in situ. They conclude that Robo1 is upregulated in stromal cells, while Robo2 is lost in epithelial cells. Fig 1a shows that Robo1 and Robo2 are expressed in acinar cells, but rarely in duct cells.
#1 This point is difficult to see due to what may be compression artefacts in Figure 1a) Fig.1d.

We hope the reviewer can be provided with the non-compressed figures. We have made adjustments in Figure 1 to increase intelligibility and now clearly separate different tissue compartments using a dotted line. In addition, we now provide quantification of RNA expression per cell type for n=6 mice in
---------They use an ex vivo explant growth model to culture exocrine cell factors (similarly to a method previously described). This model is used with exocrine cells from PDX1CRE-Robo2flox/flox and control PDX1CRE animals. Gene expression changes are shown in Figure 2d,f. Robo2f/f explants show significantly more mesenchymal, stellate cells #2 The VIM color should be made consistent between panels 2c, 2e This change has now been made.
---------The authors isolated epithelial (EpCAM+) and mesenchymal (Cd140a/Pdgfr-a+) fractions from both types of explants, and show that the mesenchymal cells do not have a floxed Robo2 : from this, they conclude that the mesenchymal fraction is not derived from an epithelial origin. They conclude it must derive from further expansion of existing fibroblastic cells. Since these are not perturbed by Robo2, the Robo2 effect must be non-autonomous.
#3 Their flow cytometry markers should be shown alongside E-Cad, Vim and aSMA (or something subset of these).

These stainings have now been added in supplementary figure 3a, showing that EpCAM co-localizes with ECad and that Cd140a co-localizes with Vimentin. We have also added to the text the reference of a previous study (Ohlund at al. J Exp Med 2017), where the same antibodies have been used to separate the epithelial and mesenchymal cell population in pancreatic tissue.
#4 The point in italics above seems to be a key element of the paper, but is not explained in sufficient clarity. A cartoon might be helpful.

In the correspondence to us there was no text in italics, presumably because of the formatting by the journal. Nevertheless, we have added a cartoon (Fig. 8) that illustrates the key findings of our work.
-----------TGF-ß pathway inhibition blocks the effects of epithelial Robo2 loss. The TGF-ß pathway is implicated by correlative evidence in Figure 3.
#5 The blot in 3g is not very convincing? (it might be fine given the experimental evidence that follows). \

The Western Blot data of n=6 Pdx1 Cre versus n=6 Robo2 F/F samples were quantified (Fig. 3d). We acknowledge that there is variability in between the individual experiments which is now visible in the individual data points in the graph. A representative image of n=3 is shown.
----------Robo1 is up-regulated, possibly a compensatory mechanism? Figure 4: panel (b) is missing (but described in the legend).

We apologize for this oversight. The legend has now been corrected.
--------In Figure 4, the authors convincingly show that TGB1 receptor inhibition via galunisertib undoes the gene expression and cellular expansion phenotype seen from Robo2 loss.The authors investigate an Caerulein-treatment acute pancreatitis model, using their Robo2f/f mice.They show stromal expansion (increased interacinar space) at D3.The authors claim that by D8, the tissue returned to baseline? #7 I'm not sure this is apparent in Figure 5a. (a relevant 'normal comparison' is in Suppl. Fig 2a).
We now put in the text: 'At D8, the epithelial tissue had also regenerated to the same extent as the Pdx1 Cre samples (Fig. 5a) (Fig. 5h,i).'

------------
They show that, like in vitro, galunisertib treatment could undo the Robo2-associated changes in the context of acute pancreatitis modelling. (Shown in Fig 5). The authors show that Slit1-/-animals recapitulate key features of the Robo2f/f model.

#8
The RNA in situ results in Suppl. Figure 9 are once again, difficult to interpret. Sup. Fig. 10.

Same as for Fig.1, we hope the reviewer can be provided with the uncompressed figures. In addition, we now provide quantification for n=3 mice in
----------------Finally, the authors show human PDAC can be subclassified using Robo1 and Robo2 expression and that Robo1 expression, in the context of Robo2 loss has a substantially worse prognosis -as predicted by both IHC or RNA-Seq scoring.
A suggestion for the RNA fish images: Figure 1a-c : label the image with dashed lines

Dashed lines have been added, as suggested by the reviewer.
Overall, interesting results -impressive use of in vitro modelling to identify an in vivo intervention.

We are grateful to the reviewer for the enthusiasm about our findings. We hope that the reviewer is pleased with the way we have now addressed his/her comments.
--

Reviewer #2:
In this manuscript from Pinho et al, the authors describe a function for ROBO2 as a stroma suppressor gene in the pancreas. The authors show that deletion of Robo2 leads to activation of Robo1 positive myofibroblasts and induction of TGF beta and Wnt pathways. They further present data associating the ROBO2low;ROBO1high phenotype as indicative for patients with the poorest survival. The in vitro data cell culture data are somewhat weak (see points below) and mostly pancreatitis related. Similar the animal model used is an acute pancreatitis model. Hence the authors discuss the role of Robo in context with pancreatic cancer. Therefore, a comparison of KC mice and KC-Robo2F/F mice would significantly strengthen the points suggested in this manuscript.
We clarify the rationale for our choice of experimental models. In Fig.1

we show that Robo2 expression is downregulated in acute pancreatitis and in tumor samples of Kras mutant mice (KPC model) epithelial Robo2 expression is lost entirely. Acute pancreatitis and Kras mutant mice represent two conditions of increased Ras signaling. We confirmed with an additional experiment that overexpression of mutant KRAS G12D in the human pancreatic duct epithelial (HPDE) cell line abolishes ROBO2 expression (see figure below).
Western blot analysis of ROBO1, ROBO2 and Beta-actin protein expression in HPDE cells transfected with mutant KRas or control empty vector.
H P D E C a p a n -2 H P A C B x P c -3 P a n c -1 P a n c 1 0 . 0 5 P a n c 8 . 1 3 P a n c 5 . 0 4 A s P c -1 P a n c 3 . 2 7 P a n c 4 . 0 3 M i a P a C a -2 H P A F I I C a p a n -1 C F P a c S u 8 6 8 6 H s 7 0 0 T S W  (Fig. 1)

, we felt that additional Robo2 loss (by crossing with the Robo2 F/F line) might not result in a different outcome. This was indeed confirmed, as we show that adult KC pancreas presents a reduced level of Robo2 expression that is similar to that of the KC or KC_ Robo2 F/F pancreas (See figure below).
RT-qPCR analysis of Robo2 mRNA expression in adult pancreas of Pdx1 Cre , Robo2 F/F , KC and KC_Robo2 F/F animals. Data is represented as Mean +/-SEM. **p<0.01, ***p<0.001. Fig. 9

and we have incorporated a separate section on this in the main text: 'Concurrent loss of Robo2 and Kras G12D activation in the pancreas does not lead to enhanced stromal changes after chronic pancreatitis.
As shown in Fig. 1 Fig. 9b,c) Fig. 9d-h).

In conclusion, Robo2 knock-out in the KC mouse background does not increase the typical phenotypic changes in epithelium and stroma. One explanation for this observation is the fact that Kras G12D mutation on its own already induced Robo2 loss.' In the original manuscript, we had already stated in the discussion, 'it would be most interesting to sequentially introduce the genetic changes to better mimic human PDAC development.' This has now been confirmed. We now elaborate a bit further on this in the discussion: 'We found that Robo2 expression is decreased early in tumor development, in samples with oncogenic Kras activation (mouse KPC model and human PanIN lesions) and is overall low in PDAC samples. Hence effects of Robo2 loss on cancer development (through TGF-b signaling and stroma activation) might depend on the timing of Robo2 inactivation. Our experiments with embryonic inactivation of Robo2, mutant Kras activation (KC; Robo2 F/F ) followed by chronic pancreatitis showed no additional changes compared to controls (not shown). It would be most interesting to sequentially introduce the genetic changes to better mimic human PDAC development.'
Figure 1 a-c needs proper quantifications. How many mice per experimental group were analyzed? A quantification of relative expression between acinar cells and ductal cells in NMP, AP and KPC as well as relative increase in expression between NMP, AP and KPC needs to be included. Fig.1d.

We have added graphs with Robo1 and 2 mRNA quantification for n=6 mice in
In Figures 2d, 2f and 3a, 3c the authors compare the pdx1cre with the Robo2F/F culture (similar in 4a-4d). However, the cell types in each culture are quite different (with mainly ADM cells in pdx1cre and mainly fibroblasts in Robo2F/F). The results shown are somewhat trivial, since it is expected that if more fibroblasts are present in one of the culture conditions that the amount of fibroblast markers increases. Is there a difference in the expression of these genes in fibroblasts from either pdx1cre or Robo2F/F (after their separation from other cells in this ADM culture)?
In Fig. 3d and 3f it is unclear if this is an IF picture from RoboF/F. if this is the case, then the control needs to be show, too. In addition, it is unclear if the quantifications shown are comparison of fibroblasts from pdx1cre or Robo2F/F (which would be the proper way to compare the role of switch in Robo isoforms in fibroblast expansion), or analyses of a heterogenous mix of cells. Fig. 2g. Smad2) (Fig. 3e)

We added in the text that goes with Figure 2: 'We note that both in Pdx1 Cre as in Robo2 F/F the Vim + cells express aSma.' Indeed, the few fibroblasts in the Pdx1 Cre present expression of the same markers and Robo2 F/F mesenchymal cells (See figure below, for reviewer only).
Immunofluorescence/RISH of Axin2, Robo1, Tgfb1 and phospho-Smad2 mesenchymal cells from Pdx1 Cre cultures. Fig. 4 Fig. 4e,f).'

We would not call our observation 'trivial' since it is a unique phenotype that we describe in this manuscript. It is an important finding because of Robo2 knock-out being specific for the epithelial cells while the key outcome is the myofibroblast activation.
Figure 4e needs quantifications. Fig. 4 Fig. 4e,f)

.'
For Figure 6e a table with the fold differences of other genes analyzed needs to be included in the supplemental data. In Figures 1-6 the authors describe a role of Robo1/2 switching for fibroblast expansion. In the tumor samples in Figure 7a none of the stromal areas express robo1 or 2. Instead both seem to only express in tumor cells. In my opinion this figure is completely disconnected to previous figures.

Staining for ROBO1 can be observed both in tumor epithelium and stromal cells. We have now added dashed lines on the inset pictures to clearly demarcate tumor epithelium and stroma.
Other points: Most of the figure legends are very superficially described. For example in Fig. 3d and 3f it is unclear if this is an IF picture from RoboF/F. Some of the figures are organized in a different order of how they are discussed in the text, which needs to be changed.
It did not really become clear why the authors call the assay in Fig. 2a an ADM culture with "exocrine cell fractions with minimal contaminating fibroblasts". This actually does not look like a typical ADM assay, but rather as a typical assay that allows the outgrowth of fibroblasts. In several figures it is unclear what the size bars indicate.

We rephrased the introduction of the cell culture model and added more references. 'An established in vitro assay with highly enriched exocrine cell fractions mimics the epithelial cell changes (acinar to ductal metaplasia) that occur during pancreatitis 15-18 '. This is a published assay in which the exocrine cells undergo acinar to ductal metaplasia that was used in several publications and has been recapitulated even with human cells (references 15-18). It is not an assay meant to have fibroblast outgrowth. We actually never saw substantial fibroblast outgrowth with any of the genetically engineered mice that we cultured the epithelial cells from. This is why the observations in the Robo2 F/F were so striking.
--Reviewer #3 (Remarks to the Author): Pinho and collaborators have studied the role of Robo1 and Robo2 receptors in pancreaitc cancer using a combination of in vitro assay and in vivo models. Their data suggest that Robo1/2 might be involved in PDAC. However, at this stage the work is not a final product, and rather an heterogeneous collection of data, many of which too preliminary. I also think that the ratio of in vitro and in vivo data is not good and that the in vivo ones should be preponderant. Although I read the papet three times and I am still unsure I understood the message. Slit1, 2 and 3 bind similarly to Robo1 and Robo2, so if Robo2 is down and Robo1 is up why is there a problem? The authors should have provided some schematics summarizing the data and working model? How does this work?
It is unsettling to find out that we failed to convey the message of our findings to the reviewer. In the revised version of the manuscript, we introduced a schematic (Fig. 8)

We respectfully disagree with the reviewer that there is a preponderance of in vitro data. We have 3 main figures on in vitro experimental analysis, 2 on in vivo experimental analysis and 1 on patient sample analysis. In addition, there are 4 supplementary figures displaying in vitro data versus 9 supplementary figures with in vivo data.
We consciously showed a mix of experimental models to substantiate our findings (which was highly appreciated by reviewer 1). All results show that epithelial Robo2 expression decreases during pancreatitis and even more in cancer, two conditions of activated Ras signaling. We started off with a simple model of cell culture where a prompt Ras activation occurs (Pinho et al, Gut 2011). Because of the apparent phenotype, we validated these findings in vivo in the caerulein model. Since stromal cell activation is mainly a topic of investigation in pancreatic cancer, we included an analysis of one of the largest datasets of human pancreatic cancer samples with clinical annotation. Following the request from reviewer 2, we have now also added data of the KC mouse model of pancreatic cancer (Suppl. Fig. 9), albeit that the timing of the genetic changes in this model is not adequate, as we anticipated, and hence no difference between Robo2 F/F and controls was found. Instead of restarting long term breeding experiments with more suitable mouse models, we felt that the human data were sufficiently supporting our findings that they could be added in without being perceived as too disperse.
To start with : Could the authors present clearly where are Robo1 and Robo2, Slit1, Slit2 and Slit3 are expressed in the normal and pathological pancreas ?
We now provide in Fig.1 and Suppl. Fig. 10

a quantification of RNA expression analysis in acinar cells, duct cells and mesenchymal cells in normal and pathological pancreas (in both acute pancreatitis and PDAC models). Islets, which are outside the scope of our study, are shown for illustration in Suppl. Fig. 2.
Could they confirm Robo1 and Robo2 expression in acinar cells using immunostaing with commerciall available anti-Robo antibodies (several have been validated and some were used for their human studies) ? This also applies to caerulein-induced pancreatitis. The in situ data are not very convincing including the colocalization of Robo2 with SMA+ cells (Fig S1b). Double immuno would help.

Nevertheless, the same antibody for Robo2 failed to work on our mouse samples used throughout the manuscript, i.e. Western Blot did not show any bands around the reported molecular weight of approx. 175 kDa whereas the human ROBO2 protein could be easily detected (see first figure below, for reviewer only). Also, in the mouse 266-6 acinar cell line, no band could be detected at the predicted molecular weight (second figure below, panel c, for reviewer only) in contrast to Robo2 detection by RT-PCR in these samples (panel a, for reviewer only) or by RISH (panel b, for reviewer only).
HEK293T cells were transiently reverse-transfected with full length c-terminally tagged ROBO1 (Flag) or ROBO2 (Myc) expression vectors with lipofectamine 2000 (100000 cells/well, 24 well-plate format) according to the manufacturer's instructions. Empty vector was used as a negative control. 48 hours posttransfection cells were lysed and subjected to western blot analysis. Left panel: The observations for these analyses confirm specificity of the ROBO1 and ROBO2 antibodies. Anti-ROBO1 antibody detects low levels of endogenous ROBO1 protein as well (see long exposure). Right panel: Separate membranes loaded with HEK293T ROBO1 and ROBO2 transfectants were incubated with ROBO1 and ROBO2 antibodies. ROBO1 and ROBO2 antibodies did not exhibit cross-reactivity for ROBO2 and ROBO1 proteins. No band is detected at the molecular weight of the positive control Slit1, 2 and 3 were all previously detected in pancreatic islets (Yang et al PNAS) which does not really fit with the current data. As Slits function redundantly how removing a single Slit (Sli1) can cause defect? Likewise, the Robo1 KO exist and is viable. It should also be studied to support the conclusions.

Our study was initiated because of the genomic aberrations found in PDAC, a cancer of the exocrine pancreas, and more specifically a loss of ROBO2 that conferred poor survival outcomes in PDAC patients (Biankin et al. Nature 2012). Hence, our experimental work focused on experimental models of the exocrine pancreas and on deletion of Robo2.
However, we do report the expression of Slits in pancreatic islets (Supp. Fig.10) Fig. 2

Other points
All the evidence supporting abnormal gene expression in Robo2FF animals with acute pancreatitis comes from RTPCR ( Figure S5).
We agree that most data come from RT-PCR. However, Figs 5, 6 and Supp. Fig S8 show data by immunostaining and RNAScope ISH in acute pancreatitis in the Robo2 F/F line. Figure S4 : why is Robo1 expression not upregulated after Robo2 silencing ? Fig.2 a, b). The major increase in Robo1 that we see in our mixed cultures comes however from the other cell type, i.e. the fibroblasts (Fig 4 d, f).

We have detected an increase in expression of Robo1in epithelial cells when Robo2 is silenced (S.
"Indicative of stromal changes, and an increase of Tgfb1, albeit only a trend (Fig.5b,c). "a tendency for increased Cd3+ T lymphocyte infiltration at D3"(S. Fig. 11c,e), "Some typical Wnt target genes showed a tendency for increased expression, albeit not statistically significant" If some data are not significant this should be clearly stated rather than suggesting that there is a difference. And the picture on the figures should also reflect the results (on S11 it seems that there is a huge increase of CD3 cells)

We tried to be as detailed and correct as possible when describing the results. The dotted graphs with the individual data points are now better reflecting the variability that often is encountered when doing in vivo analyses.
Could they also validate the specificity of the Robo2 antibody (used for human immunostaining) on Robo2F/f mice ? Were they previously validated? the anti-Robo1 reference ab72790 is an anti DLST based on the abcam site. The Robo2 ab75014 is against the mouse protein and should work on mouse pancreatic tissue. Why don't they also showRobo1/2 distribution in normal adult human pancreas? It is otherwise difficult to know what Robo2 low or high mean.
We apologize for the typo when referring to the antibody catalogue number. It should be ab7279 instead of ab72790. We would like to refer the reviewer to Suppl. Fig. 12 where we did show the staining on normal human pancreas. The validity of the human antibodies is shown above.