Glucose homeostasis is regulated by pancreatic β-cell cilia via endosomal EphA-processing

Diabetes mellitus affects one in eleven adults worldwide. Most suffer from Type 2 Diabetes which features elevated blood glucose levels and an inability to adequately secrete or respond to insulin. Insulin producing β-cells have primary cilia which are implicated in the regulation of glucose metabolism, insulin signaling and secretion. To better understand how β-cell cilia affect glucose handling, we ablate cilia from mature β-cells by deleting key cilia component Ift88. Here we report that glucose homeostasis and insulin secretion deteriorate over 12 weeks post-induction. Cilia/basal body components are required to suppress spontaneous auto-activation of EphA3 and hyper-phosphorylation of EphA receptors inhibits insulin secretion. In β-cells, loss of cilia/basal body function leads to polarity defects and epithelial-to-mesenchymal transition. Defective insulin secretion from IFT88-depleted human islets and elevated pEPHA3 in islets from diabetic donors both point to a role for cilia/basal body proteins in human glucose homeostasis.


Point-by-point response to our reviewers:
We thank the reviewers for the time and effort invested in our work, we truly feel that their comments helped improve the story and make this a better manuscript. We hope they will agree and find the manuscript revised to their satisfaction.

Reviewer 1:
In sum, this is a very comprehensive and innovative study making its points using multiple different experimental systems ranging from cell line to mouse to humans. The observation of increased EphA3 activation in islets from humans with type 2 diabetes makes this story not only interesting from a cell biological point of view, but also for scientists working on clinical aspects of this widespread human disease A few minor issues still need to be addressed: 1. Two manuscripts have been published in "Diabetes", showing that EphA-ephrinA signaling affects glucagon release. It would be nice to see whether cilia also affect glucagon release, since Pdx1-CreERT mediated cilia defects shall also be present in alpha cells, not just in beta cells.
Thank you for this valuable comment. Indeed, EphA/ Ephrin A signalling has been reported to regulate glucagon release. However, at the time of induction, the Pdx1-promoter is not transcriptionally active in α-cells. In ROSA mTmG ;Pdx1-CreER mice induced at the same time as the βICKO cohort, we did not observe GFP related fluorescence in α-cells as shown in Suppl. Fig. 3E.
The changed section reads as follows: ll.136 "To test for β-cell specificity of Pdx1-CreER driven recombination, we examined glucagon-and insulin-expressing islet cells in pancreatic sections from these reporter mice (Suppl Fig. 3E). We did not observe recombination in glucagon-positive cells of these mice, indicating that at the time of induction, Pdx1 is specific to β-cells." It might be of interest to the reviewer that we have indeed observed defects in glucagon secretion that are potentially related to defects in cell-cell-communication via EphA/ EphrinA signalling or other pathways (see below).
We are currently striving to better understand this phenotype but feel it would be beyond the scope of this study focusing on β-cells and insulin secretion.
2. The histograms shall be shown as dot plots, which enable the reader to identify possible subgroups that might otherwise be hidden.
Thank you for your suggestion. We have since replaced several graphs with dot blots to better identify individual data points. These are shown in Fig. 1A  In addition, we replaced the bar graphs for WB quantification with dot blots.

Several Western blots are overexposed. Even though this looks nice, a more scientific way is to show non-saturated Western blots wherever possible
Thank you for raising this point that of course has important consequences for WB quantification.
All of the Western Blots shown in this version of the manuscript are non-saturated. We now use ChemStudio SA 2 for WB documentation. Saturated pixels will automatically be pseudo-coloured to indicate a problem. As stated before, we only quantify and show bands that are not saturated.
4. The authors shall describe in the M&M section how they obtained islets from humans with type 2 diabetes, as this is currently missing We apologize for our omission. As is the case for the islets from normoglycemic donors, we have obtained islets from diabetic donors from the Alberta Diabetes Institute IsletCore. We have expanded the M&M section accordingly: "…Human primary islets from normoglycemic and T2DM donors were supplied by the Alberta Diabetes Institute IsletCore (Edmonton, Canada), supported by the Alberta Diabetes Foundation (ADF), the Human Organ Procurement and Exchange (HOPE) and the Trillium Gift of Life Network (TGLN) for coordinating donor organs the Islet Core…"

Reviewer 2:
Overall, this is a very interesting study reporting a novel signaling axis downstream of primary cilia in pancreatic β-cells. The biochemical evidence between Ift88 and EphA receptors is convincing, but the downstream connection with the Rac1-actin cytoskeleton is a bit tenuous and sometimes confusing -this could be expanded or removed. Additional specific comments are listed below.
Thank you for your comments. We have expanded on the connection with the actin cytoskeleton and hope we can convince you that Ift88 plays a role in actin reorganization related to Tiam1 and Rac1 activity.

Specific comments:
1) The data obtained in vivo from the βICKO animals and ex vivo KO islets are quite convincing, including GTT, insulin secretion and perifusion assays. Less convincing are the results obtained in the rescue experiments with recombinant EphrinA5-Fc chimera (Fig. 3) as well as with Tiam inhibitor NSC23766 (Fig. 5). The perifusion experiment shown in Fig. 1 shows a dramatic difference in insulin secretion between control and the βICKO islets, which cannot be appreciated anymore in the rescue experiments. Why the authors have not always included the untreated control islets as baseline? This might help to appreciate the differences between control and mutant versus the rescued insulin secretion.
We apologize for the confusion, we did not show untreated control islets as baseline in an attempt to present the data as clearly as possible and to avoid showing the same data set twice. Obviously we did not succeed. As requested, we added both induced and control untreated βICKO islets to the graphs in Fig. 3 A and 5H. We hope that the effects of EphrinA5-Fc and NSC23766 treatment respectively can be appreciated more readily now.
2) The differences in phosphorylation shown by WBs with the p-EphA3 antibody in sh-Ift88 knockdown MIN6 line treated with EphrinA5-Fc or Ptp1b inhibitor are not convincing, given the images shown in Fig. 3D  Thank you for raising this point that of course has important implications for quantification and therefore interpretation of our results. As stated previously, we are showing only unsaturated WB exposures in this version of the manuscript. We use ChemStudio SA 2 for WB documentation; saturated pixels will automatically be pseudocoloured to indicate a problem. We only quantify and show non-saturated bands.

3) Does the increased phosphorylation affect only EphA2 and EphA3 receptors or all EphA receptors? Why the authors characterize only EphA3 and nothing is shown about EphA2?
We truly regret that we are unable to answer whether or not the effect is specific to EphA2 and EphA3. The only specific antibodies to pTyr EphA are those blotted in the PathScan antibody arrays we used (Suppl. Fig. 4). Upon validation, pEphA2 antibody gave very weak bands and the EphA2 antibody did not work in immunoblotting, making validation of the PathScan result very difficult.
The bigger effect was observed for EphA3, so we focused on this receptor. Again, we apologize for not being able to give a more definitive answer, but the limited availability of suitable reagents hinders more detailed investigations of these matters. The text was revised to reflect these difficulties: ll.153: "…We confirmed EphA3 hyper-phosphorylation in Tx-treated βICKO islets by immunoblotting (Fig. 2A…). We were unable to validate excessive levels of pEphA2 due to a lack of suitable anitbodies…"

4) The IF analysis is important and should be expanded and improved. First, why in the IF
staining the authors used an antibody against EphA5 instead of EphA2 or EphA 3 (Fig. 3F)  Again, we deeply regret not giving a more straightforward and definitive answer. We were unable to find an anti-EphA3 antibody suitable for immunofluorescent staining. Instead, we had used one raised against EphA5 as a surrogate. We realize that this is unsatisfactory and removed those sections. In the absence of suitable antibodies, we overexpressed EphA3-myc in control and Ift88depleted cells. We hope that we can convince you that differences in EphA3 localization are in good agreement with the other lines of evidence and reflect defects in EphA3 internalization when Ift88 is lost. This is shown in Fig. 4F. Before addition of EphrinA5-Fc, EphA3-myc is distributed throughout the cytoplasm in control cells. In shIft88 stable cells, the pattern is less evenly distributed. In both cells, a subpopulation of EphA3-myc is presented at the plasma membrane, although it seems to be increased in shIft88 compared to controls. After addition of EphrinA5-Fc, a promiscuous ligand with the highest binding affinitiy to EphA3, control cells internalize EphA3-myc efficiently with the majority of the receptor decorating the perinuclear region. By contrast, Ift88depleted cells show myc-specific immunofluorescence in the periphery, in proximity to the plasma membrane, consistent with a failure to efficiently internalize EphA3-EphrinA5-Fc.
The text was changed accordingly and now reads as follows: ll. 282: "…None of the commercially available antibodies raised against EphA3 were suitable for immunofluorescent staining in our hands. Therefore, we used a myc-epitope tagged EphA3 expression plasmid, EphA3-myc, to visualize EphA3 localization (Fig. 4F). After stimulation with EphrinA5-Fc, EphA3-myc is efficiently internalized and predominantly found in the perinuclear region of control MIN6m9 cells. By contrast, a large subpopulation of EphA3-myc is observed in the periphery of Ift88-depleted cells, proximal to the plasma membrane. These data are in good agreement with the other lines of evidence and suggest that EphA3 internalization is dependent on Ift88 function…" As suggested by the reviewer, we quantified the co-localization of Tiam1 with CGN marker GM130 by determining the Pearson correlation coefficient. The results are shown in Fig. 5I. In the panels, white lines demarcate the CGN as marked by GM130. Correlation analysis revealed a significant decreased in Pearson coefficient, indicative of a less tight spatial correlation of Tiam1 and GM130.
The changed passage in the manuscript now reads as follows: ll. 391: "…In addition to protein levels, we also tested the localization of Tiam1 in shIft88 expressing MIN6m9 cells. We found Tiam1 immunofluorescent intensity widely distributed throughout the cytoplasm in Ift88 1 and 2 and seemingly strongly increased (Fig. 5I). In control cells, Tiam1 is associated with the GM130-positive cis-Golgi network (CGN). To better quantify the change in cellular localization, we determined Pearson's correlation coefficient and observed a significant decrease in Ift88-depleted MIN6m9 derivative cell lines. This is indicative of a less strict spatial association between Tiam1 and the CGN compared to controls…"

5) The connection of Ift88 and Rac-actin cytoskeleton is not convincing. For ex. no data is included about the actin cytoskeleton organization or manipulation of actin polymerization.
We apologize for the confusion. We hope you will agree that we have made substantial advances in our understanding of the link between loss of Ift88 function, Rac1/ Tiam1 and actin dynamics.
The role for ciliary and basal body proteins such as Ift88 or Bbs4, Bbs6 and Bbs8 in actin reorganization has been previously reported. To better understand the nature of this role, we tested the ratio of actin monomers (G-actin) to filamentous actin (F-actin) and found a shift towards higher levels of actin polymerization when MIN6m9 cells were depleted of Ift88 protein (Fig. 5A). Similarly, phalloidin labelling of F-actin revealed morphological differences between control and Ift88-depleted cells (Fig. 5B). To address the role of actin polymerization in endocytosis, we treated MIN6m9 cells in presence or absence of Ift88 with Cytochalasin D, an inhibitor of actin polymerization. We observed that EphrinA5-Fc internalization was ablated in all cells regardless of presence or absence of Ift88 (Fig. 5C). It has been shown that one of the signalling pathways regulating actin dynamics is planar cell polarity signalling, part of noncanonical, β-catenin independent Wnt signalling. We and others observed that loss of cilia/ basal body integrity affects PCP signalling and is concomitant to stabilization of β-catenin (Gerdes et al., Nat Genet, 2007). Here, we found that β-catenin protein levels are upregulated in Ift88-depleted MIN6m9 cells (Fig. 5D). We also found that Tiam1, a GEF for Rac1, is upregulated on the mRNA and protein level ( Fig. 5D and E). Tiam1 is also part of protein complex with Par3 and Par6 controlling cell polarity, indicating a polarity defect in Ift88-depleted cells. Because the shift in Fto G-actin is also observed in delaminating cells, we checked for markers of epithelialmesenchymal transition and found Snail, Slug (Snail2) and Vimentin significantly upregulated in Ift88-depleted MIN6m9 cells, indicating that the cells are gaining more mesenchymal-like properties. At the same time, E-cadherin protein levels are decreased, suggesting that these cells lose epithelial-like characteristics (Fig. 5D). The figures have been changed accordingly and the manuscript has been expanded. We added a subsection "Ift88 is involved in maintaining epithelial like polarity" (ll.326). We hope that these lines of evidence can convince you that actin dynamics are changed in Ift88-depleted MIN6m9 cells, likely due to a shift in cell polarity, that cilia-regulated actin dynamics play a role in EphrinA5-Fc internalization and that inhibition of Tiam1 can partially reverse the inhibition of insulin secretion.

The writing of the manuscript could be improved. Both Introduction and discussion are long and wordy, the important messages do not come through and it is sometime difficult to appreciate the novelties compared to previous literature on cilia in beta-cells.
We apologize for the confusion and have significantly shortened and modified both introduction and discussion of the manuscript to more clearly state our message. In addition, the manuscript was prepared after consulting a professional writer and native speaker to improve the writing in general. We hope that these measures have helped to make our point more clearly.

Reviewer 3:
These findings are novel and interesting in that they has firstly demonstrated a functional link between primary cilia and Eph-ephrin signaling and abnormal Eph signaling in human pancreas from diabetic subjects. However, they were considerably limited by a lack of control and statistis, small n numbers, and lots of errors in the manuscript and figures.
Major comments. Thank you for your comment.

To exclude any possible effects of tamoxifen on beta-cells and insulin secretion
We have used both vehicletreated βICKO mice as a control for Tamoxifen treatment and Tamoxifen treated Ift88 loxP/loxP mice to control for effects of Cre-expression as mentioned in the text: ll.91: "…To control for effects of Tx-treatment and Pdx1-CreER overexpression, both vehicletreated βICKO mice and Tx-treated Ift88 loxP/loxP mice from the starter strain served as controls…" To avoid overcrowding of the figures, we did not show both controls in the figures. We never observed significant differences in glucose tolerance or insulin secretion between vehicle-treated βICKO mice and Tamoxifen-treated Ift88 loxP/loxP mice. Above, we show GTT excursion curves at eight weeks to demonstrate there is no significant difference between the two control groups. Fig 1F, 3A, and 5D were only 2.

The n numbers of dynamic insulin secretion experiments in
How many times were these experiments repeated to conclude the findings?
The dynamic insulin secretion experiment were repeated 2 times using islets pooled from 5 different animals (N=5, two independent experiments, technical duplicates). Pooling animals should dilute differences between individuals. To increase the statistical power and therefore exclude false negative findings, we repeated the experiment 2 additional times for a total of n=4, islets pooled from 10 animals.

How does ciliary dysgenesis lead to increased Tiam-1 expression? The molecular link
mediating this changes needs to be suggested or discussed.
We deeply regret that technicial limitations make it impossible to give a direct answer, f.e. via reverse ChIP that has not been established yet. Instead, we did some literature research and found that Tiam1 and GTP-bound Rac1 enhance β-catenin/ TCF-dependent transcription in certain types of colorectal cancer. We have previously shown a link between primary cilia and noncanonical Wnt/ planar cell polarity signaling. Tiam1 forms a super-complex with the polarity complex Par/Par6/aPKC. It is possible that Tiam1 upregulation is an attempt to compensate for loss of polarity signaling. To support our hypothesis, we tested and found an increase in β-catenin that we have previously shown to be concomitant to the loss of PCP signaling (Fig. 5). β-catenin dependent Wnt signaling is one of the known inducers of Epithelial-mesenchymal transition (EMT) and we found an increase of mesenchymal markers Snail, Slug, and Vimentin, and a decrease in epithelial marker E-cadherin (Fig. 5D). Taken together, we suggest that loss or rather a shift in cell polarity leads to an increase in Tiam1 and affects actin dynamics.

The authors showed increasecd EphA phosphorylation in islets from the subjects with
diabetes. However, they did not show ciliry chanegs in these subjects. Is EphA hyperphosphorylation linked to ciliary changes in human islets?
Unfortunately that limited availability of tissues from diabetic organ donors makes is very difficult to answer this question with sufficient statistical power. Clearly, further investigations are warranted and we are in the process of contacting consortia that have collected data from larger patient cohorts. Because these studies are very time consuming, we regret that the direct answer will be beyond the scope of this manuscript. However, we were able to obtain access to Affymetrix gene expression data from normoglycemic, pre-diabetic and diabetic individuals that underwent partial pancreatectomies at the University of Tubingen. Islets were isolated by laser capture microdissection and differential gene expression was analyzed (Table1). We found that RAB3IP, a direct interaction partner of the BBSome is significantly downregulated in diabetic patients and negatively correlates with HbA1c glycosylation and fasting glucose. It correlates with insulin secretion. This provides a direct link between cilia function, vesicle transport and Type 2 diabetes.
We have added a table and amended the Results and Discussion sections accordingly. They are now worded as follows: ll. 435 (Results) "…To test for a more direct link between islet cilia/ basal body function and Type 2 Diabetes, we analyzed islet transcriptomes obtained from pancreatic resections of human donors. In a cohort (N=19) of non-diabetic, prediabetic and diabetic individuals 47 , we found significant misregulation of RAB3IP (RAB3A interacting protein), a direct interaction partner of BBSome component BBS1 and a GEF of RAB8A. 48 RAB3IP mRNA levels associate negatively with increases in fasting glucose and positively with insulin secretion, i.e. HOMA2%B 49 calculated using C-peptide and insulin (Table 1). This finding supports a functional role for cilia/ basal bodies in insulin secretion and islet function in humans…" ll.493 (Discussion) "…Finally, several lines of evidence support a role for cilia/ basal body function in insulin secretion in humans: When we depleted islets from four different donors of IFT88, we observed increased EphA3 phosphorylation levels and impaired glucose-stimulated insulin secretion, both in good agreement with our findings in Tx-treated βICKO mice. We observed increased pEPHA3 levels in islets from diabetic donors compared to normoglycemic controls. And misregulation of RAB3IP in islets from diabetic donors supports a direct link between cilia/ basal body dysfunction, vesicle trafficking and Type 2 Diabetes mellitus. Taken together, we propose that modulation of ciliary function and endosomal signal processing could offer a novel approach to therapeutic interventions in metabolic diseases such as Type 2 Diabetes mellitus…" We also took care to address the minor comments you had.
In conclusion, we hope we have been able to address the comments of all reviewers to their satisfaction.
Reviewer #2 (Remarks to the Author): In the revised version of this manuscript the authors have satisfactorily responded to all the points raised by this referee.