So welcome, everybody. Welcome to this to today’s webinar.

Our speaker today is going to be doctor Florence Oncatiel.

She’s a histology project manager at Novalyx.

And in this webinar, she’s gonna be talking about innovative imaging analysis approaches for kidney research.

At the end, we are going to have a q and a, so stay with us. And, well, welcome, Florence. Let’s get started.

Hi, everyone, and thank you for joining this seminar.

I’m going to show you some data about the PCKD mouse model we are working with in the lab and how it has been a key source for studying human ADPKD.

We have used some histology and imaging analysis approaches to better understand the disease.

But first, let me present you Novalyx.

Novalyx is a contract research organization who has been working in the drug discovery field for over twenty years and now employs more than four hundred people.

We offer an integrated workflow from target validation up to lead optimization and also preclinical work.

We have a lot of different departments that can offer services from chemistry to structural biology to pharmacology.

The pharmacology team that I’m part of is composed of a team of highly skilled scientists divided in the in vitro pharmacology team, the in vivo and histology team, and the translational science team.

We have gained a lot of experience by working with prominent, pharma companies, and we can offer more than fifty in vivo model in house that have been already optimized.

We can also perform more than two hundred study per years, and we are compliant with high standards for animal welfare.

Now let’s dive in the data.

Otosomal dominant polycystic kidney disease, or ADPKD, is the most common inherited kidney disorder.

It’s mainly caused by the mutation in either PKD1 or PKD2 genes.

It affects twelve million patients worldwide and is one of the major cause of kidney transplantation.

In terms of physiopathology, it’s characterized by the formation of cysts in the kidney, but in also in other organs such as liver or pancreas.

On the right, you may see how a cyst can appear.

So, basically, what happens is that there is some proliferation in the epithelial cells of a tubule that results in a bulge and end up into a cyst. The cyst progressively is filled with fluid, which creates some pressure on the surrounding healthy tissue and then decrease progressively renal function until the need of first dialysis and then replacement of a kidney transplantation.

So the change of this, kidney function can be monitored with the GFR, also known as glomerular filtration rate. And, progressively, this GFR will decrease until it goes up to zero when there is a need for replacement of the kidney.

There is only one available treatment right now, and it’s called Tolvaptan.

Unfortunately, it only delays the need for dialysis and kidney transplantation, and it has also important side effects.

Thus, there is still a high unmet need in the medical field. On the right image, you may observe what kidneys from a patient with ADPKD look like after they have been removed.

And to give you an idea, the size of a kidney in a human should be the size of a fist.

So here, you can see how such a disease, as ADPKD may have an impact on the lives of patients.

In order to study the disease, we established a mouse model of ADPKD based on the main mutation of the disease, which is PKD one on the PKD one gene.

And we used a Creelog system. The Creelox system needs two strains of mice, one in which the gene, the main gene, PKD1, is fluxed here between exon two and four, and another one that expressed the query recombinase. So we choose a query recombinase that was ubiquitously expressed and under a promoter of ERT two that may be activated by tamoxifen.

When these two strains are crossbred, we obtain a PKD one knockout mouse in which the cre recombinase activated by the tamoxifen administration will come and excise PKD one exon two to four in all cells.

Since, the query recombinase was ubiquitously expressed, we could expect to detect some cyst in other organs than the kidney.

So first, we verify the presence of a cyst in a liver.

And, indeed, we detected some cyst in the liver.

These cysts are in only in the PCKD group, and they usually surround the vessels.

And if you looked at the magnified version of the image, you may even recognize that this cyst probably originate from the duct.

Then we use Visioform to quantify those cysts along with the vessels and to compare the different groups.

So on the left, you have the native image. On the right, the segmented image.

And this is, at the bottom, the workflow we used to, obtain the readout.

First, we start with the native image, then we created a tissue segmentation app, then a cyst and vessel segmentation app, and finally, some readout based on the mean area, their count, or the percentage of the cyst.

On the right, you may see some results from the percentage of cyst and vessels.

And, indeed, there is a significant difference between the presence of of system vessel in the PCKD versus the intact.

The difference being the presence of assist in the PCKD group versus the intact group. Okay. So now let’s go back to the kidney.

As a matter of fact, we didn’t only established one model, but two models. And these two models are based on the disease progression, and this is related to the moment when the disease is induced, either before sixteen days of life or after sixteen days of life.

When the tamoxifen induce the carrier recombinase to excise a PKD one before sixteen days of life, then we can observe a fast progression of a disease.

The resulting disease are big medullary cysts.

When tamoxifen is given after sixteen days of life, then we observe a slow progression of a disease, meaning that you have a smaller cyst in the cortical and the medullary area.

And what is interesting with this model is that it’s closer to the human disease.

Hence, this is the model we will be presenting from now on and the results from this model. Now in order to characterize the model, the in vivo team has set up a longitudinal study to evaluate the disease evolution.

As I was mentioning before, in order to obtain this slow progression disease, the tamoxifen induction was done at day seventeen.

And then after weaning and randomization, we have two arms in the in the study, the in vivo arm and the histological arm. The in vivo arm is, based on micro CT kidney image analysis done on the same animal at width six, nine, and twelve, and in the histological arm is based on different groups of different animals, on which we obtain the kidney tissue at week four, seven, ten, and thirteen.

The animals from week thirteen on the are the same than the animals from the micro CT, image analysis.

As for the results of the micro CT, image analysis, you can see it on the bottom.

The ability of a microcity is to obtain kidney size, cyst location, and number.

In the intact group, either at early time point or late time point, there is no cyst that can be visible.

Whereas in the PCKD group, already at early time points, some cysts may be visible in the cortical area, and this cyst overwhelm the tissue at a later time point.

You can also see on the right reconstruction of, the cyst in three d to have a better idea of the location and the size of those cysts.

On the right, these are the results from the quantification of this, cyst volume by micro CT, and we can already see some increase of the cyst volume at week nine and even more, at week twelve.

In parallel, we obtained some community issues, from the from the studies, and we set up a specific immunohistochemistry for collagen three that allows us to look not only at the kidney size, its location, and number, but also at kidney fibrosis, which is also a hallmark of disease.

Here you can see some, representative image of an intact, kidney or a PCKD kidney at week thirteen.

We also have been using a positive reference, nintedanib, and I will show you some data later on.

We then quantified the cyst and fibrosis using the collagen IHC and Visioform applications.

Now I’m gonna give you an idea of a workflow we’ve been using starting by the native image, either of an intact kidney or a kidney suffering from PCKD.

First, we created a tissue segmentation app and then a cyst and collagen segmentation app with three classes, cyst, tissue, and fibrosis, and some readouts, such as deceased mean area, count, and percentage, the distant tooth capsule, or the percent of collagen in the tissue area. The distance capsule is an interesting readout because it allows to give you an idea of the location, the global location of the cyst if they are closer to the cortical area or closer to the medullary area based on the results.

Then we use a percentage of cyst area and extrapolate it to obtain the projected cyst volume that would allow us to to compare it to the syst volume obtained by the micro CT evaluation.

And what we observed is that already at seven week, we started to see some syst development, and this there was a linear increase up to ten and then thirteen weeks.

What is peculiar with histology is that, this method allows to, see, to detect really early, time points of the pathology.

For example, here at seven weeks, we were able to already detect some patch of cysts that were not detected by other methods.

Then we compared this data with other readout that have been set up by the Indivo team, such as the BUN levels.

BUN is the blood urine nitrogen, and it’s a good marker of kidney function because urea nitrogen is supposed to be cleared by the kidney. So if your kidney doesn’t function well, then the band will increase, and it’s exactly what happened in the PCKD group.

What we could also observe is that the nintedanib, our positive reference, was able to protect kidney functionality.

We also compare these results with another measure of a kidney functionality more at the end toward the end of the study, and that is GFR quantification.

GFR stands for glomerular filtration rate, and it’s a good measure of the kidney functionality.

Because when when it decrease, it means that the kidney function is not is impaired. And we have been using the method Medibicon.

So this is a transdermal measurement of an injected fluorescent tracer, the FITC synestrin, and this tracer is exclusively excreted by the kidney.

In a normal situation, this is what a trace would look like, meaning, it would be injected and then normally excreted.

When an animal has impairment of a kidney function, then the tracer takes longer to be excreted, and thus, the glomerular filtration rate resulting is decreased.

In our model on the right, you may observe that the GFR in the intact group is higher than in the PCKD group, confirming that, there is a decreased kidney functionality in the PCKD mouse model.

Then we used our collagen immuno histochemistry to evaluate the kidney fibrosis in the model.

And you can see the results here and the representative images of an intact group, PCKD groups, and nintedanib group, which is our positive reference, and the results on the right. So there is a statistical statistical difference between the intact and the PCKD group, meaning that you have a higher fibrosis index in the PCKD groups. But but what we could also observe is that nintedanib was able to protect kidneys from fibrosis development, which is what was expected from the literature.

There is another homework for ADPKD, and it’s kidney inflammation.

In ADPKD, it’s mainly the presence of macrophage that is a hallmark of a disease. And here, we decided to set up a specific staining, immunofluorescent staining of macrophage and their subset, macrophage being represented by f four eighty stain, and m one by IRF five and m twos by ARG one.

You can see here, representative images of an intact group versus a PCKD group. And already, it’s easy to detect that, there are a lot more macrophage, in the PCKD group versus the intact group.

The macrophage being usually surrounding the cyst.

In order to quantify these macrophage, we use VisioForm, first with a tissue segmentation app, then cyst segmentation app to separate the tissue from, the cyst. And then within the tissue, a nuclei detection app followed by a cell positivity app. Here, we had to use deep learning to, evaluate positivity of the cells because of the high variability of the staining.

And out of the the segmentation, we obtain different readouts such as the percentage of macrophage and their subset, the percentage of cyst, and the distance of the macrophage to cyst.

On the right part, you may see a result of, one of the readouts we obtained, which is a percentage of macrophage.

And, indeed, there is a highly significant difference between the PCKD group and the intact group. What I can also tell you and that I will not show you here is that both m one and m twos are significantly significantly increased, and, there is the ratio is, favorable toward m ones versus m twos.

And, also, what we had observed is that macrophage are usually closer to cyst when there are some remaining healthy tissue.

Finally, we wanted to evaluate the origin of a cyst in our model.

To do so, we optimized a triple staining, immunofluorescent staining, for the different segment of the tubule.

There are three different main stamen segment of the tubule in the kidney. The proximal tubule that is coming right out of the glomerulus.

Then the after the loop of ventilae, you have the distal tubule, and finally, the collecting tubule or collecting duct.

In the intact tissue, we could see that there was mainly some proximal tubule in the cortical area and, a little bit of, distal and collecting ducts as expected.

But in the PCKD group, we could observe that there were cyst of all origin.

Proximal cysts, distal cysts, collecting duct cysts, and even cysts of unknown origin like this.

So our goal was to compare the results of, this part of a study to the literature as we have done before.

And then we decided to quantify those images and the origin of the system.

So we use Visiopharm game with the workflow on the bottom.

We start with a native image, then we created a tissue segmentation, app, then assist and tubule segmentation app, and finally a tubule positivity.

Again, we had to use the deep learning, methodology because of the variability of the staining.

At the end, we obtained a server readout pertaining to the syst origin, the percentage of the syst, global versus segmental.

And what we could do also that I will not show you here, but, is that we we we could align, these different origin of assist with different segment of assist with other markers or other targets on serrated sections, which allowed us to focus on, future mode of action or future, target that may work, in this disease.

And in terms of results compared to the literature, what we found is that compared to the total amount of, cyst, the collecting ducts are the largest cyst and the largest, amount of cyst that we may find in our model, which is in accordance with the literature. Because in ADPKD, usually, the cyst arise from the collecting ducts.

In conclusion, we have shown you that we have been able to establish a ubiquitous and inducible PCKD knockout model that is representative of human ADPKD slow disease progression, both at the kidney and liver level.

We also have created some clinically relevant setups, both, at the histological, in vivo, and even translational level.

And, the what what we can say is that our model is extremely robust. The reason why is that we have carried multiple studies with similar outcome. We also have identified a positive control in tidanib.

And, what I could say is that our deep knowledge on disease mechanism arise from our ability to study disease progression, but also functional and longitudinal without, but also transcriptomics, biomarkers, or cyst origin.

We offer this model both, in a preventative and therapeutic settings and also, as I’ve shown you before, as a rapid progression model.

Thus, we think that this model is really the most relevant preclinical model for human ADPKD.

Now my presentation is, finished. I thank you for your attention, and I’m happy to take any question.

Hey, Florence. Thank you so much for the excellent presentation.

We can start with the q and a now. I will I have a question myself.

You mentioned during your presentation at some point that you had to use deep learning because of the to detect the positivity of certain cells. Can you elaborate a little bit on that?

Yes. Sure.

So, the issue sometimes when you need to define if a cell is positive or not is that you may have a lot of variability within your stain. As much effort you put in your IHC, garbage in, garbage out, meaning that sometimes you have a lot of variability to work with. And usually, what happen is that you will have your positive nucleus, then you need to define if it’s positive or negative.

Here, our issue is that the standard way was, to, define it with intensity of a of a of a channel. And when the intensity is too variable, you cannot set up a threshold to define if it’s positive or not. You may use normalization, but sometimes it’s not even enough.

So what we’ve done is that, we have created a new app in which it would would predict if an area would be positive or negative with within the channel. And then we would use this predictive map instead of the intensity threshold. We would use the predictive map, in, to, predict if a if a cell would be positive or negative. So in that way, the method is a lot more robust than just intensity, based on intensity.

Okay. Got it. Thank you. So is asking, is human ADPKD also usually seen with liver cysts?

Thank you for your question. And, yes, this is a a common physiopathology, part of a physiopathology of a disease. We usually study mostly the kidney, cyst because they have a lot of impact on the on the people quality of life. But liver cyst and even pancreas cyst are known in the in the literature.

Thank you. And Bettina asked, how did you quantify the micro CT images?

Yeah. That was a hard one. The micro CT, it’s a complete different set of image than what we used to work with. Since what we’re used to work with are large images with which, we work with Visioform.

Here, instead of having one large image, you have two hundred different images that are stacked.

And so we had to create our own tool.

And we used the, also a deep learning approach.

And we used something that is close to the unit model, and it’s called the non unit model.

And, basically, it’s sort of a, DYI sort of a deep learning model in which, you feed, some images to the model and it will define what are the parameters that are necessary to, train your images. And we did so. We trained the model and then obtain a very interesting set of data with a pretty good correlation between the prediction and the ground truth. We hit, zero point ninety seven percentage of correlation between the prediction and the and the ground truth. We had, some data about, dice, which is one another way to evaluate, how good model is, and we reach over zero point eight.

So with that, we were able, instead of having someone working over a week to work on this image, we turned it into, half a day works time.

Wow.

Yeah.

That was a work.

Big, big amount of work.

Yes. Next question, Sonia is asking, why does the band increases after the kidney volume?

Yeah. That’s another important question.

So what you need to understand with the kidney, and it also happens with the lung or the pancreas, These are these kind of organ or like that. They can really handle quite a bit of of, damage before you actually see anything happens at the at the function level. And for kidney, it’s the same. And since the bun is a measurement of a kidney function, this is the reason why it has a it increased later than the pathology is present.

So this is the reason why, for example, in the histology, it’s so sensitive, and we, sent, and we can see the, the the damage starting at week seven.

But for the ban, you can only see it a lot later during the disease.

Got it. Carrie has asked, did you have a way to exclude vessels when separating cysts by origin?

So with a lot more time, we could we decided not to because of, the kind of readouts we wanted to get out of the liver images. So it was more like a difference between the, positive, part of, doing, analysis and, the drawbacks of having to take too much time to have a readout. But yes, there is definitely a way, since the vessel do not have the same lining as the cyst, we could definitely, make a difference.

Just we needed a lot more annotation and a lot more time to do so.

Yeah.

Matt is asking, what is the base mouse being used? Lexix, OrbC, Otter? Just curious if the macrophage ranges are comparable to the human disease m one versus m two.

Yeah. It’s, black six, if I do remember well. And the macrophage ratio seems to be, similar to what we had in the human disease, the location too. But then, it would be, since mice are not human, it’s always a little bit, touchy as a as an answer to give, I would say, concerning the the the macrophage, it’s it’s always a little bit complicated, but it’s close enough, I would say, and also close enough to what we have found in the literature.

Got it. So follow-up from to follow-up on Karen’s question, is it possible that the cysts of unknown origin are lymphatic vessels? Since since lymphangenas lymphangenesis is quite common in inflammatory environments.

It’s a question we have been working on.

Our theory, it’s, is that it’s more undifferentiated type of, of of cyst. So it means it’s still tubules or part of tubule that became cyst.

One of the way to verify that and, is to is to test, for example, CD nine. That would be, one other way or, but but but the the common theory would be that these cells, regress and they differentiate. And this is a reason why we don’t see those, specific stain or specific marker, like a LRP two or, the the the common marker for the different segments.

Cool.

Well, alright. Thank you so much, Florence, for being here today and presenting this webinar.

It’s being recorded. So if anyone wants to rewatch later or share with your colleagues, we will post probably on LinkedIn, and we will also share by email with everyone. And, of course, if you wanna write to Florence and ask some questions, feel free to reach out. Thank you, Florence and the whole Novadex team as well for making this happen.

And For having me.

Have a lovely day. Yeah. Everyone. Bye.

Bye.