0:05
So next speaker is Martin Wegner is head of RND at Vivlion.
0:13
Been doing his PhD focused on innovative approaches CRISPR, you know, training and analysis.
0:25
Analysis including combinatorial applications to identify genetics interactions.
0:30
He co-invented covalently closed surplus and sized and fixed pair customer libraries, which are the basis of the technology used in New India.
0:41
Martin is currently heading the research and development activities within New India.
0:46
So again, he's going to talk about, you know, what do you call PRCISR.
0:56
Not a real word in English, but we figured it might be put a bit of fun into naming what we are actually doing.
1:06
So thank you very much.
1:07
Thank you for joining.
1:09
My name is Martin.
1:10
I work for Vivlion.
1:12
We're located in the Frankfurt Germany and we specialise in all kinds of CRISPR services and products.
1:18
And today I'd like to introduce PRCISR CRISPR, which is our CRISPR enabled discovery platform.
1:26
And I would like to line out our core technology that we use to generate uniformly distributed CRISPR libraries.
1:34
And I would also like to line out why uniformity in these libraries is important and how it has to be to open up the space for combinatorial CRISPR screening, but also how we use it in our internal screening and bioinformatics services.
1:49
And towards the end of the talk, I would like to give an outlook on aspirational road map towards genome wide gene relationship screens and datasets.
2:01
So the motivation behind Vivlion is actually we want to make the screening more efficient.
2:09
And when you look at publications, when you look at industry figures, you realise that CRISPR screening is actually quite expensive.
2:20
And this is an example for a genome wide screen.
2:22
That means you clone a plasma library and then to a lentiviral plasma backbone, you generate an lentivirus and then apply that lentivirus in a pooled fashion so that in a cell population, each cell gets a different perturbation, for example, a gene knockout.
2:38
And that can be quite expensive because typically you don't only screen 1 cell line, you do more cell lines, you do replicates, you have different experimental setups.
2:49
So the idea behind the uniformly distributed libraries is we want to reduce the cost per screen and thereby you can increase the output with the same amount of the resources.
3:00
And it also unlocked genome wide applications in more complex models, and it also unlocked combinatorial approaches.
3:06
And one of the benefits obviously that comes with that is that you improve environmental sustainability and you're much just much more efficient with your resources.
3:17
So the technology that we developed and tries to avoid a couple of steps that you typically do when you clone a plasmid library.
3:28
Typically what you do is you order an oligo pool that encodes all the guide on and sequences that you want in your screen.
3:35
Then you use the PCR you amplify this oligo pool and then using restriction and ligation cloning, you clone this set of guide RNAs into your library backbone and then amplifying bacteria.
3:47
And what you end up with is typically a distribution.
3:53
So you see a difference between highest and lowest abundant guides.
3:56
And we measure or we quantify the uniformity of that library by looking at the skew.
4:01
So we compare the 90th percentile point and the 10 percentile point of the data.
4:07
And the ratio of that is the skew.
4:08
And in this example is the Brunello library.
4:12
It's the established library that you can order by Addgene, has a skew four.
4:17
So the ideal value would be a one.
4:19
So there's no difference between highest and lowest abundant guide RNA sequence.
4:24
That's what we're trying to achieve.
4:25
Now we are in a biological system, it's really hard to achieve that.
4:28
And so we will not be able to do this properly.
4:32
But we try to be as close to 1 as possible.
4:36
So what we instead do is we order an oligo pool and directly use it in our protocol.
4:43
So what we do is we're template plasmids and we can use all kinds of plasmids.
4:48
We are more or less agnostic to the system.
4:50
One prerequisite for these plasmids is that they contain one origin of phage replication because we generate single stranded versions of these plasmids using phage. And in these plasmids we have the template sequence which we try to replace with the guide RNA sequences using the site directed mutagenesis approach.
5:10
So we site specifically and you add oligo pool to the single stranded DNA, run a polymerase and the ligase and end up with the heteroduplex.
5:20
And you can now electroporate or amplify that in bacteria with a couple of tricks that force the bacteria to use the strand of the plasma that contains the guide on is for amplifying the entire plasma pool.
5:33
And this works really well.
5:36
And this is a particularly good example of a larvae that we generated in house.
5:41
It’s minimal genome wide library which has a skew of 1.6.
5:46
And just by comparing these two distributions, I think I hope you were you appreciate that's actually a good distributed library.
5:57
Now one claim that we have is that when you have a uniform library, you can downscale the size of the experiment.
6:06
And practically that means you reduce the average representation of each of the guide on a sequences in your pooled screen.
6:14
So we have a partner Repare Therapeutics and they challenged that claim and they asked us to remake the library that they were using in house for a couple of years.
6:26
It's the Toronto Knockout version 3.
6:27
It's an established library.
6:28
It's a good library, still the standard and most on a lot of industrial settings.
6:34
So we remade that library.
6:36
It's exactly the same sequence, exactly the same library backbone, but better distributed.
6:41
And this is now data that is already three years old.
6:45
So we are now even better in making these libraries.
6:47
But what they did was they did a BRCA1 synthetic lethality screen with their library at a 400 fold and 100 fold representation of the guide RNA.
6:58
And what I hope you appreciate here is that all the control hits that they expected actually to be statistically significantly enriched only occur in the high coverage, but you don't see them in the low coverage.
7:12
And when you now screen the uniformly distributed lively with the low coverage, you see these hits shifting up into that area here and they become suddenly also statistically significant.
7:23
And this is a reduction of 400 down to 100 fold, so 4 fold reduction.
7:28
Now thinking about different screening setups, you could even go below that.
7:34
We did screens with a 20 fold coverage positive screens.
7:37
That means we look for enrichment for surviving cells and specific treatment for example, that works really well.
7:45
And the opposite if you do a negative screen, obviously you want to keep the coverage still high because you want to make sure that everything is represented in the screen.
7:54
But there it works really well.
7:57
So we are now in the process of generating a genome wide library and I just want to outline a bit the our motivation behind selecting the guide RNA sequences for that library because the quality of the sequences make a huge impact on the screen.
8:14
So what we tried to find was guide RNA sequences that were empirically shown to be functional, to be active.
8:25
So we took a publication from 2020 and what they did was they generated, they looked at all the published CRISPR screens that were available at the time and generated a library that contained 8 guide RNAs per gene.
8:39
Because we went downscale, we set up in 8 guide RNAs is too much, we want 4.
8:45
So we selected from that publication, we selected the four best guide RNAs per gene.
8:51
Now they were looking at viability screens.
8:54
So basically for an essential gene, it's really easy to decide what are the good four guide RNAs.
9:00
But if you have non-essential genes where the phenotype in the viability screen is standard around 0, it's really hard to do this.
9:08
So we also added FACS enrichment screens, for example, reporter screens.
9:13
Where you see outside of the viability phenotype you see all the phenotypes and can then start pouring the guide RNA sequences.
9:21
So we selected 4 guide RNAs per gene and whenever we added the gene to the library, we then went to establish tools to design guide RNA sequences.
9:33
The current status of the library is we have guide RNAs from almost all the genes in the ensemble database.
9:40
We're missing out on 40 genes on that.
9:42
So we are pretty much complete when it comes to protein coding genes and we have a lot of control sequences in the including non-targeting guide, safe harbour, luciferase, all kinds of controls.
9:56
And we actually screened that library already so simple viability assay and looked at for example at the controls and comparison to the Bonello which contains non targeting sequences that give a high peak.
10:11
Then around here we have two different kinds of controls in there.
10:16
We have a non-targeting sequence here and we have safe harbour targeting those guys here.
10:22
And it seems that the safe harbour targeting guides are actually a better control because they are very close to inducing no phenotype at all.
10:31
It's kind of expected because you have a double strand break, you don't have a double strand break with a non targeting sequence.
10:39
But lately, so don't have to repair anything, they'll just keep on going.
10:44
We also compared the library performance to the Toronto Knock Out version 3 and we are very close to slightly better than Toronto Knock Out version 3.
10:55
And more importantly, if you look at the number of active guide RNAs per gene, you see in blue the bars for our library and in yellow that's the control library.
11:10
So if more guide on is active per gene on average than in the other library, so statistically you can be more confident in calling your hits.
11:23
So this is basically a setup where you target a single gene and a single cell with one guide RNA.
11:27
But we have other setups, for example fixed pair and multiplexing.
11:31
These are dual targeting approaches.
11:33
Fixed pair means we have predefined guide on a combination.
11:36
So it's predefined guide on a combination flow into a cell.
11:40
And that's very useful for example, for validation libraries if you want to target the non coding regions of the genome, but also to increase the editing efficiency.
11:50
So if you use two guide RNAs on the same gene, the likelihood that you knock out that gene is actually higher.
11:56
And we have multiplexing libraries and I will show some examples for both setups.
12:04
So the idea behind fixed pair in the current libraries that we have is increasing editing efficiency.
12:13
So instead of using one guide owner per gene that's induced as an indel and ideally leads to an editing efficiency of 60-70%.
12:22
We try to design the pairs of guide RNAs in the way that it fosters the excision event typically at the very start of the gene.
12:30
And we see in an experiment that we have an editing efficiency of around 90%, sometimes even higher and it's works really well.
12:40
So again that's the claim that we have and that's the claim that has been challenged by again Repare Therapeutics.
12:48
So the way we remain that Toronto Knockout version three in a dual targeting format.
12:53
So now they have two elements per gene, and each element contains 2 guide RNAs.
12:58
And if we compare now the effect on the knockout of non-essential genes, we don't see much of the difference.
13:06
But in the viability assay, you don't expect it.
13:10
But if you now look at the essential genes, you actually see a drop out, an increased drop out of around 10 fold comparing single and dual targeting.
13:20
That's why we now also generate that, genome wide library that I mentioned in the beginning in a fixed pair version.
13:28
So you will be able to screen a library that has 40,000 elements for the reduced coverage.
13:33
And it can lead to, depending on the system that you're using, it can reduce the size of your experimental setup tenfold.
13:41
And we tested that and it works really nice actually.
13:47
We also have the multiplexing approach.
13:49
And multiplexing means, OK, if you want to combine, for example, gene families to your interest in genetic interactions, of which a prime example actually is the BRCA PARP genetic interaction, you can go for these combinatorial libraries.
14:03
Genetic interaction means if you knock out each gene individually, the cell is [happy] and lives.
14:11
But if you knock out both genes at once, then you see sometimes the most extreme form of the genetic interaction with which is synthetic lethality.
14:19
And in breast cancer, for example, we can use that because a lot of cancer cells are BRCA negative.
14:24
And if you then go in with a PARP inhibitor, you kind of use the genetic interaction to cure specifically cancer cells.
14:32
And it's a huge space.
14:33
If you think about the genome, it's 20,000 genes times 20,000 genes.
14:37
It's a huge space that has to be explored at some point.
14:42
And there are a couple of experimental stuff that you can use.
14:45
The classical 1 is probably an isogenic screen.
14:48
So you compare you a wild child type cell line and a knockout cell line.
14:53
Go in with your library and then can compare the 2 cell lines and everything that is not depleted in your wild type cell line but depleted in combination with the knockout.
15:05
Basically it's a synthetic or genetic interaction.
15:10
Now the drawback with this approach is that you have to generate clones of the cell line.
15:15
So to be able to model reality, you should actually look at different clones because the tumour environment is very heterogeneous.
15:26
And one way to circumvent this is you can directly use your cell line of interest and go on with the combinatorial CRISPR library.
15:32
And that's where the marketing libraries are very useful for.
15:36
So you are basically independent of the cell line and you can use the same library and all the cell lines that you have, you avoid these clonality effects.
15:44
And by reducing the size of the experiment, you can do actually a couple more experiments in parallel and check basically all the synthetic lethal combinations in the cell line.
15:58
Now we have a couple of libraries on stock.
16:02
We are developing actively new libraries.
16:05
And here on the left hand side, you see some examples of a library where we have 1 gene family, it's druggable gene target for the kinase, phosphatase.
16:13
Obviously all these gene families, but it can also combine these libraries into combinatorial setups.
16:22
And one thing that is for us really interesting is combining basically a genome wide screen with a set of focused genes.
16:34
So you Multiplex basically the genome with a set of genes.
16:39
And I will detail this a bit more on, I think the towards the end of the presentation, but that's a very, I would say it's an outlook into the future and not many people will be able to do it.
16:54
But as an aspiration, I would still would like to introduce that concept and later in the presentation.
17:00
So just as a view on what we are also doing.
17:09
So we also do a lot of bioinformatics analysis.
17:13
By training, I’m a bioinformatician, so we recently published a paper that takes care of processing raw sequencing data to get the recounts, does QC on the sequencing for single and also paired end sequencing.
17:30
And the idea behind our bioinformatics service is always that we do a lot of QC.
17:36
So whatever analysis we are going to start, we will not start it until we know that the data we have is of sufficient quality.
17:43
So we look at the parameters of during the sequencing, we look at metrics in the data itself, is it correlation or other statistical measurements.
17:55
And only then when we process the data and have confirmed the quality of the data, we proceed into all kinds of different analysis depending on the screen.
18:06
So gene ranking analysis, visualisation is obviously very important, and that very well summarises what we do.
18:15
Everything is very customizable depending on the setup of the experiment.
18:21
We also want to outline the screening services that we offer.
18:28
Typically what we, when we think about a CRISPR screen, we think in timelines of around 12 weeks and that includes usually making the library in close communication to the customer.
18:43
So we can use sequences or plasmid backbone provided by the customer because they very often want to use systems that they know and that they have used before.
18:53
But you also can use our guide RNA sequences.
18:58
A huge topic is always controlled in the genes, but also in the library, but also in the experimental setup.
19:05
And that's a topic that occurs basically at every point in that cycle.
19:09
And when it comes to the screen, we currently can do 200 to 250 genome wide screens with our libraries in parallel.
19:17
And at least from space wise when it comes to personnel with a small startup, we have to kind of a little bit our with our resources, but that's actually possible.
19:28
And thinking about the genome wide combinatorial screens, we can do around 5-8 currently. Yeah, some QC measurements that we make sure are maintained over the screen as we maintain coverage.
19:45
We at each time point of the screening, we collect the cells, we freeze them away.
19:52
All the libraries come with 10x capture sequences.
19:54
We can do bar coding for example, so whenever you decide to go back to one of the time points, you can actually go to the cells, do mass spec, do single cell transcriptomic readouts.
20:10
And all this is available to the customers.
20:13
Again, data quality control so that we know the data is good.
20:17
We proceed to the sequencing, and this is one example of a very large screen with.
20:24
I think around 1 million guide RNA combinations. And we did 2 replicas just viability read out, correlation is really high is the drop out of the double essential knockouts. And only when we see these kind of pictures and data, we then proceed on.
20:40
Obviously all the data that we have is then available to the customer.
20:47
Now this is the aspirational road map on the genome wide gene relationship data sets that I mentioned in the beginning.
20:55
So our idea is basically to do genome by genome wide combinatorial screening in all kinds of contexts.
21:04
And we're actually looking for partners in the industry that are interested in doing this.
21:08
So the idea is if you do a combinatorial screen, 20,000 genes on the one side combined with less than 50 to 100 genes on the other side.
21:21
We assume that with a bit of optimization of the parameters, you can go down to 200 screens if you summarise all the cells that you would have to screen.
21:36
But if you then apply a couple of other tricks, you can reduce the size of the number of screens that you would have to do.
21:43
One thing is you only have to focus on genes that are relevant in your context.
21:48
You don't need to do a genome wide screen out of the lines because one set of the genes are not even expressed, so you can actually leave them out.
21:56
We can optimise by using different systems that we have announced, for example the fixed pair approach.
22:03
A lot of companies are interested in doing this in iterative cycles.
22:07
So you do 1 screen genome by 50 genes, feed that back into some kind of AI or some kind of fertilisation pipeline, and then select the next 50 genes on the screen.
22:18
And we can also use alternative nucleases, Think about Cas13 for example.
22:24
And our idea is that we can compress the number of 200 screens down to around 25 screens.
22:33
And the timeline for that currently is for us two to three years.
22:37
But towards the end of the three years, we would like to have such a data set ready at least for a couple of cell lines.
22:46
And given that process, we will learn if there's anything else I think we can actually optimise.
22:51
So yeah, it's kind of an idea that we have.
22:53
We have not started screening that, but we obviously want to know what people think about it.
22:59
So we appreciate any feedback on that idea.
23:03
And just as the final slide, just wanted to summarise what we do.
23:08
So everything that we do is very modular, very customizable.
23:12
We have different product libraries.
23:16
We do cell engineering.
23:17
We can do pooled and arrayed libraries.
23:18
We offer services in terms of cleaning services, again, pooled or also arrayed to a certain extent, do computational analysis.
23:32
We do a lot of consulting that comes actually with all the products because we're going to make sure that any reagent that we produce works in the end of our customers.
23:42
And we also do paid R&D, for example, cell engineering, all the products that we have come with a cas9 licence.
23:49
So we don't have to take care about that.
23:53
And it extends then obviously to the use of our technology.
23:58
Yeah, we I think the main USP that we have is the uniform library design because that affects everything that we do from screening to analysis and we ensure high quality centres at every step or any project.
24:14
With that, I'm at the end, I'm here with Micheal Murphy and in an of your letter we have the more poster in the other rooms.
24:21
So if you have any questions, wanted to have anything, please reach out, be happy to take any of your questions.
24:28
Thank you very much.