79 – Quantifying perturbed SynGAP1 function caused by coding mutations

Michael Courtney, PhD

Dr. Courtney’s Bio

Dr. Courtney leads the neuronal signaling laboratory at the Turku Bioscience Centre in Finland.  He also established and is head of the Turku Screening Unit, a facility affiliated with Finnish and European infrastructures for chemical biology and early stage drug discovery. The unit offers access to lab automation instrumentation and provides services, in particular for use of live-cell high-throughput microscopy, as well as computational screening approaches from partners in structural biology teams.  

Dr. Courtney completed his PhD at the University of Dundee in 1992 and moved to Finland to develop optical methods to study neuronal cell signaling.  Such methods continue to develop rapidly and are becoming increasingly important to identify disease mechanisms.  They can also be used in phenotypic screens to help identify drugs and other compounds that may have the capacity to  restore specific cellular functions.  This talk will describe recent work to develop miniaturized live-cell optical methods to help streamline the phenotypic identification of functional defect in variants o SynGAP1 and their application to variants known or suspected to cause disease.  This work is funded by a joint grant from SRF-US, SRF-Europe and Leon and Friends e.v.

THIS IS A TRANSCRIPT ONLY:

hello everyone and welcome to today’s webinar my name is Olga Bodie and I’m a part of the team here at cengap research
0:12
fund our presentation today is perturbations of singap one function by coding sequence mutation and development
0:20
of automated methods for their quantitation I have the pleasure to introduce today’s
0:25
speaker Dr Michael Courtney Dr Courtney leads the neuronal signal Lab at the
0:30
turkey bioscience Center in Finland we also established and is head of the turkey screening unit a facility
0:37
affiliated with Finnish and European infrastructures for chemical biology and early stage drug discovery
0:43
the unit offers access to lab automation information instrumentation and provide services in particular for use of live
0:51
cell High throughput microscopy as well as computational screening approaches from Partners in structural biology
0:59
teams Dr Courtney completed his PhD at the University of Dundee in 1992 and moved
1:06
to Finland to develop Optical methods to study neuronal signal cell signaling
1:11
such methods continue to develop rapidly and are becoming increasingly important to identify disease mechanisms
1:18
they can also be used in phenotypic screens to help identify drugs and other compounds that may have capacity to
1:24
restore specific cellular functions this talk will describe recent were to
1:29
develop miniaturized lifestyle Optical methods to help streamline the phenotypic identification of functional
1:36
defects in variants of syngak1 and their application to variants known are
1:41
suspected to cause disease this work is funded by a joint Grant from syngap research fund U.S srf Europe
1:49
and Leon and France a recorded version of this webinar will be available on the srf website under
1:56
the resource tab and by the end of this presentation you will have the opportunity to get the
2:02
answer to your questions and we’d love to hear from you so please write your questions in the Q a below
2:08
for those just joining us welcome and again our speaker today is Dr Michael Courtney to discuss perturbations of
2:14
syngetwin function by coding sequence mutation and development of automated methods for their quantitation
2:21
it’s now my pleasure to turn things over to Dr Courtney thank you
2:27
thank you very much I’ll go for that introduction and thank you to srf to give me an opportunity once again to
2:33
tell about our work related to singap so I’m sharing my screen now can you see
2:40
that yes yep good okay
2:47
okay good so what I’m I’d like to cover in this presentation is uh kind of to
2:53
explain what it is that we’re working on a relation to singap one and uh why we are taking the approaches that we are uh
3:00
and as I would have pointed out this is a project ongoing project uh funded by
3:05
srf Indian and Friends uh and focused on Miss censorians of Singapore one we also
3:10
have a New Horizon Euro project on neurological disorders related to rest which includes Cinder
3:15
which is a large 15 partner Consortium project I’m not going to talk about that today it’s just starting but for this
3:23
presentation I would like to introduce uh some Concepts that are important for for what we I’m going to try and present
3:29
and then to explain how we’re trying to efficiently measure sync up one function and uh to determine whether these
3:37
measures that we are obtaining are robust enough for future restorative drug screens that could be performed
3:45
um so I think this community is quite familiar with Singaporean disease obviously so what I’d like to uh kind of
3:52
skip some of these Basics and get to the point here that missense variants are are currently categorized in clean bar
4:00
as a pathogenic or likely pathogenic and I know likely benign and of Uncertain significance so one of the questions
4:06
that arises is uh how are these Miss sense variants categorized for example
4:12
as pathogenic and I think this is uh based primarily on a
4:17
prediction uh because we know quite a lot about protein structure uh on the
4:23
other hand we are there’s a lot we still don’t know about protein structure so this is a representation of uh
4:31
structure of syngap from alpha full that’s maybe not the very best representation but it’s considered to be
4:38
not bad and the color coding here uh is showing how confident they are about the
4:45
positions of different parts of a singap one protein so the blue the dark blue
4:51
and light blue shows a pretty high level of confidence of the average positions
4:57
of uh amino acids uh forming the protein uh particularly in this region which has
5:05
a various well-known globular domains uh pH domain C2 domain
5:10
and uh Gap domain and also this protocol Helix now uh in the particular case of
5:18
syngap this isn’t uh just single one this isn’t just a prediction it actually is
5:23
supported also by crystallographic data this is data obtained from from a crystal of the purified protein uh
5:31
which which basically confirms these blue Origins uh having said that this is
5:37
a very static view of the protein but in a Cell a protein is very Dynamic there’s a lot of flexibility and this is not
5:43
captured either by this model nor by the crystal structure or in fact crystal structure indicates
5:48
uh flexibility and Dynamics by not being visible in certain parts of the
5:54
structure uh but what I’d like to highlight here is all these orange and yellow regions where the confidence is
6:00
extremely low it means we really don’t know what this part of the sequence is where it is positioned what kind of
6:07
forms it may take and it may very well be quite unstructured which means that
6:12
it’s very flexible it can take different confirmations it may have it’s often the case that these unstructured regions
6:18
form specific interactions with Partners in which case they then become structured in the presence of the
6:25
partners but if we don’t know where the partners are we won’t know what their structures are so we really don’t have a
6:30
lot of information about this the rest of the sequence it kind of looks like spaghetti here which means that the
6:36
prediction is very poor in its reasons so if we don’t have a possibility to
6:42
predict everything about the uh structure that’s based on the sequence
6:48
then perhaps we need to look at experimental evidence and this is what the focus of this presentation will be
6:54
what kind of experimental evidence can we get about uh what is the consequence
6:59
of a mid-sense variance in the thin Gap sequence and it’s is this evidence
7:05
useful for other things specifically for direct discovery okay so another question that arises is
7:12
uh whether all pathogenic uh missense variants are acting via hapler
7:19
insufficiency so in the case of the the most common known forms of uh syngap uh
7:28
there is a nonsense mutation that means an early stop codon in the coding DNA
7:36
and thus the RNA produced from the DNA in in one allele
7:41
one gene copy of syngap one and uh there are mechanisms in cells called nonsense
7:48
mediated Decay which are recognizing these early stop codons and destroying
7:54
messenger RNA that contains these early stop codons and the
7:59
consequence of this is usually a protective mechanism to make sure that
8:05
fragments of proteins and half proteins aren’t produced because they may interfere with the function of the normal full-length protein which is
8:11
encoded on the other of the two copies that we have of every Gene and
8:17
unfortunately singap one is one of those genes where a single copy is insufficient for function and that’s why
8:25
it’s called Epsilon sufficiency so the second copy is described as loss of functions so that’s very clear in the
8:31
case of each nonsense mutations but when it comes to sync up in a sense variance
8:36
uh the question is does this substitution I mean there are some parts of the protein where you if you take out
8:42
one amino acid and put in another one instead the whole structure will become unstable and there are again mechanisms in the cell that will in this case
8:50
degrade an unstable unfolded protein so we may just have no protein again so this is clearly a loss of function like
8:57
a haplan sufficiency like the nonsense mutation on the other hand it may be that some
9:02
missense variants lead to a protein that still can’t fold but perhaps it’s
9:07
exposing some hydrophobic regions because of this failure to fold and these hydrophobic regions can form
9:14
Aggregates and there are many neurological disorders where protein aggregates are known to play a role so
9:21
it this could be one of the outcomes of a given missense variation it could be
9:27
on the other hand that there is expression and there’s a solubility of the Sim Gap protein but it’s simply in
9:33
the wrong place index should in part be at the synapse and if it doesn’t get to
9:39
the synapse there will be some problems on the other hand it could be at the synapse that unable to iPhone’s targets
9:44
or perhaps they can act on its targets but it doesn’t respond to input says a lot of different reasons why the
9:50
midstance variation could lead to a disturbed or loss of function um
9:56
uh in each case we’d have to take a look to kind of determine what is what is the
10:01
issue and then there’s a different kind of conceptually different consequence uh
10:07
this possible term incense variant uh if it is expressed but not functional uh because it could take the place of the
10:14
wild type protein uh produced from the wild type allele and this is called the dominant negative function where it
10:20
could be a simple displacement where uh the synapse for example there’ll be maybe a missense rather than the wild
10:26
type and therefore the activity will not be present as it should be on the other
10:31
hand syngap has this special property to form primers and it may be that substitution of just one of the three in
10:38
the primer of uh wild type with a missense could actually poison the
10:43
entire oligomer but we actually don’t really know enough about this triangers to really be sure about that so that’s a
10:49
bit speculation at this time but it’s something to be to investigate okay so why does all of this matter well
10:55
uh if a mission really does act as a dominant negative uh this is a problem
11:01
if we now think about the therapy is kind of currently under development to counter hapler insufficiency
11:07
uh because these are designed to boost expression of same gut protein from the genes this means the normal wild type
11:13
allele and it also means the uh variant allele so if the variant allele is is
11:20
actually a dominant negative suppressing the activity of the world type then uh these kind of therapies uh may even be
11:26
harmful uh in which case the Innocence syngap uh
11:32
those within something Gap will be excluded from the trials and will be able to get any benefit cuts for their
11:37
own safety however if we can determine what is wrong with the missense variant and
11:44
actually put numbers on it and quantify it in comparison with the wild thought it means we could potentially set up a
11:50
screen for compounds to try and find compounds which make the variant more
11:55
like the wild type so a word about compound screening there are basically two different kinds of
12:01
compound screening it’s a big Pharma tend to go for this uh screening of normal novel chemical entities and in
12:06
this case they go for very large libraries uh they often hold libraries of more than a million compounds that
12:13
could be tested one by one or on the other hand there are these resources like DNA encoded libraries that allow a
12:18
pooled screen of more than a billion different chemical entities uh and the idea here is to find a number
12:26
of hips uh which themselves are never going to be drugs but they can be optimized Medicinal Chemistry can be
12:32
applied whereas safety studies need to be applied it’s a very long and expensive process but in the end a novel
12:38
compound may come out of it that can be used as a drug uh uh and clinical trials
12:44
uh which itself uh poses additional challenges for for rare diseases if
12:50
there are not all that many uh known patients with the disease so this is not
12:56
really what we’re talking about in this case on the other hand what we can do is instead of searching a novel chemical
13:01
entity we can search for existing drugs this can be either agents launched uh for other for treating other diseases or
13:09
or at least have been in a clinical trial because for these compounds there
13:14
is already safety information there’s already clinical data so it’s already in principle possible and safe to or given
13:22
certain conditions uh or someone to take these uh compounds under guidance of a
13:28
clinician uh so in this case the number of compounds is way smaller we’re talking
13:33
about something like four or five thousand that most uh that can be screened uh there are still regulatory
13:40
challenges uh they’re still uh costs for phase three trials and uh the big problem here is that uh these drugs are
13:47
already patented for one or another purpose uh some of them are uh the payments are expired so the the IP
13:53
potential is very low in these cases and the investment prospects are therefore very low so getting this funded is a
13:58
real difficulty uh however if there is evidence that these existing drugs could
14:04
be useful in the case of uh for example on this sense then a clinician can
14:11
decide whether or not it’s reasonable to use it off label and this has actually already been done
14:17
uh we can give us an example uh statins which have been which have been prescribed for syngap uh and as far as I
14:23
understand in the complete absence of data for whether or not starting has
14:28
could really be useful for uh syngap itself it’s more the statins have a
14:34
particular uh effect on lipid metabolism which was thought to be potentially
14:39
useful and and that’s been used in epilepsies before
14:45
okay so our project is called identifying targetable defects in
14:52
singaporeans and the reason for this is that we are trying to make possible these small molecule screens
14:59
particularly with a drug Library screen in mind to identify those drugs that
15:04
render us and get variant function more like the wild type so what we’re talking about here is an assay development
15:10
project we’re trying to get to the stage where we can run a drug screen we wanted to try to include the bucketing of the
15:16
touchscreen into the project but without having the assays without knowing what that entails it was actually not
15:23
possible to to kind of Define a budget where we could be sure we could go ahead with that so this project needs to get
15:29
to that stage to have those assays ready and hopefully uh proceed with a drug
15:36
screen in a future project okay so we have three aims the first aim is to monitor the localization
15:42
expression and dynamics of that variants compared with the wild type the second one is to try to understand the
15:48
mechanisms of mislocalization because we kind of know that there is some level of this liberalization and to look out for
15:55
aberrant gain of function because if the protein is not in the right place it may start to interact with proteins or maybe
16:01
because it’s interacting with other proteins that we actually don’t know about at this point and by uh applying
16:07
appropriate methods we may be able to identify those and they may that interaction may actually be a druggable
16:13
Target to follow up but we have to identify those first and the third aim is to apply structural
16:19
methods because these are still important uh to understand the uh
16:25
disruption of of protein structure and syndactness sense variance with a view
16:30
to possibly looking out for restoration based on the structural data however I have to say that the tools for
16:39
pretty instructor and ligand docking and so on are not really ideal or
16:44
identifying uh structural restoration approaches so this is really a bit of a
16:49
long shot but we thought while we’re looking at the protein structure it’s definitely something that we need to consider and take a look at
16:55
and in our project we’re uh which is really a kind of proof of concept for this that’s assay development we are
17:01
looking at these three Miss sanitarians so we don’t have too wide arranged to consider okay
17:11
it’s stuck okay here’s the next slide so the first question is really what are the functions of thing up one that we
17:19
can take a look at and how can we measure them in a scalable way because we want to do these screens in the
17:24
future we want to measure uh use hundreds or even thousands of different drugs uh so we need to be able to assay
17:31
hundreds or thousands of different samples uh in parallel okay so first of all simply got one
17:37
protein function so uh think up one is a gap it’s a GDP is activating protein so
17:43
what it does it it enables or promotes the gtpase activity of these small gdpas
17:50
as a Ras family such as Ras itself which is normally bound to a GTP and it’s in
17:56
active form and then uh after gdpi’s action this will form a
18:02
GDP and rest boundary DP is in a inactive form so while singer is a gtps
18:09
activating protein what it’s doing is in fact inactivating the rust it’s turning
18:14
it from this form to this form okay but this is just the the Gap domain there’s a lot of other parts of the
18:20
protein and to go into what the rest of this protein does we’d have to consider not simply the action of of syngap on on
18:28
another protein on brass but on the cellular context so here we move over
18:33
here so syngap wine is a synaptic gap it’s primarily localized to excitatory
18:39
synapses in the brain in other places too but are we not going into that now uh so I’m drawing a synapse very
18:46
schematically over here so normally we have syngap tethered uh to The receptors
18:51
to near the membrane synaptic membrane rashes up the membrane and what syngap
18:57
is normally doing is earning rust into an inactive State uh but when there is
19:03
synaptic activity there’s a relocalization of this in gap away from
19:08
the synapse which enables rest to have a chance to be activated to act on its targets such as the Earth pathway which
19:16
then in turn promotes the in the externalization of neurotransmitter
19:22
receptors an increase in neurotransmitter receptors it’s a membrane means that the synapse is now
19:27
more sensitive to a neurotransmitter release so there’s increased connectivity in increased firing of the
19:33
neurons and synapses and the promotion of network activity overall which can
19:39
then further promote the relocalization of syngap out of the sign up so you can see there’s a positive feedback loop
19:45
going on here so it’s perhaps no no particular surprise that if something goes wrong and there isn’t enough syngap
19:51
to reduce the activity of rasp then we can head towards our an epileptic state
19:58
or seizure activity so if we think about brain function for example this can be measured in house models of syngap we
20:05
can start to look at the cognitive roles the syngap run an activity and so on so
20:10
this is great for validation of drugs should we get to that state but it’s not ideal for screening we’re not going to
20:16
be testing hundreds or thousands of drugs in Mouse models but we can consider this in cell-based systems so
20:23
we can stick with the subway systems here so I want to highlight also that each of these steps leads On To The Next
20:30
Step so we should be able to read if there’s an impact on Singapore and localization we expect to have impacts
20:36
Downstream the relocalization is is salty then we should also have these Downstream effects so we can measure
20:42
this at multiple levels maybe we should measure multiple levels but even one level will already tell us if
20:48
something’s wrong okay so in our approach actually aim one is to set up an essays for localization
20:55
relocalization expression stability solubility and uh the activity of syngap
21:00
variance on signaling such as ERC once we have these
21:06
assays set up uh it should become fairly straightforward to then ask okay so this
21:12
is the the wild type function now does the presence of the variant which maybe has a disrupted function does it
21:18
actually prevent even the wild type from playing its normal role this is the the dominant negative actions we will look
21:25
at okay name two uh sorry aim one is so this is mainly uh has been carried out
21:31
so far by in the lab with uh well from Barbara and myself uh aim two is where
21:37
we start to work with uh the neighboring proteomics facility of Otto Coco and his
21:44
colleagues we will set up a proximity
21:49
interaction screen so Lily is very familiar with proximity labeling methods
21:55
already and to put this into the proteomic context means that we can now uh you know
22:02
uh unbiased manner evaluate all the proteins in the neurons and ask which
22:08
one of these are interacting with singap are there new proteins interacting with uh Miss dense variance of syngap uh that
22:17
are not normally interacting with the wild types in gap because they may well be uh targets that we can follow up okay
22:24
and then as I mentioned already we have the syngap modeling team uh to predict
22:29
uh the structural perturbations caused by the Miss sensor this is Alia ali um
22:36
like a postula in only pentagonist group in our University okay so before we go into the results
22:43
that we have been obtaining I’m just I want to show later on some preliminary results uh please bear in mind they’re
22:49
preliminary we haven’t finished our project with the work is still going on but we have uh some indications of where
22:57
it’s going and what’s working uh but before we go there I want to a little bit take a look at the literature
23:03
with you to explain why we are taking the approaches we are and why we’re not taking another Approach at least not as
23:09
a main uh path let’s say so uh thing up is a gap is it uh
23:18
causes uh inactivation of of gdpases of the RAS
23:23
family such as wrap one uh so it’s possible to express the Gap domain alone actually that doesn’t work for some
23:29
reason but the C2 and the Gap together can be expressed just this fragment the same Gap can be expressed in bacteria
23:35
purified to a high level and put into a tube together with rub one in this case
23:42
and an optical assay system so we’re talking sorry I’m throwing the structure here because we’re talking just about
23:47
this core domain our syngap so none of this these other parts are present in
23:52
this assay this is work that’s been published by minky science lab and it
23:58
you can see very nicely in this chemically defined purified system we can easily measure or they can easily
24:04
measure the activity of syngap C2 Gap fragment on a substrate wrap 1B
24:12
and if they look at a missense mutant in the Gap domain uh it turns out this
24:17
residue is fairly important for the Gap activity because you see the green line here uh from this in this sense uh it’s
24:24
much less active on Route 1D uh and then they also looked at a couple
24:30
of missions variants on the C2 domain uh these have a lesser impact but anyway
24:36
quite a measurable effect on the wild type function so this is very clear it’s
24:42
very nice simple readout um there are there are some issues to bear in mind though so this is completely
24:49
cell free uh so it’s lacking the cellular environment so whatever it is a cell might do to this system it’s not
24:55
measured here uh uh preparing these proteins in a
25:00
purified form it’s actually a little bit laborious so if we would have a lot of missense variants to look at it could be
25:06
quite tough to go through them all uh we are only considering the this fragment alone the core uh G-tube Gap the main
25:14
tandem domain and ignoring the rest of the protein which may modulate this function and the expression conditions
25:21
are artificial I said these are expressed in bacteria so uh one of the limitations that we want to avoid in in
25:28
our approach so these kinds of methods actually require quite a lot of protein so we cannot
25:35
simply extract the protein from neurons or from brain we have to use a so-called expression systems and bacteria are the
25:43
easiest example but there are plenty of others um and we actually already know it’s well
25:49
known that some wild type proteins from human or from other cells they they fail
25:54
to ex to express and fold properly but sufficient levels in these expression systems because the the proteins are not
26:02
that easy to express in high levels the cells actually have trouble doing that uh
26:08
so that doesn’t mean there’s anything wrong with those proteins I’m talking in now about wild type protein so it’s like
26:13
full length and GAP is one such example you can’t express full-length syngap in
26:19
a native form in in E coli because it uh it isn’t soluble
26:24
but you can express this domain as as we’ve shown in the previous slide so so the failure to to express uh is is not
26:32
actually an indication of of uh of a problem with the function of the
26:37
protein it could be a problem of the expression system or their problem with trying to produce
26:43
too much protein uh so so that’s that’s a complication if we want to use the system to understand misens variance a
26:50
second issue is that if we have activity or let’s say if we don’t activity have activity in the ice with the isolated
26:58
protein is that the same as what’s going to happen in Impact cells and uh here we have to think about chat rooms so
27:04
chaperones are proteins in intact cells that promote folding uh human cells have them E coli also have them they’re not
27:11
the same chaperones in fact uh so we may have for a protein that is that’s requiring
27:17
human sufferance to fold properly uh or let’s say a missense variant may still
27:23
fold and be functional in a human cell or in a neuron and if we isolate it from
27:29
bacteria it actually may not work so there are some of these unknowns that are going to confuse us a little bit uh
27:35
so there is a really valuable method for targeted screens let’s say we we see
27:40
that the missense is not working well and perhaps we want to try and find a molecule that improves that we’re taking
27:47
a more phenotypic approach we’re trying to understand in the context of a neuron isn’t it is a misens variant working
27:53
well or is there something going on so that’s why we’re not going for this but having said that there are some specific
28:00
questions that we may need to address using recombinant protein and it is a quite a lot of work like I said but
28:07
we’re very lucky that uh these questions have come up for us at a time when our university has to set up a protein core
28:13
uh for which they can provide services to us researchers uh basically to
28:20
generate proteins for us um and of course we need to pay for that but in this early phase they want to do
28:27
trial runs to demonstrate their services and they’re not charging this for that so this is perfect we can get them to
28:32
produce some zincap protein for us all the C2 Gap domain in this case and Pedro
28:38
at the Toca protein course has been doing just that so that’s really good so we’ve been able to uh make use of this
28:44
opportunity to help us in this single project okay so one step up from recombinant
28:50
protein is to look at syngap in a model cell line uh there are various ones that
28:57
could be used many of them are human cells or even better not just a millennial cell the British human cell line but they have their own Gap they
29:03
have their own kinators and so on uh uh and the hex 300 203 cells so we like
29:10
them they’re really easy to use we use them a lot in the lab it’s very easy to express put in something
29:15
and then we can do all kinds of measurements so in fact there’s a very nice study uh very comprehensive study
29:21
from Kurt houses lab that has done just this in act 293 cells with 57 different sync upon variants it’s pretty amazing
29:27
work uh and he has uh or his lab has shown amongst other things uh how the
29:34
stability of the protein looks uh in the case of all these variants in comparison with The Wild type shown in black hair
29:40
emergency activity of various Pathways including let’s look at the Earth pathway here of
29:47
the wild type compared with the different variants now not only black is
29:52
a wild type but there’s a few groups here and so we have to look at the blue columns which are misans varians
30:00
associated with disease and the green ones are actually
30:06
misconference not associated with disease so we can see for example in the
30:12
stability plot there’s actually not a huge difference with a couple of exceptions so actually all of these proteins all of these variants are able
30:18
to express so it’s not like the the nonsense nonsense media to Decay at all
30:24
there’s actually a pretty good expression of most of them um but the
30:29
a disease-associated variants well some of them they have a slightly reduced
30:34
expression and some of them have a slightly increased expression so we can’t say that okay that’s simply not
30:40
enough protein around uh at least not not in most cases uh if we look at the error Pathways so
30:47
uh we can see that this effect is actually a little bit on the
30:52
small side but certainly the wild Types on the left-hand side meaning it’s more active than most of the others but many
30:58
of them are not really very different from the wild type and those are the those which are most affected here
31:04
include some of these uh misens herons that are not associated with the disease
31:10
uh so it’s I’m sorry a neat system a lot of data here to look at
31:16
um there are disadvantages it’s not a urinal system once again we don’t have
31:21
the neural compartments we can’t ask about Singapore synapse because they’re in the synapses this is actually a
31:27
kidney derived cell it’s a very convenient system but not a neuron and this is an over expression of course
31:34
that is that uh these are normal cells and there are methods used to over Express syngap because there’s no syngap
31:41
in these cells under normal conditions so uh if we go to these cells and we
31:47
transfect them with zincap protein so you can see the cells here in blue in the microscope and we can see the finger
31:56
expression in magenta we see the higher levels at least here and what we notice is there are pretty large Aggregates now
32:02
to be fair if we look at neurons and we stay in for singap we do see Hunter we
32:09
it’s not like we have a smooth distribution we see Zynga in puncta but some of these puncture or blobs are
32:16
really huge in the case of the 393 cells they’re almost halfway across the cell
32:22
so these are perhaps a large Aggregates that are that’s not what we would normally see in
32:28
the neuron now we can control the size of these Aggregates by by adjusting the conditions by maybe trying to reduce the
32:34
level of expression but it’s hard to know what is not over expression in these cells because they normally don’t
32:40
express any singular so so this introduces a little bit of uncertainty here and again it’s a reason why we
32:47
don’t really want to use these cells in this uh analysis of missense variants we
32:52
actually want to take a look at the neuronal systems even though the 203s
32:59
are a great method a great system to work with okay and
33:06
so we were largely inspired by this early work from Kevin Rambo who showed
33:11
that if you take uh neurons and culture them from mice or if you take neurons
33:18
from uh these are wild type mice and these are mice uh embryos Mouse embryos
33:23
that have a complete knockout of syngap you can very easily see a big difference in the uh Oak activity
33:32
so what we’re looking for in our case is uh High throughput approach uh it starts
33:37
to get complicated if we have to have different Mouse lines especially the full narcotics actually embryonic lethal but what we can do is we can in this
33:44
case we can take right neurons we can put them into our high throughput microscope system automated system we
33:49
can include an arc reporter uh shown in blue here when we stimulate synaptic
33:55
activity in the cells at this point Time Zero then we see an increase in Earth
34:00
activity that’s easily measured and we can introduce a reagents uh
34:07
Administration aav to knock down uh syngap it’s an rmei and we can see an
34:14
increase in Earth activity so this all looks good and it’s as I said it’s automated so we have the authentic
34:20
neuronal environment we’re talking about the endogenous protein this is real thing up in a cell that normally
34:27
expresses syngap uh neuronal compartments exist there are synapses here we can go in if we run to and
34:34
introduce Miss sensing Gap nevertheless there’s still some issues here so for example uh this is in gap yes it’s
34:40
single but it’s not human sync up uh we have the wild type there we don’t have all the missense uh introduce uh having
34:47
rats that already expressed uh missense singap instead of the wild
34:53
type would be a problematic uh and difficult to achieve but we have to bring them in on top of the endogenous
35:01
thing Gap and then these are actually wholesale measurements both are in Gavin’s paper and in our study uh so the synapses are
35:08
there but uh what we’re actually seeing is that Arc activity all over the whole
35:14
cell is being increased uh when we reduce the levels of sync app so we’re
35:19
not getting a synapse specific readout of singap function here
35:25
uh we can also go to human cells and I’m taking this example uh from
35:30
government who kindly provided us with his uh cell lines so we can produce human neurons in our lab as well
35:37
um and here’s a one of the readouts from from his paper on this so you can see in
35:44
this particular graph that there’s a difference between the red and the blue and lines so there’s an increase in synaptic activity in The Knockout lines
35:52
compared or The Knockout neurons compared with the wild type neurons but the effects are a little bit modest I
35:57
would say and it’s only through uh combining two different lines of wild types two different lines of Knockouts
36:05
uh the statistical significance Institute with this particular measure at this particular title so there’s other data some of it’s a little bit
36:11
clearer uh some of it’s a lot clearer in fact but depending on what you measure uh some of the effects are a little bit
36:17
on the small side uh but it is a real human neuron system it’s a human syncope
36:23
that’s being expressed here uh there are neural compartments once again um but we thought that this perhaps may not
36:30
be the the best system for screening it may be good for validation for testing particular points but uh and that’s what
36:37
we plan to do uh we will try and use human neurons for validation but not for screens in this case for the misan’s
36:45
parents I want to highlight the the mouse models obviously we’re not talking about Mouse models for screening uh thousands of
36:52
drugs uh but uh there are some very useful models have now become uh have
36:59
been developed by the kubernetes lab and Kevin rundberg’s lab uh in these cases
37:05
they are expressing uh these human variants in this case there’s a frame
37:12
shift and the point mutation leading to a fermentation
37:18
and these will do allow behavioral cognitive measures which is what we care
37:23
about in the end after all and investigation of brain circuits and so on uh so for valid at the validation
37:29
stage should we come up with some molecules such models could be very useful but what I want to highlight here
37:34
as well is that these Studies have already shown that it’s the syngap alpha I is a form that appears to be
37:41
particularly important for restoring higher functions such as cognitive measures and that’s important because
37:48
it’s the authorizer form that we’re studying in our research as well okay so we’re talking about setting up
37:54
screening essays it means automation it means that we need to have the singer missense variants expressed in all or
38:02
most of the cells we need efficient Gene transfer and the best way to do this is to it seems to us is to use uh I don’t
38:10
know Associated virus system a gene transfer it has low toxicity prolonged expression it’s very efficient very easy
38:16
to work with everything’s good except the genome size is very small uh this is
38:22
a problem uh with that we had to deal with but the first thing we had to deal with so the the typical of the self aav
38:28
Vector that you can get hold of is not even big enough to put in full links and
38:33
go so then we can’t study the wild type compared with any sense for instance because we cannot put in sync up into
38:39
DAV uh and that’s before we think about adding a fluorescent tag because we want
38:45
to look at the living cells to look at Dynamics and so on so the first thing we had to wrestle with was how to fit
38:50
singap into an area we can we can fit into the aav uh
38:57
plasmid but aav particles will not form unless the
39:02
cassette between the two atrs inverted terminal repeats is small enough so the
39:08
whole thing can be about 5 000 bases but not more uh so it’s kind of long story short uh
39:14
we try to optimize all these uh sorry this one all these non-coding regions uh
39:20
so that to enable us to introduce a large Trend stream which is induct in our case and we were able to save about
39:26
1 000 bases of space compared with the typical commonly used a
39:32
Vector while still retaining most of the expression levels
39:37
or let’s say about half of the expression levels okay and then that is good but we still
39:44
didn’t have room for the fluorescent protein which is another 750 base pairs that we have to introduce uh so we can
39:51
put in singular but we want to see where it is so we want some kind of fluorescent marker of it so for this
39:57
return to fluorescence complementation and in this case we have uh
40:02
made use of the observation that you can cut a fluorescent protein which looks something like this a kind of barrel
40:08
shape into two parts and in the best case we can have a single strand here of about 15 amino acids and the other 250
40:15
220 amino acids or so over here and if they
40:21
uh interact together with one another then we recover fluorescence so all we
40:26
need is to have a SIM gut a very large syngap protein with this very short tag and if this uh rest of the thing of the
40:35
fluorescent protein which is introduced with another virus which is present in the cell then we
40:42
will get fluorescence where our singap is in order to ensure this interaction takes place we actually need to have an
40:47
affinity tag here so we actually have to have both this orange part and the red
40:53
part but anyway it saves us about another 650 base pairs of space and then
40:59
finally it means that we are able to introduce syngap with not a fluorescent
41:05
tag but a fluorogenic tag into neurons with aav and if we also introduce the
41:11
rest of the fluorescent protein with this Affinity construct then we can
41:18
generate of fluorescent detectable thing Gap so this
41:24
is an immuno block to show that the of the box here is in the way but
41:29
um the this is just the uninfected cells with the endogenous in gap right neurons
41:35
normally Express syngap and it’s about 130 kilonewtons if we then introduce the
41:41
aav system to express the thin Gap with the fluorogenic tag then we have this
41:46
larger form of Zinger which has the fluorogenic tag covalently attached and we also however do see an increase over
41:54
here so this formation of the fluorescent form isn’t 100 efficient unfortunately and this may cause some
42:01
complications in some cases but it’s anyway useful enough that we can carry
42:06
out some assays and these are the very first kinds of images we got with the very initial constructs they are not the
42:14
fully optimized and corrected syngap sequences so I’m putting wild type hearing rotations
42:20
instance one and this sense two so these are anyway showing that the constructs
42:26
are allow us to visualize in living cells a localization that looks quite
42:32
normal for thin Gap along processors a little bit in the cell body whereas the
42:37
misense variants seem to be somewhat different let’s say although
42:43
there are puncta here that has just fewer of them you know okay
42:49
so now we’ve we know that the system basically works and we can put in all this optimization uh and we need to make
42:56
sure that the sync up sequences we are using are absolutely perfectly correct fortunately there are new full plasmid
43:02
sequencing services available nowadays we don’t need to go for singer sequencing and we can check what we have is really correct
43:09
um and we’ve been able to produce and validate uh six constructs now and use
43:15
them this is the syngot one Alpha One wild type three mid sense variants that
43:21
we that are part of our project are nonsense a variant as a comparison I
43:27
want to highlight that normally if we if one would have this nonsense uh mutation in a human cell it
43:36
would be degraded like I mentioned by muscles mediated decay but because we’re just expressing the
43:42
coding sequence we don’t have the extra uh untranslated regions in our aav we
43:49
don’t have the nonsense mediated Decay pathway kicking in here so what we will
43:54
see with this variant is a an expression of syngap as it would look
44:00
uh if the singap one was truncated um uh without them nonsense media okay
44:08
and we also have the same thing that went up to uh World type as well uh
44:14
We’ve designed this in a modular way so it’s very easy to for example put in the mutation these mutations into the one
44:21
after two we can switch the uh split fluorescence protein tags and have alternate Affinity tags for the double
44:28
labeling approaches that we will need for the dominant negative investigations
44:33
uh we want to have alternative complementation strategies to try to
44:39
achieve higher efficiency of labeling which could be important in some cases we’re working on the proximity labeling
44:47
tags that we need for the proximity proteomics we name two and we have a
44:52
variety of other variants that we’re working on so we have quite a number of
44:57
other constructs ongoing which have not yet reached the stage of full validation but
45:04
they’ll soon be available to us and we’ll be able to use them in our project okay but I want to give you an example
45:10
of what it looks like in terms of the localization which we’re measuring together with the longitudinal stability
45:16
which I will explain so just to remind what it is that we’re trying to do here we’re trying to look at the missense
45:21
singap and understand uh what is wrong in this case identify the localization of the syndropin compared with in
45:28
comparison of the wild type and it’s very hard to kind of manually look at these images and say okay this looks
45:34
normal this doesn’t look normal we can kind of say that but if we start to now introduce drugs and maybe there’s a
45:40
shift from one state to another we need to really numerically quantify that and that’s what I want what I want to show
45:46
and once we have this numerical quantification then we can start to go for the rectification attempts
45:52
uh and yes what we’re trying to do is to have some kind of system which allows us to screen uh up to four thousand drugs
46:00
and we have to bear in mind that some readouts may just be too noisy and it will not really be possible that other
46:06
readouts hopefully will be less noisy and robust enough to do that so this is what we need to do can we find a way to
46:13
make the readout sufficiently robust for this so we are using complex cells which are
46:19
but not that easy to work with certainly not as easy as the two or three cells or cell three assays so ten thousand is
46:26
actually not very many samples in a screen for a cell free assay but for complex cell models it’s actually quite
46:31
a lot uh but so we need a simple assay fully automated High throughput uh with
46:38
uh we need to demonstrate highly high reproducibility of the readout ideally and it’s commonly done with a robust
46:45
screen you just do the screen once and look at the hips and work with them uh we may have to do the screen twice or
46:51
maybe three times and just make sure we’re not generating the worst case that
46:57
can generate a lot of false positives that is uh find compounds or drugs which we think are useful and then when we go
47:03
to do all the validations we find out actually no that was wrong we had some variation in our screen which made us uh
47:09
the fullest so we really don’t want a lot of false positives because we’ll spend the rest of our lives trying to figure out uh or find out that they’re
47:16
not useful okay so as an example of of the development
47:21
that we’re doing here I want to show you uh one of the tests we ran so we had
47:26
seven constructs that is one wild type three missionaries a nonsense variant or some additional controls and then we
47:33
were testing out four different detection methods using these split proteins and we had six different replicas in each case so all this
47:40
together leads to 168 different samples we put them in a three or four well
47:45
plate which is I played about this size which uh has little sample wells in them
47:51
about three millimeters across each and for each sample we put microscopy
47:56
Imaging Fields nine of them a neighboring Fields uh with nine to
48:02
thirteen different uh stead positions so we can get a three-dimensional representation of about 10 to 30 different cells and what we wanted to
48:10
know was was this measurement that we’re making reproducible
48:15
uh so that if we measure today and we measure it tomorrow would we get the
48:20
same result it’s no good if if we get one result one day but if we measure it another day we get a different result that would that’s something that we’d
48:26
need to know so that’s why in part why we’re working with living cells here so the cells are still alive tomorrow are still alive the next day we’re using aav
48:33
so we we can actually see expression for weeks in the cells which is very convenient so we can do this so we have
48:40
measured actually five time points in our original data set uh to check whether or not the readout is the same
48:47
day after day uh so this particular data set comes to about a hundred thousand image frames uh so the image analysis
48:54
obviously we’re not doing this manually drawing circles or counting spots so we need to find a way to automate this uh
49:00
and what we noticed and what you might have seen in the earlier side these little spots of singer they’re really
49:06
quite diverse you can see the bright bright ones there are sometimes some really bright bigger ones and then there’s some very dim ones and that
49:13
means it’s actually quite difficult for conventional segmentation approaches used to detect spots to find the dimmest
49:20
ones and the brighter ones and all very various different kinds and that’s why we went for uh training a
49:26
deep learning model uh perhaps I don’t go into this unless someone’s interested to hear more about it uh but we were
49:32
able to develop a deep learning model that could be used in uh imaging software that’s commonly used called Fiji or image K and our model in the end
49:41
that we were able to train to identify this thoughts were in a sense too good uh so on the one hand everything that it
49:49
detected we took a look at some examples and it really was the puncture that it was detecting
49:54
we couldn’t say at any point okay it’s taking something that definitely isn’t a punter but it was detecting even those
50:00
really tiniest ones that we wouldn’t normally include in our analysis so I think that’s okay because we can just
50:07
gather all the data and we can say okay now we don’t want the smallest ones because there’s no way they can be signed up since it’s just way too small
50:13
uh so so we need to add filters to our data during the downstream analysis so
50:20
we were able to do this identification of the puncta we want to separate those puncture that are in the
50:26
cell bodies of the neurons uh to from the ones which are along the dendrites and so on because we’re interested in
50:32
synapses and we can’t be sure that’s what’s going on in the cell body but in the along the dendrites uh we can be
50:39
more sure that we’re dealing with synaptic puncture uh and in the end from this data set that I described we had a total of uh 30
50:48
about 35 features puncture leading to about 1 billion values
50:54
measured overall and this is quite a lot obviously uh so what I can say is the
51:00
segmentation was actually pretty efficient and very fast the model is very good and worked well but the feature collection
51:06
collecting all this data was really quite slow
51:12
um and this could become a bottleneck if we want to measure from hundreds or
51:18
thousands of trucks uh so this is definitely a concern uh but we’re quite
51:23
fortunate in Finland that we have this organization whose job is in part to
51:28
provide researchers with free access to supercomputers so we can without charge
51:34
use supercomputers for for running this kind of getting through these kind of
51:40
bottlenecks and not only that but the the third fastest supercomputer on the
51:45
planet according to uh measure last year is in Finland looked after by these guys it’s owned by a European Consulting with
51:52
10 countries but we also have access to this one um and what’s more this uh these guys are
52:00
Lux have recognized that image processing is a growing area requiring
52:06
uh the compute uh high performance Computing as they call it and so they
52:11
have put aside some resources that is themselves to help us bring our uh
52:19
image quantification scripts into uh a form that’s applicable to the
52:24
supercomputer because we can’t simply simply put those scripts onto the super computer they have to be converted into
52:31
in a particular way and that’s something that is something these airlines are very good at and they’re helping us do
52:37
that basically for free from our site so that’s another resource we’re bringing into the syngap project uh luckily
52:44
okay so I just want to give you an example of the kind of data we end up with from these billion uh
52:50
data points so we can just pick one feature let’s say we want to look at the mean intensity of all these uh puncture
52:56
we’ve detected we can make an average for every sample we had 158 different samples uh we’re just considering those
53:03
with one detection method and now we can compare the six replicates of the wild
53:08
type with the different missense variant variants and it’s pretty obvious there’s a difference and let’s do it so we can
53:14
see that even measuring the mean intensity there’s a clear difference which is what we need but actually for a
53:20
drug screen this this little level of variability is actually even a little bit High
53:25
uh we want to to have something more robust but we have all this other data so the question is what do we do with
53:31
all this data how do we how do we handle all these features and how do we do filtering to get the most robust
53:37
possible readout that is we want to get these error bars as smaller as possible we want to it to be as clear as possible
53:44
uh what the differences between a wild type and a missense variant
53:49
um and ultimately is the data so reproducible that just by looking at those numbers uh can we see okay that’s
53:56
that’s this form of uh missed them for that’s the truncation mutant or that’s the wild type because once we get to
54:03
that stage then we can say does this or that or another compound
54:08
bring the results towards the wild type state from the original Nissan state
54:15
uh so what we want to do is make use of all the relevant data we could talk about what is relevant and what is not relevant among the features but this is
54:22
what it looks at as a kind of summary looks like as a summary so what we did here was ask the computer to look at all
54:29
the data and represent those features uh for all the data
54:35
without telling it which ones are the wild Pirates which ones are the truncation regions which ones are the missense variants
54:41
um just position every sample according to the properties of the puncture on a
54:47
2d space and this is called the umap and so what we did when we looked at the result we found that in fact all the
54:54
truncation data from the truncation variants are all over here all the data
55:00
from the missense variants are over here and all the while type ones are grouped over here and they’re not overlapping at
55:06
all uh so this is good this is just the first step uh it gets better but um the
55:13
other thing we wanted to do I mentioned we were measuring repeatedly over multiple time points to check the
55:19
stability so what we want to know is uh are these data points very noisy and going from one place to another over
55:26
time so if we simply draw a line between the points so here we have the first data point second data point third data
55:32
point fourth data point for one particular sample one particular well in that three or four well played uh one
55:38
particular field of neurons and we can see that by and large those samples are
55:44
kind of keeping within their little space which means they’re staying very similar uh over time
55:50
there are a couple of cases where there’s a bit of a jump here in this sample and a couple of minutes and
55:56
switching from one group to another so there’s essentially two groups here but by and large they’re really staying in
56:02
their space over time and this is important for drug screening because we know that if we were teslaction for constructs it will take a long time and
56:10
we do not want our assay to suddenly show something different later on compared with how it was earlier on so
56:16
we know we can put our plates in the machine for five days and uh we everything keeps in the same space
56:22
that’s actually very important and what am I filtering so we didn’t filter anything we we just allowed uh
56:29
use of all these little tiny spots as well which are not really what we probably want to look at so there’s a question where are the cutoffs that we
56:35
should use um so this is the result of putting in a cutoff uh and the cutoff was not
56:43
with was based on a kind of experimental evidence uh not it wasn’t us that well we chose
56:50
it but we chose it based on on empirical evidence so that’s important this is not any kind of supervised machine learning
56:56
we’re not overfitting here what we’re doing is we’re saying okay we actually realized the small spots are not what we
57:02
want we only look at medium-sized spots in the data and we are we asked the
57:08
computer to again sort out all the samples according to the properties of those spots again without telling the
57:13
computer what are these different variants and we can see we get really good uh separation between wild types
57:21
the truncation mutant and the missense mutants there’s definitely overlap error between them it’s kind of interesting I
57:27
mean it could have it could have been very different but it turns out that they’re actually very similar even though there are different mutations
57:34
um it seems like the this one I think is the LR mutants a bit off to the side but
57:40
um but I think this tells us already that we’re kind of ready for a drug screen
57:47
that this kind of uh readout because that’s such good separation between the
57:52
replicas okay so what else can we do so it’s not
57:58
enough for the Zynga to be at the synapse it actually has to disperse in response to activity and that’s why we’re using living cells because we can
58:04
ask that question so really beautiful work from huben has lab I showed this uh dispersion response uh so the question
58:11
is can we actually see this in our system in an automated fashion and uh sorry I’ve got the time scale labeled
58:19
wrong here but uh under Baseline conditions we have this kind of distribution of syngap this is the wild
58:25
type and we can we can induce a synaptic activity and what we see here the
58:31
movie’s not working are supposed to be a movie now uh sorry about that but yeah good morning okay now it’s running so
58:38
you can see this little spots kind of get a bit fainter and the neighboring regions get a bit brighter so this is
58:44
Quantified here so we can see at the end of the time course it’s not really very much difference between uh where the
58:50
puncture was and the neighboring region uh so this is what we expect it means that okay that we can in an automated
58:56
fashion you automated pipetting on our samples and read this out from all of our samples and we can do this in our
59:03
throughput uh and this is uh some data it’s just a very preliminary uh data
59:09
it’s definitely a little bit noisy and much noisier than we would like and there are things that we can optimize but we can clearly see with the wild
59:16
type there’s a clear dispersion as a response the number of hunger is reducing over time
59:23
um and this is where we realized actually it’s the medium anchor that we want to look at because if we look at the solar panel they’re not really
59:28
responding but in terms of missense the let’s say a preliminary conclusion we
59:33
had to be careful here we need to do more work to be sure but it looks very
59:38
much like there’s all those variants they’re also responding they’re also capable of localizing capable of
59:44
responding maybe the ALR is a bit defective in the response uh maybe this is a little bit steep but we really need
59:52
to do more experiments to be sure uh as a small poetry that’s kind of interesting they seem to have bigger responses in the variants and in the
59:59
wild type publicly so whether or not this is going to be a screenable readout I don’t know
1:00:04
but I wanted to highlight one more thing um it’s a bit cumbersome as in the last example to actually have to put a pet on
1:00:12
a stimulus in a high throughput manner to monitor the response so in the first
1:00:19
case we were simply doing Imaging in the second case we were doing Imaging during the petting
1:00:26
uh using robotics on the imager uh it’s durable but it’s not so scalable so it
1:00:33
would be much neater if we could use optigenetics to cause the translocation so the recruitment lab has demonstrated
1:00:40
what the mechanism behind the translocation uh which helps us to to
1:00:45
develop an optogenetic method and here’s the outcome of this uh in some preliminary experiments some setup
1:00:51
experiments so the magenta here uh you can see the spots of syngap at the
1:00:57
beginning of the experiment you can see there’s little spots along with dendrite where the cyan is the location of the
1:01:02
dendrite and in fact I’ve already given away the result because it’s actually the final location of syngap so
1:01:09
in this experiment we are simply using light to switch on a
1:01:14
uh the the fusion or the dispersion of the mechanism causing dispersion of
1:01:20
syngap and we can see in this movie here hopefully you can see those magenta spots disappear pretty quickly
1:01:27
um okay so uh one last slide uh this was from our uh singap modeling team uh the
1:01:34
capacilla provided this slide for me today so uh what they’ve been looking at
1:01:39
is using molecular Dynamics simulations do you remember I was saying that when we look at the predictions of syngap
1:01:45
structure there are some regions we can predict some regions we can’t predict but among those regions we can predict uh and even the crystallographic data
1:01:52
it’s very static it’s just I’ve kind of Frozen condition of of the protein and in
1:01:59
reality the protein has flexibility is moving around it has to move around to to to do what it’s supposed to do to
1:02:06
have its functions to be a gap for example uh and so what they what they
1:02:12
modeling team did was to use molecular Dynamics simulations software called Amber to consider how does the wild Type
1:02:20
move compared with the missense variance that we’re looking at and they can see
1:02:27
quite rapidly that uh within a few nanoseconds there are really significant
1:02:32
conformational effects of putting in these uh
1:02:37
Miss transference and that perhaps could be predicted already because these are quite large changes but critical reasons
1:02:44
in the structure are nevertheless these are actually quite localized
1:02:49
deformations of the structure uh and what’s seen because a whole uh C2 Gap
1:02:55
domain a PhD to get domain is modeled here there’s no large scale unfolding so
1:03:01
this is kind of important it’s not that these uh variants are causing the whole protein to collapse because or even the
1:03:07
C2 movement to collapse uh but it’s just kind of causing a local disruption and
1:03:12
the rest of the domain seems to be able to cope with it now it doesn’t mean that the uh the domain is functional and
1:03:18
that’s another question that has to be addressed but at least what we know is that it’s not driving misfolding of the
1:03:23
protein okay now with that uh I would like to acknowledge the people who have been behind this work so for the
1:03:30
experimental work uh it was mainly carried about Lily with the help from Barbara myself uh we’ve just uh
1:03:36
recruited Pierre to the lab he’s going to join us next week to now we have all
1:03:41
these additional uh uh tools the syngap that I described in
1:03:47
aav we have a lot of work to do now and it’s we really need more hands on deck and uh Pearson help us with that in the
1:03:55
future and uh the singap modeling team that is
1:04:00
Alia Ali and olivensing kindness lab I forgot to say they also have a new
1:04:06
syngap sorry a new single project funded by srf to look not just at those three
1:04:13
missed Instagrams but to do predictions on a much wider range of syngap uh
1:04:18
missense variants and develop a web portal that will help clinicians make decisions as to whether
1:04:25
variants are are potentially pathogenic or not and the proteomics will be done
1:04:32
with Auto calco and his team I mentioned the Lakshmi helping us with the super
1:04:37
Computing and look and Johnny help us with some additional computational issues and Pedro I mentioned where they
1:04:44
took operating corn and this work wouldn’t have been possible without sync up research fund and the other Friends
1:04:50
funding so thank you very much for listening I can I’d like to take some
1:04:55
questions if you haven’t yeah that was great thank you so much that was really good
1:05:01
um there’s one question in the Q a real quick that I can you see it or I can read it
1:05:07
um says very interesting talk in your aav vector did you use full links and get
1:05:14
ones and what promoter did you use for this and get one expression construct
1:05:19
right that’s a good question uh yes so yes we did useful links thing up one the
1:05:24
downside is the space is definitely constrained so we’re not able to introduce uh untranslated regions which
1:05:30
are potentially important uh help regulate the localization for
1:05:36
example they could be local translation at the synapse so we unfortunately aren’t able to introduce some of these
1:05:43
additional aspects uh with a of a h okay and then the expression constructs that
1:05:50
we’re using in the end we decided to go for uh EFS ef1s or say it’s a there’s a
1:05:59
quite a commonly used promoter uh for housekeeping team uh elongation Factor
1:06:04
one uh it’s about one and a half kilobases normally but there’s a extra short version that’s been used uh cas9
1:06:11
which is huge as well uh which is only about 250 base as long uh the downside
1:06:17
is that it’s not particularly neuron specific but we are using neuron
1:06:22
specific promoters for the uh the complementation part uh so that we will we only get
1:06:29
fluorescence complementation in neurons because we have a synapse promoter for example in the complementation area
1:06:37
hopefully that answers question
1:06:43
um if anybody else has questions too yeah go ahead or if they are I know she’s here or if anybody wants me to
1:06:49
promote you to uh panelists I can definitely do that too and you can answer or ask yourself
1:06:58
um yeah hi thanks Lauren hi Dr Courtney
1:07:03
thank you so much for that amazing talk and Lauren please um you know interrupt me and have
1:07:08
anybody I’d love it for you to continue with the managing the questions but I just wanted to for sure
1:07:14
say a few things first off wow thank you for such amazing work and
1:07:22
um I really loved how you went into really really good detail on different
1:07:29
types of assays and why you know what why some are appropriate and why some are sufficient or not sufficient that
1:07:36
was really really helpful um because I think a lot of us as parents I’m a parent a lot of us are
1:07:44
wondering like why well what what why isn’t there an essay why can’t we do you know what’s what’s going on let’s just
1:07:49
look at this why why are we having to raise money why are we having to give so many cell lines
1:07:55
um and I think you really gave a lot of detail around what the what the tough issues are and
1:08:01
how why it’s tough to figure this stuff out um your data where you were showing the uh
1:08:10
the colored spots coalesce if you could kind of um
1:08:15
maybe go back to that a little bit and tell us just just kind of remind us really simply what what’s going on here
1:08:23
so I think you’re looking at I understand the wild types I understand the not the RX is the nonsense right
1:08:30
yeah is that nonsense um is that one that’s not going to be not
1:08:37
going to have uh nonsense mediated Decay is that right uh normally it would have been submitted
1:08:43
to Decay if it was in the the normal genomic context uh but in our case
1:08:48
because we don’t have these uh untranslated regions and splicing going on then it is being expressed aberrantly
1:08:55
if you like right and so so is there are the the wild types are being expressed the same
1:09:02
way as the in the sense right that all these all these versions are being expressed the same way is that right
1:09:07
they’re all they’re all being delivered to the same neuron culture I mean there’s a neuron culture which we put
1:09:14
into a multi-world plate and we have some wells where we put in the wild type the virus encoding the wild type and in
1:09:20
fact we have two different if you’re wondering what this purple’s bottom paint spot is one of them is the
1:09:25
corrected wire attack this wild Type e is correct but there was a silent mutation
1:09:31
which was just in our the DNA that we were working with and although the silent mutation means that the protein
1:09:37
should be correct uh we just thought maybe it’s better to have the completely correct sequence
1:09:45
yeah so you never know so um in fact we’ve even seen some silent
1:09:51
I’ve seen one silent mutation and singap be categorized as pathogenic and I don’t
1:09:58
understand why like I understand I understand possible reasons why right I understand possible reasons why I don’t
1:10:04
know why in particular that one was or whose decision it was or what they were seeing that made them think that but um
1:10:10
yeah okay so I guess there isn’t a way with this system to get to the nonsense
1:10:16
mediated decay uh that’s we haven’t really thought about that so how would we get to it I
1:10:24
mean I mean I think well you’d have to have yourselves you’d have to have yourselves start with it
1:10:29
I think yeah well I guess not because you have time yeah I haven’t thought about that at all uh so yeah I mean basically we’re
1:10:36
working within the constraints of the AV system so I think it’s probably going to be tough if we went the full-length uh
1:10:42
syngap and we want to have those introns that are participating in the nonsense media mechanism
1:10:47
uh it just doesn’t fit so we could work with fragments if we want to
1:10:53
specifically look at that it depends what the question is I mean it’s the idea to ask whether there’s a way to
1:10:59
specifically Express the nonsense media decay of a syngap variant or I think suppressing
1:11:07
the nonsense mediated care overall probably isn’t a good idea right because it’s a kind of safety mechanism right I
1:11:12
guess okay I have an idea I think it would be uh expressing the wild type and co-expressing an untagged
1:11:18
um a non-tag nonsense I think that’s what it would be just to see if it changes the world type
1:11:25
yeah um with the nonsense media to Decay same
1:11:30
place yeah yeah um yeah I think we in principle we could
1:11:36
go in that direction with with lentivirus the problem with lentive I mean lentivirus has a lot more space and
1:11:41
it works and we use it for the human neurons for example over time uh but the problem with antivirus is it has a
1:11:48
different safety category we have to work with it in a special lab and with the with the human neurons they go
1:11:53
through all these medium changes and we start with a big plate of undifferentiated cells and yeah yeah
1:11:59
differentiated neurons that are kind of validated to be lenti free
1:12:06
well played yeah yeah yeah I I don’t mean I didn’t mean for I
1:12:11
didn’t mean for your uh I I meant just to do it once not to do it as yeah not
1:12:17
to do it as a um in a test yeah okay so so that’s definitely something I think that could be done
1:12:23
um with some thought about it much but uh it’s yeah potentially doable so if
1:12:28
you have a specific do you have a specific nonsense in mind or this this no no not a specific no not
1:12:35
a specific one I was just wondering how that spot would change and how that spot might change the wild type because we’re
1:12:41
looking we’re thinking like does that the whole idea of the protein truncating
1:12:47
variants not being dominant negatives is that they’re not there yeah but in this system they are there so I was wondering
1:12:54
if this is looking like some kind of dominant negative that doesn’t normally happen in
1:13:01
wild type and maybe it doesn’t matter maybe it doesn’t matter if if it’s just if the spots don’t overlap in this sense
1:13:07
maybe it just doesn’t matter I just was thinking for certain if you’re going to test a whole lot of missense you might
1:13:13
want to know if any of them look like dominant negative is what I was thinking or if any of them look like or if any of
1:13:19
them look like how fluid sufficiency and so those those are the kind of the two right those are kind of the two
1:13:25
things you want to know do they look yeah like have someone do they look wild type do they look like dominant negative
1:13:31
do they look like half low insufficiency do they look like hypomorphs do they look like neomorphs
1:13:37
right and to see that you’re going to see spots in different areas yeah that’s correct yeah so so in a sense except we
1:13:44
haven’t gone we haven’t made the constructs in that way but we kind of can almost immediately do what you’re
1:13:49
suggesting which is basically to let’s say we express the wild type and we put in an unlabeled Mission so we don’t even
1:13:55
see it but we know we put it there and we can ask whether the world type spots are kind of drifting away right negative
1:14:02
effect what we tried to do is to go for a double labeling approach so we’ll receive both at the same time yeah which
1:14:08
is a bit more challenging and so we’re slower getting there uh it’s on its way but but yes we could go for this kind of
1:14:15
part labeled part unlabeled approach yeah that’s a good idea well since your spots are so tight you can yeah yep yeah
1:14:22
definitely okay what what are you most excited about here
1:14:29
um like as far as what do you like are you most excited about a particular assay as being you know because as you
1:14:35
said you’re trying all these assays yeah you don’t need all the essays to work you need some you know you need one or a
1:14:41
couple to be really robust assays yeah I I kind of like the way this came
1:14:47
out it was I mean I didn’t know from the beginning how easy or difficult it would be to do the segmentation and to run the
1:14:52
analysis I think I was very happy with the Deep learning uh model to to identify the the puncta and I was also
1:14:59
very happy to see this kind of come out without kind of having to Resort any supervised learning because I’m quite
1:15:04
worried that I mean I also dabbled with super supervised learning methods and you can see it just really easily
1:15:10
overfits the data right because it already knows you know which ones which yeah that’s not what we want to do we
1:15:17
want to we want the properties of the proteins to tell us what’s going on
1:15:22
um so so I’m yeah I’m really happy with the separation we have already uh there’s still some some tricks we could
1:15:27
play to try and improve it uh but maybe we’re already there and we can already
1:15:33
make plans for the drug screens as it is yeah uh I like the optigenetic tool for
1:15:39
inducing dispersion I think that’s going to be just so much easier than having to
1:15:45
handle the fluidics in a high throughput Manner and so that can can you remind us that happens over a very fast time
1:15:51
period right uh I I had an incorrect time scale so I don’t really remember the time period I think this is probably
1:15:58
minutes not seconds sorry about that uh right but still that’s still that’s very fast as far as like it is very fast yeah
1:16:05
yeah if we look at recognize data I had that I stole that for this one sorry I’m
1:16:10
blocking the screen for myself at least um so we can see in his experiments which are endpoint measurements I think
1:16:16
they’re fixed in the cells and they’re taking a look they also see a very fast effect I I don’t know if that first time point is
1:16:22
five minutes uh here it’s 10 minutes over here but it’s it’s yeah it’s a really fast effect
1:16:28
uh it’s nice to have the live cell reader you know I think most labs are not
1:16:34
using the the lifestyle reader because of course it’s more convenient to fix samples and go when you can to the uh
1:16:40
microscope and do more detailed studies but what we’re always trying to do is seeing what is it we can see with
1:16:46
lifestyle because the Dynamics often tells us a lot more than just uh yeah so
1:16:52
in the in the lifestyle dispersion is that something you repeat or is that something you do at once with the up the
1:16:58
the optical role so we’ve done this kind of experiment before not within Gap and it was really
1:17:04
convenient that we could repeat it because it’s a bit like the electricity studies quite often they just depolarize
1:17:10
cells multiple times under different conditions and build up a whole lot of data uh you know almost automatically in
1:17:17
very little time so ideally we would be able to repeat this multiple times on the same sample and build up a kind of
1:17:23
average and get much more precise information unfortunately what we’re seeing is that this is
1:17:30
the reversibility here is pretty slow so I don’t know I can’t tell you how
1:17:36
slow it is and it may be if we have a lot of samples that’s fine we just need to do this experiment on each sample and
1:17:43
then we can come back when we finish and doing all over again and do it all over again and we would be able to get this
1:17:48
kind of repeated measure that would give us more confidence and less noise in our
1:17:53
measurements which would make it more of a screening assay it will take time but it would still be doable
1:17:59
um so all I can say right now is that it certainly doesn’t reverse in in a matter
1:18:05
of a few minutes which is kind of interesting
1:18:10
yeah even with wild tape absolutely
1:18:15
yeah I didn’t expect that yeah we are we are providing quite a
1:18:21
strong input uh other as was recognize that so same
1:18:27
idea here have you thought of it yeah have you thought about doing less of an input and seeing what sure yeah yeah I
1:18:34
mean that’s quite easy with the opportunity method it’s very easy you can you know set up or go into different
1:18:39
stimulus levels what seems to happen is you just get a nursery response but we need to do a bit more work we’ll see
1:18:46
um yeah um and then I I had a question about the very small puncta that you were
1:18:53
realizing had nothing to do with the synapses potentially they’re in the cell
1:18:59
body or in the you know other another part that’s not the synapse is that right yeah but it could also be noise I
1:19:05
mean we’re a bit constrained by the amount of light we can put into our samples because they’re living samples uh light is for the toxic so we have to
1:19:13
be a little bit careful so we can’t you know that’s another reason why people like to fix cells and they go to their comfortable microscopes and they can
1:19:19
really use almost any laser power and they get away with it but in our case the cells will be paid after that so
1:19:25
right we wouldn’t be able to do it so we have noise which will give us little spots uh so that is part of it uh yeah
1:19:32
there’s also I mean if we look let’s go back to this Dynamic measure here
1:19:38
um it’s it’s a bit noisy but I think you can see in our data and you can probably see
1:19:44
I think well it’s kind of hard here but in uh in in the Iraqi paper of 2015 they also
1:19:52
have some data I think kind of indicating that the the location localization in the
1:19:57
dendrites isn’t necessarily totally smooth it’s a little bit pancake there as well so perhaps you can see it
1:20:05
here what I’ve done here is to it’s the same data but I’ve applied a de-convolution algorithm this is the raw
1:20:11
data so you perhaps you can see these tiny spots in the shaft
1:20:16
yeah it’s like looking at the night sky it’s like dim Stars right that’s what it
1:20:22
looks like yeah yeah um the other thing I I didn’t get to maybe I can show it here is that um that
1:20:30
if we can I can just manually Move Along here then you can see at the end all the
1:20:35
puncture that were here disappeared most of them but these have stayed very strong so we also have larger punter of
1:20:42
singap that are not going anywhere with the stimulation and I don’t think it’s because the optic genetics isn’t in this
1:20:49
cell I think they’re there’s just kind of larger puncture uh what they are I don’t really know I mean there’s always
1:20:55
a risk that we have a bit of over expression in some cells and we have some artificial Aggregates or perhaps
1:21:01
it’s just the way there there’s a lot of there is also singap in the Soma in the cell body uh
1:21:07
which can form this fairly large puncture I I’m not sure if they’re epsomatic
1:21:13
synapses I sort of doubt it but so there are other structures here as well
1:21:19
sorry I interrupted oh no I mean the reason I was asking this at the beginning I think is I think
1:21:26
you’re probably gonna I think I know the answer now but I was gonna ask if you were going to take that data of the smaller stuff and somehow separately try
1:21:33
to analyze that for the cell bodies but it sounds like if it’s mostly noise that’s not gonna you know absolutely oh
1:21:40
you did try okay yeah yeah we tried different size ranges and did that help do you think it was nice
1:21:46
um yes it did but we know there’s going to be noise and yes of course of course
1:21:52
model that uh yeah I mean what we can do with these kind of the computational
1:21:57
methods we can basically generate synthetic data we can generate synthetic noise and run all the algorithms on that
1:22:03
we’ll see what happens so we can help kind of fine-tune our methods with that
1:22:08
kind of approach yeah yeah there’s a lot of great tools out there that have been developed by others that we can just pick them up and
1:22:14
use them that really helps well thank you very much
1:22:22
thank you that was really good do you anyone else have any questions or
1:22:29
we did go have lunch or dinner um probably all right guys well thank
1:22:36
you very much for doing this thanks to Jr for coming and asking questions well actually I have one more question
1:22:44
that’s not about your work it’s about our work so as a patient advocacy group can you can you give us any advice on
1:22:51
what what we should be doing so we should just be I mean from from my perspective I think the SR perspective
1:22:58
in general is find as many patients as we can figure out who they are figure out what their
1:23:03
variants are figure out where we see in the sense figure out all
1:23:09
the you know where we see fuss try to see if we figure if we can figure out especially with these single amino acid
1:23:16
changes right whether they’re whether they’re not sure what’s happening whether they’re tolerated whether they’re pathogenic
1:23:22
um and is there is there anything we can do besides try to find patients and make
1:23:30
make cell lines uh so there’s one thing that I mean in
1:23:35
this particular project we picked we were already working with two uh missense variants and then uh the g344s
1:23:44
was suggested to us because it’s actually a very common one which makes a lot of sense but it does it’s also that they all happen to be in the C2 domain
1:23:50
and so what we see here is that these variants uh
1:23:56
they okay they’re in the same domain but still they are not obviously in places
1:24:03
that influence the localization of the protein so the C2 domain has previously been described as somehow important by
1:24:10
unknown for unknown reasons to be uh promoting the Gap activity
1:24:16
but no we’re not looking at the Gap activity here we’re looking at the localization so
1:24:22
so the the way we’ve selected these transference is their Institute domain
1:24:27
we just thought that would make sense I mean biochemist by training so we think about domains but now if you think about
1:24:32
the the clinical side and the the patient side then then obviously it’s the symptoms of the important things not
1:24:39
the not really the the domains and the residues so what’s I what I I can’t see very easily
1:24:46
for example if I look at the clinvar data it kind of said yes there are you know there’s some their
1:24:53
symptoms here or no there are no symptoms here but some kind of classification to link the misense
1:24:59
variants to the symptoms would be really helpful so for example if we would look at different misidence variants that
1:25:06
appear to result in similar uh symptomology that would be
1:25:13
kind of useful yeah is that how but I understand there
1:25:19
you know they’re not that many yeah there are many different Miss answerings and not that many patients
1:25:25
who have one or other misense Varian but at least they could maybe be grouped in this way that would that work do you
1:25:31
think well I mean that’s definitely a major goal for every you know every
1:25:37
pediatric geneticist I think but the yes but the um I I think that what we’re seeing in the
1:25:44
patient population is that while 80 of them are protein truncating
1:25:49
variants they have a range of symptoms and that the Miss sense the ones the
1:25:57
single amino acid changes that have been categorized as pathogenic so those those
1:26:03
that we’re considering Miss sense are mostly in domains that have that originally had a partial crystal
1:26:09
structure right first off and that so there’s a little bit so there’s some
1:26:14
that’s not a um I don’t want to say bias but the sample
1:26:19
is you know not all it’s not everyone and then there’s
1:26:25
this basically the same range of symptoms that you see in a protein truncating variant so what we’re what
1:26:32
we’re not seeing is oh all the people with variants right in this region have
1:26:37
the same symptoms and all of it so it’s more like there’s a big range of symptoms that you know everybody has some amount
1:26:45
of of all the symptoms but some kids have way earlier seizure
1:26:51
burden some kids have uh way or way more severe intellectual
1:26:59
disability burden some kids have a much more difficult sleep burden you
1:27:05
know they can’t sleep but others kids can sleep right so so there’s these different sort of silos of
1:27:12
of interrelated problems because maybe the seizures are what’s happening for the sleep and maybe the seizures are
1:27:19
happening for the ID right I mean they’re interrelated but they’re things that we kind of count differently
1:27:24
and we’re we’re not seeing really great patterns emerge where whereas on this slide we
1:27:33
see really great patterns emerge here right like these are we can tell the
1:27:38
protein truncating from the Miss sense right here absolutely we’re not really seeing that in the patient data but I
1:27:46
will say it is still somewhat early days and there are people working on
1:27:52
um all the citizen data so getting all our all our patient medical records
1:27:58
put into a form that they can be digitally compared um because when parents talk like first
1:28:05
off when parents talk it’s really interesting we’re like oh my kid loves water oh my kid loves water oh my kid lives water right so we’re kind of
1:28:11
noticing some stuff we’re saying oh my kid can see really well in the dark oh my good can see really well in the dark
1:28:16
oh my kid could see really ball in the dark right so parents have some interesting information
1:28:22
that is really useful but at the same time we’ll say oh my kid has a hard time with Transitions and oh my kid has a
1:28:28
hard time with transitions well those can look completely different one can be having huge amounts and and first off
1:28:35
they’re at different ages right so so but but they can just be ultimately different classes even if even if they
1:28:43
have the same thing so we have to have this sort of systematic way to understand and then things like liking
1:28:48
water and seeing well in the dark those aren’t really in our medical records so um anyway I just sort of telling you the
1:28:56
complexity we would all love to see this play out a certain way but it’s not happening like it does in some diseases
1:29:04
where the channel Works more you know over time or it works not as much right those
1:29:09
are two different directions for a protein to go we’re not really seeing that yet with Singapore
1:29:14
the two things that come to mind that for me so one is that uh like I said it would be quite nice to kind of do this
1:29:21
kind of work with picking a bunch of uh let’s say a bunch of
1:29:27
instance that some kind of similar symptoms and another punch that have other similar
1:29:32
symptoms so they’re you know two different groups do they full integrated from groups there or not regardless of
1:29:38
whether they’re in the C2 domain or something else and then the other thing is what you said that the different
1:29:44
protein concating variants are giving different symptoms right so well
1:29:52
well I’m saying I’m saying they have a different profile like we’re talking about the same symptoms but a different profile within that yeah yeah
1:29:59
but that means that the nonsense needed to Decay is not working completely right otherwise
1:30:05
they there wouldn’t be anything it’s really not clear it’s it’s just so
1:30:11
not clear right like I am hard to say because it’s hard like I
1:30:17
think with the haploid they should see since there is one gene working [Music]
1:30:22
you know with a lot of with most genes in the human genome one’s enough you know so there’s they’re fewer genes for which
1:30:29
one one you know there’s way more recessive diseases than dominant right is that
1:30:35
right or am I am I making that up and so yeah so that so there is good
1:30:40
protein in there there’s just not enough and then the question is how much how
1:30:45
much is the other protein just a disrupter or just irrelevant
1:30:51
right that is is that is the is the pathogenic variant irrelevant
1:30:59
like it’s just missing it doesn’t look just like a huge you know multi-genic deletion would look where this right or
1:31:06
even just just maybe not multi-genic but you know the full Gene being deleted um right so there’s nothing there’s
1:31:12
nothing there to be a disrupter with a full Gene deletion right yeah yeah so we
1:31:17
kind of want to compare that to everything else but
1:31:22
um so you don’t we don’t know at the moment whether the difference
1:31:28
uh in the profile of symptoms uh is because of
1:31:35
some other genetic background effect or because these transgression regions are not actually being completely that’s
1:31:42
right and I think I think what we’re seeing though is that I so I I’m not exactly sure why
1:31:50
um I’m not exactly sure what the data is on this but we’re what I’ve heard and
1:31:56
what I’ve kind of instinctively believe is that syngap one
1:32:02
variants that are pathogenic have very high penetrance and that they have a little bit of
1:32:08
variable expressivity so I think it’s really high penetrance
1:32:14
and so I yes I would think um that
1:32:20
perturbations in other genes even even just like allele even tolerated alleles
1:32:26
and other genes in this background might have an effect right so even if you had a a shank three or a you know dlg four
1:32:34
which is which we know there’s a dlg4 binding site on sync app one so
1:32:40
potentially a change in dlg4 that’s normally tolerated would
1:32:46
be different in a Sim Gap one haploin sufficiency background right that would that might uncover some change in
1:32:53
another Gene and so there’s a lot of genes that it can interact with so um yeah so I think it’s just you and me
1:33:00
talking at this point so sorry to keep you so long I’ve I obviously