Scientist Stories: Michael Lin, Creating Interfaces to the Brain with Fluorescent Proteins

Hello. So first, I'd like to thank Feng Zhang and the McGovern Institute for inviting me here. It's a real honor to be here. And I'm excited to share with you some of our recent developments on optical sensors and control elements. So as Jeff Lichtman has elegantly Joshua Elkington just talked about, technology has really always driven our understanding of the brain. And in fact, it's— the field of neuroscience is at any given time defined by the types of experiments we can do because the brain is such a large and complex organism. So disruptive technologies, what you can consider disruptive technologies, have really driven what we can do in neuroscience. And of course that started out with the Golgi stain, as we just talked about, being used by Ramon y Cajal. And then later on, of course, because the brain is an electrical organ, the ability to actually record electrical signals gave us great insight into what the brain was actually doing. And then later on, the ability to patch clamp the membrane and measure the contributions of particular channels to the action potential gave us a more mechanistic understanding of how that electrical signal was being generated. And most recently in the past 20 years or so, optical methods have become— have revolutionized our ability to investigate the brain. And many of those optical probes are based on fluorescent proteins. And as we know, fluorescent proteins are very versatile because they are a genetically encoded light source.. And all you really need to get that light source is to take your gene, which encodes your fluorescent protein, and express it in your cell type of interest. For example, in the very first demonstration of heterologous expression of GFP in the worm nervous system, or fuse it to a protein. And you can do more complicated things, like you can make sensors of particular signals in the nervous system based on fluorescent protein. So as many of us here use, the calcium sensor GCaMP is a green fluorescent protein on which has been attached a calcium-binding domain, calmodulin, in such a way that when in the absence of calcium, the green fluorescent protein is dim, but in the presence of calcium, it assumes its more normal fold and becomes fluorescent. And the details of that have been investigated by labs such as Lorenz, who unfortunately can't be here today. And of course, that's very useful because we can now visualize in real time action potentials in circuits in living brains. So that's on the optical sensing side. Of course, even more recently has been the revolution in optical control led by Ed and Feng and Karl, and with this nice picture from our hosts. The idea is apparent. You, as we know, you can generate action potentials or you can hyperpolarize neurons using light. And of course that gives us— and then with subtype specificity imposed by the genetics. And so that gives us now control of the nervous system firing. So where are we going next in terms of technology development? Well, I think a lot of us have a common goal of putting as much of the nervous system under optical control as possible. And that's because of the great complexity of the nervous system. And so having optical control, which can be temporally and spatially precise, is a major advantage experimentally. And of course, there's always a need for more better sensors. And so I would say on the sensing side, Our lab has identified some holes in our toolbox that we would like to fill. Besides GCaMP, which is a sensor of activity but uses calcium as a proxy for electrical activity, we would like to make sensors that are very robust and have millisecond kinetics to really look at the fine timing of neuronal electrical impulses. And then we'd also, on this time scale of events where electrical activity is fast and protein activity is of moderate speed on the order of minutes, and permanent changes or more stable changes in nervous systems such as structural changes to the synapse are on longer time scales of hours, we have some general methods such as FRET reporters to look at protein activities, but we don't really have well-established methods to look at long-term structural changes. And on the control side of the equation, we have the rhodopsins to perform optogenetics at the— on electrical activity at the millisecond and second time scale. But we don't have great methods for controlling protein activity or controlling long-term changes in the nervous system. So I'll be telling you about 3 probes that we've been working on in our lab. The first one for looking at structural synaptic plasticity the second one for looking at millisecond timescale electrical activity, and the third one for optical control of protein functions and potentially synaptic structure as well. So we'll start with the structural synaptic plasticity reporter. As many of you here are experts, and much more expert than me, we know that synaptic growth underlies many learned network properties. For example, in the hippocampus, everyone's favorite system, place cells emerge during exploration of a new space, and that emergence is thought to be due to synaptic strengthening at particular synapses, which is somewhat like LTP as we study in hippocampus slice. In the hippocampus, of course, the primary excitatory neurons are the pyramidal cells, for example, the CA1 pyramidal cell. And on the pyramidal cell dendrites are thousands of synapses, and those synapses can, to a large degree, be modulated in terms of their strength and their size individually. And we know that modulation that's required for long-term potentiation, for many kinds of long-term potentiation, requires new protein synthesis. And much of that new protein synthesis probably happens locally because dendrites are are filled with mRNAs and ribosomes, and you can see them in close proximity to the synapses, even within spines in this EM by Kristen Harris. And of course, this was known for a long time since experiments by Kandel, in which protein synthesis inhibitors were applied to slice as LTP was performed. In control slices, LTP stays up, over 2 hours, but in the presence of inhibitors of protein synthesis, LTP decays after about an hour. And so together these kinds of data have suggested a model whereby activation, strong activation of a synapse through release of neurotransmitters induces the translation of RNAs that are preexisting in the dendrite into proteins.. And those proteins are then taken up by the activated synapse. And those proteins might include structural components that would help to make the synapse grow and stabilize synaptic growth. And so this is a pretty commonly accepted hypothesis, but the problem we saw was that it was really hard to actually find strong evidence for this hypothesis. And the hypothesis would predict that there are new proteins that are delivered specifically to growing synapses. Another way to frame this would be through alternative language of this hypothesis, that synaptic tags, newly synthesized proteins that specifically label stimulated synapses, exist. And if those tags exist, can we use new protein distributions to mark stably potentiated synapses, to serve as a reporter of stably potentiated synapses? So to do this, we wanted a way to specifically label only newly synthesized proteins and not the total protein pool. We could do that in a live neuron. And so we devised a reporter that's based on a fluorescent protein, but basically is a drug-inducible fluorescent protein with a drug giving you control over whether or not you have fluorescence. So we want drug to gate protein fluorescence, but to gate it in a time-dependent manner. So the probe we came up with is this, which is a little bit complicated to understand, but it's pretty easy to use. So what we do is we fuse a protein of interest, for example, like PSD95, the major synaptic structural protein, to a tag. And a tag consists of half of GFP, then a protease, which is surrounded by cognate protease sites, and then the other half of the fluorescent protein. The protease is— by default will cleave itself out, and we use hepatitis C virus proteases. Non-toxic. You can express it in mammalian neurons without causing any effects. And it has a very high sequence specificity for its substrate. So it doesn't cleave— it doesn't cleave well in trans and it doesn't cleave many other substrates. So it will cleave itself out. It would actually then get degraded. But this will break the GFP before it has a chance to mature and so that you never get fluorescence. But the other reason we chose this protease is that it has very good cell-permeable inhibitors that have developed to treat hepatitis C virus. So we can take one of those inhibitors, apply it to our neurons, and then from that point on, newly synthesized proteins don't suffer the degradation of the GFP. So the two halves of the GFP are held close enough together because the proteolysis is blocked., and then that GFP has a chance to mature and become fluorescent. Um, so basically it's very easy to use. In the absence of adding a drug, everything that was made is not fluorescent. But after you add the drug, everything that's made from that time on is fluorescent. So here's a movie showing it working. Time 0 is before we add the drug. And, um, over the next 21 hours, uh, we're just looking at PSD95 fused to this tag, which we call timestamp YFP., and we can see where newly synthesized PSD95 goes. So PSD95 was always there in the culture in the neurons. It's just before we add the drug, we can't see it. So after we add the drug, you're seeing new PSD95 localized throughout the dendritic tree. But it seems to go earlier to some synapses and not to others. And it's precisely that kind of heterogeneity that we're interested in capturing. So what we wanted next to do was to specifically stimulate some synapses on the neuron but not others, and then ask whether new PSD95 goes specifically to the stimulated synapses. And I should say that we hypothesized that PSD95 would be one of these proteins that are newly translated in response to synaptic stimulation because it was identified as a potential target for FMRP, an RNA-binding protein that's mutated in fragile X mental retardation. And is localized to— and the RNA has been shown to be localized to dendrites. So we wanted to ask whether PSEN95 would be synthesized and go specifically to stimulated synapses. So we cultured neurons in these culture chambers that are made out of PDMS. They're basically very simple microfluidic devices. And the chambers have two compartments. Which are separated by a wall. The wall is 50 microns wide, and through the wall run these little tunnels that are 7 microns wide. And so that's wide enough for neurons to send their dendrites and their axons through, but too narrow for the cell bodies to migrate through. So you can see a neuron stuck here trying to get through. It can only get its dendrite through. And that allows us to apply a stimulus to one side and hit some neurons that have extended dendrites through, and thereby preferentially stimulate these synapses on this neuron, not synapses elsewhere. Of course, we could try local perfusion as well, but we found this to be a more reliable way to do this. So in this experiment, we've added BDNF as a model for synaptic growth and synaptic stimulation. So BDNF is sufficient for a synaptic change that looks like LTP, and some groups believe that— it's also necessary for endogenous LTP. So BDNF is applied on this side, and it's going to diffuse down these tunnels where it encounters synapses on dendrites of a neuron that's actually lying on the other side of the barrier. So you can see— and we now look at new PSD95 after we add the drug to both sides. So we can see PSD95 wherever it comes up. And this is the time course here. So we could see what looks like new PST95 enriched at stimulated regions at the point of closest encounter to the BDNF that's coming down the tunnels. So indeed it does look like PST95 goes preferentially to stimulated synapses. So we then explored other stimuli that might be a little bit more physiological. If we stimulate mGluRs with mGluR agonist, DHPG— so mGluRs also gate LTP. At high doses, as people here are very familiar with, it causes LTD. But at low doses, it facilitates LTP. So it can go either way. So DHPG applied to the right chamber will diffuse down through the tunnels, and in this location encounters synapses of dendrite from this neuron. And you can see here at the early time point, newly synthesized PSD95 individual synapses of that neuron. We can also— this is another agonist of metabotropic glutamate receptors. Again, we see preferential localization of UPS95 to the stimulated portions of the dendrite. It's very obvious here compared to an mCherry cell fill. And we've finally done this recently. Those data were published in our Nature Neuroscience paper. We've done this recently with glycine LTP, which we thought was important to show since in the previous cases we were using pharmacological agents. In this case we're using glycine to induce LTP that's triggered by release of endogenous synaptic vesicles. And this is an NMDA receptor-dependent LTP. So we see on the side that has glycine, on the side that has no glycine, so on this side the NMDA receptor is being facilitated, we can see what looks like early and more abundant new PSD95 on this side. So is this new, is this due to local translation, is it due to PSD95 that's made in the cell body and then is brought to the synapse? We can't really rule out the latter possibility. We favor the former possibility. And the reason we favor it is if we remove the 3' UTR, as we've done in this experiment, we don't see local, early local PST95 accumulation at the stimulated site. Instead, you see what looks like a gradient, a movement of PST95 from the cell body. So I hope that's apparent. It's a little fast, but it wasn't so fast before I upgraded my computer. But anyway, so you can see what looks like a wave of nuclease 95 coming from the cell body. So in the absence of 3'-UTR, the PSC95 RNA is not localized in the dendrite, it remains in the cell body. And so we think this is indirect evidence that the effect actually does require local translation of PSC95. The other evidence that it might require local translation of PSC95 would be if we fail to see this in an FMRP-deficient mouse, for example. And so we've been doing those experiments, and actually very recent data, very preliminary data, we actually do see a defect in new PSD95 localization to the stimulated side in PSD95-deficient mouse neurons. So, something for this part is that using this reporter of newly synthesized proteins, of specific proteins that you're interested in, we have what we believe is direct evidence of a newly translated synaptic tag. And this effect requires a 3' untranslated region, which is the region that contains elements for localization and repression. The effect is reduced in FMRP knockouts. We believe that the hypothesis of local translation mediated by activity in FMRP does hold true for the case of PSD95. Ongoing efforts will be looking at really establishing the severity of the defect in FMRP knockouts, finding other proteins that also may be targets,, and then using our system to go into mice and then visualize the distribution of new proteins in dendritic trees in, for example, CA1 neurons during learning. And to do that, we had to make a knock-in mouse. And so we've started doing that for both PSD95 and another gene, Arc. And we're further along with Arc because we started it knowing that Arc would be a positive control. So Arc, ARC, as many of you are aware, is a gene whose RNA is synthesized, is transcribed in response to activity, but its RNA is also translated into protein in response to activity. And so we tagged it with timestamp to serve as a positive control for our later experiments with PSD95. So to do the tagging is pretty standard. We've put in this timestamp YFP cassette at the C-terminus of ARC. We also have this Frt-flanked PGK-neo-positive selection cassette. At the ESL stage, it looks like that. And then after we make mice that go germline, we perform the flip recombination to take out the PGK-neo cassette. So we've actually just got those mice. They're homozygous viable, which is nice. But previously, we had a homozygous viable FRT-neo-FRT cassette containing mice, and we started just verifying that those neurons actually do have drug-dependent Arc signal. Those are some chimeric mice. They're cute as mice should be. Then we took out the neurons from the mice and plated them in culture. You don't see anything here because the neurons were not treated with drugs, so we don't see any signal as we would want. If we treat with BDNF alone, most of the time you don't see signal, but— oh yeah, you don't see signal at all because there's no drug. If you treat with the BMS protease inhibitor alone, most of the time you don't see signal, but you see signal in a couple of neurons. One is here, which seemed to have some ongoing Arc synthesis and then upregulates its Arc synthesis at a particular time. There's also a neuron here that you could see more weakly, where there's some low level of Arc comes up. But this is just with the inhibitor alone, no stimulation. And then if we add the inhibitor plus BDNF to stimulate Arc synthesis, then you can see very robust Arc, new Arc expression in many neurons in culture. So it does seem to work, and so we will be using those mice in the near future. So that's the first part about structural synaptic plasticity, so I'll move on to the second part, which is electrical activity. So as we were talking about, fluorescent proteins allow real-time optical sensing of neuronal activity with calcium reporters. And that's been used for a long, long time. Interestingly, the first calcium reporter was— it was a long time ago. I think it was in 1997. Since then, calcium reporters have gone through many iterations of development. They have been worked on by probably at least 5 labs continuously for over a decade. We're talking about more than 50, maybe 100 people years of work going into developing better and more sensitive calcium sensors. That's led us now to GCaMP6. This was GCaMP3 that I was showing here, but we now have GCaMP6, which gives you these really large signals in response to single action potentials. So you can visualize single action potentials with calcium sensing. However, calcium, as many of you are aware, is not everything in the nervous system. It's a proxy of neuronal electrical activity. And it's a slow and not always direct proxy. The voltage signal itself is much richer than the calcium signal. It's faster because the action potential is up and down within 2 milliseconds, whereas the calcium signal persists in the cell for hundreds of milliseconds. You can have substantial changes that don't lead to calcium entry into the neuron. Those could be either hyperpolarizing or depolarizing. So if you want to look at up and down states, for example, that's completely silent in terms of the calcium response, but there's definitely a voltage change. So people have been interested in voltage sensors. And when we started our work, the best voltage sensor was probably ArcLight. And ArcLight has a large fluorescence response, about 40% from -70 to +30 millivolts. And so it has the largest voltage-dependent fluorescence change of any sensor. And so people are quite interested in it. And ArcLight is a fusion of pH-sensitive GFP fluorin to the C-terminus of this four-helix bundle, which is the voltage-sensing domain from voltage-sensing phosphatase. So voltage-sensing channels and voltage-sensing phosphatases use this four-helix bundle, which— on which the— on the fourth helix reside positive charges, which sense the transmembrane voltage. So in the polarized state of the neuron, the interior is more negative than the positive, and this helix is down. In the depolarized state, that helix moves upwards. For reasons that are still not understood, the fluorine, when fused to the C-terminus of this voltage-sensing domain, becomes dim in the depolarized state when it's moved upwards. The problem with ArcLight has been its kinetics. The original ArcLight had on kinetics about 15 milliseconds and off kinetics about 30 milliseconds, which is an order of magnitude slower than action potential itself. And so it wasn't very sensitive for action potentials. That's been improved very recently by the Pierobon lab. So they've changed the voltage sensing domain from C squared siona to the CHIKN. And that is a faster voltage sensing domain. And so the kinetics has been sped up to— I don't quite remember, but something in the high single digits of milliseconds. But it's still a little bit— it's still slow, as you can see from the time scale here. That these— the widths of these fluorescence transients are about 20 milliseconds, still much wider than action potential itself. So we wanted to make a new design for a membrane voltage sensor that could possibly be faster. And so we started looking at what was known about the structure of the voltage sensing domain. At that time we started, there was no structure of one domain in the up and the down states, but there was a molecular dynamic simulation based on a potassium channel voltage sensing domain. In both the up and down states. So actually I'm running this movie from this paper, which I run backwards, because they were actually investigating the closing of the channel. The original structure was in the up state. So I ran it backwards, so now the movie is starting with the channel in the down state, and now it's moving up. And so something you see that's very interesting, so we'll see it again, it starts down, moves up. It doesn't look like much of a change, but what's happening is the S4 helix, the red helix, is moving upwards as everyone expects. But not only that, but something very interesting is happening in the S3-S4 loop, extracellular loop of this protein. This loose linker is actually forming an extra helical turn after it moves upwards. So I think you can see that now. It's loose and now moves upwards and this extra turn forms. Right, so it's kind of like putting a telephone cable back. And so we thought that maybe this location would be a good place to insert a conformationally sensitive GFP. And this kind of change is very recently, just this year, been seen again in the siona voltage sensing domain in the structure by Eduardo Perozo. And in this structure, It's a little hard to tell, but the black one is the resting down state, and the white form is the up state. And they were able to get up and down structures for one protein, the siona phosphatase VSD. And in the up state, you can see what looks like an extra new helix. So it's a little bit different from this. This new helix is not attached to S3.. But there is an extra coil. And so we thought that would be a good place to try to link conformational changes in the voltage sensing domain to conformational changes in GFP. So we took the circularly permuted GFP, same one that's used in GCaMP. We tried several ones. In the end, we used a GFP that wasn't from GCaMP, but it's the same circular permutation that everyone uses to make conformationally sensitive GFP reporters. And we inserted that in the S3-S4 loop. And we screened just a few constructs, and many of them showed voltage sensitivity, and we just used the one that showed the largest change and had the fastest kinetics. And what we see with these sensors is that actually, again, they're bright at baseline, but they're actually dimmer upon depolarization. That's consistent with more strain being put on the GFP barrel with the extra coiling of the alpha helix on the extracellular loop. So we call this, because it's very fast, it has on and off kinetics of 2 milliseconds each, we call this accelerated sensor of action potentials, or ASAP. So ASAP has a dynamic range, a total dynamic range that's very large, over 50 millivolts, from the physiological resting state of -70 to depolarization of positive 30 has about a 15% change in fluorescence. And its kinetics are 2 milliseconds up and down, as I alluded to, which is order of magnitude faster than the original arc light. And so when you apply a voltage— when you apply an action potential waveform, the fluorescence trace of ASAP1 follows the shape of the voltage trace quite well, whereas ARC-LITE is blunted due to its time averaging effects. And you can see what— you can see mock mini-EPSPs and IPSPs well tracked with ASAP-1, not so well tracked with ARC-LITE. Same is true with the 3AP burst. And because of its fast kinetics, you could take— you can take action potential trains up to 200 Hz, and we believe up to 300 Hz, although we haven't tried, With ASAP-1, you could see in single trials without averaging, you can follow every single spike, whereas with ArcLight, you get this flat baseline due to time averaging after 100 Hz. In neurons, you can tell in electrical trace where you have minis and APs, you can see every mini, you can see every AP with high signal noise, whereas with ArcLight, you can't be sure if the APs have happened or not. So basically, we think it's pretty good. It gives you the large variety of electrical events in neurons that you'd like. You'd like to be able to differentiate minis. You'd like to see minis. You'd like to be able to see your APs. You'd like to know how many APs had happened. So ASAP will give you, I think, most of those things under the right illumination conditions. The one thing that you would want to be mindful of, of course, though, is that because ASAP is so fast, you have to image at high repetition rate. So with GCaMP, because the signal persists for about half a second, you could image it at 30 hertz and capture APs when they happen. But in voltage, because the signal is gone in a few milliseconds, you really should be acquiring at 500 hertz or 1,000 hertz. And so that means you just need to get a fast CMOS camera. They're not too expensive these days, like the Flash IV. Just $16,000, and you need to use enough— they did— well, yeah, well, it used to be much more expensive than that. I'm not being paid by Hamamatsu either. And you need to apply enough light to your sample that you can see signal over noise within your 1 or 2 millisecond exposure. But it's doable. We're within the doable range. For example, in this case, we've taken a neuron in culture,. And we put enough light on it imaging at 1,000 Hz that we can see action potentials with a signal-to-noise ratio of 4 or so. And you can see leaving the light on for 15 minutes— this axis is on the order of seconds, and this is at the beginning of the experiment. 15 minutes later, which is 900 seconds later, they're still going. The neuron's still putting out action potentials, and we could still read them with similar signal-to-noise. So you can image for maybe half an hour, maybe more, with the right imaging conditions. So unfortunately we don't have pretty neuron movies. We have neuron movies, but they're not pretty because we do them for high sensitivity and binning. But we do have nice hex cell images. So this is a hex cell that's been patched and then taken through voltage steps. And you could see with 1,000 Hz imaging, the fluorescence responding to voltage. All right, so anyway, we're— that's it for the voltage sensor, but we're of course trying to make it even better. We have some ideas about how we could make the dynamic response even larger. So last, I will tell you about optical control of protein functions. So visualizing time and location of protein activity has provided a lot of insight into mechanisms of synaptic plasticity. For example, on the left is a movie from Ryohei Yasuda's lab looking at a CaMKII sensor. Looking— and in this movie they're able to show that CaMKII is activated in individual spines in response to neurotransmitter stimulation, but that activation is very transient. But of course CaMKII activity is required for LTP development. If you put on a kinase inhibitor, you fail to get LTP. So this shows that chem kinase— the requirement for chem kinase is very early and transient. And his lab has gone on to do FRET reporting for many other signal transduction molecules and worked out their logic in the circuit of synaptic plasticity. Our lab was also looking at using FRET sensors to look at RhoA activity during axon guidance in this developing neuron. To show that events happen locally and quickly in the growth cone. But besides visualizing protein activities, can we control the time and location of protein activity? That would be really nice because then we could test hypotheses about what's really required and where and when it's required. But of course, there's no easy, generalizable way to control protein activities quickly or locally. Transcriptional regulation is too slow. Chemical agonists are often not specific. If they are, that's if they're available. They're often not available. And then neither of these methods allow spatial control. So what we want ideally is to have a method for making proteins that are photoregulatable. And the method should be generalizable. And it would be great if it were 100% genetically encoded like GFP without requiring chemical cofactors. And it would be most convenient if you could just make one protein that would be optically controlled without having to co-express something else. And of course we don't want to use UV light, we want to use non-toxic wavelengths of light. So when we first started this, we considered all the natural photoregulated systems from plants and bacteria. And of course many other labs have been pursuing this as well. And so other labs have taken systems like bacteriophytochromes UVRA, phototropin, rhodopsins, and link them up, link their natural light-responsive functions up to mammalian protein functions. But in none of those systems are the four criteria fully satisfied. The four criteria: generalizability, cofactor independence, single-chain construction, and non-phototoxicity. And in terms of the generalizability, what we were most interested in was creating a system that looked like this. Conceptually. So if we could fuse to the two termini of a protein of interest, domains X and Y, which bind to themselves at baseline, then that might be sufficient to inactivate the protein, especially if the NNC termini are flanking the active site or the binding site of the protein. But then when we put on light, if light dissociates X and Y, then that would open up the protein. So the only problem is what should X and Y be? We thought about adapting some of these natural proteins to become X and Y, like reversing their inducibility by light. But in the end, we decided to engineer a completely new protein-light-regulated interaction. And that was by none other than our old friend fluorescent proteins. And that's, yeah, because we, you know, not only do we love fluorescent proteins, but they actually have they already meet the non-cofactor requirements, no co-factor requirement. And we actually— they have structural features that are very similar to the natural photoregulatory proteins from plants and bacteria. Namely, they undergo a light-dependent conformational change. So we use a fluorescent protein named DROMPA. Some of you may use it for its photo-switching properties. It's green fluorescent at baseline, but it's switched off by the cyan light that's used to illuminate it.. And then it's switched on again by violet light. So cyan light is 500, violet light is 400 nanometers. And other labs have shown that this photo switching corresponds with the cis-trans isomerization of the two rings of the chromophore. So originally the chromophore is in the bright cis state. When it's illuminated by 500 nanometer light, it flips to the other side and it becomes dark. And then it can be flipped back again by 400 nanometer light. But on this side, NMR studies have shown that there's a loss of structure along this barrel surface of the protein. And we simply noticed that that barrel surface is the primary dimerizing surface in dimeric and tetrameric forms of fluorescent proteins. So we asked if you were to make a tetrameric form of DROMPA, which you can make by single point mutation, residue 145 to asparagine, would you have the following situation where the Bright form of the, of DROMPA is a tetramer, but when you shine cyan light on it, it monomerizes, and then when you shine violet light on it, it goes back to tetramer. This was really easy to show, actually. So DROMPA tetramer is tetrameric. DROMPA145N is tetrameric before light treatment. After cyan light illumination, it runs as a monomer on this native gel, and after violet light The same sample will now run as a tetramer again. So it's fully reversible and it's quite robust. And we can then now do this protein sandwich that we wanted to make. We put DROMPA145N at both ends of a protein of interest. We chose Intersectin because its N and C-termini flank its active site. And so the expectation is that at baseline, DROMPA will form a tetramer and then this will create a large steric hindrance to the active site and occlude it. But when you shine cyan light on it, then the tetramer would dissociate, and then Intersectin would be able to interact with its substrate, which is Cdc42. So Intersectin is a GEF. It's a guanine-nucleotide exchange factor, which activates GDP for GTP exchange on Cdc42, thereby activating it. Cdc42, as many of you know, mediates the formation of filopodia., which is involved in chemotaxis and synaptogenesis. So to show that this would work, we put this intersectin fusion, which we call a FLIP for fluorescent light inducible protein, to the membrane where it can be in position to activate Cdc42. And then we shine light on cells expressing this construct. And what's nice about this system is the DROMPA fluorescence gives you an idea how much light you need to shine. It gives you a direct feedback on the activity state of the protein. So in this case, we put on enough light to bleach drompa fluorescence to about 33% of previously, which is about 30 seconds. So you see the drompa fluorescence going down. And then after that happens, we look at an mNeptune-FasZn fusion. So this is an RFP fusion to FasZn, a filopodia marker, to look at the formation of filopodia. Prior to illumination, there are very few and only short filopodia, and after illumination, there are many long filopodia. It does appear to work. Here's a movie— I'm out of time. Here's a movie which is not playing for showing the induction of filopodia after light stimulation. Light stimulation was at zero, and after that, you see the formation of long filopodia. It's generalizable as we want it to be, so we generalize it to a protease, If the protease is activated, it will cleave away a cherry substrate from the membrane. So you see the cherry starting at the membrane, and after photoactivation, you see cherry filling up the cytoplasm. So we think that we can apply this potentially to induce structural plasticity in a spatially specific manner. So because filopodia are believed to be predecessors of synapses, we took a developing neuron at 14 days in vitro and illuminated it. Part of its dendritic tree. I mean, you can see the formation of filopodia after illumination. Of course, there's a lot of filopodia formation going on at baseline, and so to do this well, we had to do it in a blinded manner. So we had to get a couple— it took, I think, 2 high school students and 1 spouse, postdoc spouse in the end, to do the quantitation. But In the end, we were— we see what looks like a 50% increase in the— above the basal rate in filopodia formation within the illuminated region compared to the non-illuminated region. Of course, we don't tell the people when they're scoring for formation what was done, so they don't know what was illuminated or what was not illuminated. They just score all the new filopodia, and so we see this increase in filopodia formation. And it seems to translate to increase in spine formation as well. And so we believe that eventually translate to an increase in synapse formation. So the advantages of our fluorescent light-inducible proteins over other methods, we believe, is that it's potentially more generalizable to your particular protein of interest. And it's single chain, it's cofactor independent, and uses cyan light, actually, which is not used by any of the natural systems. And it's self-reporting because it's a fluorescent protein. And also because it's based on a fluorescent protein, we think it's potentially extensible to other wavelengths because there are red switchable fluorescent proteins as well. And so we'd like to apply this to more different proteins, more colors as mentioned, optimize our performance, and then seek new applications. And so in terms of photoregulatory proteins in biology, to take the big picture in terms of photoregulation, There are many natural proteins that are involved in sensing light and changing biology. But I think we can add to that this first example of an unnatural protein. We don't think this happens in natural fluorescent proteins. We think this is an engineered feature of fluorescent proteins. It's been engineered in the lab. So I think this is the first engineered protein, light-dependent protein interaction. So just to sort of give you a little motivation, you can learn a lot by by looking at things happen, like watching TV. And so we've had TV for 60, 70 years now. But you can learn even more by doing. If you had a computer, you can actually move things around. That really is much more powerful than looking. So if you were— [MUSIC] You could watch somebody play the piano, and you know that there's this thing called a piano, and they can make music. Kind of like, you know, there's a brain and it can do something. But if you really want to learn how that works, then it would be nice to actually try to play piano. And so I think with protein photocontrol, we could start trying to learn to play the piano of the brain.

So—

so, of course, you start with the easy songs. So in summary, what I've been showing you and talking about is our efforts to fill in this matrix of optical control and optical sensing. I mean, what's interesting is how much of the matrix— to me, what's interesting is how much of the matrix is filled by fluorescent proteins. So this one class of proteins is really quite versatile, even more versatile than we imagined just a few years ago. So that brings me to acknowledgments. These people helped— these people performed the timestamp work. Francois and Ying performed the ASAP work in collaboration with Mark Schnitzer's lab, and Xin did most of the FLIP work.

Thank you for your attention. Any further questions? Hey, Mike.

The—

going back to your timestamp, it just occurred to me, what's special about that degradable protein? How does it have to be— because I remember the inhibitor actually is a protease inhibitor, right? It's hard to penetrate, hard to deliver. So how does it compare to, say, the destabilized domain developed by Tom Wando's?

Right, so I think there are two components to the question. One is protein degradation and the other one is drug permeability. So this protein degrades due to the placement of an N-end rule inducing amino acid at the cleavage site. So there's a rule by which the first amino acid at the N-terminus determines protein stability. If it's a positive charge, it's very unstable. So we put a positive charge right after cleavage site, so this gets degraded away. But when in this conformation, that cleavage site is not exposed, so then the N-end rule is not invoked. The drug is very cell-permeable, and the nice thing is it gets— the drug class gets better and better, so every year there are better versions of these drugs that are less toxic. And now there are 3 approved drugs for humans, so you could just go— and Santa Cruz has started selling them, so you can just go buy them now. It's very convenient. It doesn't penetrate the blood-brain barrier. The drug is 600 Daltons. Usually above 200 Daltons, you don't get good penetration. Actually, Roger's lab has been working on a new method to get these hydrophobic drugs across the blood-brain barrier, which apparently shows pretty decent results. And so There actually may be a way to do it. It's aided by complexation with nanoparticles, which I guess get— enough of them get into the endothelial cells that form the blood-brain barrier to get onto the other side, to release the drug on the other side. But for now, the best way to use it, and the way we're planning on using it, is via cannulation. So in the hippocampus, where we were interested in looking at CA1 distribution, distributions in new arc. We need to cannulate anyway to put an endoscopic microscope there, so we could just go down the same bore with drug. But can you replace it with the destabilized domain? Oh yeah, so we didn't want to do that because if you were to destabilize the— okay, so one idea is why don't you just put destabilized GFP on there, and it's going to it's going to turn over very fast, and so in the basal state you don't have very much signal. And then when you induce new protein synthesis, you have more signal. And this is true. People have used destabilized GFP on its own. Aaron Schumann's lab has used it on its own to show that new protein synthesis occurs and is stimulus-induced. But we didn't want to tag something with it. We're interested in seeing where the proteins go, and we didn't want to tag PSC95 with that. Because then you would be constantly degrading PSD95, and that might have effects on synapses. Also, that's not an all-or-nothing thing. It's not binary. So you can see fold differences, right? But what we want is something as sharp as possible so we could get the best signal-over-background induction.

But it's a good question. So for the FLIP system, you showed nicely that you can induce it very well. I was wondering, so you showed those native cells, and it seems like the violet light cannot bring everything back to tetramer.

So how well can you switch the system off? Yeah, it's a good question. In vitro, it's very efficient.— this one that I showed you is, I would say, representative. But sometimes we get even more complete restoration. So in this gel, you see some monomer at baseline anyway in the— you see some monomer at baseline anyway before illumination. And so this brings back everything to within experimental variability. So I think it's complete. The— and that goes with the— that corresponds with the previous observation that DROMPA is a very robust photoswitcher. So people who use it for tracking purposes, they can get complete switching as well of the fluorescence. So we just see this correlates nicely with the fluorescence. And in other experiments where we use less protein, we don't see this at all. Now this is 100 micromolar protein since we have— So we're talking about concentrations. There's 100 micromolar protein, so that shows that the Kd of this interaction is greater than 100. However, the Kd of this interaction has been measured at less than 10. So we're getting a more than 10-fold difference in affinity on the switch. [INAUDIBLE] [INAUDIBLE] Right. You're absolutely right. Yeah. So the question was for the kinetics of activation and inactivation. So the kinetics— I mean, we know the kinetics of photobleaching here, of photo switching here. And they're not super fast. And we've actually made mutants of DROMPA that do switch faster. I mean, the kinetics I think are within a usable range. So within 10 seconds you can get your— more than 50% of your protein activated. But it's true then the protein then needs to go find its partners and do its thing, and that just will depend on the particular protein. And so with the Cdc42 with filopodia formation, we seem to get a delay of— like 15 minutes before we— we see some futility right away, but we don't see really huge effects until 15 minutes later. And we think that's just a matter of the protein finding its substrates. Now you can imagine that with the— holding these two domains on either end, the protein is not going to find its substrates as easily as a normal protein. But I think that unfortunately is an experimental limitation. That you have to model into your interpretation. Thank you.