Tip:
Highlight text to annotate it
X
Our next speaker is another presidential award winner and is now part of the intramural program
of the NIBIB to show how versatile the institute has become. Hari Shroff will be our guest
speaker, and he’ll be speaking about ‘New Technologies for High Spatial and Temporal
Resolution Imaging of Cells and Organisms.’ Hari, it’s great to have you.
Thank you.
All right, so before I get started here, I just want to thank Rod for letting me be part
of this 10th anniversary – 10th anniversary celebration. It’s a great honor to be here,
and especially in the company of such great speakers.
So my – I’m a microscopist by training. And I try and think about ways of improving
microscopes that are commercially available. Most of my work is not on sort of large organisms
like humans or mice, although I seem to be sort of moving in that direction because I
started out studying single cells. I’m going to tell you a bit today about my work on C.
elegans, and I’ve been also starting to work on zebrafish a bit.
So half of my lab works on mechanisms for improving the spatial resolution of microscopes,
in particular getting past the diffraction limit of light, which is about 250 nanometers
in size. And although I have lots of pretty pictures from that side of the lab, I thought
instead I’d tell you sort of a different story – I’d tell you a bit about imaging
noninvasively, and in particular how to avoid cell torture.
So one of the things that I’ve – that I learned early in my tenure as a microscopist
is that it’s very easy to fry a single cell. And here you see a fiberglass cultured on
a single – on a – on a single cover slip. And you’re looking at it in brightfield
modality, but I was also shining a laser on this – on this cell. And you can sort of
see it curling up. And eventually I – if I – if I were to keep showing you the movie,
the cell would lift off the cover slip and sort of die a watery death.
Hari, did we use this?
Can people hear me or – all right. Here we go.
OK.
(Cross talk.)
Ah, OK.
Mandatory.
So the reason for this is that most microscopes that you can buy are remarkably inefficient.
They dose the sample volumetrically. They dose the entire volume of the cell, even though
we’re recording this image with a CCD camera, which is a two-dimensional detector. So I’m
going to tell you about a microscopy technology that we and others have developed that is
– that is far better and lets much – lets you image effectively much, much longer.
So hold that thought in your mind, because now I want to sort of switch gears and motivate
the importance of this from a biological perspective. Another one of the goals of my lab other than
improving microscopy is to better understand neurodevelopment in the brain. And so if you
think about the human brain, it’s of course this fantastically complicated organ – you
know, billions of neurons, orders of magnitude more synapses. And despite this large number
of parts, the brain somehow manages to wire correctly the huge amount of the time.
How this precise connectivity happens in vivo is still not very well understood. But one
thing should be clear from the sort of mismatching numbers. This kind of complexity is somehow
coded for by about 25,000 gene products, even though there are trillions of sort of functional
parts that make up the nervous system. Merely cataloging the genes and their functions is
not going to be enough to sort of understand how this thing wires. What you really want
is sort of a movie of how the entire brain forms at subcellular resolution. But unfortunately,
with today’s technology we can’t do this in the human brain.
So if you jump down a few orders of magnitude and complexity and think about a much simpler
organism, the C. elegans worm, this is the beast that has only 302 neurons; about 5,000
chemical synapses, most of them – most of which have been mapped at the ion level of
resolution; and the sort of gene number here is about 20,000. This is a simple enough system.
It’s transparent. Its lineage has been worked out. It’s something that you could – you
could really ask with a microscope how these genes act in vivo to direct wiring. You can
ask where are the neurons, how did they get there, where are the axons.
If you want to look for general principles, I would argue the worm embryo is a good place
to start. And for those of you that think that the worm embryo is too far removed from
the human, I would just point out that there have been two, arguably three, Nobel prizes
that have involved this worm: one for sort of programmed cell death – you know, one
kind of involving GFP; and then I should also say that technologies like RNA interference
have – are also very easy to deploy in this – in this organism.
So what I’m really after is sort of a four-dimensional atlas of neurodevelopment, sort of a dynamic
Google Map for the worm where I can optically reconstruct where are all the cells at every
point in development, both in space and in time; where are all the processes. And if
you had such an atlas of cell motions and cell processes, you could then map onto this
atlas a transcription factor – expression dynamics.
And I should say this is a collaboration with a neuroscientist at Yale and a – and a developmental
biologist at Sloan-Kettering.
So this is sort of the goal. How do we get to this goal? Well, given the fact that we
have these awesome tools like green fluorescent protein, and given the great genetic control
we have in C. elegans, you can actually label all the neurons in the worm. And the problem
is that you label typically all of the neurons, and that creates a contrast problem. So one
of the things you need is a way of selectively marking sparse subsets of neurons. Otherwise
you have – you have all of them light up, and you can’t distinguish one from another.
Even if you have worm lines – genetically modified worm lines that give you a few transcription
factors expressed in a few select neurons, you don’t always know which neurons those
are. And so what you need to be able to do for this atlas is identify which cell is which.
This is actually not a crazy idea, because you can track nine out of 10 rounds of cell
division already just by following the pattern of cell divisions in the worm embryo. And
you can do this computationally. And so this is a computationally-derived lineage tree.
And then the reason that I’m interested in this problem is that you need a way of
imaging this fast, volumetrically and without killing the worm. And that’s quite a difficult
thing.
So why is this sort of a challenge? If you look at sort of a cartoon of the C. elegans
embryo in cross-section, it’s about the size of a large cultured cell. So it’s about
40 to 50 microns laterally and about 20 or 30 microns thick. And what I’m after is
an atlas that gets you the position of all the cells through subcellular resolution.
So what I’d like to do is sort of carve this embryo up optically into diffraction-limited
voxels of less than a micron in size. And I’d like to do this over the 14-hour developmental
time period of the worm.
So how might you illuminate the worm to get this sort of information back in your fluorescence
microscope? One thing you can do is simply illuminate the entire volume of the worm,
and that’s called wide-field imaging or epifluorescence imaging. So if you shine a
laser beam throughout the volume, you get fluorescence everywhere through the volume.
And if the worms were very thin, this might be the way to go. There’s probably no better
way for very thin samples. It’s very fast, because all you have to do is move the plane
of focus through the volume and build up a volumetric stack.
But you have a problem with out-of-focus blur. There’s no optical sectioning in this illumination
strategy. And you also have a tremendous problem with out-of-focus bleaching and damage. So
you might be visualizing a plane down here, but if you’re illuminating the entire stack,
then you’re bleaching the fluorescence that occurs up here.
Sort of another thing you might think about doing is using a confocal microscope. And
just to review this, the idea is you have a laser beam; you focus it into the sample,
and then you scan this laser beam around in the sample. And for every – for every position
of the laser beam, you can – you record the fluorescence signal. You use a confocal
pinhole to reject the out-of-focus light except in the vicinity of focus. And this – this
is sort of a workhorse tool. You can buy one. It’s prevalent everywhere. And unlike the
wide-field microscope, it gives you optical sectioning and it – and it can give you
near-diffraction-limited resolution.
The problem is that if you – if you look at what this looks like in the context of
the worm embryo, it, like the wide-field microscope, also bleaches out of focus. So you also have
– the only useful information you gain at any instant of time is for the vicinity of
focus. But you have this sort of cone of – this hourglass shape of execution that wastes the
fluorescence up and – up and below the focal point.
This is also intrinsically slow, because it’s a point-scanning technique. So you record
one piece of information here, but then you have to scan this execution this way – up
and down and into – and into and out of the page. And then in order to make that fast,
you usually have to take the laser and turn the intensity up. And that frequently fries
your sample.
So conceptually anyway, you can do much better if you just illuminate one plane at a time
and you use a light sheet to illuminate only the plane that you’re detecting. And then
to build up a volume, you can scan this light sheet in one direction that’s very fast.
If you tailor the light sheet the right way, you can minimize out-of-focus bleaching and
damage.
The idea of using light sheets in microscopy is actually nothing new. The Germans were
doing it in 1903 and 1904, investigating colloidal suspension. But the – this technique has
undergone a renaissance in the last decade ago – decade or so, especially in the context
of developmental biology. And they sort of undergo – they go by this acronym: selective
plane illumination microscope techniques, or SPIM techniques.
Just to give you an idea of what these microscopes have looked like historically, some of the
early work was done in fish. And so you take your fish and embed it in an agar or a cylinder,
put that agar or cylinder right in a water-filled chamber, and then come in from one side with
a light sheet. You detect the fluorescence in this perpendicular detection, and you scan
the light sheet very fast in this direction and build up this optically-sectioned volume.
And so you can – this movie is from almost – you know, from quite a while ago, but
it’s still pretty illustrative.
If you were to use – do the same imaging when you illuminate the entire volume, you’ll
have this – all this out-of-focus blur. But if you do this light-sheet-based volumetric
imaging, you can get very optically clear images of, in this case, this fish embryo.
And if you have the time to rotate the sample around, you can really fill in this volume
very beautifully and with a minimum of damage to the sample.
So this technology has been around for a number of years, but there are very few of these
actually around, and it has not yet been deployed in the market. And part of the reason for
this is that this is really a pain in the *** to build. So as a tool developer, as a
microscopist, I started thinking about ways of making this sort of easier to use. And
the solution we came up with is something we called inverted microscope base SIM (sic)
– or base SPIM, or iSPIM.
So the idea that we had is to just implement this geometry in place of the brightfield
illumination pillar of a conventional epifluorescence microscope. So you have these two perpendicular
objectives. In my case, the worm embryo sits over here on a glass cover slip with the focus
of both objectives. And then you bring in the light sheet via this objective. And you
stand it in this direction, and the fluorescence gets sent to a camera.
The advantage of this geometry is that you can use conventional sample-mounting protocols
like glass cover slips, and you can investigate the sample before you do the SPIM imaging
using the lower, conventional optical microscope train. So you can look at it at different
magnifications if you want. You can do photoactivation. You can – you can do lots of different things
using this flexible approach.
So we sort of built this system, and then we started to prototype it on C. elegans embryos.
And I’m going to show you just a few movies. This is – this is one example of a C. elegans
embryo from the two stage, stage eventually all the way to hatching. And what I’m – what
is displayed here are the positions of the nuclei. The nuclei have been marked with a
GFB histomarker. And using this light sheet microscope, we can collect about 25,000 volumes
over the course of embryogenesis. We can image every 2 seconds over about 14 hours without
frying this worm.
And this was a good control for us, because all of the cell divisions have been studied
in great detail in the ‘80s. And we detect no difference using this imaging modality.
In the – in the pattern that sort of stereotyped cell divisions, the embryos all hatch on time.
This is about 30 times faster than a confocal spinning disc, which is sort of the state-of-the-art,
fast optical microscope you might – you might buy commercially.
One of the things we noticed that is a little bit of an imaging nightmare is that after
the muscles form, the embryo starts to twitch pretty rapidly. So you have to image super-fast
to be able to still resolve these individual cells. But we can do that – for the most
part we can track these individual cells, although there is a bit of motion blur. And
if I kind of advance this movie over here, what you find repeatedly is that eventually
the worm hatches out of the eggshell and goes on to make more worms. So it is indeed noninvasive.
And so this was sort of a good proof of concept for us.
What I’m really after though is the neurons. And so as a proof of concept – another proof
of concept, we built a worm strain that had only a few neurons highlighted. And again,
I’ve sort of taken this four-dimensional movie – and in the interests of time sort
of broken it up into four different time segments. In this particular movie, there are about
four neurons that you can kind of see over here. They start out at the anterior end of
the embryo. And if I advance this movie, you see that they kind of crawl from the anterior
end to the posterior end. And once the embryo starts to twitch, all hell breaks loose – but
you can still, even after the muscles twitch, observe sort of by eye the motion of these
individual cells.
A little later, you can see that two of these cells form axons, which are these sort of
long structures over here. And if I – maybe if I pause the movie, I can even point out
the growth cones at the very end of these axons will appear maybe in this movie over
here. So you can actually see submicron-type structures in this microscope. And you can
actually follow and track these cells, you know, using this noninvasive microscope.
And I also want to point out that this is a pretty dynamic system. So the neurons start
out in a morphology that is nowhere near where they end up. And I would argue that you really
want to do this kind of dynamic imaging to see how the neurodevelopment happens, instead
of fixing an embryo and looking at many thousands of fixed embryos. It’s much easier to see
what’s going on if you have one embryo and can follow the sort of developmental progression
throughout time.
Now, when we built this strain, we didn’t actually know what the identity of these neurons
were. And in compiling this atlas, that’s sort of a necessary component. So just as
– just as – just to show that that’s possible, you can – you can do multiple-color
imaging, where you label all the nuclei in red, let’s say, and neurons in green.
And then you can – you can – if you can correlate the position of the nuclei with
the neurons, then you can – you can – so for example, this particular cell over here
– it’s moving a little fast; let me – let me – let me bring it back. This guy over
here, you can sort of see – you can correlate where the nucleus is and the – and the cytoplasm
– (inaudible). So you can build up these lineage trees and actually identify these
neurons as these particular names, which really don’t mean much to me. But at least in principle,
you can do this sort of analysis for all of the neurons in this worm and then identify
each one from the data itself.
Now given this kind of data, you can do sort of cell biological studies. You can ask how
long are these neurites as a function of time. You know, where are they in relationship to
another? I’m just going to illustrate one piece of information that we learn from these
movies. If you take one of these neurons, this so-called ALA neuron that you find if
you look at wormatlas.org, which documents where all these things end up in the adult
worm – you find that this neurite sort of kinks back and then goes towards the tail
of the – of the embryo, although it projects first the other way.
And we can actually see this in our – in our movies. So I’ve marked over here this
ALA neuron at about seven hours post-fertilization. And if you look a little bit further, at about
seven hours 15 minutes you find these kind of outgrowths of this neuron. If you look
a little bit further, you find that these outgrowths bifurcate; you find this forked
structure over here. This forked structure persists for a little while, and then it commits.
And then one of the forks disappears and this neurite goes always in the other direction.
So at least in principle, you can use this system to study neurodevelopmental events
in vivo as they happen.
All right. I’m going to finish just by discussing sort of where we’re going with this technology.
One of the things we learned in the course of doing this is that although imaging every
two seconds an entire volume of a worm is pretty fast, it’s actually not quite fast
enough all the time. And so we’re trying to push this feed further.
We’ve also – we’ve also discovered that there’s a way of actually increasing the
isotropic resolution of this microscope. All of the movies that I showed you were always
XY views as opposed to XZ views. And that’s because this microscope, as in – like any
single-view microscope, has worse resolution along the optical axis than perpendicular
to the axis. There are less angles that give you resolution along the optic axis.
One solution to this is to rapidly flex two perpendicular views. And this 90-degree objective
geometry actually gives you those. So what we’ve done is we now mount a camera on each
arm of this selective plane illumination microscope, record the views from each perpendicular direction,
and then we can fuse those and get a – get a volumetric time series that is – that
is more isotropic.
So here’s an example of that. We’re collecting each volume now in less than a second in both
views. And then you can see the XY and the XZ views. What I like about this movie is
that it’s actually quite difficult to say which direction is which. So performing this
multiview imaging really does even out the isotropy of the – of the movie. And again,
you know, this is not so damaging to the worm that they don’t hatch on time. All the cell
divisions also seem to happen on time.
These are – these are two areas that we need help on. So, you know, the neural strain
that we built is one of only a few that has just a few neurons on. We need – we need
to sort of basically develop lines that have all of the 222 neurons marked sparsely. One
way that we’ve thought about doing this is to employ photoactivatable markers that
label all the neurons and then go in specifically with a different color laser and activate
just a few neurons.
The other thing we need to do to compile all of this volumetric data into a digital atlas
is to straighten the worm embryos, because although the embryos themselves have a – have
a stereotyped developmental program. The twitching of course is stochastic. So if you want to
compare one embryo to another embryo, you have to align it along the common body axis.
So if you have ideas about this, please email us. We’re happy to talk about this.
And then finally, I’ll just say that you can use this system to image many other phenomena
besides worms. So you can image things like the zebrafish lateral line. We’ve started
to image interactions between mouse egg and ***. So we’d like to observe the events
that happen after the *** binds to the egg and penetrates the zona pellucida. You can
investigate in microtubule dynamics, influenza, infection of cells, vesicle transport – anything
that you need to apply volumetric images pretty fast and without killing the organism, this
microscope is excellent at doing.
So with that, I think I’ll just stop by thanking Richard Leapman and Hank Eden, the
– my bosses who hired me, very supportive both inside the NIH and outside and in – and
in my lab. The vast majority of this work was done by Yicong Wu, who unfortunately is
– couldn’t be here, but fortunately is having a baby. (Laughter.) And then Alireza
Ghitani was a postbac who also helped him. So thank you. (Applause.)
Perhaps time for one question for Dr. Shroff? All right. It’s as clear as mud. (Laughter.)
It’s exciting. Oh yes, please.
So of course you know about vessel plane illumination. And how much does that help?
So I can’t – I can’t sustain these imaging rates with vessel plane illumination because
I fry the embryo.
Really?
So over – because you put in energy throughout the volume in that thing, and you rely on
these tricks to get rid of the side lobes. It turns out for the embryo it’s too much.
Yeah. That was one of the first things we tried. Yeah.
Because of your former – it was your former mentor – (off mic) – who was pushing vessel
plane.
Right; right.
OK.
Thank you very much.