Top
SCIENCE. STORIES. IMPACT.

iBiology

Bringing the World's Best Biology to You

Super-Resolution: Structured Illumination Microscopy (SIM)

00:00:13.01 Hi there, I am David Agard from UCSF
00:00:17.09 and Howard Hughes Medical Institute.
00:00:18.28 And in this lecture I want to talk
00:00:20.16 about going beyond the diffraction limit
00:00:22.04 using conventional microscopes, and it's a
00:00:25.22 methodology known as both Structured
00:00:29.23 Illumination Microscopy as well as
00:00:32.08 Interference Microscopy. And we'll be
00:00:34.24 talking about both these in terms of
00:00:37.00 what their potential is for biological
00:00:38.03 imaging.
00:00:42.26 So as you all know that the
00:00:45.16 fundamental limit in light microscopy is
00:00:48.07 the wavelength of light in a numerical
00:00:48.08 aperture of the objective lens, and what
00:00:52.18 we see here is a coverslip and objective
00:00:55.00 lens, and this angle that defines the
00:00:57.19 maximum rate that can go into the objective
00:00:59.00 lens really defines not only the
00:01:00.29 light-gathering power of the lens but
00:01:02.12 what the resolution is. And in a previous
00:01:06.16 lecture on deconvolution microscopy,
00:01:07.20 we talked about the point spread function
00:01:08.12 and the details and mathematics of how
00:01:12.20 resolution is defined and what the
00:01:15.03 consequences of this limitation are, but
00:01:18.01 now what we want to talk about is how do
00:01:19.23 we go beyond what seemed to be these
00:01:22.18 inviolate physical principles of
00:01:26.29 both a finite wavelength of light
00:01:27.14 and the finite numerical aperture of the
00:01:29.00 lens. And effectively what there
00:01:33.10 are is there are two methods where we
00:01:33.26 can cheat physics and that's what we'll
00:01:35.12 be talking about in this lecture. And the
00:01:39.22 first strategy is to collect more of the
00:01:41.16 information. So why use just one
00:01:44.07 objective lens?
00:01:45.12 How about all this light down here which
00:01:48.04 is being lost? If we can capture that we
00:01:49.19 should be able to capture the
00:01:51.21 information from that and effectively
00:01:52.25 have a much higher NA coming from the
00:01:55.26 lens.
00:01:56.29 The second strategy that I'll focus on
00:01:58.08 is using structured illumination to
00:02:01.17 improve resolution, and that is as you'll
00:02:03.09 see instead of illuminating the
00:02:06.08 sample with uniform field of light, we'll
00:02:06.26 actually put pattern in that and look at
00:02:10.15 what the consequences of that are and how
00:02:11.15 we can then get to use the computer to
00:02:13.01 extract a higher-resolution information.
00:02:16.20 Ok, so as we talked about before in
00:02:20.03 terms of what the transfer function of
00:02:23.10 the microscope is I now want to talk
00:02:25.19 about what happens when we
00:02:27.12 have two objective lenses and so if you
00:02:31.01 imagine a simple microscope which now
00:02:33.01 has the sample waves in-between two
00:02:37.05 objective lenses, what we do is we
00:02:40.02 illuminate with light that comes in
00:02:42.00 through one objective lens but the
00:02:43.10 emitted light that actually contains all
00:02:44.10 the sample information goes through both
00:02:46.21 lenses - both that one and this one - and
00:02:50.07 comes around this little prism and
00:02:51.26 interfere with one another before being
00:02:53.27 collected on the CCD and that
00:02:55.27 interference pattern will be a way of
00:02:58.05 finding out much more high-resolution
00:02:59.18 information about precisely where the
00:03:01.21 light came from because as you can
00:03:02.20 imagine only light that really has the
00:03:05.12 balanced path between this path and that
00:03:07.21 path ends up constructively interfering
00:03:09.24 on the imaging detector ok
00:03:13.19 And so the way we can think about that
00:03:14.01 is here's the light-filled that's
00:03:18.06 coming out of the sampler going
00:03:23.04 into the sample what makes it through
00:03:24.25 the objective lens is just as a bit that
00:03:26.04 will make it through angle alpha as we
00:03:28.09 see here this this this wedge part and
00:03:31.13 then the total information that we can
00:03:33.23 see is the convolution of this with
00:03:38.04 itself because this is the wave and that
00:03:39.28 what we measure is intensities and
00:03:41.15 that's the self convolution and that's
00:03:43.16 this large region of support like in the
00:03:46.01 xy-plane ok, but here when we have this
00:03:47.22 convolved with itself, we get this pattern
00:03:50.29 which is what we showed before and just
00:03:52.24 doughnut-like royal structure in the
00:03:54.23 deconvolution lecture showing what the
00:03:59.01 three-dimensional imaging profile is of a
00:03:59.19 light microscope but if we have two
00:04:01.26 objective lenses, then we have this
00:04:05.27 convolved with itself which obviously
00:04:07.27 contains way more information than this.
00:04:09.16 And you can see that we come much better
00:04:11.14 at approximating the full information
00:04:13.07 field that we would have had if we had
00:04:18.03 essentially an infinite numerical
00:04:18.15 aperture for our objective lens. It's still
00:04:20.17 limited by the wavelength which defines
00:04:22.21 the circle, but nonetheless we get to
00:04:24.15 collect essentially all this information
00:04:27.08 providing much higher resolution
00:04:30.27 information, ok.
00:04:32.06 Now this is what you get. It turns
00:04:35.06 out either whether you illuminate
00:04:36.25 through both objective lenses or you
00:04:38.15 observe through both objective lenses,
00:04:40.21 but you know wait there's more we can do.
00:04:45.18 We can both illuminate through both
00:04:46.19 objective lenses and observe for both
00:04:48.20 objective lenses, so we create
00:04:49.00 interference patterns for the light
00:04:51.14 going in, and we create interference
00:04:54.15 pattern for the light coming out, and
00:04:56.28 that does even more for us as we can see
00:04:58.19 here, so here is that simple microscope
00:04:59.13 where the light comes in. It goes through
00:05:04.09 one objective lens. It also goes through
00:05:06.27 the other objective lens. Similarly, the
00:05:08.28 light coming out of this lens comes back
00:05:10.14 up and then through this dichroic goes to
00:05:12.01 the detector and as well as
00:05:14.19 through that way.
00:05:17.04 Ok, so through both paths, and the result
00:05:18.00 is not just this which I showed you
00:05:20.02 before in terms of kind of coming up to
00:05:23.08 the full circle of comparable
00:05:26.28 resolution in the z-direction the XY-
00:05:27.00 direction but you can now get double
00:05:30.03 the resolution in the Z direction than
00:05:33.20 possible in the XY and so you can do get
00:05:37.20 very very high-resolution information
00:05:39.18 along the axial direction which was
00:05:40.09 normally something very badly behaved in
00:05:44.12 any kind of conventional microscope.
00:05:46.28 Moreover, you see that all the missing
00:05:48.18 regions both the missing cone right here
00:05:50.11 and then the missing regions in
00:05:55.20 the if you just illuminate or excite
00:05:57.10 through our or look at the mission
00:06:00.09 through both lenses, and you can see all
00:06:01.07 that gets filled in very nicely, so
00:06:03.19 there's a huge increase in lateral
00:06:05.05 resolution and now everything is
00:06:08.28 observed so that we don't have to try to
00:06:09.19 use very powerful methods that recreate
00:06:11.28 the missing data, we can just deal with
00:06:16.00 the fact that this intensity is lower
00:06:17.06 than this intensity etc, ok. And the
00:06:19.03 result is here, so this is, for example,
00:06:23.03 a single fluorescent bead imaged in
00:06:27.04 the XZ direction, so this is the focal
00:06:28.27 direction you see the characteristic
00:06:30.19 elongated pattern that you get
00:06:34.13 as the out-of-focus light
00:06:35.11 accumulates as you go up and down. Now
00:06:40.20 with the two lenses using both the
00:06:42.24 detection and illumination what you see
00:06:44.28 is you get this complex pattern which is
00:06:46.21 an interference pattern but basically
00:06:51.12 shows very sharp detail especially
00:06:53.14 in the Z direction, and then using either
00:06:56.08 the Wiener filtering that I talked about
00:06:59.04 in the deconvolution section where you
00:07:00.21 just divide by the intensity differences
00:07:02.26 or an even more powerful energy
00:07:06.07 deconvolution approach, you can see you
00:07:06.16 can get the very high-resolution where
00:07:08.28 the possible resolutions now are like 70
00:07:11.25 nanometers in the Z direction instead of
00:07:14.25 something like 600-700 which would be
00:07:18.06 for a conventional lens, and it's
00:07:20.28 actually much better in yet in the
00:07:21.14 z-direction than XY.
00:07:22.06 Ok, and so this just gives an example
00:07:25.18 of a cell HeLa cell where the
00:07:29.06 microtubules have been stained with
00:07:30.22 fluorescent antibodies, and then what you can
00:07:35.01 see is here would be if we took this
00:07:38.01 image
00:07:39.09 normal XY a set of sections and
00:07:41.18 deconvolved it, we see we get rid of a
00:07:43.11 lot of the blur, but nonetheless that's a
00:07:45.15 radical difference from what we get
00:07:46.04 using this 2 lens interference approach.
00:07:48.15 And now you can see very fine detail
00:07:51.14 throughout the entire depth of this of
00:07:54.09 the sample, ok.
00:07:55.02 So this is a very powerful approach.
00:07:56.17 The kind of drawback from it is that you
00:08:00.07 actually need to have your sample wedged
00:08:00.15 in-between two objective lenses and
00:08:04.05 everything has to be perfectly matched and
00:08:05.27 the path lines have to be very tightly
00:08:08.22 controlled, and so it's an
00:08:10.27 experimental nightmare at times, but
00:08:14.06 the results are really quite
00:08:15.27 extraordinary, and it's much more
00:08:17.22 sensitive because again you collect all
00:08:18.14 the light from both sides, and you get to
00:08:20.27 utilize that in making the image.
00:08:21.09 There's a confocal
00:08:26.24 version of this known as 4Pi imaging,
00:08:27.03 and there's a commercial instrument from
00:08:29.22 Leica based on this technology that does
00:08:33.02 this in a confocal degree, but that is
00:08:36.02 actually the
00:08:39.03 commercial instruments are more like
00:08:42.28 this where you're only doing one of
00:08:45.09 these illumination directions and not
00:08:46.02 simultaneous illumination and ignition
00:08:50.22 through the two objective lenses just
00:08:52.22 because that's much more demanding
00:08:54.04 technically, ok. The other strategy which
00:08:57.04 is a much more i think universal
00:09:00.06 strategy and and although both of these
00:09:01.05 lend themselves to live imaging is
00:09:05.26 what's known as Structured Illumination
00:09:06.21 microscopy, and in this microscopy
00:09:10.03 instead of illuminating the sample with
00:09:13.28 the uniform field we illuminate with a
00:09:14.04 pattern like a grading pattern. And the
00:09:18.17 concept of it and why it does some good
00:09:20.03 is shown here in the simple slide, and this
00:09:22.15 is you know many of you I'm sure
00:09:25.08 have seen a moiré pattern. If you take
00:09:26.29 two screens from like your screen
00:09:30.14 door and you overlap them on one another,
00:09:33.00 you'll see these interference patterns
00:09:34.07 where they constructively and
00:09:36.03 destructively interfere, and really what
00:09:39.06 that is is that the pattern of one
00:09:40.20 screen is not precisely matching the
00:09:42.11 pattern of the other. Why we care about
00:09:45.00 this is the fact that if we imagine that
00:09:47.15 the screen in the back is the object and
00:09:49.17 this one in the front is illuminating
00:09:53.00 light that we illuminate with an
00:09:55.08 integrating pattern, what we see is that
00:09:56.08 the fringe patterns that we observe are
00:09:59.15 much lower frequency. They are much coarser
00:10:01.25 than the actual screen pattern itself,
00:10:03.11 and this provides a way of
00:10:06.09 getting high frequency information that
00:10:07.01 would be beyond what the objective lens
00:10:09.02 could see and putting it back into the
00:10:10.02 objective lens and capturing that, ok. And
00:10:13.23 so experimentally what we do is we
00:10:16.08 illuminate the sample with a
00:10:20.04 pattern of light actually you know we
00:10:21.14 use a literally use a grading and
00:10:24.03 we observe these interference fringes
00:10:25.07 between structure in the sample
00:10:27.28 and the interference pattern and the
00:10:30.10 grading that we are illuminating with. Now
00:10:32.05 when you look at the image, it looks like
00:10:34.15 you've made a crummy image that has all
00:10:36.29 kinds of patterning on it, and so but
00:10:37.14 there's information in there from beyond
00:10:42.03 what would fit through the objective
00:10:43.19 lens, and so this methodology requires
00:10:45.07 computers to reconstruct the image and
00:10:48.14 extract all the information that's
00:10:49.00 present and unscramble it so that we can
00:10:51.27 actually make sense of it. And so I want
00:10:53.19 to briefly describe that in a little
00:10:55.29 more detail. So if we imagine that you
00:11:01.05 know here's the observable region,
00:11:01.23 essentially the limit of what light
00:11:05.21 can pass through the
00:11:07.15 finite aperture of the objective lens, but
00:11:09.16 there's all this other space out here -
00:11:11.23 information that is unobserved from our
00:11:14.26 sample, ok, and what we're trying to do is
00:11:17.29 to figure out a way of capturing all
00:11:19.09 this information and understanding
00:11:21.00 it because that's the high-resolution tail
00:11:22.11 that's lost by both the finite
00:11:25.14 objective aperture and wavelength of
00:11:27.15 the lens, and so when we illuminate with
00:11:29.09 a grading pattern, it turns out that's
00:11:33.19 equivalent in the sort of diffraction
00:11:35.13 space for a space of illuminating with
00:11:38.19 information that has three spots - one
00:11:42.03 here, one here, and one here, and where the
00:11:44.09 frequency of that pattern is determined
00:11:47.16 by sort of the radius of this spot from
00:11:51.10 the center of the image and this again
00:11:54.19 is a convolution just like we talked
00:11:55.26 about before, but now this is the
00:11:57.00 convolution - the multiplication of the
00:12:00.25 illumination with the sample in real
00:12:02.17 space becomes a convolution in
00:12:06.01 diffraction space. And what that has the
00:12:07.16 benefit of doing is it makes three
00:12:10.29 copies, in this case each one for each of
00:12:14.06 these, so effectively what we're doing is
00:12:15.25 we're taking this whole information
00:12:16.05 regime and replacing it at this center,
00:12:20.23 this center, and that center so that they
00:12:22.05 all overlap. Now when they overlap like
00:12:25.05 this what we see is that we get
00:12:27.21 information from down here, comes through
00:12:30.00 in the guy that shifted up and makes it
00:12:32.16 into the observable region information
00:12:35.20 that was up here in the guy that
00:12:38.07 this one that gets shifted down also
00:12:40.23 maps into the observable region, and so
00:12:42.24 we have all this extra information that
00:12:44.01 makes it into the observable region. But
00:12:47.02 it's all added together looks like you
00:12:48.08 know a bit of a mess, but it turns out
00:12:50.08 simply by recording three images where
00:12:54.21 we shift the image a little bit
00:12:58.01 120 degrees, 240 degrees,
00:13:01.23 you know 360 degrees, we can sort out
00:13:03.29 those three components precisely just by
00:13:06.05 using some simple algebra, and so
00:13:10.06 that allows us to figure out how to
00:13:13.07 extract from all this combined
00:13:13.20 information here the information that
00:13:14.24 belongs up here and belongs down there.
00:13:18.15 Ok, so again here is the observed
00:13:21.05 region. The spots indicate the
00:13:26.01 frequency of the pattern that we are
00:13:29.01 illuminating with, and as you can see
00:13:30.01 what we do in practice is we try to make
00:13:31.19 absolutely the finest pattern that can
00:13:33.05 just squeak through the objective lens.
00:13:36.14 That's right here at the limit of what'll get
00:13:37.13 through the objective lens. Anything out
00:13:40.21 here won't make it through; anything in
00:13:42.04 here we can see but you know it's at
00:13:45.26 lower resolution, and so by making the
00:13:47.13 finest pattern, we can get the most
00:13:49.21 information, and what we end up getting
00:13:50.18 is from the grading in one position we
00:13:55.09 get all the information here and all the
00:13:56.22 information here, and we get to map that
00:14:00.20 into the objective lens. And then we can
00:14:05.09 repeat this by rotating the grading into
00:14:07.08 different directions - two directions
00:14:07.23 or three directions, and we can then do a
00:14:10.20 very good job of filling up this space so
00:14:13.14 that we now get out to twice the
00:14:14.16 resolution.
00:14:16.07 Ok, and it turns out that was just sort of
00:14:20.05 what we do in the XY plane, but we can
00:14:24.02 also create interference patterns in the
00:14:25.27 Z direction as well and that has the
00:14:27.12 benefit of not just extending
00:14:30.19 the lateral resolution
00:14:31.23 in the xy-plane but also takes care of
00:14:36.16 this missing cone region. You see our
00:14:37.12 little you know as in this structure you see our
00:14:40.24 little trail doughnut but it
00:14:43.12 convolves that out, and so we get much
00:14:45.00 higher resolution in the Z direction as
00:14:48.07 well as filling in all the missing
00:14:49.12 regions, ok. And this is done very
00:14:52.13 simply; the microscope is really quite
00:14:54.14 simple. We have light coming in through a
00:14:57.06 fiberoptic; its focus goes through a
00:15:00.01 polarizer focused on the grading. That
00:15:02.23 grading gives three diffraction orders
00:15:04.15 and these then get placed where the the
00:15:08.08 limit guys are imaged right at the limit
00:15:12.16 of the back focal plane plus the central
00:15:13.11 term. They then form structure on the
00:15:18.02 sample in x, y, and z you can see here's
00:15:20.02 Z interference terms & XY interference
00:15:22.25 terms on the sample, and then we record
00:15:26.29 those images. And what that allows us to
00:15:29.25 do is to build off these big
00:15:32.02 regions of Fourier space where only this
00:15:34.25 little you know this doughnut here
00:15:37.04 would have been observable with the
00:15:40.03 conventional microscope, ok.
00:15:40.16 So huge increase in resolution in
00:15:44.20 X, Y, and Z over conventional microscope.
00:15:50.06 And so by way of example here's an
00:15:52.09 actin stained HeLa cell. You can see all
00:15:55.29 the stress fibers in the cell under
00:15:57.27 normal illumination. Down here is where
00:16:01.04 we illuminate with one of the structured
00:16:01.20 illumination patterns. If you look very
00:16:03.15 closely, you can see little stripey
00:16:04.25 patterns along the sample which is
00:16:08.07 essentially an interference of the
00:16:09.29 diffraction grading with the
00:16:14.05 actin filament structure. We take the
00:16:16.24 Fourier transform of this we see you
00:16:18.15 know the structure of the actin
00:16:19.22 bundles, but you know when we
00:16:25.04 use the structured illumination, now we
00:16:27.02 see new structure - this little spot
00:16:27.11 here and these ray's coming off like
00:16:29.18 that and similarly down here that
00:16:34.07 represent the new information that was
00:16:34.22 displaced really from outside of this
00:16:36.26 circle. It gets \mapped into this circle
00:16:39.15 that we can observe, and so when we do
00:16:42.15 this, for example, we have a seven
00:16:45.27 different components. There's a central
00:16:48.04 component like you would normally
00:16:49.09 observe and each of the displaced
00:16:51.10 components. This is done with three
00:16:51.17 different orientation angles where you
00:16:54.12 can see new components that have
00:16:56.24 been reconstructed that go to much
00:16:59.12 higher resolution. And when we put all
00:17:01.04 that together, we now see we have a
00:17:04.14 Fourier transform that extends twice as far
00:17:07.21 as what would have been observable in a
00:17:09.23 conventional microscope, and when you
00:17:11.06 look at the consequences for an image,
00:17:14.14 the results are pretty stunning, and so
00:17:16.14 you go from an image like this to an
00:17:18.09 image like this, and so clearly much much
00:17:21.09 sharper detail. You can see actin filaments
00:17:24.23 have gotten extraordinarily close to one
00:17:27.24 another.
00:17:28.23 Here's the blow-up of that, and so where
00:17:32.16 you normally you would just see one big
00:17:33.23 blur, you can see completely resolved
00:17:36.27 separate actin filaments that are very
00:17:39.05 close to one another.
00:17:39.25 Similarly in three dimensions this is
00:17:40.01 just with fluorescent spheres, you can see
00:17:42.16 instead of having the normal horrible
00:17:45.13 out-of-focus blur in the Z direction that
00:17:48.25 you would normally have then you
00:17:51.25 see that tightens up quite dramatically
00:17:53.02 without having to do computer processing
00:17:55.20 in terms of deconvolution but just
00:17:58.24 because we now observe much more
00:17:59.16 information in the Z direction and we
00:18:02.11 see you know a superb XY spatial
00:18:07.24 resolution. Now it turns out this method can
00:18:10.12 actually be combined with that
00:18:10.13 interference methodology I talked about
00:18:14.17 in the beginning and actually get three
00:18:15.14 dimensional resolution all the way
00:18:17.29 around this uniform at very high
00:18:20.24 resolution. But you know here's just a
00:18:23.21 couple of examples where you can see
00:18:26.11 quite dramatic differences. These are
00:18:28.10 neurons, and these are synaptic boutons
00:18:30.15 from drosophila neurons and you can see
00:18:33.01 the difference between you know the
00:18:33.26 before and after using structured
00:18:37.23 illumination, and this methodology lends
00:18:40.24 itself very nicely to live imaging as
00:18:42.18 well and totally limited by how fast you
00:18:44.16 can form those patterns, and recent work
00:18:46.15 both at Janelia and here at UCSF has
00:18:50.00 used solid state devices to generate the
00:18:53.29 patterns essentially like in a video
00:18:55.01 projector instead of having to have a
00:18:57.20 physical grading that you rotate and
00:19:01.12 translate, and that allows us to do a
00:19:03.23 complete three-dimensional stack
00:19:04.19 within a second on live samples, so it's
00:19:09.15 quite exciting. Here is just a few pretty
00:19:12.25 examples of showing only color imaging
00:19:14.09 using structured illumination, so here's
00:19:19.04 you know here's chromatin in the cells
00:19:21.06 nuclear envelope being stained as well
00:19:23.11 as microtubule cytoskeleto, and then
00:19:26.22 you know you can zoom into the nucleus
00:19:29.05 and look at deformations in the nuclear
00:19:33.05 envelope around the chromatin, and you
00:19:34.15 can, in fact, if you look very closely in
00:19:37.12 this, you can see the nuclear pores, and
00:19:38.24 you can see where the nuclear pores
00:19:41.06 actually go in and push the DNA out of
00:19:43.04 the way all with light microscopic
00:19:46.04 imaging, ok. Now what are the limits, right?
00:19:47.24 I mean, it would seem from this that of
00:19:53.07 you know the best we can do is a
00:19:54.21 factor of 2 in resolution, and so the
00:19:57.04 reason is here the observed region,
00:20:00.11 and then we create a pattern, and
00:20:06.02 then you know that allows us access
00:20:08.08 to information out here, but what if we
00:20:11.12 could somehow create a pattern that had
00:20:13.18 frequencies that were much higher
00:20:13.28 frequency than defined by the
00:20:16.06 diffraction limit of the objective lens.
00:20:18.19 If we could do that, then we could get
00:20:20.20 information about
00:20:24.12 arbitrary sample information at any
00:20:26.20 resolution you wanted. The problem is
00:20:28.23 that how do we ever create a pattern
00:20:30.28 that has higher frequency content that
00:20:33.11 can fit through the objective lens, and
00:20:35.27 linear theory says that this is
00:20:36.05 impossible that you're limited the best
00:20:37.11 you can do which is I showed you before
00:20:38.28 you put the pattern of illumination
00:20:43.01 pattern right at the limit of the
00:20:44.02 numerical aperture of the objective lens,
00:20:46.16 but it turns out that there are
00:20:49.08 strategies where we can exploit
00:20:51.06 nonlinear processes of the dyes
00:20:53.03 themselves to get much higher resolution,
00:20:54.24 and two approaches have been
00:20:57.05 used. One is to actually use
00:21:02.24 non-linearity in the photoresponse of
00:21:03.03 the light of the dye itself or to use
00:21:07.04 photo switchable dyes where we can turn them
00:21:08.01 on and turn them off, and the turning on and
00:21:10.04 off then is an inherently nonlinear
00:21:12.27 process, and just by way of
00:21:16.27 illustration, here is an example from
00:21:19.14 trying to do this just by cranking
00:21:25.16 up the excitation light until now you
00:21:28.15 get a nonlinear response. That is, here's
00:21:31.03 the illumination, here's the emission
00:21:32.15 from the fluorophore; this could be from any
00:21:35.20 fluorophore. It doesn't matter, and that low
00:21:37.23 illumination levels, you get a very
00:21:39.00 linear response. You double the input
00:21:40.26 light you get double the output light,
00:21:42.03 but as you go up higher and higher in
00:21:45.26 illumination density, that stops
00:21:48.02 happening, and the reason is that you
00:21:48.07 start saturating the raising of
00:21:51.02 essentially exciting the fluorophore to the
00:21:54.07 excited state. You have a finite excited
00:21:55.16 state lifetime, and you can't get more
00:21:58.02 than one photon out before that decays,
00:22:01.23 and as a consequence this then starts
00:22:04.07 rolling over, and so it means if you create a
00:22:06.20 sine wave that you illuminate with at
00:22:09.09 high intensity, it's no longer linear and
00:22:12.27 so it will generate harmonics just like
00:22:14.11 if you take not just the principle wave
00:22:18.27 but you add second, third, and fourth
00:22:19.22 order harmonics you can build up higher
00:22:22.02 and higher frequency information, and if
00:22:23.21 we look at that here's a dye where we
00:22:25.28 crank up the intensity and we can see
00:22:28.08 not just the primary frequency but more
00:22:29.29 and more harmonics being a form as a
00:22:35.16 consequence of saturating the excitation
00:22:38.23 light, and each of these harmonics then
00:22:41.12 gives you access to further and further
00:22:42.28 and further information out about your
00:22:45.25 sample, and then as it turns out there's
00:22:49.07 no theoretical limit to what resolution
00:22:52.11 you could actually get with the light
00:22:53.19 microscope, so despite aspects of finite
00:22:55.12 numerical aperture and limited
00:22:59.06 wavelength, it turns out you can get
00:23:02.19 access to infinite amounts of
00:23:03.08 information.
00:23:04.00 The problem is a practical problem
00:23:07.02 of signal-to-noise photo destruction
00:23:08.02 of the fluorophore, those sorts of things, but not
00:23:12.03 an underlying physical limitation, and so
00:23:15.09 we can see that where conventional
00:23:16.00 microscopy gives us access to this sort
00:23:18.25 of region of Fourier space and
00:23:20.17 information about the sample, linear
00:23:22.00 structured illumination doubles that very
00:23:25.29 nicely by creating a pattern that is the
00:23:29.25 finest thing we can fit through the
00:23:30.04 objective lens by doing non linear
00:23:31.12 structured illumination either by
00:23:35.08 the saturation method or with photo-
00:23:35.17 switchable dyes we can get an arbitrary
00:23:39.15 number of harmonics that just again as I
00:23:40.22 said it's the signal-to-noise issue and
00:23:41.29 a dye issue of how far out you can go.
00:23:46.09 Now as you imagined, as you go further
00:23:47.18 and further out you're getting a smaller
00:23:49.28 and smaller region of all the
00:23:52.05 information because your porthole, your
00:23:53.05 viewport into that area, is always the
00:23:55.24 same size. It's always the same size as in
00:23:57.25 the conventional microscope, but as you
00:24:00.15 go further out you need more angles to
00:24:03.24 be able to fill this in and get a very
00:24:04.17 nice view of what the space
00:24:08.07 looks like, but you know with just two
00:24:11.00 new harmonics now recording eight
00:24:13.10 directions instead of three directions
00:24:16.10 you can do a very good job of filling up
00:24:17.23 Fourier space to high resolution, and
00:24:19.20 here's an example with fluorescent
00:24:22.18 microspheres where you can see you go
00:24:24.22 from a conventional image to the linear
00:24:27.02 structured illumination image the
00:24:28.21 saturated image and you can see you get
00:24:30.22 you know 46 - 50 nanometer resolution
00:24:33.20 and in principle this can go on and on
00:24:34.17 and on, and so the problem with
00:24:37.26 super-resolution then becomes one of
00:24:39.27 optimal dye development and better
00:24:46.05 detectors for example and not one of
00:24:47.17 fundamental physics. I want to thank the
00:24:49.14 many many people who have been involved in
00:24:52.25 various aspects of this work over the
00:24:53.20 years. This has been a wonderful
00:24:56.00 collaboration between my lab and Johnson
00:24:58.22 at Glad at UCSF, and I especially want
00:25:00.22 to thank Mats Gustafsson, the late
00:25:02.26 Mats Gustafsson who is really the inventor of a
00:25:04.02 number of the super-resolution
00:25:05.04 technologies.
00:25:07.15 Thank you.

This Talk
Speaker: David Agard
Audience:
  • Researcher
Recorded: July 2012
More Talks in Microscopy Series
All Talks in Microscopy Series

Talk Overview

This lecture describes several methods for approximately doubling the resolution of the light microscope: 1) illuminating and detecting through two objectives, 2) structured illumination microscopy (SIM) with a patterned light source, and 3) saturating (high intensity) structured illumination to provide further resolution extension. The methods and examples of images are presented.

Questions

  1. What physical property(ies) limit the resolution of microscopes?
    1. Wavelength of the detected light
    2. Numerical aperture of the objective lens
    3. Magnification of the objective lens
    4. Wavelength of the light used for fluorescent excitation
    5. A, B, C, and D
    6. A and B
  2. The use of 2 opposing objective lenses can increase resolution by
    1. Having a larger effective numerical aperture than possible with 1 objective
    2. Interference of the emitted light collected through the 2 lenses
    3. Interference of the excitation light going through both lenses
    4. A, B, C and D
  3. The resolution of I5m microscopy is best in
    1. XY (lateral)
    2. Z (axial)
    3. Same laterally and axially when using interference
  4. Structured illumination microscopy:
    1. Uses patterned light to illuminate the sample
    2. Produces images that can be directly observed through the eye-piece
    3. Produces images that can only be observed after computer-reconstruction of the raw images
    4. Can not be used for fluorescence microscopy
    5. A and B
    6. A. and C
  5. he resolution of Structured Illumination microscopy is best in
    1. XY (lateral)
    2. Z (axial)
    3. Same laterally and axially when using interference
  6. The resolution of Structured Illumination microscopy
    1. Is at best twice the resolution of a conventional microscope
    2. Can be arbitrarily high when using non-linear phenomena
    3. Depends on the magnification of the objective lens
    4. Is a function of the computer algorithm used for image reconstruction
  7. Live cell imaging using structured illumination
    1. Can only be done using non-linear phenomena such as dye saturation
    2. Is best executed with a system that can move the illumination pattern very quickly so that the sample does not change in between images
    3. Does not work
    4. Results in less photo-bleaching than conventional microscopy

Answers

View Answers

Speaker Bio

David Agard

David A. Agard

David Agard is Professor of Biochemistry and Biophysics at the University of California, San Francisco, an Investigator of the Howard Hughes Medical Institute and a member of the National Academy of Sciences. Agard’s lab studies the relationship between the structure and the function of macro and supramolecular complexes such as centrosomes, chromosomes, and HSP 90…. Continue Reading

Playlist: Microscopy Series

Reader Interactions

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *

iBiology and iBiology Courses are part of the Science Communication Lab (SCL). Our mission remains the same, to connect people to science. However, our focus has shifted to producing and evaluating cinematic films for education and the public, which you can find on the Science Communication Lab website. For more information, please see this blog post!