23.01.2009 Review
Diffraction-unlimited optical microscopy
Peter Dedeckera, Johan Hofkensa and Jun-ichi Hottaa, 
aDepartment of Chemistry, Katholieke Universiteit Leuven, Celestijnenlaan 200F, B-3001 Heverlee, Belgium
Available online 21 January 2009.
Optical microscopy, and fluorescence microscopy in particular, has emerged as one of the most powerful and convenient microscopic tools available today. This power does come at a price, however, in terms of a limited spatial resolution: traditionally fluorescence microscopy has been limited by diffraction to a resolution of a few hundred nanometers, far too large to discern nanostructuring in biological or material samples. Recent conceptual advances have emerged that challenge this once-thought ‘unbreakable’ barrier, and fluorescence microscopy with nanometer resolution is now within reach. In this review we highlight some of the approaches that have made this paradigm shift possible.
One of the great paradoxes of life is that in order to truly understand the everyday issues and concepts that affect and fascinate us, one invariably has to focus on the microscopic world around us. This idea, while no longer particularly surprising, was realized as soon as the first lenses and microscopes were developed, and has only become stronger as our ability to observe smaller and smaller details improved. While this idea is usually brought up in the context of the life sciences, it is no different for the materials of today and tomorrow. Designing new materials tailored to specific needs and properties requires both study and manipulation of their micro- and nanoscale structuring, and microscopes are invaluable tools in modern-day research.
Not all approaches to visualize small details are equally useful for a given sample or problem, however, and a wide range of microscopy techniques have been developed. Each of these has its share of strengths and weaknesses, but one of the most convenient is optical microscopy, which, in an ideal case, requires only the illumination of the sample. Because of this, optical microscopy tends to be relatively noninvasive, nondestructive, and convenient to use. Fluorescence microscopy in particular has proven to be one of the most powerful microscopy tools available today, by combining the advantages of optical microscopy with the selectivity and extreme sensitivity of fluorescence emission. Indeed, fluorescence microscopy allows for measurements down to the level of individual molecules[1], [2] and [3], and has been successfully applied in the study of, for example, polymer dynamics[4], [5], [6], [7], [8] and [9], catalysts[10], [11], [12], [13], [14] and [15], dendrimers16.
Nevertheless, revealing nanoscale structuring and ordering remains one of the great challenges of science. Indeed, one aspect that is true for every microscopy technique is that there is a lower limit to the sizes of the details that can discerned, that is, the spatial resolution is limited. Intuitively, if the sample structure becomes too small, then it cannot be revealed. There are several possible reasons why this is so, ranging from technical aspects, such as instrument imperfections, vibrations, and measurement noise, to limitations imposed by the fundamental laws of nature. While the former limitations are, to some extent, specific to a particular instrument or experiment, the latter present limiting barriers that are hard to overcome.
In the case of optical microscopy, this fundamental limitation stems directly from the wave-like character of light: no matter what kind of optical components we use, we cannot focus the light waves to an infinitely small point, only to a finite region. This might seem surprising at first: when we collect the fluorescence emission from a single, isolated molecule and image it through a lens system (such as a microscope), the resulting image will not be a point but rather a three-dimensional shape of much larger dimensions (hundreds of nanometers; see Fig. 1). The same holds true for the excitation light: even if the light source is infinitely small, we can only focus the light to a diffraction-limited region and cannot confine it further. This is known as point spreading, and is an immutable law of nature.
Fig. 1. Point spreading: after imaging, an object of negligible size (such as a single molecule) will show up as a distribution of much bigger dimensions. The centroids of these distributions are indicated by the green and blue markers. If the emitters are spaced close together, as is the case for the molecules in the blue circle, then they cannot be distinguished and the structural information is lost. (Adapted from 83. © Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.)
Point spreading is problematical if we want to image a sample that contains fluorescent molecules in close proximity, such as in a finely structured sample. In this case the emission distributions from closely spaced fluorophores overlap in the resulting fluorescence image, making it impossible to distinguish between molecules that are close together (Fig. 1). Hence point spreading imposes a fundamental limitation on the spatial resolution that can be attained. (We will not go into further detail here; more extended, yet accessible, discussions have been published elsewhere[17], [18] and [19].)
While imperfections in the imaging system worsen the point spreading beyond the theoretical minimum, the optics in today's microscopes and objectives have advanced to the point where the level of aberrations is so low that their contribution is negligible. Therefore little gain in resolution can be expected from further improvements in the quality of the microscope optics. While it is possible to modify the optical system in such a way that the point spreading is strongly reduced, e.g. by including a second objective[20] and [21], these approaches tend to be limited by the fact that the instrument design and alignment become increasingly complex.
It thus appears that the spatial resolution in optical microscopy is limited to a few hundred nanometers, and that details at smaller distances are fundamentally out of reach. Fortunately it turns out that it is possible to obtain an imaging resolution that is fundamentally unlimited by diffraction, and a variety of techniques that aim to achieve this have recently appeared. Intuitively these techniques work by making fluorescence emission into a rare event: if neither the excitation illumination nor the fluorescence emission can be confined further, then we have to make sure that only a small fraction of the molecules emits fluorescence even though all of them may experience a similar intensity excitation. Alternatively, if we process the images mathematically, then we can invert the problem, and try to minimize the number of molecules that are not emitting beyond what is normally allowed by diffraction. It turns out that the answer to this problem lies in the properties of the fluorescent molecules themselves, and in the fact that fluorescence is a dynamic phenomenon[22], [23] and [24].
Nonlinearity
When we think about fluorescence emission, we tend to think about it as a fairly well-behaved, constant phenomenon. If we take a fluorescent molecule and irradiate it with excitation light, we expect to see a continuous, and predictable, fluorescence emission. Moreover, if we double the excitation intensity then we expect to see a twofold increase in fluorescence emission, and so on. This seems intuitive, because it agrees with the observation of emission from a dye-loaded solution in a fluorescence cuvette.
This apparent proportionality, however, only holds at the ensemble (averaged) level. One of the most fascinating findings from single-molecule experiments is that fluorescence emission is in fact a very dynamic phenomenon: individual molecules are seen to rapidly cycle between emissive and nonemissive states, displaying a highly complex and fascinating dynamic behavior[25], [26], [27], [28] and [29] (Fig. 2). The origin of this on–off ‘blinking’ behavior is not always clear, though it can sometimes be attributed to, for example, the formation of triplet states, photoisomerization, or electron transfer. It is also known to be virtually ubiquitous, being observed in wide variety of systems, including organic molecules[25], [26], [27] and [29], semiconductor nanocrystals28, and fluorescent proteins30.
Fig. 2. The emission of a single molecule of a fluorescent protein, clearly showing complex fluorescence dynamics, and the resulting nonlinearity in the fluorescence emission. The inset shows an expansion of the emission trace.
Apart from transient nonfluorescent or ‘dark state’ formation, interruptions in the fluorescence emission can arise for other reasons as well, including saturation (the fact that there is an upper limit for the emission rate due to the finite excited-state lifetime), and phenomena such as photoswitching or photoactivation[31], [32], [33], [34], [35], [36], [37], [38], [39], [40] and [41].
The net result is that the fluorescence emission is no longer strictly proportional to the excitation intensity. For a given sample and experimental setup the emission can depend on time, on the excitation intensity or intensities, or, by applying a spatial intensity distribution, on the position of the molecule with respect to the microscope's focus.
Every approach to achieve a diffraction-unlimited resolution makes use of this fact in some way. It is possible to roughly divide these approaches based on whether they make use of nonlinearities in space (that is, try to exploit the nonlinearity to increase the spatial confinement of the fluorescence emission) or time (that is, try to separate the emission of the different labels in time). In what follows we will discuss each of these approaches, and mention some possible advantages and disadvantages of each.
Nonlinearity in space
For a more detailed example of how nonlinearities can be put to work, let us look at the process of fluorescence emission itself. The initial absorption of an excitation photon causes the molecule to enter an electronically excited state, where it remains for (usually) a few nanoseconds, before returning to the ground state through the emission of a fluorescence photon. The central issue here is that there is always some nonzero delay between the absorption and emission events. In other words, the emission rate can only increase up to the point where the molecule is constantly in the excited state, when it reaches a plateau. At this point, increasing the excitation intensity does not increase the emission rate any further, and the fluorescence emission is said to be saturated (and strongly nonlinear).
Fig. 3 shows an example of this. In this figure a cross-section of the fluorescence emission rate along the focus of a confocal fluorescence microscope is shown, at different excitation powers. The shape of the diffraction-limited excitation intensity distribution is shown by the red curve in the figure, and does not change throughout the experiment. However, as the excitation power is increased, the saturation of the emission drastically alters the resulting emission distribution.
Fig. 3. The effect of saturation: he relative emission rate at the focus of a confocal fluorescence microscope and different excitation powers. The effect of saturation of the fluorescence emission is clearly visible. The relative emission powers, from red to purple, are 1, 5, 20, 100, and 1000, respectively.
The most crucial aspect of Fig. 3 is that it demonstrates that saturation enables us to create emission distributions with sharp edges and strongly confined nonemissive regions, even though these were not allowed in linear systems by diffraction.
This simple example shows how using fluorescence nonlinearities can allow us to bypass the limitations of diffraction.
Saturation of the emission rate is not the only possibility, however, as any transition to a state with different emissive properties can be used as long as it is saturable (that is, its light-induced formation is much faster than its recovery) [22]. Approaches that make use of this principle to create ‘sharp’ distributions in emission are known as RESOLFT microscopy, standing for ‘reversible, saturable optically linear fluorescence transitions’ [23].
By itself, the example shown in Fig. 3 is not directly useful for attaining a higher imaging resolution, since the fluorescence confinement is not improved but, rather, worsened compared to the ‘standard’ diffraction-limited case. Hence we must either modify the imaging to use the saturation in a different way or process the images mathematically. A number of techniques have been developed that attempt both approaches, and we will expand on these in what follows.
STED microscopy
While the focusing of the excitation light will always lead to some spatial distribution, the exact shape of this distribution can be modified into different ‘modes’ by manipulating the relative phases of the input laser beam. Especially interesting are ‘donut modes’, which are characterized by a sharp intensity zero at the maximum, surrounded by a region of intense irradiation[42], [43] and [44] (Fig. 4). Only at the exact center of the microscope focus is the intensity truly – mathematically – zero, while there is some irradiation intensity everywhere else in the focal volume.
|
