Emerging Tools for Single-Cell Analysis

this power. If the spot is not diffraction limited because of specimen effects, then more power can be used because the light is spread over a larger area. It is recommended to stay at least a factor of 10 below saturation, preferably a factor of 100.

Effects of Singlet-State Saturation

Because fluorescence saturation affects any dye molecules in the crossover, those in high-concentration (bright) pixels are affected the same as those in low-concentration (dim) pixels. As a result, the contrast of the final image may seem unaffected: it does not look “saturated” in the same way that a video image does when many pixels are recorded as 256 because the signal level is too high. Therefore, one cannot feel confident that one is avoiding fluorescence saturation by the casual evaluation of collected image data.

Rather than distorting the intensity contrast, fluorescence saturation causes the signal from any particular pixel to become a function of variables other than dye concentration. These variables include the local fluorescent decay time, which can depend strongly on the molecular environment, and local defocus effects that may affect the peak intensity in the focused spot and hence the degree of saturation. It follows from this last point that, as saturation occurs mainly in the focused spot, relatively more signal will be produced (and detected) from adjacent planes when one operates near saturation. This effect will marginally reduce the z-resolution.

However, the most important reason that saturation should be avoided is that it exposes the specimen to additional damaging illumination without producing proportionally more data. In fact, this bleaching effect may be more than linearly proportional to the “wasted” illumination: it may change the damage mechanism. Although, photodegradation mechanisms are as yet poorly understood, it now seems likely that the absorption of a second photon by a molecule still in the singlet-excited state may turn out to be a common bleaching mechanism (Sanchez et al., 1997). If this occurs, then it follows that above some intensity level, the bleach rate will be proportional to the square of the light flux, and operation near saturation would increase the effective bleach rate/excitation photon as well as reducing the signal-to-illumina- tion ratio.

Fluorescence saturation is a fundamental limitation on the rate at which confocal fluorescent data can be acquired. It can be side-stepped only if the beam is focused into a large spot or if the illuminating light is formed into more than one focused spot on the specimen. These conditions are present only in the disk-scanning instruments or laser line scanners (Hell et al., 1996; Buist et al., 1998; White and Amos, 1995).

Problem of Interactions

Limits on the performance of the CLSM do not act alone but can combine in complex and subtle ways. To highlight these interactions, let us define a characteristic microscopical problem shown schematically in Figure 11.2. The specimen is a living cell in which some of the protein subunits making up bundles of cytoskeletal fibers have been replaced with fluorescent analogs. The diameter of each fiber is about 10 times

smaller than the resolution limit of light optics, and excitation of the fluorescent dye causes it to bleach, a process that is toxic to the cell. Because the object is to observe the formation and movement of the linear cytoskeletal elements within the living cell, we will need to collect many images over time.

The important features of this example are as follows:

1.Observations will be improved by high spatial resolution in all three dimensions.

2.The ability to quantitate intensity of stain/unit length may permit determination of the number of subresolution, linear polymers bundled together to make up each visible fiber.

3.A number of images must be recorded at different times in order to show change/motion.

4.Each measurement will cause bleaching and cellular toxicity.

Clearly, these conditions contain an inherent contradiction: to obtain more quantitative temporal or spatial accuracy, we must pass more light through the sample, but this will produce more fading and cytotoxicity. As a result, the “improved” images may be of a dying cell rather than a living one, and, what we might call the biological reliability of the measurement may be inversely proportional to the physical accuracy set by counting statistics.

F i g . 11.2. Diagram of a notional specimen: a living cell process containing a filamentous structure smaller than the resolution limit and sparsely stained with fluorescent analog molecules.

There are other interactions among these parameters. High spatial resolution and Nyquist sampling imply that the microvolumes (pixels) excited by the beam and sampled by the detector must be partially overlapping. The measurement at each pixel is really the detection of that small fraction (usually 1%) of the fluorescent excitations in the specimen that actually produce signal in the detector (Sandison et al., 1994).

Higher spatial resolution implies smaller pixels (or voxels in 3D) and hence the need to count more photons from any given volume of the specimen. In addition, smaller features show less contrast. Therefore, if small features are to be resolved, even more events must be detected in order to reduce the statistical noise enough to visualize lower contrast features.

The increases in detected signal required for improved resolution are not insignificant. Just maintaining the statistical accuracy of the measurements when the size of the voxels is reduced by a factor of 2 requires four times more signal from the same volume of specimen in order to image a single plane and eight times more if a 3D volume is to be sampled.

Although this calculation accounts for the increase in the number of voxels needed, it still grossly underestimates the increase in dose needed to visualize smaller features. Assuming that the contrast of small features is inversely proportional to their size, the signal required from each voxel to “visualize” a feature is therefore inversely proportional to the square of the contrast. Therefore, if one accounts for both the additional voxels and the reduced contrast, the dose of excitation required to both sample and “resolve” a small feature increases to the fourth power of its spacings.

Fortunately, the x–y resolution, dx-y, is defined by the Abbé equation,

and the only way to actually increase it by two times is to increase the NA by two times. Because this also increases the fraction of the photons emitted from the specimen that contribute to the signal by a factor of ( NA)2, or four times, improved “resolution” can be attained with only a moderate dose penalty.

The most commonly used expression for resolution in the z-direction, dz, is

2

d 0 2 (11.4)

z ( obj)

This argument introduces the importance of keeping the pixel size appropriate to the operating resolution. Microscopists who normally record wide-field (WF) images on film may be less familiar with the idea that pixel size is an explicit experimental variable. It is important to remember that, on most commercial CLSMs, the “zoom” magnification factor used to change the size of the area scanned on the specimen usually does so by changing the pixel size (on the specimen). Therefore, although the ability to vary the display magnification by a factor of about 10 : 1 may seem to be a great convenience, with a given specific and NA, only one zoom setting provides optimal Nyquist sampling of the optical image data. All other zoom settings must nec-

essarily either overor undersample the object and produce either more beam damage or less resolution than they should.

Given the interrelated constraints highlighted above, the two parameters of a confocal microscope that must be optimized to approach its ultimate performance are:

1.Photon Efficiency. The system must count as many as possible of the photons scattered or emitted by the sample at the plane of focus.

2.Spatial and Temporal Resolution. Although, in practice, spatial resolution in fluorescence confocal microscopy is more likely to be limited by insufficient signal than by optics per se, it is still important to increase the x, y, and z optical resolution as much as possible by using the appropriate pixel size and the highest NA available and by scanning the beam no faster than is required to answer the biological question.

PRACTICAL PHOTON EFFICIENCY

Photon efficiency ( ) is the fraction of the signal photons generated by the action of the laser beam on the specimen that is accurately represented in the final image data set. Although photons can be lost both between the light source and the specimen or between the specimen and the detector, those lost before reaching the sample can usually be “replaced” with relative ease by an increase in laser power. However, photons lost after leaving the sample represent a fundamental loss: they carry information obtained at the expense of radiation damage to the specimen.

Light generated inside the specimen can be lost through several mechanisms:

1.absorption or scattering by either the objective lens or the medium that couples it to the focus plane or by the fixed and/or moving mirrors and transfer optics needed to scan the beam;

2.incorrect alignment of the optical system resulting in the improper placement or orientation of the pinhole;

3.low quantum efficiency (QE) of the photon detector; or

4.imprecise digitization of the output of the photon detector.

These subjects will be covered here as a background to the descriptions of practical tests of performance that follow.

Losses in the Optical System

Objectives. To a first approximation, fluorescent light proceeds from the site of its generation equally in all directions. As a result, only a small fraction even strikes the objective lens: ~30% of the total for NA 1.4 (assuming that the specimen is mounted in medium with an index of refraction of 1.515). The fraction of this light that emerges above the objective depends on its transmittance. Microscope manufacturers

T A B L E 11.1. Transmission of Selected Representative Objectives at Various Wavelengths

are now more willing to provide transmission data for certain modern lenses as a function of wavelength (Table 1). Instructions on making comparative measurements of the transmission of objectives can be found in Centonze and Pawley (1995).

Mirrors. A simple setup can be used to test the performance of the internal mirrors on most microscopes. All that is needed is a photodiode light sensor (one without an air space between it and the clear “window”) linked to a sensitive current meter (or better, a basic photometer) and a front-surfaced mirror. After adjusting the instrument to measure backscattered light (BSL) with low laser power, align the instrument if necessary; then switch to a high zoom setting and measure the light emerging from the objective with the sensor. Call this light reading Ia. Be sure to prevent stray room light from reaching the sensor and to couple the lens directly to the sensor with immersion oil if appropriate.

Now place a front-surfaced mirror on the specimen stage and set up the microscope to image the reflecting surface using the BSL signal. Then the beam scanning is stopped (or made to scan a very small raster) when focused on the mirror surface (brightest). Be careful to keep the illumination very low so as not to damage the PMT in the microscope and to realign the instrument if reducing the laser power means you have to add any neutral density (ND) filters in front of the laser. Focus initially on dust on the mirror specimen; then adjust the focus and alignment to give maximum reflected signal.

If your microscope produces specular reflection artifacts (bright blobs having nothing to do with the specimen) when used in the BSL mode, make your measurements in a part of the image field unaffected by these reflections.

After turning off the PMT(!), make a second reading (Pa) with the same photodiode placed just in front of the pinhole, making sure that all the light strikes the sensitive part of the sensor. To check that the sensor is protected from room light, obscure or turn off the laser, which should make the reading go to zero.

The Pa/Ia ratio will usually be depressingly small (~10–20%), but the difference covers losses at the various mirror and filter surfaces, including the beamsplitter (up to 70% loss for one pass through a so-called 50/50 reflected light beamsplitter), as well as those at the eyepiece, the various transfer lenses, and the objective.

Though indirect and somewhat cumbersome, such measurements can be useful for two reasons: (1) as a rough method of comparing the performance of different instruments (or different types of mirrors fitted to the same instrument) and (2) to monitor performance of a specific instrument over time. Performance degrades as dust and/or hydrocarbons and other atmospheric vapors condense onto the surfaces of mirrors and other optical components. In instruments having up to 11 reflective surfaces in the detector chain, a change in reflectance of even 1% can have marked effects on total photon efficiency.

Pinhole. In a confocal microscope, the pinhole prevents fluorescent light from reaching the detector unless it originates from the plane of focus. The pinhole is mounted in an image plane, and if it is misaligned or if its size is reduced beyond that corresponding to a diffraction-limited spot in that plane, then it will severely reduce the number of photons reaching the detector while producing only a marginal improvement in x–y resolution. Making the pinhole larger than the diffraction-limited spot allows detection of more photons; but as almost all of those originating from the plane of focus were already being collected, most of the additional signal comes from adjacent planes, reducing the z-resolution. Choice of the appropriate size and shape for the pinhole is a sensitive function of and of the objective lens NA and magnification. However, even an “optimum” aperture will exclude at least the 18% of photons that are in the outer rings of the Airy disk.

Multiphoton Excitation. Now that we have considered the loss produced by the pinhole, it is perhaps appropriate to insert a brief discussion of multiphoton fluorescent excitation—the only method of laser scanning light microscopy that can produce optical section data without the use of a pinhole. Earlier we discussed singlet-state saturation, a problem that occurs only when the extreme optical brightness of the milliwatt laser is focused into a diffraction-limited spot. Two-photon or multiphoton microscopy depends for its operation on a peak light intensity almost one million times higher again (Denk et al., 1990; see Chapter 10).

Intensity levels this high can produce nonlinear interactions, that is, interactions in which the output event is not linearly dependent on the input intensity. One such interaction is two-photon excitation. In this process, the density of photons in the “vicinity” of a fluorescent molecule becomes so great that occasionally two are absorbed simultaneously, almost as if they were a single photon having half the wavelength and twice the energy. Because this process requires the presence of two photons, the likelihood of its occurring varies with the square of the intensity of the photon flux.

Assuming negligible absorption, the intensity of light at any level in the focused cone emerging from a high-NA objective varies with the square of the distance from the focus plane. Consequently, the probability of exciting a fluorescent photon by a two-photon process drops off with the fourth power of the distance from the focus plane, and light is effectively excited only from very near the focus plane.

This is a very useful characteristic. It means that the dye is not subject to bleaching above or below the focus plane and that optical sectioning occurs automatically, without the need for using a pinhole in front of the photodetector. It is well suited to looking deep into living tissue slices because the lack of a pinhole permits one to collect and utilize as a signal even fluorescent light that has been scattered between the focus plane and the objective lens. In addition, UV-excited dyes can now be excited using normal optics and near-IR illumination, the design of the dichroic is simplified by the large “Stokes shift” between the wavelength of the illumination and that of the fluorescent light, and the longer-wavelength, near-IR illumination is scattered less by cellular structures.

There are also some disadvantages. To obtain the required intensity without cooking the specimen, one normally uses a laser capable of making pulses about 10-13 s in duration at a repetition rate of about 100 MHz (i.e., the laser is on only about 0.001% of the time, but the peak power is about 105 times greater than the average power). Such pulses require complex and expensive lasers that usually must be coupled to the scan head directly rather than via an optical fiber. In addition, the linear resolution is about two times larger (and the volume resolution eight times larger) because of the longer wavelength used; because average laser power levels are often 10–100 times greater than those used in single-photon excitation, cells are likely to be severely damaged if they happen to contain any pigments that absorb at the illumination wavelength. Finally, although the total amount of bleaching is reduced because it occurs only at the plane of focus, it seems that two-photon excitation may be about twice as damaging per excited photon as single-photon excitation. (Sanchez et al., 1997).

In like manner, it is possible in some cases to use similar equipment to perform three-photon excitation (Hell et al., 1996).

Because the fluorescent light is confined to a single plane by the excitation process, multiphoton instruments can in principle be operated not only without pinholes but also without descanning the fluorescent light through the scanning mirrors at all (Wokosin et al., 1998). However, as with normal confocal, the final image is still dependent on the performance of the photodetector and the digitizing system employed.

Detection and Measurement Losses

The Detector. The detector characteristics of importance to 3D microscopy have been reviewed by several authors (Pawley, 1994; Sheppard et al., 1992; Sandison et al., 1994; Stelzer, 1997). The characteristics of most importance to photon efficiency are:

1.Quantum Efficiency (QE). The proportion of the photons arriving at the detector that actually contribute to its output signal. (This may be a strong function of the wavelength of the detected photons; see Figure 11.3.)

the blue/green, this still means that 70% of the photons produce no signal. At 565 nm, only about 15% of photons striking the outer surface of a selected, S-20, end-window PMT are absorbed in the thin photocathode (PC) layer evaporated onto the inner surface of the window. The remaining 85% are either transmitted or reflected. As only absorbed photons can produce photoelectrons (PEs) to be subsequently amplified in the multiplier section, it is clear that any mechanism that reduces PC transmission or reflection losses may substantially improve PMT QE. This can be done by introducing collimated light into the PMT at an angle such that when it strikes the PC, it is totally internally reflected (Gunter et al., 1970; Fig. 11.4A). The light not absorbed when it first reaches the PC reflects off the outer surface of the window and then strikes the PC again. This process continues until the light either is absorbed or reaches the far side of the PC. Using such an optical enhancer can increase the QE by 60% at 520 nm, 180% at 690 nm, and 220% at 820 nm (Fig. 11.4B; Pawley et al., 1993b).

To take advantage of this optical effect, manufacturers have introduced tubes in which the PC material has been evaporated onto a “prismatic” surface made up of contiguous 45° pyramids. This allows a parallel, axial light beam to interact with the PC twice, a process that improves the average QE at a cost of pronounced (50%) variations in local, effective QE: depending on where on the pyramid a particular photon arrives, the resulting PE may or may not be able to avoid colliding with the walls of the “valleys.”

Transmission losses can also be eliminated by employing a solid photocathode, and recent side-window PMTs such as the Hamamatsu R-3896 have markedly higher QE, especially in the red. Although a significant achievement, such improved PC performance in the red is usually accompanied by significantly (10 ) higher dark-count rates. While the dark current of a PMT is usually low enough to be inconsequential, it is a strong function of temperature. In the heated interior of some commercial instruments, the dark count may not always be small compared to the signal level of a weakly fluorescent sample, and this is especially true when excitation levels have been reduced to permit photon counting without pulse pileup (see below).

Of more concern is the fact that the PMT output pulses produced by individual PEs can vary in size by an order of magnitude because of statistical variations in the small number of particles present in the early stages of the electron multiplier (Fig. 11.5A). The distribution in pulse height of these single-PE pulses is shown for several representative PMTs in Figure 11.5B.

As long as all photons make an equivalent contribution to the output current, the current remains exactly proportional to the number of photons. However, in practice, there is no way of knowing if the current sensed during a single digitizing interval corresponds to (for instance) one photon that happened to have made a large pulse or three photons that have made smaller ones. This uncertainty contributes to an equivalent noise term, called multiplicative noise, that (depending on the singlephotoelectron response of the tube) can add 15–30% to the intrinsic, statistical, or photon noise. If not remedied by the use of pulse-counting techniques, the effect of this additional noise can be removed only by increasing the counting time by (1.15)2 132% to (1.30)2 169%. In other words, operating a PMT in the analog mode reduces the effective QE of the PMT by between 25 and 45%!