Review: Video-enhanced light microscopy and its applications in cell biology

The major inherent advantage of the light microscope (LM) over the electron microscope (EM), namely its utility for the study of dynamic processes in living cells, has until recently been severely compromised by its inability to make visible small cellular structures with the contrast-generation methods available for unstained cytoplasm. One consequence of this has been the dominance of electron-microscopic investigations of fixed cells and tissues during the postwar development of cell biology. However, within the last few years the combination of novel optical microscopic techniques with advanced video and digital image-processing technology has resulted in dramatic improvements in the quality of light-microscopic images, leading to a true renaissance in the use of light microscopy for the imaging of living biological specimens. These advances in what has come to be known as video-enhanced light microscopy (VELM) fall into two related categories.

The first is video-intensified fluorescence microscopy (VIFM). Here a highly sensitive low-light-level video camera is attached to a conventional fluorescence microscope in order to amplify the signal obtained from the fluorescently labelled specimen, permitting the visualization of faint fluorescence that could not be clearly seen by eye nor recorded by direct photomicrography. Frequently image averaging, using a digital image processor, is employed to enhance the signal, in which case the procedure is often alternatively described as digital imaging fluorescence microscopy.

Such is the exquisite sensitivity of VIFM systems that it is now possible to observe cells over extended periods of time at such low fluorescence excitation illuminating intensities that fluorochrome photobleaching ceases to be a problem. This allows one to study, for instance, the redistribution of labelled surface ligands or intracellular molecules over extended periods of time, and separately makes possible three-dimensional reconstructions from serial optical sections of static fluorescently labelled structures observed within single intact cells.

The second new technique is that of video-enhanced contrast microscopy (VECM). This is a complementary technique to VIFM, in which a high-resolution high-light-level video camera receives its input from an unconventionally adjusted light microscope, usually employing Nomarski optics. Both analogue and digital contrast-enhancement procedures and background image subtraction are used to produce major gains of image quality, permitting for the first time objects as small as individual microtubules and rapid processes such as fast axonal transport to be clearly seen in real time with high contrast, despite the fact that the objects observed may have dimensions an order of magnitude smaller than the resolution limit of the light microscope.

VECM is thus far more than the simple use of a video camera to observe and to permit the video recording of a conventional light microscope image, as an economical alternative to cinematography. Rather, it represents a radically new way of studying the dynamics of microscopic structures. Indeed, such is its power that VECM has recently played a central indispensable role in one of the major advances in modern cell biology, namely the initial characterization of a third universal cellular motile system involved in the transport of intracellular membranous organelles upon single microtubules. In addition, it seems poised to revolutionize the study of macromolecular localization in living cells using colloidal gold labels.

The purpose of this review is twofold: first, to describe the principles by which such dramatic improvements in the quality of light-microscopic images may be obtained and to outline the equipment and procedures employed, and additionally to highlight selected applications of VELM from leading laboratories in this field that have resulted in major advances in our understanding of cellular structure and function.

While applications of VELM can be neatly divided into those concerned with contrast enhancement of transmitted or reflected light-microscopic images of relatively high intensity, and those using high-sensitivity video cameras to enhance epi-fluorescence microscopic images of extremely low intensity, certain investigations require a combination of both experimental approaches. Since, with the exception of a few dedicated optical components and the different video cameras necessary for each approach, they employ a great deal of common optical and electronic apparatus, this combination is both straightforward and economical. A typical setup suitable for both applications, but for simplicity drawn with a non-inverted light microscope and only a single video camera, is shown diagrammatically in Fig. 1.

The heart of any VELM system is a large, stable, high-quality optical microscope, ideally mounted upon a vibration-isolating table in a dust-free environment. Microscopes having a direct optical path free from prisms and accessory lenses are generally to be preferred for fluorescence studies, since such additional optical components introduce unwanted surface reflections, while inverted microscope configurations have found widespread favour for VECM studies using Nomarski optics, largely for reasons of mechanical stability at the high magnifications employed, and of experimental convenience in accessing the specimen.

The objectives used usually combine high magnification with the highest possible numerical aperture (NA), in order to maximize both resolution and fluorescence image brightness. When used with transmitted light, they are matched with a condenser of at least equal numerical aperture to avoid resolution impairment. In contrast to conventional practice, a high-pressure mercury arc lamp with a d.c. power supply is generally employed to provide a stable light source of very high illuminating intensity for Nomarski and polarization applications, while tungsten or quartz iodide lamps may be used for video-intensified fluorescence studies, because of the low fluorescence excitation intensity required.

However, the mercury arc lamp, lamp housing, collector lens and illuminating relay optics provided as standard microscope equipment generally fail to provide even illumination across the entire aperture of the microscope condenser, seriously contributing to the problem of background ‘mottle’ in VECM images discussed below. Ellis (1985) has recently described a novel illuminator using a single optical fibre (0’75 mm diameter) to integrate the light from a mercury arc into a featureless luminous disc of light, which may then be imaged to illuminate evenly the full area of the condenser aperture. Examples of images obtained in this way are given by Inoué (1986). As a further development of this method, B. J. Schnapp (unpublished) described the use of a zoom lens and an achromat objective as a relay pair between the glass fibre and the condenser, and furthermore recommended the use of an Oriel (Stamford, Connecticut) mercury arc lamp and stabilized power supply to overcome the temporal instability experienced with standard microscope mercury arc illuminators. These methods of improving specimen illumination have led to further significant gains in VECM image quality, as described below (see Fig. 6).

To the microscope are attached the one or two video cameras (or other imaging devices), ideally arranged so that 100% of the light may be directed to the selected camera, using a light-tight optical interface that typically incorporates an eyepiece and a video camera lens.

It is often naively claimed that video cameras lack sufficient resolution to record microscopic images accurately. While it is self-evident that present-day video images are of more limited absolute resolution than photographic prints of comparable size, this claim of insufficient resolution is not necessarily true, but depends upon the magnification at which the optical image of the microscopic specimen is projected onto the target of the video camera. If the magnification of the optical interface between the microscope’s primary image plane and the camera target, termed the transfer factor, is made sufficiently large, the optical image of even a high numerical aperture objective will be oversampled by the video camera, and the resulting video image will remain diffraction-limited.

The appropriate transfer factor, typically ×6·3 for Nomarski studies with a ×63/l·4 NA objective, may be simply evaluated by considering the limiting spatial frequency resolvable by the optical system at the specimen plane in relation to the limiting horizontal and vertical spatial frequencies detectable on the TV camera target, and may be conveniently achieved by using a ×25 eyepiece and a 63 mm video camera lens (Schnapp, 1987). With such a high transfer factor, the resulting video image may be viewed without ‘empty magnification’ from a moderate distance of about 2 m at a video monitor screen magnification in the region of × 10000. Alternatively, an eyepiece of appropriate magnification may be used to project the intermediate image directly onto the camera target, an arrangement that generates less mottle (B. J. Schnapp, personal communication).

The trade-off for this necessarily high magnification and the over-sampling of the optical image by the video system is, of course, a small field of view, typically only 20 μm across. Consequently, some workers have found it useful to substitute a zoom lens system in place of a single high-magnification eyepiece, such that the maximum transfer factor remains unchanged for detailed high-resolution study, but can easily be decreased to obtain a larger field of view for survey work, without the necessity of changing either objective or eyepiece. Inoué (1986) illustrated the use of a Leitz motorized zoom ocular (×1·6 to ×6·3) for this purpose, in combination with a 300 mm telephoto lens (transfer factor ×l·92 to ×7·56), while Schnapp (unpublished) has found the manual Leitz zoom ocular (×5 to × 12·5) in combination with a 128 mm video camera lens (transfer factor range ×2·56 to ×6·4) to be a particularly favourable relay pair, since it contributes less to the mottle pattern seen in VECM images than the ×25 eyepiece/63 mm lens pair described above (see Fig. 6, below).

For video-intensified fluorescence microscopy it is necessary to strike an empirical balance, based upon the degree of labelling within one’s specimen and the sensitivity of the low-light-level video camera employed, between a high transfer factor, which maximizes image resolution, and a lower one, which increases image brightness. A transfer factor of ×l, for instance, which may be conveniently achieved by using a trinocular head and a direct, non-magnifying C-mount video adaptor on a conventional epi-fluorescence microscope, such that the primary image plane of the microscope falls directly upon the video camera tube target, will give a 40-fold enhancement of image brightness over a transfer factor of ×6·3. Although the optical image is consequently undersampled by the video system, for certain specimens the resulting loss in absolute resolving power may not be a serious problem. Fluorescently labelled features in the specimen are self-luminous against a dark background, and hence can be clearly seen at any magnification (although not necessarily resolved from one another), provided they are of sufficient brightness to be detected, just as can stars in the night sky, which subtend a solid angle on the retina smaller than the angular resolution limit of the unaided eye.

Most monochrome video cameras used for VELM produce conventional analogue video signals similar to those displayed on domestic televisions, generally complying with either the American and Japanese 525 lines per frame, 30 frames per second RS-170 (NTSC colour) standard, the British CCIR (PAL colour) standard, or the differently encoded SECAM standard used in France and elsewhere, the latter two both operating at 625 lines per frame and 25 frames per second. Triple standard (NTSC/PAL/SECAM) video cassette recorders (VCRs) enable one conveniently to play, although not to re-record, material taped using a different video standard, but otherwise the three systems are, for all practical low-budget purposes, incompatible. In both systems, each video frame is divided into two interlaced fields, displayed sequentially: the first comprising the odd, and the second the even numbered lines of the full frame. This system ensures a field repetition rate of 60 or 50 Hz, slightly above the flicker fusion frequency of the human eye, thus giving a steady image despite the fact that the image is being updated at only half that frequency.

Solid-state charge-coupled device (CCD) array detectors, which have the virtues of low geometric distortion, shading and lag, and also high-definition digital TV systems, are about to revolutionize both the nature and the quality of the primary video image. At present, however, for high-light-level VECM applications video cameras are usually equipped with monochrome analogue 1 inch Newvicon or Chainicon image tubes, which have relatively high light sensitivity (300–500 nA lux⁻¹) and good horizontal resolution (700–800 TV lines). Here the intent, using procedures described below, is to optimize the image quality of individual video frames, thus permitting real-time video recording of rapid cellular events.

For VIFM, in contrast, one of several types of image-intensifying video tube is usually used to detect low-intensity fluorescence emissions. Silicon intensifier target (SIT) image tubes of very high sensitivity (typically 40 000 nA lux⁻¹) and moderate horizontal resolution (approx. 600 lines at 10⁻²lux faceplate illumination, falling to 200 lines at 2×10⁻⁵lux) are most commonly employed. Intensified SIT (ISIT) tubes increase this sensitivity by another order of magnitude, although their resolution is somewhat lower. Such tubes, particularly the latter, are inherently noisy, both because of the statistical fluctuations in photon emission from very dim fluorescent specimens and, more importantly, because of the presence of substantial electronic ‘white’ noise generated by the SIT and ISIT intensifying devices themselves. Improved signal-to-noise ratios (SNR) in these video images may be achieved by digital signal averaging, as explained below. This limits their usefulness for the study of very fast transient fluorescent events, but is well suited for the study of stationary objects or of slowly changing phenomena such as cell capping.

The possibilities of low-light-level imaging have been further extended by the present availability of ultra-high sensitivity video cameras, incorporating a new generation of two or three-stage multi-channel plate (MCP) image intensifiers, which are less noisy and less sensitive to ‘blooming’ (oversaturation) on image bright spots. In these, the optical microscopic image is focussed onto a photocathode, causing the emission of electrons, which are in turn focussed onto the MCP image intensifier, in which they are amplified some million-fold before striking a phosphor screen. The intensified image so produced is then optically projected onto the target of a conventional high-resolution video camera or CCD camera, usually by direct fibre optics coupling, to produce a standard video signal with an overall enhancement in sensitivity that is maximally more than two orders of magnitude greater than that of a SIT camera, albeit at present with a maximum horizontal resolution of only some 400 TV lines, limited by the structure of the MCP itself.

Even more recently, the advent of slow scan cooled CCD array cameras, in which dark current noise is virtually eliminated, enables prolonged integrating exposures to detect low photon fluxes, with an extremely high signal-to-noise ratio. Current improvements in image scan frequency, from about 0·1 Hz to better than 4 Hz, now make these imaging devices additionally attractive for the study of slow dynamic processes of well-stained specimens.

The usefulness of ultra-high-sensitivity intensified cameras in detecting extremely weakly fluorescent specimens has yet to be established in practice, since at these extremes of sensitivity the autofluorescence of optical components, immersion oil and the specimen itself may lead to a severe reduction of specimen contrast.

The output signal from the selected TV camera is usually fed via a time-date generator to a monochrome video monitor, to allow observation of the initial ‘raw’ video image. For this purpose it is very useful to display also a graphical overlay from an analogue video analyser showing how the intensity fluctuations of the image lie within the full dynamic range of the video signal from black to white, so that this may first be optimized by adjustment of the analogue video black level and gain controls, as described below. The optimized video signal is then passed to a digital image processor. These devices, originally designed for the analysis of LANDSAT satellite images and similar data, are well suited for the enhancement of light-microscopic images. They consist of hardwired electronic components dedicated to storing and manipulating at high speeds the vast volumes of data presented by the incoming video signal, and are controlled by a microprocessor responsible for the selection and synchronization of the processing functions. An image processor may be either configured as a stand-alone unit, with its own keyboard by which the operator inputs instructions, or, in more sophisticated versions, may be connected by a high-speed direct memory access (DMA) link to a host micro- or mini-computer, whose central processor and magnetic or optical disc storage facilities can be used to process further and archive the digital image data.

Typically, the digital image processor will contain a video-rate (25 or 30 frames s⁻¹) 8-bit analogue-to-digital converter (ADC), capable of transforming the incoming video signal into a series of 8-bit numbers, each representing one of 256 possible grey levels for that particular picture element (pixel). Digital contrast enhancement of the live VELM images may be achieved ‘on the fly’ at this stage by non-linear greylevel optimization through an 8-bit input look-up table, in which the grey levels of the incoming signal are modified according to a pre-assigned function to enhance selectively the contrast of a particular greylevel zone, or to occupy more efficiently the full dynamic range from black to white. The image processor can store at least one full video frame as a two-dimensional array of such numbers in a dedicated random-access memory (RAM) termed an image memory or frame store. For all but the simplest applications it is desirable to have an image processor with a minimum of two such frame stores, each storing one digitized 8-bit video image of at least 5I2×5I2 pixels and thus occupying at least a quarter of a megabyte of RAM. In addition, the image processor will usually contain an arithmetic logic unit capable of image averaging, of subtracting a stored (background) image from each frame of an incoming live image, and of a number of other contrast-enhancement processes pre-selected by the operator. A variety of spatial filtering and edge sharpening processes are usually also available as standard functions, but very few image processors are capable of these pixel group processes at video frame rates. For more detailed discussions of digital image processing, readers are referred to the excellent primer by Baxes (1984) and to the authoritative treatises by Pratt (1978), Gonzalez & Wintz (1979) and Castleman (1979).

The contrast of the final enhanced image is optimized by passage through an 8-bit output lookup table, and is then converted back to an analogue monochrome video signal by an 8-bit digital to analogue converter (DAC) within the image processor. Alternatively, a pseudo-colour RGB image may be generated via a programmable 24-bit look-up table in which each monochrome image grey level may be assigned an arbitrarily different colour, and three such DACs, one for each 8-bit colour channel. The resultant video image is then displayed on a high-resolution monochrome or RGB colour monitor.

In addition, for recording the processed real-time video observations, a high-quality 3/4-inch video cassette recorder (VCR) is generally employed, into which it is useful to have a microphone input for the simultaneous verbal recording of experimental details and observations made during the course of one’s studies onto one of the audio channels. Digital information (for instance, concerning the microscope focus position or the specimen temperature) may optionally be recorded onto the second audio channel.

The dynamic bandwidth of present-day semi-professional ‘low band’ VCRs is limited, so that the resolution of the video image becomes restricted upon recording to about 340 TV lines. This then becomes the limiting resolution when the recorded images are played back. However, it is possible at reasonable cost to have such VCRs re-engineered by a video specialist to increase their monochrome resolution to 430 lines (5·5 MHz; ‘wide band’), thus more nearly approaching the resolution of the rest of the system.

For studies of slow dynamic processes such as cell division, a special time-lapse VCR is usually required. Unfortunately, most presently available time-lapse VCRs designed for security surveillance have limited resolution and give noise bars across the image upon dynamic replay, and also use non-standard tape formats such that videotapes recorded on them cannot be replayed on standard VCRs. Recently, however, two groups (De Brabander et al. 1985; Allen, 1987) have reported the successful use of a video animation controller to permit time-lapse recording onto a standard or wide-band 3/4-inch U-matic VCR, with no impairment of image quality.

Digital optical memory disc recorders (OMDRs) with enormous storage capacity of up to 2 gigabytes (2000 Mbytes) are now available, and look set to revolutionize the time-lapse recording and archiving of high-resolution digitized video images, particularly when erasable optical discs and digital recorders capable of video-rate recording and replay come onto the market. A 2 Gbyte disc can store 8000 full images each of 512² pixels by 8 bits, or substantially more if digital image compression algorithms are employed. If each image represented a digitized video frame, the storage of 8000 such images would represent over 2 h of time-lapse recording made at 1/25th of normal speed, i.e. one frame per second. OMDRs have a further substantial advantage of giving high-quality full-frame still replay, compared with the reduced vertical resolution of the single-field ‘freeze’ replay mode of a VCR.

The availability of high-sensitivity video cameras now makes routine the use of video-intensified fluorescence microscopy (VIFM) pioneered by Reynolds (1968, 1972), Willingham & Pastan (1978) and Reynolds & Taylor (1980). The obvious advantage of such highly sensitive imaging systems is that of being able to reduce the conventionally intense excitatory illumination incident upon the fluorescent specimen to such low levels that radiation damage to the specimen, and more importantly photobleaching of the fluorescent chromophore, the bane of conventional fluorescence microscopy, are virtually abolished. When used with an electronic shutter to prevent unnecessary illumination between observations, this permits video observation of the same living specimen over extended periods of time.

For this purpose, conventional epi-illumination fluorescence optics are used with a high NA objective, but the incident excitatory beam is heavily attenuated by neutral density filters. For many applications, direct video image intensification is all that is required, making this the simplest form of video-enhanced light microscopy. The output from the high-sensitivity video camera is simply recorded and used directly, without further enhancement, as described in detail by Willingham & Pastan (1983).

However, digital image averaging is extremely valuable for improving the poor signal-to-noise ratio (SNR) of the video image typically obtained in such applications, as exemplified in Fig. 2. This image-processing procedure is conceptually straightforward: a number of consecutive video frames are averaged, reinforcing the temporally unchanging signal of the specimen image while tending to cancel out the randomly varying noise. This results in an enhancement of the signal-to-noise ratio in proportion to the square root of the number of frames averaged. Such digital signal averaging is obviously compatible with the use of timelapse recording, which may be required to follow slow cellular processes, a single averaged video frame being recorded at the end of each averaging period. In practice, mechanical vibrations, specimen drift and other experimental factors usually limit the period over which averaging may be usefully employed to some 10 S (256 frames).

For fast-moving cellular processes, the benefits of noise reduction of such averaging procedures have to be balanced against the blur resulting from object movement during the averaging period. A compromise between blurring and noise reduction may be achieved by recursive filtering, i.e. the calculation of a running average. In this, the output is a weighted sum of all previous frames, with the most recently input frame having the largest weight and the preceding frames being given successively less weights in the averaging procedure. The smaller the value of the recursive filtering constant, the steeper this exponential decrease and thus the shorter the effective time constant over which the running average is made, a situation suitable for more rapidly moving specimens. Such recursive filtering has the great advantage of producing a continuous output image, in contrast to the discontinuous images produced by the simple averaging procedure, and is similar in effect to a very long persistence phosphor on a video screen.

Alternatively, the Kalman filter (Erasmus, 1982) may be employed. Like the recursive filter, this has the advantage of giving a continuous display. However, rather than giving a weighted average, it calculates the true average of all the image frames from the moment the averaging process started, thus maximizing the SNR without the necessity of predefining the number of frames to be averaged, and can be terminated when the averaged image looks satisfactory.

Examples from two fields will serve to illustrate the power of video intensification techniques in permitting extended observation of dynamic cellular processes, which previously could be seen only as isolated ‘snap shots’ of different cells taken by conventional photomicrography at what the experimentalist hoped were appropriate time intervals, each requiring such intense illumination and prolonged exposure as to invalidate subsequent study of that particular cell.

The receptor-mediated endocytosis and subsequent fate of fluorescently labelled macromolecules through the endosomal pathway was originally studied using video-intensification microscopy by Willingham & Pastan (1978). More recently, similar elegant studies by Herman & Albertini (1984) of the internalization of fluorescently labelled concanavalin A and low-density lipoprotein by cultured rat ovarian granulosa cells well illustrate the power of this technique, and have given further details of the movements of the endocytic organelles involved. These workers found that receptor-mediated endocytosis led to endosome formation within 10 min of warming the cells to 37°C, after preincubation with the ligands. Initial saltatory motion of the endosomes so formed in the cortical cytoplasm lasted for some 20 min, and was followed by their subsequent non-saltatory migration towards the centre of the cell, in association, surprisingly, with mitochondria. Primary lysosomes, which in untreated cells were dispersed throughout the cytoplasm, were seen to make a saltatory migration to the perinuclear region during the initial phase of saltatory movement of the endosomes. Upon subsequent arrival of the endosomes, fusion of the two types of organelle was seen to occur. Both lysosomal and endosomal movements were dependent upon intact microtubules, being abolished by the microtubule-depolymerizing drug nocodazole. It is now clear that such directed intracellular translocation of membranous organelles is closely related to the phenomenon of fast axonal transport, which has been studied by VECM and is described below.

As an alternative to studying the fates of exogenous molecules, the distributions of selected endogenous molecules or organelles within living cells may be observed by the technique of fluorescence analogue cytochemistry, pioneered by Taylor & Wang (1978) and Wehland & Weber (1980), and reviewed by Taylor & Wang (1980), Wang et al. (1982) and Taylor et al. (1984, 1986a). In this, a normal intracellular component is first biochemically purified and labelled with a fluorochrome, and then reincorporated into living cells by direct microinjection, by fusion with liposomes or erythrocyte ghosts, or by one of a variety of more esoteric methods. Once the cells have been thus loaded and checked for viability, VIFM may be used to study the incorporation of the fluorescent analogue into normal cellular structures, and its subsequent redistribution or behaviour in response to physiological and pharmacological stimuli. Alternatively, one may study the intracellular distributions of fluorescent dyes that label specific intracellular components, for instance NBD ceramide, which labels the Golgi complex (Lipsky & Pagano, 1985), rhodamine 123 and other dyes, which selectively stain mitochondria (Johnson et al. 1980, 1981), or rhodamine phalloidin, which binds to F-actin but not G-actin (Amato et al. 1983), or one may observe the disruption and recovery of normal cellular function following the intracellular microinjections of fluorescently labelled antibody to cytoskeletal components (Wehland & Weber, 1980).

The full promise of this approach has recently been demonstrated by DeBiasio et al. (1987), who have used the capability of a cooled CCD array camera to image from the blue into the near infrared (i.e. well beyond the range of the human eye) to undertake multiparameter fluorescence imaging on living cells. Five chromophores with non-overlapping fluorescence emission spectra, including a newly synthesized dye that emits in the infrared, staining different components within the cell, were sequentially imaged at 420, 542, 590, 675 and 785 nm within the same cell by changing the wavelength-selection filters.

One of the most dramatic applications of fluorescence analogue cytochemistry has been that of Sanger and his colleagues (Sanger et al. 1984, 1986a,b), who have studied the process of sarcomere formation in developing chick cardiac myotubules injected with fluorescently labelled α-actinin. This protein becomes incorporated into the Z-bands, and can be used to follow the development within individual myofibrils of labelled ‘minisarcomeres’ into sarcomeres of mature length (Fig. 3). In the video-recordings, the Z-bands of such mature sarcomeres can be seen to move closer together during the rhythmic contractions of the myotubes (Fig. 4A), while immature fibrils with more closely spaced α-actinin labelled cross-striations within the same cell show no signs of contractile activity (Fig.4B).

The possibility of subsequent immunogold labelling with, for instance, colloidal gold-conjugated anti-fluorescein antibodies, permits a direct correlation to be made between the fluorescence images obtained by light microscopy and the subsequent ultrastructural localization of the labelled compounds in the electron microscope (Luby-Phelps et al. 1984).

VIFM may also be used to great effect to determine translational mobility and membrane permeability by fluorescence recovery after photobleaching (FRAP) studies (Peters, 1985), or, by fluorescence ratio imaging (Tsien & Poenie, 1986; Bright et al. 1987), to quantify intracellular free calcium concentrations (Keith el al. 1985; Williams et al. 1985) or intracellular pH (Tanasugarn et al. 1984; Bright et al. 1987) within different regions of individual cells, particularly using the new generation of fluorescent calcium-sensitive probes (Grynkiewicz et al. 1985).

The use of fluorescence analogue cytochemistry in conjunction with quantitative digital image processing has been discussed by Taylor et al. (1984, 1986a), and further details of this and other applications of VIFM may be found in an excellent wide-ranging review by Arndt-Jovin et al. (1985) and by Taylor et al. (1986b).

In a quite separate use of VIFM procedures, Agard, Sedat and their colleagues (Agard & Sedat, 1983; Agard, 1984; Mathog et al. 1984, 1985) pioneered the use of cellular tomography, i.e. non-invasive optical sectioning of fluorescently labelled intact cells, for the three-dimensional computer reconstructions of cellular structures, to aid their analysis of the folding of giant polytene chromosomes within the nuclei of intact salivary’ gland cells. Their procedure is as follows: the chromosomes, stained by either of the DNA-specific non-intercalating fluorescent dyes Hoechst 33258 or 4’,6-diamidino-2-phenylindole (DAP1), are observed within intact 30 μ m diameter nuclei under epi-illumination fluorescence optics using a SIT camera or, more recently, a cooled CCD array camera. For each focal plane, 256 consecutive video frames are each digitized into 512² 8-bit pixels and averaged using an image processor, the resulting averaged image being stored on magnetic disc. Under computer control, the microscope focus is then advanced by a known distance (typically l · 3 μ m) and another optical section is collected. A complete data set of 24 sections encompassing the entire nucleus takes about 4min to collect, and is stored as 6 megabytes of digital data. As in any conventional fluorescence microscope image, the in-focus contributions arising from the selected focal plane within the specimen are partially degraded by out-of-focus contributions due to fluorescence emissions of light from other illuminated regions of the specimen above and below the focal plane, resulting in blurring of the observed image. The magnitudes of these out-of-focus contributions may be calculated, to a reasonable approximation, from in-focus images of these other regions, using a knowledge of the out-of-focus point spread function of the objective lens, as described in detail by Agard (1984). Thus, if one has collected a through-focal series of fluorescence images obtained by optically sectioning a single selectively labelled cell or nucleus, it is possible to estimate, for each image plane, the out-of-focus contributions made by the other planes in the object. By subtracting the sum of these contributions from each observed image, one can derive a set of ‘deblurred’ optical section images of enhanced clarity.

Agard, Sedat and their colleagues have implemented this computationally intensive blur-deconvolution procedure, giving striking improvement in image quality of individual optical sections of the polytene chromosomes (Agard & Sedat, 1983; Mathog et al. 1984, 1985; Gruenbaum et al. 1984). The three-dimensional conformations of the chromosomes thus revealed were then analysed by studying groups of sections from the entire stack, displayed stereoscopically on a high-resolution raster graphics monitor, with the aid of interactive modelling computer programs (Mathog et al. 1984, 1985; Mathog, 1985; Gruenbaum et al. 1984).

An alternative procedure, confocal scanning fluorescence microscopy, now enables comparable blur-free fluorescence images to be obtained directly from the microscope without computational blur deconvolution (White et al. 1987). Since this does not involve video techniques, it falls beyond the scope of this review, but will be discussed in detail elsewhere.

Long before the advent of VELM, significant improvements in the understanding of phase (as opposed to amplitude) contrast generation in transmission optical microscopy had led to the development of several highly successful optical systems for viewing transparent unstained biological specimens. Of these Zernike’s (1955) phase-contrast method and the Nomarski differential interference contrast (DIC) method (Allen et al. 1969) are now the most widely used. Ellis’s single sideband interference contrast (Ellis, 1978) and Hoffman’s modulation contrast (Hoffman, 1977) procedures are related asymmetric illumination systems that are alternatives to DIC, more suitable for particular applications but involving certain trade-offs in the intensity and/or the resolution of the image, albeit at a considerable saving in the cost of optical components.

In addition, the use of polarizing optics for the study of dichroic biological structures has been long appreciated, and has been developed and exploited with particular elegance by Inoué (1961, 1981b), while interference reflection contrast microscopy (Curtis, 1964; Izzard & Lochner, 1976) has been widely used to provide a means of studying the close appositions made between cultured cells and their substrata.

In principle, light microscopes employing any of these optical systems can benefit to some degree from video contrast-enhancement techniques, but to date these have been extensively used only in conjunction with transmitted light DIC and polarization microscopy (Allen et al. 1981a,b; Inoué, 1981a; Inoué & Tilney, 1982; Allen & Allen, 1983; Steponkus et al. 1984). In addition, Kachar (1985) has described a simple and effective asymmetric illumination contrast procedure employing VECM, and De Brabander et al. (1985, 1986) have employed VECM to permit the visualization of individual colloidal gold particles by conventional bright field and by epi-polarization optics, as described below.

Nomarski optics employ high-extinction polarizer and analyser elements, which in conventional use are set close to extinction, resulting in an image of relatively low light intensity in which the specimen contrast is optimized for the human eye. As pointed out by Allen et al. (1981b) (whose formulae I have here modified slightly to enhance their clarity), to a first approximation this perceived contrast, C, equals I |I_S– I_B|/I_B, where |I_S– I_B| is the absolute value of the difference between the intensity of the specimen, I_s, and that of the background, I_B, as it does in any observed scene. Under the conditions of conventional Nomarski observation, individual microtubules and similar minute cellular structures are totally invisible, since they scatter so little light that the value of |I_S– I_B| is imperceptibly small (Figs 5A, 6A). The sequential process of DIC contrast enhancement used in VECM, which results in these structures being clearly imaged with high contrast, involves three distinct and complementary stages, as follows.

The process starts with unconventional adjustments to the light microscope itself. First, the light level is maximized so as to increase the absolute value of the brightness difference between the specimen and its background, |I_S– I_B|, by using a stable d.c. mercury arc light source as described above, and by employing a condenser and an objective of the highest possible numerical aperture (NA l·3 or l·4) operated with a fully open condenser diaphragm.

An additional crucial step is now employed. As first discovered accidentally by Allen and subsequently validated theoretically (Allen et al. 1981a,b), the magnitude of the signal |I_S– I_B| may be further increased by adjusting the Nomarski analyser and polarizer away from the high extinction setting normally employed. Allen et al. (1981a,b) achieved this by inserting a quarter wave plate before the analyser, this combination forming a de Sénarmont compensator in which rotation of the analyser produces a bias retardation of 1/180th of a wavelength per degree (de Sénarmont, 1840; Bennett, 1950). One can more simply but less accurately achieve the same end by offsetting the objective Wollaston prism. Allen used large bias retardations of between 1/4 and 1/9 of a wavelength away from extinction, and named this procedure for optimizing the light microscope image to the video camera’s capabilities Allen Video Enhanced Contrast microscopy, using the acronyms AVEC-DIC and AVEC-POL to describe its applications to Nomarski DIC and to polarization optics, respectively. However, the exact values of bias retardation that best match the optical conditions to the varying saturation characteristics of the video tubes used in different laboratories is a matter of keen debate and experimental optimization among the cognoscenti, and bias retardations of lower values, about 1/15th of a wavelength, have been widely used by other workers (Inoué, 1981a; Inoué & Tilney, 1983; Valeet al. 1985a) and validated experimentally (Schnapp, 1987).

Important additional advantages of the use of VECM for polarization microscopy, described by Allen et al. (19816), are that specimens that scatter or depolarize light too much for observation under conventional high-extinction polarization conditions may now’ be studied, and that costly strain-free objectives and rectifying polarizers (Inoué, 1961), while still desirable (Inoué, 1986), are no longer essential. Furthermore, the resultant high-contrast images produced by video enhancement are sufficiently bright to permit real-time video recording, with frame exposure times some three orders of magnitude less than those required for photography under conventional high-extinction conditions.

In Kachar’s (1985) asymmetric illumination video contrast method, the microscope is again used at full condenser aperture, but with otherwise conventional bright-field optics. The lack of polarizers, phase apertures, half-stops or other interruptions to the light path results in very bright images from a conventional tungsten lamp illuminator. Contrast is generated by the use of asymmetric specimen illumination from this light source, resulting in unequal contributions to the left and right sidebands of the specimen’s diffraction pattern in the back focal place of the objective. The phase offsets of these side bands relative to the undiffracted zero-order wave have been shown by Ellis (1978) to be equal and opposite, so that when both are present with equal intensity they cancel out, generating no phase contrast from transparent objects illuminated under conventional bright-held conditions. However, introduction of sideband inequality, by the simple expedient of offsetting the illumination to eliminate one sideband, or by modifying one half of either of the conjugate images at the aperture iris of the condenser and at the back focal plane of the objective, results in the generation of phase contrast between the remaining sideband and the undiffracted beam. This procedure has long been known to light microscopists, but since the effect is very small relative to the overall intensity of the undiffracted beam, it is usually invisible to the unaided eye except on very favourable specimens. The use of the analogue video contrast-enhancement procedure described below rescues the weak signal, and generates a high-contrast image similar in many respects to a DIC image, but more suitable for the study of highly refractile specimens such as muscle.

These examples illustrate how conventional optical microscopes may be re-adjusted to optimize their contrast and image-generating potential to match the capabilities of analogue video processing using modern high-resolution video cameras.

While maximizing the signal |I_S– I_B|, these adjustments also unfortunately increase the overall intensity of the image, I_B, to such an extent that it is now far too bright to be viewable by the human eye, which perceives only glare. However, high-resolution video cameras fitted with Chainicon or Newwicon tubes have a much larger dynamic range than the eye, and can faithfully record the bright images produced by these apparently counter-productive microscope adjustments.

Once this bright image has been converted into an analogue video signal (Fig. 5B), the second stage of contrast enhancement is easily achieved by electronic adjustments akin to those of altering the brightness and contrast controls of a domestic television receiver. First, the overall background intensity of the video image is reduced by using the black level control (also termed the offset, bias or pedestal control), which simply adds a negative d.c. bias voltage to the overall video signal (Fig. 5C; see also Fig. 2C). Then the gain control is employed to amplify the resulting low-voltage (i.e. dark, low intensity) signal to occupy more fully the entire dynamic voltage range of the video signal from black to white, giving a final image of greatly enhanced contrast (Fig. 5D; see also Fig. 2D). For this, it is important that the video camera be equipped with manual black level and gain controls covering the full dynamic range of the video signal, ideally housed in a separate control box rather than on the camera itself (Fig. 1). The modified formula for video image contrast is thus: C = A·|I_S– I_B|/(I_B+I_V), where A is the video amplification factor set by the gain control and I_V is the negative d.c. bias voltage introduced by the black level adjustment.

Although the above procedure results in an enormous improvement of contrast, such that previously invisible microtubules (Fig. 6A) now become visible (Fig. 6B), it has one unfortunate consequence. While barely noticeable in conventional DIC images, background image blemishes generated primarily by unevenness in the illumination provided by standard microscope mercury arc illuminators, by lens imperfections, particularly in the optical interface with the video camera, and by dust particles upon inaccessible optical components, become amplified to troublesome levels during the analogue video processing, producing a background ‘mottle’, which almost totally obscures the specimen image (Fig. 6B). Fortunately, this may easily be removed by subtraction of a specimen-free background image of the mottle pattern alone (Fig. 5E), obtained either in the absence of the specimen or, more conveniently, by shifting the narrow DIC focal plane of the microscope away from the specimen plane. This, the third stage of contrast enhancement, utilizes the digital image processor, in which this digitized averaged background image is initially stored. Then, without other alterations of the microscope or video camera setting, one observes the refocussed specimen while digitally subtracting the background mottle image in real time from each incoming video frame of the specimen image (Allen & Allen, 1983; Alien, 1985; Schnapp, 1987; unpublished), to generate a final live VELM image of the specimen upon an almost featureless grey background (Figs5F, 6C). Further digital contrast enhancement may then be applied through the output lookup table, to ensure that the resulting image fully occupies the dynamic range of the video signal, or to accentuate detail within a certain grey-level zone.

Thus, by virtue of the extraordinary capabilities of the VECM technique to enhance the contrast generated by the scattering of minute amounts of light, tiny cellular structures such as microtubules (25 nm diameter) may now be seen with considerable clarity in living cytoplasm, despite the fact that their dimensions are an order of magnitude less than the Abbe resolution limit of the light microscope, their apparent diameters being inflated to that of the Airy disc (approx. 200 nm) by diffraction effects (Allen et al. 1981a,b; Allen et al. 1982; Brady et al. 1982; Hayden & Allen, 1984; Valeet al. 1985a; Schnapp et al. 1985; Allen et al. 1985; Koonce & Schliwa, 1985). The fact that these visualized filaments are indeed individual microtubules, rather than bundles of several microtubules, has been unequivocally established by electron microscopy after initial VECM observation. In dissociated squid axoplasm this has been achieved by conventional transmission electron microscopy after deep etching and rotary shadowing (Schnapp et al. 1985), while in intact cytoplasmic strands of the freshwater amoeba, Reticulomyxa filosa, individual microtubules have been clearly seen by high-voltage electron microscopy after fixation and critical point drying (Koonce & Schliwa, 1985).

While digital background mottle subtraction works well, the generation of video images initially free from mottle would be even more advantageous, since large mottle variations within an image ultimately limit the degree of analogue contrast enhancement that may be applied. It is thus highly significant that Schnapp (unpublished), by employing the novel optical fibre illumination system of Ellis (1985) and the improved relay optics described above, has dramatically reduced the mottle of his DIC images to such an extent that single microtubules may now be viewed by VECM without the need for digital background image subtraction (Fig. 6D).

Subtraction of such VECM images of a single specimen recorded at different times from one another provides a powerful method of detecting and quantifying motion, since all static elements of the specimen vanish, leaving only positive (new position) and negative (old position) images of any moving component.

For further information on the theoretical and practical aspects of VECM see Allen (1985), Schnapp (1987; unpublished) & Inoué (1986).

While many cellular phenomena have by now been observed by VECM, the one to which this technique has made the most dramatic contribution is that of fast axonal transport. This is the process by which membranous organelles are moved at rates of up to 400 mm per day along living axons, supplying the nerve terminal with new mitochondria, Golgi-derived vesicles and biosynthetic products from the cell body (orthograde or anterograde transport), and carrying senescent mitochondria as multilamellar vesicles and endocytosed material within multivesicular bodies back to the cell body for lysosomal destruction or recycling (retrograde transport). The giant axon of the squid has provided an ideal system for the study of fast axonal transport by VECM, since the entire axonal cytoplasm may conveniently be extruded from the axonal membrane and its associated cortical cytoskeletal sheath. This naked cytoplasm continues to support fast axonal transport for prolonged periods under the appropriate ionic conditions, and may easily be dissociated to allow observation of the individual structural components involved.

Using VECM, small membranous organelles and larger mitochondria may be seen moving in the dissociated squid axoplasm along single microtubules (Allen et al. 1982; Brady et al. 1982, 1985; Martz et al. 1984; Allen et al. 1985; Schnapp et al. 1985; Vale et al. 1985a). These are able to support both orthograde and retrograde transport simultaneously (Fig. 7), and in dissociated axoplasm, where viscosity ceases to be a limiting factor, it has been shown that large and small organelles travel at the same velocity, indicating common molecular mechanisms.

Both vesicles and mitochondria are capable of ‘changing tracks’ to another microtubule where it crosses the one currently being travelled (the microtubules being haphazardly oriented in the dissociated axoplasm), but in all cases they maintain the correct polarity of their movement relative to that of the microtubules themselves. Individual mitochondria may occasionally be physically torn into two halves, if by chance opposite ends become attached to two differently oriented microtubules and move off in different directions! Static micrographs can in no way do justice to these dynamic phenomena, which are best appreciated by viewing the original real-time video recordings.

More recently, VECM has played an indispensible part in permitting the direct functional assay of biochemical fractions for the protein components responsible for vesicle translocation upon purified microtubules. This has led to the isolation and partial characterization of a previously unknown protein, an ATP-hydrolysing enzyme named kinesin, which appears to be the motor responsible for orthograde transport (Vale et al. 1985b,c;Amos, 1987). Surprisingly, kinesin also permits the translocation of exogenous vesicles and of negatively charged polystyrene beads along microtubules, and promotes the axial gliding of sheared fragments of microtubules upon a glass surface, as shown in Fig. 8 (Allen et al. 1985; Vale et al. 19856; Weiss, 1986). An analogous protein has been isolated from calf brain (Brady, 1985), while a separate, larger protein, which may be the motor for retrograde transport, has been identified both in squid optic lobes and in bovine spinal cord axons (Vale et al. 1985d; Hollenbeck & Chapman, 1986).

These studies have thus permitted the initial characterization of a third unique class of cellular motile proteins that drive fast axonal transport by powering the translocation of membranous organelles upon individual microtubules. This hitherto unknown family of molecular motors is quite distinct from the myosins and dyneins involved in the more familiar forms of cellular motility, which have been best characterized in muscle and cilia, respectively. Furthermore, there is now strong circumstantial evidence that this mechanism of organelle translocation is also of widespread occurrence in non-neuronal cell types from amoebae to cultured mammalian cell lines, rather than being unique to fast axonal transport (see, e.g., Koonce & Schliwa, 1985; Scholey et al. 1985; De Brabander et al. 1985).

This heady rush of exciting new results clearly illustrates the importance of VECM, which now provides the biologist with a straightforward method of observing dynamic phenomena in living cytoplasm with great clarity, and of elucidating their molecular basis and pharmacological responses in suitable model systems.

In another major breakthrough, De Brabander et al. (1985, 1986) have recently combined the methodology of VECM with that of colloidal gold immunocytochemical labelling, which itself has revolutionized the electron-microscopic localization of known cellular macromolecules (reviewed by De Mey, 1983; and Beesley, 1985). Using conventional bright-field or epipolarization optics, and the contrast-enhancement techniques described above, De Brabander and his colleagues have been able to visualize individual stationary colloidal gold particles as small as 5 nm diameter in the light microscope, these particles appearing the size of the Airy disc (approx. 200 nm in diameter) because of diffraction effects (Fig. 9). Fig. 10 shows polylysine-coated 40 nm diameter gold particles after addition to PTK₂ cells, visualized by a variety of video-enhanced optical illumination systems.

Small immunogold particles may be used to stain microtubules or other cellular structures, rendering them a delicate pink colour in conventional bright-field microscopy (De Mey et al. 1982). The contrast of such immunogold staining may be dramatically enhanced by VECM (De Brabanderet al. 1986) as shown in Fig. 11. Here the individual gold particles are not resolved, but their binding clearly delineates the microtubules. In a similar application of colloidal gold labelling and epipolarization VECM, Inoué et al. (1985a,b) and Bajer et al. (1985) collected serial optical sections at 0·5μm intervals through Haemanthus katherinae endosperm cells whose microtubules had been indirectly immunolabelled with 5 nm gold particles, and they used these to generate high-resolution stereoscopic images of the three-dimensional distribution of microtubules within the mitotic spindle. Since immuno-gold labelling can be undertaken simultaneously with immunofluorescence or fluorescence analogue labelling, as exemplified in Fig. 12, the use of a microscope equipped with both high- and low-light-level video cameras for VECM and VIFM now makes possible sophisticated double or multiple labelling experiments.

The ability to coat gold particles with any chosen antibody, cellular protein or ligand, and then to follow changes in their individual locations after incubation with or microinjection or electroporation into living cells, represents a revolutionary step towards a new cell biology, namely the use of the light microscope for the dynamic study of known macromolecular interactions within their natural environment, the living cytoplasm.

Furthermore, such gold-labelled cells may be subsequently processed by a variety of EM preparative techniques (e.g. label-fracture, deep-etching, cryoultramicrotomy, freeze substitution followed by plastic embedding and thin sectioning, or freeze substitution followed by critical point drying and whole-mount observation), permitting the subsequent transmission EM localization of these very same gold particles by transmission EM, as De Brabander et al. (1985) have elegantly demonstrated (Fig. 13).

The potential provided by video-enhanced light microscopy of directly visualizing and of video-recording the movements of cellular organelles and cytochemical probes, and the possibility of subsequently determining the ultrastructural locations of these organelles, fluorochromes or individual gold particles bound to particular protein molecules, fulfils a previously unobtainable dream of making direct correlations between dynamic processes observable only in the living cell and specific macromolecular interactions revealed only by high-resolution electron microscopy. This permits a truly integrated experimental approach to many of the current problems of cell biology, and illustrates more clearly than ever before the complementary natures of optical and electron microscopy.

This paper is dedicated in memory of Robert Day Allen, whose death has been a great loss for biological microscopists worldwide. I am deeply indebted both to him and to my friends and colleagues Shinya Inoué, Bachara Kachar, Phil Presley (of Carl Zeiss, Inc.), Bruce Schnapp, Lans Taylor and Dieter Weiss for stimulating discussions and valuable practical advice concerning VECM, and for allowing me to study their experimental systems at the Woods Hole Marine Biological Laboratory, Massachusetts; and to David Agard, David Mathog and John Sedat of the Department of Biochemistry and Biophysics, University of California Medical School in San Francisco, for enthusiastically sharing with me their expertise in the area of VIFM cellular tomography during a short working visit to their laboratories. 1 am also most grateful to all those who have kindly supplied micrographs of their work to illustrate this review, and to Helen Saibil for her critical reading of the manuscript. This study was made possible by travel grants from the Marshall and Orr Bequest (administered by the Council of the Royal Society), the Nuffield Foundation and the Company of Biologists, to whom I am most grateful. An article covering some of the topics contained in this review in a condensed form has previously appeared in the Proceedings of the Royal Microscopical Society (Shotton, 1987), and common material is included here with permission from that Society.