Principles of microscopy for ophthalmologists

Abstract

This short review begins with the theories of Airy, Rayleigh and Abbe on microscope resolution. Next, the principal developments in microscopy in the last half‐century are examined for relevance to ophthalmology: confocal microscopy, photoactivation light microscopy (PALM), stochastic optical reconstruction microscopy (STORM), stimulated emission depletion (STED), structured illumination (SI), 2‐photon and multiphoton excitation microscopy with a focused beam. Except for confocal, these are difficult to apply to the eye in vivo, as are the interference methods available in microscopes. However, interferometry in the form of coherence tomography is now a major ophthalmic method which has diverged from microscopy. Multiphoton excitation microscopy with an unfocussed beam is a new, low‐damage microscope method so‐far not exploited in ophthalmoscopy. The Mesolens, which throws off the historic limitations in microscopy set by the human eye, is described as a possible future aid to ophthalmology of the anterior eye.

Introduction

The optical microscope and ophthalmoscope were 19th century developments which, after a period of relative stasis, have shot forward in recent decades. The development of confocal laser scanning microscopes was triggered by the expansion of use of fluorescent probes in the 1970s, where unprecedented specific chemical information suddenly became available in the optical microscope, but detail was often obscured by the glare from out‐of‐focus fluorescent objects.

Efficient confocal imaging was made possible by the development of lasers and computers [1] and was quickly incorporated into ophthalmoscopes [2]. Multiphoton excitation of fluorescence enabled the collection of scanned images resembling those of confocal microscopes [3–6], but this method has been applied to the human eye in vivo only in recent years. New types of camera made it easy to image single fluorescent molecules in the microscope, if immobile, and this led to several methods for super‐resolution [7–9], which have not yet been applied in ophthalmology. The available designs for interference microscopes have also proved of little ophthalmic value: by contrast, the use of the ophthalmoscope has been extended enormously by the method of optical coherence tomography [10], developed independently of microscopy. The designs for the objectives (the prime lenses of optical microscopes) were largely perfected by 1890, but a so‐called Mesolens design, described in detail in a publication in 2016 [11], but not yet adopted by the major microscope manufacturers, has potential use for the examination of the anterior of the human eye in vivo.

Basic principles of microscopy

In the early 19th century, understanding of the optical microscope came from two sources. One was from the mathematician/ astronomer George Airy, who in 1835 published his calculation of the exact form of the image of a star [12]. His work was a tour de force of mathematical integration, using the newly‐discovered mathematical functions of F.W. Bessel, which made it possible to predict the exact size and shape of the image as a function of the diameter of the focussing lens, its focal length and the wavelength in vacuo. Airy’s image calculations apply to telescopes, microscopes, and all other optical devices with circular apertures. This gave an answer to the question of the highest resolution obtainable in a perfect optical system, using wave theory rather than the ray concept (which sets no limit to resolution). The elemental image, of which all microscope images are composed, is a disk of light, into which more than 80% of the energy from a point object is collected. This is surrounded by a series of rings so faint that they often need photographic overexposure of the disk to record them. The intensity distribution is as shown in Fig. 1, after Born and Wolf [13].

Fig. 1. Intensity variation across an Airy disk.

The optical units incorporate the numerical aperture and wavelength, making the diagram applicable to all perfect lenses and mirrors.

To suggest an agreed definition of resolution, Lord Rayleigh proposed the criterion that two objects should be described as resolved if their Airy distributions were separated far enough that the peak of one Airy disk fell over the first dark ring of the other. With this criterion, the basic equation of optical microscopy is:

$$\mathrm{r}=0.61\lambda /\mathrm{n}\sin \mathrm{\theta }\left(\mathrm{The}\mathrm{Airy}/\mathrm{Rayleigh}\mathrm{equation}\right)$$

where r is the resolution distance (equal to the radius of the Airy disk), 0.61 is a factor derived from the Bessel mathematics, λ is the wavelength in vacuo in the same units (usually μm) as the disk radius. The n sin θ part of the equation reflects the work, described next, of Ernst Abbe [14], who developed a different measure of resolution by practical experiments with microscopes, using much simpler mathematics. θ in the above equation is half of the total angle of capture of light from the specimen into the objective lens of a microscope and n is the refractive index of the fluid or air between the lens and the specimen. Abbe made finely ruled diffraction gratings to analyse the resolving power of the microscope and illuminated them with a parallel beam of light from a distant source. With the illumination on‐axis, part of the light went straight through the grating and passed as a parallel beam into the objective lens, remaining parallel to the optical axis of the lens. This was called a zero‐order beam. Two so‐called first‐order diffracted beams emerged from the illuminated part of the grating at equal angles from the axis on either side of the central zero order beam. If the angle of the first order beams was too great, so they missed the objective lens altogether, Abbe observed that the period of the slits in the grating could then not be detected in the microscope image. However, the periodic structure could be restored to view by sending in the illumination beam obliquely as shown in Fig. 2.

Fig. 2. Abbe’s method for obtaining highest resolution of a grating.

The illuminating beam, entering from the bottom of the diagram, is parallel to AB. AB is tilted to a specific angle, so the zero‐order beam enters at the extreme edge of the lens (right hand side in the diagram of the lens) and only one of the first order beams enters (to the left). In the diagram the two beams are shown by solid lines ending in arrows and the dashed lines represent wavefronts. The spacing of the slits shown in the grating is d. Only three slits are shown but the real number illuminated is much larger: d is small relative to the width of the objective lens.

Abbe’s mathematical analysis was based on the formation of wavefronts by combination of waves emerging from adjacent slits in the grating. All the marked angles in Fig. 2 are either right angles or equal to θ (the semi‐angle of acceptance of the objective). Along the wavefront of the first order beam, the phase near point C must be the same as that near the left‐hand slit, so AB + BC must be equal to one wavelength. From this, d (the period of the finest grating that could be resolved as periodic) can be shown by trigonometry to be equal to ½ λ / sin θ.

If a fluid, such as oil, of a higher refractive index, n fills the space between specimen and lens, the wavelength is reduced, and the angles of diffraction became smaller and the first order beam can be more easily captured, so the resolution is improved. Abbe modified his equation by adding the refractive index, n, because the real wavelength is λ /n, where λ is the wavelength in a vacuum (closely similar to that in air) whereas that in oil may be approximately 30% shorter. His equation became:

$$\mathrm{d}=\frac{1}{2}\lambda /\mathrm{n}\sin \mathrm{\theta }.$$

(This is Abbe’s equation for the period, d, of a just‐resolved grating in a medium of refractive index n)

He called the quantity n sin θ the ‘numerical aperture’ (N.A.) of the lens. This is now used universally as a figure of merit for the resolving ability of microscope objectives and is usually engraved on them, often with the fluid for which they are designed (e.g. water, glycerol, or an oil of closely defined refractive index and dispersion. Although d is not quite the same measure of resolution as the Airy/Rayleigh r, the numerical aperture (N.A.) or n sin θ is incorporated into the Airy/Rayleigh equation, which is often written:

$$\mathrm{r}=0.61\lambda /\mathrm{N}.\mathrm{A}.,\mathrm{where}\mathrm{N}.\mathrm{A}.=\mathrm{n}\sin \mathrm{\theta }.$$

Airy disks are not normally observed with microscopes (or slit lamps), since lens designers adjust the ratio of magnification to N.A. so that the disks are just too small to see. However, Fig. 3 shows a photograph of a region of the cephalothorax of a spider, showing four of the eyes, taken with a lens with a low numerical aperture, but using a camera with a very high pixel number. The lower panel shows a region of the image above it enlarged ten times. Pale spots in the enlarged image can be seen, which are the Airy disks. These are large enough to see because of the low numerical aperture of the lens. If a lens with a higher NA is used, each of the disks is transformed into an individual bright hair, identical in morphology to the hairs in the upper image but smaller. This shows that all microscope images are artefacts, in the sense that the artifice of the lens designer controls the size of the disks: all images are convolutions of the disk, which is an optical artefact, with the real structure.

Fig. 3. Photograph of eyes of a spider, and the space between them.

This photograph was taken with a lens of very low numerical aperture, but with a camera with a large number of pixels.

The Airy/Rayleigh equation has many uses. For example, it provides a means of calculating the maximum possible resolution of the fundus of an eye by an ophthalmoscope, assuming that the focussing by cornea and lens is perfect. If the pupil of a human eye is dilated to 5 mm and the distance from pupil to retina is 25 mm and the refractive index of the vitreous body is taken as that of water (1.313) the N.A. of the eye is 0.13. From this, it follows that in green light (where λ = 0.5 um) the resolution of an ophthalmoscope image of the fundus is 2.3 um. In this eye, assuming a perfectly refracting cornea and lens, it should be easy to image the densest mosaic of retinal cones, where the distance between adjacent cones is 6.2 μm, but more difficult to image the densest arrays of rods, which are only 2.7 um apart [15]. Adaptive optics, mentioned below, are beneficial with real eyes in both cases.

By estimation from the length from cornea to retina and pupil diameter, the N.A.of any camera-type eye can be calculated. Such predictions from the Airy/Raleigh equation assume perfect optics: for the mouse eye an N.A. of 0.5 is calculated [16], with even higher values for the chicken, where the pupil may be of human size (5 mm diameter) in an eye of much smaller diameter. Despite these simplistic high N.A. calculations the refractive performance of these eyes is inferior to that of humans, so their resolution is much inferior [16].

The focus of a lens is three‐dimensional, with the Airy disk being a section in a plane perpendicular to the optic axis (the xy plane) of a prolate spheroid, surrounded by shells of much lower intensity [13]. At high N.A. (e.g. 1.4) the volume of maximum intensity extends along the optical axis about three times its diameter, but at very low NA the volume (which is also known as the point spread function) extends to form an axially‐elongated needle‐shape, being inversely proportional in length to the square of the NA, whereas the Airy disk diameter is proportional to the inverse of the NA. The variations with N.A. of the resolutions in the xy plane and in depth (z) are shown in Fig. 4.

Fig. 4. Variation with numerical aperture of the resolution of a lens in the normal sense (i.e. lateral) and in depth.

Note the tolerable reduction in lateral resolution (blue) down to N.A. 0.1 often found in commercial microscope objectives of low magnification and the catastrophic reduction in depth resolution (red) below N.A. 0.5, which is only partially improved by confocal and multiphoton imaging.

New methods in optical microscopy of actual or potential benefit in ophthalmology

The following methods will be considered:

1. Confocal microscopy

2. Photoactivation light microscopy (PALM)

3. Stochastic optical reconstruction microscopy (STORM)

4. Stimulated emission depletion (STED)

5. Structured illumination (SI)

6. 2‐photon and multiphoton excitation microscopy with a focused beam

7. 2‐photon and multiphoton excitation microscopy with an unfocused beam

8. Interference microscopy.

Confocal microscopy

In a simple confocal layout (Fig. 5) laser light is first focussed and then refocussed into the specimen by the objective lens to a single spot which is scanned raster‐fashion in two dimensions by means of oscillating mirrors [1]. Other configurations have been developed [17]. In the few nanoseconds that it takes to absorb and re‐emit fluorescence the mirror movement is negligible, so, with the setup of Fig. 5, the fluorescence would be focussed back into the laser, but is separated by a spectrally‐specific reflector, almost always (incorrectly) called a ‘dichroic’, which allows the longer wavelength emission to pass instead to a focus in a pinhole positioned in an opaque screen so that most of the beam is transmitted through the pinhole to a detector. The sole purpose of this refocussing into a pinhole is to reduce the out‐of‐focus glare. This happens because if there is fluorescent material at a point on the optical axis at the level marked ‘too high’ in Fig. 5 it will be doubly excluded from detection because, firstly, the illumination of the specimen will be less bright (because spread over a disk‐shaped area at that level) and, secondly, the emitted fluorescence will be distributed over the opaque screen as a broad unfocussed disk of light, with only a minor amount passing through the pinhole. The same will occur if the material is too low. In such a single‐spot laser scanning microscope (LSM) there is only a small improvement in lateral resolution (of a factor of 1.414 under ideal conditions with a very small pinhole, but the glare reduction is the great advantage: it gives a startling clarity to the image. If the focus knob is adjusted slightly, some regions of a fluorescently labelled specimen disappear into darkness, while others at a different level light up. In such an apparatus, scanning a single spot and detecting only from that moving spot, there is no optical image: the image, which is built up over time from the varying signal from a unitary detector such as a photomultiplier tube, exists only in computer memory.

Fig. 5. Simplified diagram of a confocal laser‐scanning microscope.

The scanning mechanism is not shown: in its simplest (though  least practical) form the specimen is moved relative to a stationary light spot. For biological specimens, beam scanning by means of mirrors is preferred, particularly when immersion lenses are required.

While the confocal LSM was being eagerly adopted in cell‐, neuro‐ and developmental biology in the 80 s, the confocal principle was being applied also to laser scanning ophthalmoscopes [2]. Modern instruments may now allow not just imaging of the fundus with a scanning spot but the intensity of the spot can be modulated to produce a visible pattern or text that the patient can see in any part of the retinal field, or fail to see because of damage to the retina when the spot is moved to a blind region. Also, the same optical system can be used to irradiate a bleeding vessel and photocoagulate the blood.

The fact that the scan can be made at speeds up to video rate, or beyond, while the reflected or emitted light remains as a stationary focal point, makes possible adaptive imaging [18]. In the simplest form of this, the light from the stationary focus is spread and converted by a positive lens into a collimated beam, which is shone on to a Shack‐Hartmann imaging array. This apparatus allows the errors in flatness of the wavefronts in the beam to be measured in real time and used either to provide information for corneal surgery to correct refractive errors or to modulate the input beam to provide a better image of the fundus, using, for example a deformable mirror to modify the input laser beam. The latter method has been used to achieve great improvements, particularly in the imaging of the photoreceptor mosaic in the eye.

Photoactivation light microscopy (PALM) (ref. [7])

With modern cameras it is now easy to record the position of individual fluorescent molecules, provided they are immobilized or, at least, confined to 2D diffusion (e.g. in a lipid bilayer). However, any specimen bright enough for photography tends to contain so many fluorescing molecules that the Airy disks generated around each molecule overlap, and an area of uniform brightness is seen. The PALM method solves this problem by using fluorescent photoproteins derived originally from the green fluorescent protein of the jellyfish Aequorea and other coelenterates. A wide variety of bacterial, plant and animal cells can be genetically modified to produce these photoproteins and even to synthesize hybrid proteins consisting of one molecule of a protein that is under study covalently linked to one fluorescent photoprotein molecule. The advantage of the photoproteins over ordinary fluorescent stains is that they can be photoactivated, i.e. converted from non‐fluorescent into a fluorescent state. In photoactivation, light of one wavelength, e.g. blue, can switch a photoprotein, initially non‐fluorescent, into a state where it can be made to fluoresce with green light, emitting red fluorescence. The PALM imaging is done by first activating a tiny proportion of the photoprotein molecules that are present with a blue light of low intensity. Then a green light of high intensity is shone on the specimen and kept on for a long time, until the previously activated molecules have been taken through many cycles of absorption of green photons and emission of red ones, which invariably ends with the destruction of these molecules by bleaching, brought about by the free radicles created by the process of fluorescence. At this point the detector has recorded all the emitted photons of red light, and the image is of strong bright Airy disks with a high signal ‐to‐noise ratio and with few overlaps, because so few of the photoprotein molecules present were activated by the blue flash. The bulk of the molecules may be completely unaffected at this stage.

The photographic image is stored and the position of the centroid of every Airy disk is calculated and stored. The blue light is again applied, and the process of activation, fluorescent emission, photography and calculation of centroids is continued hundreds of times. When the centroids are combined in a projection, the result is an image with a resolution ranging from 2 to 25 nm: ten to a hundred times higher than the normal Airy/Rayleigh resolution.

The downside of this method is that the specimen must be absolutely immobile at the molecular level for a total imaging time that can be hours or days. The first PALM experiments used a frozen section of tissue. Also, it requires the tissue or cell line to be genetic modified so that it synthesises the photoprotein. These two factors make PALM unlikely to be useful in ophthalmic research, except with ex‐vivo material.

STORM (ref. [7]) is similar to PALM in that the fluorescence of a subset of the fluorescent molecules can be recorded, but uses organic dyes rather than photoproteins. Unfortunately, it is equally unlikely to find use in ophthalmic work although there has been recent success in applying it to living cells.

STED (stimulated emission depletion) (ref. [9]) This method shared a Nobel Prize with PALM, but uses a totally different principle, which is made possible by the phenomenon of stimulated emission, which forms the basis of laser action. A STED microscope resembles a single‐point confocal laser scanning microscope in that a laser beam is focussed on the specimen and the emission is collected and passed through a limiting aperture to a detector. Since only very high magnification images are required, a piezo‐driver XY stage rather than scanning mirrors can be used. There are, however, two lasers in use. One is focused by a high NA objective to give a normal scan of an Airy disk of light on the specimen. This results in fluorescent molecules throughout the Airy disk being raised to a high energy level, from which each molecule may fall by the emission of a photon of a slightly lower energy than the one that was absorbed. The difference from ordinary laser scanning is that there is the second laser which has a longer wavelength and its focus, though concentric with that of the first laser is converted (by special optics) into a doughnut shape. The laser foci are exactly superimposed and scanned simultaneously. The first laser brings the fluorescent molecules lying within its Airy disk focus into an excited state. The second laser produces stimulated emission: a very special process in which the longer‐wavelength photons produce a kind of resonance in the excited molecules which makes them emit a photon of exactly the same wavelength as the stimulating photons. By placing a short‐pass filter in front of the detector, these longer‐wavelength photons are blocked and the image only the light emitted by the scanning of the central area of the Airy disk, where the doughnut beam intensity falls to zero. The doughnut beam effect is saturable, so outside a sharp circular boundary the shorter‐wavelength fluorescent emission can be suppressed totally, and the effect is as if the specimen were being scanned with a spot far smaller than the Airy disk. This is genuine super‐resolution, whereas the centroid‐calculation methods are better regarded as computational localization. With biological specimens, STED achieves a resolution of 10‐70 nm, cf the 200 nm of ordinary optical resolution.

STED can work at video rate, but it is of value only at very high magnifications and only certain special dyes can be used, so it seems unlikely to find use soon in ophthalmology.

Structured illumination [8]

Structured illumination microscopy can best be explained by a thought experiment. Imagine that the specimen being examined is a shiny metallic grating, as in Abbe’s experiments. The illumination is by means of reflected light. Further imagine that the grating is so fine that the periodicity cannot be resolved in the microscope with reflected light. Then imagine the use of some device to illuminate the specimen with bright parallel lines, parallel with the slits in the grating with a period that is only just resolved. If the grating period is, say, 20% shorter than the period of the bright lines, the image will show a moire effect, due to the periodic coincidences in the two sets of lines. In other words, the structure of the illumination can reveal the presence of a periodicity that could not be resolved with uniform illumination.

A structured illumination microscope [8] uses optics that create an interference pattern of parallel bright lines at the limit of resolution over the entire microscope field. To record one super‐resolved two‐dimensional image least nine images are recorded, with the illuminated lines at three different angles across the field and, at each angle, the illumination fringes moved to three different positions. The images are digitized and fed into a computer, where their Fourier transforms are calculated and combined. The computed transforms contain all the information in each image, including the phase information, so an inverse transform can be calculated. The result of the combined nine transforms is an image with an improvement in XY resolution by a factor of up to two relative to the Airy/Rayleigh resolution and an even greater improvement factor in the axial resolution. This improvement in resolution is obtained even with non‐periodic biological specimens.

Since, like PALM, this requires an immobile specimen while all the images are taken, the method is fine for cell biology, where it gives excellent information from fixed specimens, but there is no obvious value in ophthalmology.

Two‐photon and multiphoton excitation with a focused beam [3–6]

In a confocal microscope, the total amount of beam damage (including photobleaching) is the same at all the levels shown in Fig. 5 as in the in‐focus level. This is because the beam is wider the further it is from the focal plane so any point will be irradiated longer during scanning, and this counterbalances the falloff of intensity in the cone of light away from the focal plane. It was pointed out by Sheppard [3] that the damaging absorption above and below the plane of focus in fluorescent biological specimens in a microscope might be avoided by using the 2‐photon absorption process. This process does not appear naturally except in the interior of stars, but can be observed in the focus of high‐powered lasers. Remarkably, Göppert‐Mayer [4] discussed it as a physical possibility decades before the invention of the laser. The advantage for microscopy predicted by Sheppard [3] was demonstrated experimentally by Denk, Stricker and Webb [5] in 1990 using the newly‐available commercial confocal scanning microscope but with a mode‐locked dye laser instead of the normal continuous‐wave laser. Soon after this, a prototype mode‐locked titanium sapphire laser proved to be a more convenient light source for this purpose and such lasers were adopted world‐wide for 2‐photon microscopy [6].

The key point about two‐photon absorption is that it is not proportional to the intensity of the laser beam, but to the square of the intensity, because two photons must arrive almost simultaneously at the fluorescent molecule for any absorption to occur. This means that significant absorption occurs only at the focus and the rest of the laser beam is not absorbed (see Fig. 6). Fluorescence thus occurs only in a submicron‐thick layer, observable as an ‘optical section’ without the need for a confocal detection pinhole. This improves the efficiency of the detector, although a seldom‐discussed improvement in resolution can often be seen if a confocal pinhole is used. To achieve 2‐photon excitation, a mode‐locked laser is essential: this type of laser emits pulses with powers of megawatts per square centimeter at the focus of a lens of high N.A. lasting only a few tens of femtoseconds (a femtosecond is 10^‐12 seconds) with relatively long dark periods (lasting 10⁻⁹ seconds) between pulses. Three‐photon imaging in a microscope was first discovered by Wokosin et al. [19]. The key observation was that fluorescence of a commonly used ultraviolet‐excited dye, the specific DNA label (DAPI) could be excited by using a mode‐locked laser of higher infra‐red wavelength than previously used. The three‐photon nature of the excitation was proved by measurements that the intensity of emission was proportion to the cube of the infra‐red intensity rather than the square. The term ‘multiphoton microscopy’ is now used to include cases of absorption of 2 or more photons. 3‐photon excitation can be obtained with a wide variety of dyes and also with natural metabolites such as NADH (nicotinamide adenine dinucleotide in reduced form).

Fig. 6. Comparison of fluorescence induced by single and 2‐photon excitation.

A continuous‐wavegreen laser beam (helium neon, wavelength 543 nm) is focused by the lens on the right into the upper part of a cuvette of dye solution (safranine in water) and induces yellow fluorescence throughout the length of the beam. This is the expected pattern for ordinary single‐photon fluorescence. The lens on the left focuses an invisible pulsed infra‐red beam into the lower region, producing a tiny spot of yellow fluorescence (arrowed) restricted to the focal region of the beam, due to 2‐photon absorption.

The advantages of multiphoton microscopy are that the infra‐red beam penetrates up to a factor of two deeper into most biological specimens than does visible light and photodamage is confined to a single plane of focus. Also, tissues often show characteristic autofluorescence colours and collagen can be imaged because of 2‐photon scattering, a non‐fluorescence process in which the scattered light is of a wavelength exactly half that of the incident infra‐red wavelength.

The chief disadvantages are the high cost of the mode‐locked laser, the need to determine the two‐photon absorption spectra of fluorescent probes (since this is not simply related to the single photon spectrum) and the damage to the specimen if highly absorbing materials such as melanin (which is present in most human tissues, including the iris of the eye) are hit by the scanning beam. All practitioners of 2‐photon imaging learn that melanin granules are burnt by the beam at normal imaging intensities, each producing a flash of light and leaving a bubble of carbon dioxide. Surprisingly, 2‐photon excitation of autofluorescence in the anterior [20] and even the fundus [21] of the eye of human volunteers has been carried out, allowing, in the latter case, study of visual pigment intermediates.

2‐photon microscopy with an unfocussed beam

In the normal practice of 2‐photon microscopy, the pulsed beam from the mode‐locked laser must be focused as sharply as possible within the specimen to produce sufficient fluorescence to allow recording during the pixel dwell time of the scanner, which is typically a few microseconds.

However, it has been shown that even when the pulsed beam is spread over an area tens of microns in diameter, emission from fluorescent dyes, including the fluorescent calcium indicator Fluo‐4 AM can be captured by a camera, and even recorded at rates of 100 frames per sec [22]. This widefield method has been greeted with a dumbfounded silence by cell physiologists, whose microscopes, even if equipped with an expensive mode‐locked laser source, do not allow the beam to be spread in this way. The authors have explained the result quantitatively in that the huge reduction in intensity in the spread beam relative to the focused is counterbalanced by an equally huge increase in exposure time and an increase in the thickness of the layer of excitation. Although no optical sectioning occurs, because the beam is unfocussed, the procedure was found to show a great advantage when compared with the ubiquitous wide‐field use of conventional single‐photon sources: photobleaching was reduced drastically, allowing video and timelapse recording. Even at 100 frames per second, camera recording of intracellular calcium transients could be continued for 60 s without photobleaching a frequently‐used probe dye (Fluo‐4), whereas, with the conventional blue single‐photon excitation from a mercury lamp, more than 70% of the fluorescence vanished because of photobleaching within this period.

These observations, which have drawn little comment from the physiological community, may perhaps lead to a means of gentle 2‐photon imaging of the fundus of the eye, without need for scanning and with advantage of better penetration and lower scattering by the cornea, lens and vitreous body. However, the safety of focussing an infra‐red pulsed beam anywhere in the eye must be considered carefully.

Interference microscopy

A wide range of interference microscopes has been developed, to detect and measure changes in refractive index or of the level of reflective layers in specimens, which cause changes in phase of the transmitted or reflected light. In all of them, the image is formed by splitting the illumination beam into two and recombining the two beams and observing the changes in intensity due to interference. In biological research the current dominant type is the differential interference contrast microscope (DIC) where the splitting and recombining is done by means of birefringent prisms. Biological DIC uses one beamsplitter in the illumination path and a second beamsplitter, acting as a combiner, in the path between objective and eyepiece of the microscope [23]. Such a transmission configuration is of little value for ophthalmology. But DIC microscopes for materials science have been available for at least half a century and might seem more promising for ophthalmic use since they work as epireflection instruments. They contain only one beam‐splitter used both to split the beams before they enter the objective lens from within the microscope tube and recombine them after reflection from the specimen and passage back through the same objective. It is very likely that these have been tried for ophthalmology of the anterior eye and rejected, because they work well for examination of single reflective surfaces and do not yield informative images of volume specimens such as the corneal stroma. They might, however, yield interesting results by the faint reflection from the conjunctiva with water immersion. The author has imaged corneal cells in the stroma and the basement membrane of the cornea at the behest of Professor Peter Watson who brought donor corneas from Addenbrookes Hospital for examination with a prototype confocal microscope in the MRC Laboratory of Molecular Biology. Confocal reflection is not an interferometric method but worked very well with a high numerical aperture water immersion objective applied to the cornea, revealing subcellular detail. The cells showed a characteristic shrinkage of the cytoplasm into radial strands, as a result of ultraviolet irradiation delivered in an attempt to reduce the rate of rejection by graft recipients.

For ophthalmology, a far more useful array of methods collectively known as optical coherence tomography (OCT) (ref. 10) has been developed, using a confocal ophthalmoscope containing a miniaturized Michelson interferometer. The first devices of this type were ‘time domain’ instruments, in which a single depth scan was made by shining one beam of the interferometer from a beamsplitter into the eye and and varying the optical path length of a comparison beam, which was reflected back to the beamsplitter by a moveable mirror and combined with the beam from the eye. At a certain length of the comparison beam fringes like Newton’s rings were observed, and this was an indication of the depth of a reflecting layer within the eye. The beam in the eye was then shifted in the XY plane and the depth at an adjacent region was detected. An image of a section perpendicular to the surface of the retina could thus be built up, showing normal and pathological changes in retinal contour. Advances in electro‐optics including frequency‐scanning lasers have now made possible far faster and more convenient instruments, working in the frequency rather than the time domain, as has been well explained [10]. Modern OCT instruments can even measure velocity of blood flow in retinal vessels by an optical Doppler method like those used in ultrasound scanning. It is clear that the designers in this field have little to learn from interference microscopy.

The Mesolens: a possible tool for wide‐field imaging of the anterior eye at high resolution

Wide imaging, for example of a field of 5 mm diameter to cover the whole of the pupil, can be done easily by means of a standard zoom stereomicroscope or a slit lamp. The image may look perfect to the eye, but actually has quite limited resolution because of low N.A. optics. Even modern zoom systems which are advertised as having ‘N.A. 0.5 at a magnification of 2x’ do not have these two specifications simultaneously: the N.A. falls to 0.1 as the magnification is reduced to 2x [24].

The range of objective lenses available from the principal manufacturers covers magnifications from 100x to 4x but it is never explained why the N.A. is approximately 1/40 of the magnification for all these lenses. Figure 7 provides the answer, which is that this ratio of N.A. to magnification produces the same apparent size of Airy disk for all the objectives when the image is viewed by eye through the same eyepiece. The Fig.  7 shows, for each objective, the M‐shaped intensity profiles of a pair of subresolution objects just resolved according to the Rayleigh criterion. The intensity profiles are scaled to the size in which the Airy patterns would appear in a standard 10x eyepiece. Also, to the same scale, are shown two lines corresponding to a psychophysical test measurement of the acuity of a human eye. The distance between the two lines is adjusted so that they subtend one minute of arc when viewed in a 10x eyepiece. This is the accepted acuity of an eye with 20:20 vision [25]. It is clear that the N.A. and magnification of every objective except that on the extreme right has been made to show the Airy disk radius (i.e. the Airy/Rayleigh resolution) at a size that matches the acuity of the human eye. It is likely that, historically, the NA was maximized for the 100x and 60x lenses at 1.4 (where θ, the semi‐angle of capture is approximately 70 degrees (near the practical limit of lens design which was achieved in the 19th century). Then, all of the other lenses were designed to give the same resolution in image space (which could be achieved easily since lenses of low N.A. need less compensation for aberrations and are easily designed and cheaply manufactured. There would have been no incentive to design a better 4x lens, since the better resolution could not be detected by the eye using the standard 10x eyepiece and eyepieces of higher magnification would sacrifice field. But modern cameras can have greater acuity than the human eye.

Fig. 7. Intensity profiles of pairs of just‐resolved objects as seen in the microscope image using a standard 10x eyepiece.

The Mesolens forms much smaller Airy disks than the standard lenses and has almost three times better XY resolution and 8.6x better Z resolution (FWHM) than the best available 4x lenses, which have an N.A. of 0.2.

Throughout the first phase of widespread use of confocal microscopy, some users encountered the problem that their microscopes did not give thin optical sectioning when low‐magnification objectives were used, as needed, for large specimens, e.g. a 12.5 day old mouse embryo 5–6 mm long and 3 mm thick. The intensity profile, to the right of Fig. 7, shows a new lens, which we have called the Mesolens, on which I have been working for the last 15 years, with the lens designer Esmond Reid and the laser physicist Professor Gail McConnell (University of Strathclyde) and members of her research group [11, 26], to make this type of imaging possible. With reference to Fig. 4, the prototype Mesolens was designed as a 4x objective to cover the entire body of a mouse embryo 5 mm long, which also required the working distance to be 3 mm, since this is the thickness of the embryo. This working distance and field size made it inevitable that the lens was large (over 50 cm long and approximately 65 mm in diameter in the mid‐region). It was found that a lens with a numerical aperture of 0.47 could be designed, which reduces the optical section thickness almost 4‐fold relative to a 4x objective of the highest available numerical aperture (N.A.0.2). The most difficult part was the full colour correction over the large range of 400 to 700 nm, which necessitated several unusual glass types and high precision in making and mounting the many lens elements.

The raison d’etre of the Mesolens was to permit confocal microscopy to be extended to large specimens. Confocal imaging depends crucially on the N.A. of the objective: standard low‐power objectives show severe smearing of XZ images along the Z axis as well as a high noise level because of their small throughput of light. A standard 4x lens (as used on clinical microscopes) has N.A. = 0.1 and has only 1/20 of the light throughput of the Mesolens. Because of its higher XY and dramatically higher axial resolution the Mesolens is able to collect 115 times as much information from a given volume of the specimen than the standard 4x lenses of N.A. 0.1, which have the same field size.

The large NA and large working distance of the Mesolens require a large aperture size, which means that the Mesolens cannot be screwed on to a regular microscope: McConnell had to build a custom microscope (see Fig. 8) around the giant objective lens, with correspondingly large ancillary optics such as filters and scanning mirrors for confocal operation. A piezo‐driven chip‐shifting camera was used, which shifted each pixel into a 3 × 3 array of positions and then combined all 9 images to give one 250‐megapixel image. This is still barely adequate for the required Nyquist sampling of the Mesolens image.

Fig. 8. Prof McConnell with the Mesolens system, equipped for confocal LSM or camera imaging.

The eyepieces are used for finding areas in the specimen. A laser bank supplies light for confocal imaging and the conical unit at the top allows the introduction of LED lighting for camera imaging. Transmission imaging, including darkfield and polarizing optics can be used and scanned transmission imaging using a differential phase detector can be performed in register with scanned confocal fluorescence or reflection.

It has not yet proved possible to convert the Mesolens into a commercial product: it requires skilled manufacture to tighter tolerances than for conventional microscopes. The Mesolens has, however, fulfilled our hopes of capturing single medium‐magnification images with a detail previously not available, so rare events can be counted with statistical validity, such as cell division in what were thought to be quiescent tissues. In a tissue, bacteria or other pathogens can be individually located in relation to the overall tissue architecture of lung or brain. Imaging all the cells makes it easier to avoid unconscious bias, e.g. in selecting a ‘typical’ or ‘representative’ cell. And what is recorded in a single imaging session is not just a few megabytes of images, but can be multiple terabytes, resulting in a dataset that can be archived and could potentially be analysed in future by methods that have not yet been invented, using, perhaps, new forms of artificial intelligence. Figure 9 shows a mouse embryo specimen: the type of specimen for which the Mesolens was designed. Only one image of a series of hundreds taken through the same specimen is shown. A video showing focussing through this series has been published [11]. Figure 9a shows the full‐sized image. 9b shows the region of the developing eye, in which the brightest tissue is the presumptive retina. 9b is merely a software‐enlarged portion of image 9a: the Mesolens is not a zoom lens. A diverse range of specimens and videos is available on a Strathclyde website [26].

Fig. 9. Mouse embryo fixed, stained with acridine orange and immersed in a high-refractive‐index fluid to render it transparent.

The fluorescence, chiefly confined to nuclei is imaged by confocal LSM, using a microscope designed around a Mesolens. The embryo is more than 5 mm long and 2–3 mm thick. The extracted and software-enlarged region, b, of image a, shows the eye with the presumptive retina. When this retinal layer was followed proximally towards the brain, it became the optic nerve (not shown in this image).

Discussion

It is no surprise that the chief optical discovery of the 19th century, namely that the limit of resolution of telescopes and microscopes was set by the wave nature of light, applies also to ophthalmoscopes. However, the living eye is a much more difficult specimen than the paraffin sections, fixed and cleared or the monolayers of cultured living or fixed cells normally examined with clinical transmission microscopes. This is probably why, as described here, so few of the spectacular advances in microscopy in the last five decades have been transferred to ophthalmology.

Laser scanning confocal optics have been well implemented in ophthalmoscopes, but few of the new super‐resolution microscopic methods such as PALM, STORM, STED and structured illumination have proved easy to transfer. The optical probes such as stains or photoproteins are difficult or impossible to use in vivo in the eye and there is no prospect of immobilizing eyes with the required precision of nanometers for super‐resolution. The 2‐photon methods are only just appearing in ophthalmology in vivo, more than 30 years after their implementation in microscopes, probably because of fears about beam‐damage. This may perhaps be mitigated by the spread‐beam method discussed here.

The mesolens has not yet been applied in ophthalmology. It is difficult to manufacture and few instruments are available: there are currently only four, all of which are in the UK, plus a more advanced model now being manufactured. However, the Mesolens can be used with an aqueous contact fluid and may prove to yield valuable information from the anterior eye because of its higher resolution than other instruments with equal field coverage (up to 6 mm diameter).

The rate of innovation in both microscopy and ophthalmology is accelerating because of advances in lasers, electro‐optical devices, cameras, computers, and software. Practitioners in both fields have now become developers of instruments, but the innovation is not neck‐and‐neck in the two fields. This is particularly clear in interferometry, where the only interference microscopes in general biological use are of the differential interference type, perfected in the 1970s, but optical coherence tomography has recently become a standard clinical tool and is being improved rapidly in speed and resolution by the use of completely new technology such as frequency‐scanning lasers.

Funding

Leverhulme Trust - 91232 [AMOS] Based on a paper presented at the 51st Cambridge Ophthalmological Symposium‐Engineering and the Eye at St John’s College, Cambridge 7 September 2023 27.

Competing interests

W.B. Amos is a director of Mesolens Ltd but has to date received no financial benefit or salary from the company. He has held an Emeritus Fellowship of the Leverhulme Foundation for research as a visiting Professor at the University of Strathclyde.