Until the late 1980s, a majority of life science researchers investigated the intricate details of biological structure by capturing single snapshots of various cytological features using fixed and stained (in effect, non-living) specimens. Over the past couple of decades, however, research in the biological and medical sciences has largely shifted focus to investigating dynamic processes that occur on the molecular, cellular, and entire organism level in living systems on a vast array of timescales ranging from milliseconds to hours. The drivers of this transition to imaging living cells have been advanced developments in microscopy instrumentation and more sensitive digital cameras, as well as new synthetic and genetically-encoded fluorophores that are able to target specific organelles with high precision. Thus, the era of single-shot photomicrographs and digital images of fixed, dead cells has largely yielded to live-cell imaging using a wide spectrum of synthetic probes, quantum dots, and fluorescent proteins, where maintaining cells in focus on the microscope stage for extended periods of time is a mission-critical factor.

The increasing application of live-cell time-lapse imaging techniques in transmitted light and fluorescence microscopy is perhaps best evidenced by the voluminous number of research reports appearing almost on a daily basis in the scientific literature. In most cases, cellular activity is monitored in tissue culture imaging chambers specifically designed for the microscope stage using fluorescent probe technology coupled to sequential image capture. Advanced contrast-enhancing techniques, such as differential interference contrast (DIC), phase contrast, darkfield, fluorescence, and Hoffman modulated contrast (HMC) can be employed for recording the dynamics in a wide spectrum of specimens as continuous sequences over lengthy time periods. Similarly, more advanced fluorescence techniques, including laser scanning confocal, spinning disk, multiphoton, and total internal reflection (TIRF) microscopy are increasingly being used to monitor intracellular processes over time.

Aside from its importance in cell biology, time-lapse imaging can also be used to study a wide spectrum of liquid samples and solid materials in chemical, industrial, and geological systems, as well as liquid crystalline phase transitions and structural analysis of new and advanced materials in metallography. In the basic life sciences, time-lapse imaging (also termed cinemicrography) has proven to be an effective tool for investigating particle motion, intermolecular interactions, cell migration, cell division, organelle dynamics, apoptosis, differentiation, and neural process outgrowth, among a host of new and promising applications. More recently, the revolution in development of fluorescent proteins spanning the entire visible spectrum has resulted in the discovery of countless dynamic intracellular events that can only be investigated by observation over time.

Presented in Figure 1 are examples from several of the more useful and popular contrast-enhancing techniques employed for time-lapse imaging in live-cell microscopy. The adherent Indian Muntjac deer skin fibroblast cells illustrated in Figure 1(a) were imaged in DIC mode to visualize stress fiber outgrowth of the cytoskeleton. Phase contrast imaging of rat kangaroo kidney epithelial cells (PtK1 line) in Figure 1(b) reveals a bi-nucleated cell arising from aborted cell division, while the successful division of an individual human osteosarcoma (U2OS line) cell is captured with HMC in Figure 1(c). These powerful transmitted light techniques are widely utilized for time-lapse live-cell imaging. Likewise the fluorescence techniques of laser scanning and spinning disk confocal microscopy (Figure 1(d), dEos fluorescent protein in Chinese hamster ovary cells; and Figure 1(e), EGFP-labeled endoplasmic reticulum in HeLa cells) can be employed for a number of observations, including photoactivation, resonance energy transfer (FRET), photobleaching techniques, motility, and the dynamics of fluorescent protein distribution. Total internal reflection fluorescence microscopy (TIRFM; Figure 2(f), illustrating mCherry fluorescent protein fused to vinculin in fox lung fibroblast cells) is employed to observe events that occur in the cell membrane near the coverslip.

Historical Perspective on Time-Lapse Imaging

The first application of time-lapse imaging in living cells was reported over 100 years ago (approximately five years before Charlie Chaplin made his first movie) by a young French graduate student who was examining motility in syphilis-producing spirochaetes. The student, Jean Comandon, captured image sequences using an enormous cinema film camera coupled to a much smaller darkfield microscope, a sophisticated configuration for the period, but considered very crude by modern standards. During the latter half of the twentieth century, time-lapse image sequences were typically recorded using compact 16-millimeter cameras on microscopes equipped with phase contrast illumination. The time intervals between successive image captures were controlled by bulky auxiliary intervalometers, devices that turn the lamp on and off between images in order to avoid excessive illumination and heating of the specimens. Electronic light source shutters for microscopes were not commercially available until the 1980s.

Early time-lapse imaging techniques relying on film cameras were subject to a number of pitfalls. Foremost, the film had to be sent out (in most cases) for commercial processing, which meant that the evaluation of experimental results could be delayed by days or even weeks. In addition, the use of film was costly and subject to a host of processing variations that usually led to inconsistent results. For example, exposure mistakes might not be discovered until weeks after the experiment had been concluded, and the common problem of focus drift was often not detectable until the processed film was viewed. Although many of the artifacts associated with time-lapse imaging, such as focus drift, uneven specimen heating, and vibration, still hamper the final results when using modern digital cameras, the technique has significantly matured over the past decade.

The first major advance in time-lapse cinemicrography occurred when video tube cameras and frame grabber computer cards were finally integrated with the microscope, beginning in the 1970s. Although the images produced by a video camera are lower in resolution than traditional emulsion-based film, the uncertainty about exposure settings associated with film cameras are dramatically reduced. As an added benefit, the image sequence can be viewed during acquisition and the results of an extended time-lapse experiment are immediately available for review and editing using the videotape output. In comparison, the cost of video cameras coupled to a tape recorder was significantly less than a film cinema camera when the expense of film processing is considered. Duplication costs for videotapes versus film are also much less, and tapes containing bad sequences can be erased or overwritten and placed back into production.

An example of live-cell dynamics is presented in Figure 2 using a photoconvertible optical highlighter fluorescent protein to label the mitochondria. The specimen is an adherent rabbit kidney epithelial cell culture (RK-13 line) expressing a fusion of the red-to-green photoconvertable dEos fluorescent protein to a mitochondrial targeting peptide sequence. Upon illumination with a 488-nanometer laser, abundant green fluorescent mitochondria are observed throughout the cytoplasm (Figure 2(a)). A single mitochondrion located in a long filopodium is photoconverted using a brief pulse from a 405-nanometer laser in the first frame. The newly red fluorescent organelle is captured after a 20-minute interval as an unconverted mitochondrion approaches (Figure 2(b)). After fragmenting, the unconverted mitochondrion makes a close approach to the photoconverted neighbor (Figure 2(c)), and the two mitochondria subsequently fuse (Figure 2(d)) with the photoconverted fluorescent protein tag being distributed between the organelles. After a second fragmentation event (Figure 2(e)), a small portion of the photoconverted mitochondrion approaches a second unconverted mitochondrion, but instead of fusing, forms a toroid (Figure 2(f)). The frames illustrated in Figure 2 were selected from a time-lapse sequence containing over 1000 images gathered during a 2 hour period.

Although an increasing number of investigators are becoming involved with live-cell time-lapse sequence acquisition using fluorescent proteins, digital still images continue to be widely employed for documenting dynamic events in cell cultures over extended periods of time. In many cases, intermittent single images can be captured at higher resolution with improved signal-to-noise using illumination levels greater than are feasible with serial time-lapse imaging. Nevertheless, rapid technological developments in computer architecture and data storage capacity have propelled time-lapse cinemicrography to a new level of sophistication. Previous limitations in primary memory, fast caches, processor speed, and hard drive capacity have been overcome so that complex image sequences containing thousands of frames and composing several gigabytes can be stored locally on a host computer. Hardware advances have been paralleled by new and sophisticated software packages capable of controlling virtually all aspects of high-speed image acquisition, including stage motion, illumination intensity, exposure time (and other camera parameters), auxiliary optics (such as placing DIC prisms in the light path) and wavelength. The most advanced software can also perform post-acquisition image processing to enhance video features. In simpler systems, computers of medium performance can be employed to control a scientific grade digital camera via a number of interfaces.

Modern advances in specialized fluorescence microscopy techniques, including laser scanning confocal, spinning disk, TIRF, and multiphoton, coupled to the latest detector technologies, have enabled scientists to observe a wide spectrum of dynamic events utilizing newly developed fluorescent probes targeted at specific cellular compartments, biomolecules, and receptor proteins. By employing this advanced methodology, multiple events can simultaneously be recorded in four or more dimensions (laterally, axially, temporally, and spectrally) to provide a non-invasive window into intracellular activity over a selected period of time. Furthermore, live-cell imaging techniques have helped to lead the current revolution in cell biology, which is providing scientists with spatial and temporal information that was previously unavailable.

Origins of Focus Drift

Despite the wide variety of technological advances that have occurred in optical microscopy during the past few years, axial fluctuations manifested by slow changes to specimen focus over the course of time-lapse imaging remains a significant problem. The term focus drift is often used to describe the inability of a microscope to maintain the selected focal plane over an extended period of time. This artifact occurs independently of the natural motion in living specimens and is primarily affected by a number of contributing factors, listed below. In general, focus drift is more of a problem when using high magnification and numerical aperture oil immersion objectives (having a very shallow depth of focus) than it is for lower magnification (10x and 20x) objectives with wider focal depths.

• Thermal Drift - Temperature variations are perhaps the most common source of focus drift. Fluctuations in temperature produced as a result of air conditioners and central heating units in the laboratory, intense illumination sources on the microscope, as well as unevenly heated objectives and stages are usually the primary culprits in focus drift. In addition, differential expansion and contraction rates in materials used to construct the culture chamber and/or microscope optical train can result in a change of the distance between the objective front lens and the coverslip, leading to a loss of focus. With the highest magnification objectives, a change of just one degree Celsius is often sufficient to produce a shift of between 0.5 and 1.0 micrometers in the focal plane.
• Coverslip Flex - Thermal gradients and other variations in culture chamber temperature, as well as artifacts associated with forcing fluids through the imaging chamber during perfusion, can produce the diaphragm effect where the specimen bounces out of focus as the coverslip flexes. This artifact is often evident in chambers where the perfusion system is poorly controlled. Coverslip flexing may not present significant problems in cases where the image capture sequence can be interlaced between perfusion sessions, but it can seriously affect those studies that require relative high temporal resolution (2 seconds or less). Coverslip flex can be minimized by reducing the diameter of the perfusion input port in flow chambers, and by equipping the microscope with an objective heater to compensate for fluctuations in chamber temperature. Small, but reproducible, movement of the coverslip can also occur in culture chambers that feature a conductive, transparent coating to maintain temperature. In this case, applying current loads to the coated coverslip can generate contractions or expansions that temporarily upset focus. The correct focal plane will generally be re-established once the current load is removed.
• Imaging Chamber Heating Inhomogeneities - Live-cell imaging chambers are produced in a wide variety of configurations that often utilize entirely different technologies for maintaining the specimen at a constant temperature. For example, chambers featuring coverslips that are heated with a conductive coating (Figure 3(a)) often produce superior thermal consistency across the glass surface than do chambers that heat only the metal surrounding the periphery of the coverslip. In the latter case thermal gradients can be substantial (Figure 3(b)). Furthermore, high numerical aperture oil or water immersion objectives can act as a sink to conduct heat away from the specimen chamber and into the metallic objective, thus creating a thermal gradient in the region of the immersion medium. Other artifacts, such as current variations from the power supply and significant fluctuations in room temperature can also be responsible for irregularities in specimen chamber heating.

• Vibration - A common artifact with numerous origins, vibrations are often produced by a variety of elements associated with the instrumental configuration, in addition to sources that arise in the surrounding environment. Within the microscope and accessories, filter wheels, shutters, automatic stages, filter turrets, and other electromechanical motion of components is a common source of vibration that interferes with stable focus. Common external (non-microscope) vibration sources are air handling systems, foot traffic, refrigerator compressors, elevators, and heavy equipment motion near the imaging setup. Isolation tables, damping pads, and benchtop platforms, available from a wide variety of distributors, are useful to minimize vibration.
• Mechanical Instability - An nosepiece fully loaded with objectives can weigh between 3 and 5 pounds, which often represents a significant gravitational strain on the mechanical components involved in the microscope focus mechanism. As a result, focus drift can occur simply due to the pull of gravity on the nosepiece. This artifact can be alleviated to a degree by only using a single objective to record time-lapse videos. In addition, it is wise to ensure (if possible) that the microscope is equipped with a precision focusing system having a short mechanical length, and to maintain the proper setting of the focus tension control. Other sources of mechanical instability include loose gear sets in the focus rack or condenser pillar, as well as auxiliary components.
• Immersion Media Fluctuations - In high resolution live-cell investigations, the microscope objectives employed for time-lapse experiments may require an immersion medium consisting of oil, water, or glycerin. Over the extended periods of time necessary to gather sufficient data, viscosity fluctuations and chemical breakdown of the imaging medium can affect the properties and spread of immersion media. The viscosity and refractive index of immersion oils are often temperature dependent and water is prone to evaporation, factors that must be considered and monitored to avoid focus drift. Newly developed high performance organic oils are available in a wide range of refractive indices (including values that are identical to water and glycerin) and offer an excellent solution to using traditional immersion media. In some cases (depending upon the imaging chamber configuration), a gasket or tubing can be utilized to form a seal between the objective and coverslip to reduce the level of water evaporation.
• Adding Reagents - Many investigations require changing of culture media or the addition of chemical reagents during time-lapse imaging sequences. The physical act of introducing reagents into the culture chamber can generate mechanical shock that will result in focus drift, and adding reagents that are not at the same temperature as the imaging medium can have a similar effect. In heated perfusion systems, the introduction of new media or chemicals generally produces insignificant changes to the axial focus plane, but adding liquids to an open chamber using a pipette or syringe can disturb focus.
• Lateral Stage Movement - In some cases, room temperature fluctuations or vibration will produce a small lateral (x-y) translation of the mechanical stage, which (in some cases) leads to axial focus drift. This is generally an insignificant problem, but can become severe if an open duct is propelling cold or warm air directly on the microscope. If the stage is equipped with a clamp, it should be enabled before commencing time-lapse imaging, but otherwise there are few remedies. Advanced microscope stages using stepper motors are not generally affected by lateral movement artifacts.
• Insecure Specimen Chamber - Movement of the specimen imaging chamber during a time-lapse sequence can result in focus drift due to the new position of the coverslip. In stage adapters that secure a glass-bottom Petri dish in the stage aperture, changes in the dish position can occur as the result of clamp flex or mechanical artifacts such as disturbing the electrical wires and gas/liquid tubing connecting the chamber to controllers and other auxiliary components. Additionally, multi-well imaging chambers often suffer from slight placement variations when separate coverslips are attached to each well, leading to focus error while scanning the plate. Chamber movement and coverslip height inconsistencies are a common artifact that must be addressed on a regular basis.
• Specimen Movement - An unavoidable artifact in live-cell imaging is movement of the specimen away from the microscope focal plane. This often occurs when gathering time-lapse sequences using adherent cells that are undergoing mitosis, where the cells dramatically change shape in metaphase. In other cases, the specimen can move completely out of the viewfield, either by detaching from the coverslip or through motility. Other than using gels and reagents such as fibronectin to anchor cells and tissues to the coverslip, there are no mechanical, chemical, or electrical remedies for specimen movement.

Solutions to Focus Drift

For many years, the best solution for correcting focus drift was to station a student or technician near the microscope in order to provide manual re-focusing when necessary. Unfortunately, this is a relatively cost-prohibitive and largely ineffective solution that certainly is not suited to long-term imaging experiments. The problem of focus drift has been addressed through a variety of avenues, including software algorithms, dedicated microscope hardware, anti-vibration pads and tables, and enclosing the microscope in a protective environment (as illustrated in Figure 4). These methods have afforded varying levels of success, but none have provided a universal solution for focus drift that can be extended to virtually any live-cell imaging scenario.

Thermal instability, which is perhaps the most significant source of focus drift, can largely be eliminated with large Plexiglas enclosures (termed environmental chambers; Figure 4(c)) that insulate the microscope from its external environment and also provide excellent conditions to promote log-phase cell growth. The incubator-style enclosure surrounds the microscope stage, objectives, fluorescence filters, and transmitted light condenser, virtually encompassing the entire microscope as well as the specimen. These chambers can be used with a variety of culture vessels, including standard culture bottles, Petri dishes, microscope slides with mounted coverslips and a variety of open and closed imaging chambers. Temperature is maintained with an external heating unit (usually forced air) and the carbon dioxide concentration is controlled with a sensing unit coupled to a regulator that is fed by a cylinder of pure gas.

Environmental chambers can also be equipped with humidity control and several designs provide rubber glove access to avoid disturbing the environmental equilibrium when manipulating the cells during imaging. In order to maintain a high degree of temperature control and to avoid focus drift, several of the more sophisticated incubator chambers enclose virtually the entire microscope with the exception of the eyepieces, camera, and lamphouses. On the downside, environmental chambers can impede rapid access to the specimen and are cumbersome when repeated manipulation is necessary. In addition, the high humidity level inside the chamber can add to the expense of maintaining the instrument due to premature degradation of gear lubricants and the oxidation of metal surfaces and lens coatings. As an alternative, objective heaters coupled to stage-top incubators can be used, but these devices are not always compatible with immersion objectives having protective barrels that are difficult to remove.

Over the past couple of decades a number of software algorithms have been devised to address focus drift based on contrast or edge detection. Many of these have eventually been incorporated into scientific imaging software packages designed to also process images and control microscopes. Software control of focus drift requires the microscope to be fitted with a digital camera and a motorized nosepiece under computer control. Typically, a focus-finding software-based microscope requires two complementary algorithms to achieve focus control. The first establishes a correspondence between the axial (z) position of the objective and true focus. The second is a search algorithm (Figure 5) that samples the focal plane by capturing an image to determine the location of the maximum focus "score" (optimum position). Most of the algorithms developed for software control of focus have, however, been created using static specimens featuring stable geometry and are far less effective when applied to dynamic living cells in time-lapse investigations. Furthermore, the algorithms are usually optimized for a particular illumination scenario (brightfield, phase contrast, etc.) and can fail when applied to a contrast-enhancing technique (such as fluorescence) for which they were not designed.

Software focusing algorithms generally rely on the fact that a well-focused image contains more contrast or finer detail (in effect, much higher frequencies) than images that are out of focus. As an example, when using a Laplace operator to estimate focus, the algorithm computes the second spatial derivative of a sample image, moves the objective up or down by a pre-determined amount, and then repeats the process until the best fit is determined. Increasingly smaller steps are usually taken in both directions during the analysis. High output values from the filter indicate areas of large intensity change, which usually signify an edge within the image. In any image, the highest frequency components occur at the edges and should be the most prominent feature when the specimen is optimally focused. As the focus drifts, the edges blur to produce lower second derivatives across the image. Sampling is repeated until the best fit is determined. Often this can take several minutes, which relegates software focus routines to only those time-lapse experiments that require using long intervals between image capture.

In confocal time-lapse microscopy, where every image remains in sharp focus regardless of whether the instrument is experiencing axial drift, focus correction can be accomplished by periodically collecting a series of z-stacks (optical sections) from the specimen, irrespective of the thickness. The individual stacks are then analyzed using software to determine the Pearson's correlation coefficient between each image in the z-stack and a reference image recorded at the beginning of the sequence. After analyzing the stack, the image having the maximal correlation coefficient of pixel intensities is used to identify the correct focal plane (corresponding to the plane of the reference image). The co-localization information can then be used to reset the focal plane of the objective. Although this technique should work quite well provided the software and hardware are interfaced properly, there are currently no commercial applications.

In theory, using software algorithms on fluorescent specimens should produce excellent results due to the extremely high contrast between the brightly fluorescing structures and the dark background. In practice, however, software-based compensation techniques require calculations that are based on the capture of several images of the fluorescent species, which can lead to photobleaching and phototoxicity effects that are aggravated with each exposure of the specimen to light. Furthermore, the algorithms are virtually useless for determining focus using instruments that produce crisp in-focus optical sections regardless of the axial focal position (such as confocal, multiphoton, and DIC). Another pitfall when using software to maintain focus is that the algorithms can often determine an optimum focal plane that is different from the initial plane chosen at the beginning of the time-lapse sequence. When using fluorescence in combination with software focus-determining algorithms, an alternative is to perform the focus analysis steps using a transmitted light image of the specimen. However, this requires the application of mechanical shutters for both the fluorescence and transmitted light path, which can introduce additional vibration.

Illustrated in Figure 6 are two time-lapse sequences conducted either without any focus drift correction (Figure 6(a), 6(b), and 6(c)) or with a hardware-based focus maintenance system, as discussed below (Figure 6(d), 6(e), and 6(f)). The specimen is a culture of fox lung (FoLuline) fibroblast cells expressing a fusion of mPlum fluorescent protein and a mitochondrial targeting sequence to produce brightly fluorescing organelles. Images were gathered with a confocal microscope operating in combination fluorescence and DIC mode in 30-second intervals over a period of 7 hours. Without an autofocus system, the microscope quickly loses focus, as evidenced by the successive loss of fluorescence intensity in Figures 6(b) and 6(c), which occurred in less than 45 minutes. In contrast, hardware autofocus manages to produce sharp, well-focused images (Figure 6(d) through Figure 6(f)) for the entire investigation.

Focus Drift Compensation Systems

Over the years, a number of hardware-based solutions designed to compensate for focus drift have been developed. In most cases, these devices attempt to measure the distance between the objective and the specimen holder or stage in order to keep the specimen in focus for long periods of time. For example, an eddy current sensor attached to the microscope nosepiece was designed by one inventor to monitor the distance to the stage and correct fluctuations using a DC motor coupled to the focus mechanism. Another technique takes advantage of the distance-dependent capacitance between the specimen holder and a collar attached to the objective in order to provide an error signal for piezoelectric elements supporting the specimen holder. A third, more sophisticated methodology relies on placing a holographic grating in the objective pupil to keep the specimen in focus. The specialized grating produces two defocused first-order images that are detected by separate sensors to generate focus correction signals. Finally, a precursor to modern hardware solutions incorporated focus-dependent changes in the back-reflection of an off-axis helium-neon laser beam that was directed to a small prism placed in the optical train and monitored using a two-photodiode array. Spontaneous drifts in focus are detected as a nonzero difference signal, which is then used to drive a simple motor connected to the fine focus mechanism. Although these hardware focus-compensating solutions were successful to varying degrees, none have seen duty in commercial microscope systems.

As a general approach to improving microscope focus stability, the manufacturers have expended considerable effort to generate engineering improvements that have reduced focus drift to a minimum. For example, several of the embedded mechanical systems (focus assemblies, shutters, motorized stages, turret rotation) were improved and problems with heavy objectives were addressed. Instrument frames are now fabricated with specialized alloys that are resistant to thermal expansion, whereas auxiliary components such as the nosepiece are now made with heat-resistant polymers. Such improvements have been coupled with advanced linear encoders, which are motorized assemblies that can relocate the stage or nosepiece to a precise location with better than 50-nanometer accuracy. Even though these improvements have resulted in advanced microscopes that can maintain focus within a micrometer or two for extended time periods, they still have been unable to overcome the fact that heated specimen chambers (and their associated coverslips) are subject to thermal expansion and contraction, leading to back to the original focus drift problems.

In order to address thermal drift artifacts inherent in virtually all coverslip-based imaging chambers, the microscope and aftermarket manufacturers have recently introduced new hardware solutions to compensate for focus drift by employing several somewhat different, but still closely related approaches. An important point to note is that, with immersion objectives, most of these devices operate based on the assumption that the relationship between the specimen under observation and the aqueous surface of the coverslip is fixed (therefore, the cells or thin tissues are not freely floating in the culture medium). Instead, the cells or tissues (such as Drosophila melanogaster egg chambers) are either naturally adherent to the glass or are attached to the coverslip via thin layers of laminin, fibronectin, poly-lysine, or glycoproteins. All of the commercially available systems locate the upper external coverslip surface (the interface between glass and tissue culture medium or buffer) by measuring the shaped line pattern location of reflected light emitted by a weak near-infrared laser or light-emitting diode (LED; as illustrated in Figure 7). In contrast, for dry objectives, the reflection of light from the lower coverslip surface (in contact with air) is used to determine whether the specimen is in focus.

The illustration presented in Figure 7 diagrams how focus drift compensation systems are able to take advantage of the reflection of light from the coverslip to gauge relative focal position. A region of interest in the specimen is used by the operator to set the initial offset from the upper coverslip surface, which establishes the focal plane and serves as the basis for compensation (Figure 7(a)). When focus drift occurs (here, as the result of axial movement of the coverslip closer to the objective; Figure 7(b)), the detected intensity of laser or diode reflectance from the coverslip provides the information required to reposition the objective and resume the offset compensation. Thus, the motorized axial focus unit receives feedback from the compensation system to correct for the drift detected at the coverslip, and then restores offset to the predefined axial value, providing sharp focus for the specimen (Figures 7(c) and 7(d)). Note that the specimen itself is not involved in focus correction, rather the interference pattern generated by the reflection at the glass-water interface controls these devices.

One of the popular, low-cost aftermarket commercial focus correction systems operates by measuring the reflectance of a near-infrared LED source using total internal reflection from the coverslip. This design operates quite efficiently when using high numerical aperture objectives (1.4 or greater) and can be retrofitted along with a focus control motor to virtually any microscope frame. The primary limitation in using total internal reflection for gauging focus is the absolute requirement for very high numerical aperture oil immersion objectives in order to achieve the critical angle of incidence for the focus-measuring light beam. Dry and water immersion objectives having much smaller numerical apertures are unsuitable for this modification due to the fact that the objective numerical aperture must be significantly higher than the refractive index of the specimen and its surrounding medium (approximately 1.33 to 1.38 for living cells bathed in growth or imaging medium). Thus, while suitable for a majority of high resolution live-cell imaging investigations, this device is not applicable to scenarios where water immersion is critical or for high-throughput examinations involving multi-well plates and dry objectives. Another aftermarket product uses a high resolution sensor (with 5-nanometer accuracy) to measure the spacing between the objective front lens and the specimen in order to determine focus. In this device, the objective is mounted in a piezoelectric nanopositioner mounted between the objective threads and the nosepiece. The range of motion of the nanopositioner is approximately 100 micrometers, which is sufficient to provide a broad focal range.

Several commercial focus drift compensation devices have been introduced by the major microscope manufacturers in the past several years. All of these systems are designed to take advantage of near-infrared light reflection from the coverslip-media interface in live cell imaging chambers to determine the distance between the objective and the coverslip, but none are suitable for fixed specimens using media of high refractive index (approaching 1.5, the value for most coverslips). During operation, line-shaped light originating from a laser or LED is passed through a dedicated optical system that intersects with the primary microscope optical train, and is focused by the objective on the interface between the coverslip and the medium supporting the cells under observation. Light reflected from this interface is then routed back into the focus correction optical system and onto the surface of a dedicated, integrated charge-coupled device (CCD). Based on the intensity and position of the reflected light captured by the CCD, the software feedback system establishes the relationship between the glass-water interface and the objective focal point.

After activating the focus correction hardware, the operator can introduce an offset that moves the objective focal point to a selected region within the specimen while maintaining the same reference point at the interface, thus providing sharp focus for the specimen at a specific distance from the interface (see Figure 7). The offset value is therefore defined as the distance between the objective focal plane and the water-glass (or glass-air) interface boundary. Depending upon the manufacturer, these focus compensating systems are capable of monitoring and correcting focus drift in varying timescales, ranging from a few milliseconds to 10 or more seconds. This provides a distinct advantage over the software-based focus compensation routines. Furthermore, in most cases focus drift compensation parameters can be defined in the microscope control software to coordinate drift correction with image acquisition, as well as integrating operation with other peripheral devices, such as shutters, filter wheels, and motorized stages.

An example of long term focus drift before and after hardware compensation is presented in Figure 8(a). The specimen is an adherent culture of human carcinoma cells (HeLa line) labeled with a fusion of EGFP to human alpha-tubulin. Images were captured with a 60x oil immersion objective in 4-minute intervals for four hours under fluorescence illumination in a spinning disk confocal microscope. The cells were housed in a resistive heating chamber in which the coverslip temperature is maintained by an element embedded in the metal surrounding the coverslip. Temperature in the microscope laboratory was dropped at the rate of one degree Celsius per hour. Over a period of 240 minutes, without hardware compensation, axial displacement of the focal plane was approximately 2.8 micrometers. In contrast, with hardware focus drift correction, the focal plane was maintained to a precision of approximately 200 nanometers for the same period of time. Figure 8(b) illustrates the effects of opening the access door of an environmental chamber equilibrated to 37 degrees Celsius for a period of 1 minute. Note the sudden dip in axial displacement (approximately 3.4 micrometers) that occurs as the result of the temperature shock.

An example of focus drift compensation hardware is the Nikon Perfect Focus System (PFS), which is integrated into the axial control hardware of the Ti-E inverted microscope series (illustrated in Figure 9). A wide range of dry and oil immersion objectives, from 4x to 100x with varying numerical apertures, can be employed with Ti-E-PFS instruments for focus compensation. The working distance range of objectives available for PFS covers 120 micrometers to over 15 millimeters, affording a high level of versatility. The near-infrared 870-nanometer LED and CCD line sensor utilized by the PFS are housed in a specialized nosepiece unit (Figure 9) that does not require infinity space and enables the primary microscope optical train to remain dedicated for imaging. Among the most advanced features of the Nikon PFS is the 5 millisecond (200 Hz) sampling rate, which is independent of microscope and camera control software, and considerably faster than other systems that must repeatedly probe both the coverslip interface and the focal plane of interest. Utilization of the long-wavelength LED enables the Ti-E-PFS to be used with fluorophores emitting in the wavelength range between 340 and 750 nanometers. Additionally, the PFS can be used with a large variety of contrast-enhancing imaging modes, including brightfield, phase contrast, Hoffman modulation contrast, DIC, widefield fluorescence, confocal, TIRFM, spinning disk, and line-scanning swept-field. The most state-of-the-art version of Nikon's PFS is the Perfect Focus System 4 for the Ti2-E inverted microscope.

The focus drift correction systems described above are designed primarily to image living cells housed in specialized imaging chambers equipped with a coverslip ranging in thickness from 150 to 180 micrometers (#1 or #1.5 coverslips). Other specimens may be far more difficult to observe with focus drift correction due to weak infrared reflectivity or excessive scattered light. These include fixed specimens that are mounted in a high refractive index medium (that more closely matches that of the coverslip). In this case, the amount of reflected light from the focus monitoring system may be insufficient to detect the interface surface. Likewise, thick tissue specimens, which scatter a considerable amount of light, are difficult to use with focus drift correction. Thick glass coverslips (greater than 180 micrometers) or plastic tissue culture dishes are also not recommended, as the boundary surface may not be detectable due to insufficient offset. Finally, dust and debris on the coverslip surface can degrade the precision of boundary detection, leading to excessive hunting or errors in focus correction.

Although the currently available focus drift correction systems are designed to measure reflectance from the coverslip interface, a number of variations from one design to another provide a multitude of both choices and potential pitfalls for researchers conducting long-term time-lapse imaging. Of primary interest is the substantial variation in price among the commercial systems, which enables researchers to, in some cases, select a solution that fits more closely with their budgets than their actual live-cell imaging needs. As discussed above, several aftermarket manufacturers provide auxiliary focus correction units that can be retrofitted for use on existing motorized (or non-motorized) microscopes. In contrast, the more refined turnkey systems offered by the microscope manufacturers are fully integrated into their inverted frames and therefore do not require modification of the microscope or use of an optical port for installation. All of the focus correction systems are capable of imaging multiple features at different lateral and axial locations (in effect, for every viewfield captured at each time point).

A primary consideration when evaluating commercial focus drift correction systems is the frequency with which the feedback loop monitors and adjusts focus. The Nikon PFS system offers a continuously active mechanism, which runs independently of microscope and camera control software, while others undertake the focus detection and correction process immediately before each image is captured. There are advantages and disadvantages to each approach. For example, if the experiment involves real-time image capture, then the Nikon PFS system employing a continuous feed mechanism may be more useful because it will correct for drift independently of a software call. However, for experiments where capturing images intermittently over long time periods is the key consideration, slower systems that evaluate focus before each image capture may be sufficient. In general, the primary consideration in choosing a focus compensation system is the speed at which images are captured. The faster systems perform well for capturing images at long time intervals, whereas the slower systems will be less useful for image capture at intervals shorter than 3 to 5 seconds.

In typical time-lapse scenarios, where a significant delay occurs between image captures, the short interval required to measure reflectance and then offset to a specific axial position above the coverslip is inconsequential. However, when gathering streams of images in rapid bursts of 30 frames per second or more, the fastest compensation systems will display superior performance. However, it should be noted that, depending upon design, a continuous feedback system might not be able to complete the compensation step between every exposure and therefore might be at risk of capturing an image while the system is still hunting for the reference plane. In such cases, resulting videos often suffer from jitterartifacts (unpredictable pixel shifts along one of the lateral axes that occur intermittently every several frames) upon playback. Another variable that should be considered in selecting between continuous and capture-driven focus compensation systems is the total experiment time. Time-lapse experiments can range in lengths from several seconds to hours or even days. For experiments that last several hours or days, continuous feedback systems that rely on piezoelectric actuators might be subject to mechanical drift or positioning inconsistencies due to hysteresis or creep. In short-term experiments, the effects of these artifacts are generally negligible, but over prolonged periods of acquisition, their contributions could be significant.

In conclusion, solving the problem of focus drift is perhaps the most significant advance in live-cell imaging that has occurred in recent history. This artifact must always be addressed during configuration of an experimental system designed for time-lapse imaging and is far more pronounced when using high numerical aperture objectives where the narrow depth of focus can easily shift with the slightest vibration or change in temperature. Focus drift is also a key factor (as well as a considerable challenge) to eliminate when imaging a complex sequence of lateral and axial positions within a single specimen at each time interval. In most cases, the new focus drift compensation systems available from the microscope manufacturers are able to provide excellent correction, albeit at varying levels of speed. Regardless of the instrument level of sophistication, however, when conducting time-lapse experiments with living cells, close attention must be paid to the drift factors listed above, including the thermal environment, specimen imaging chambers, vibration, immersion media, and mechanical stability. Once all of the fundamental details have been addressed, the cautious investigator should be rewarded with excellent results.