Personal cytometers: Slow flow or no flow?

Abstract

Background:

Although some manufacturers have optimistically described instruments with prices in the US$40,000 range as “personal cytometers”, analogy with the personal computer suggests that the target price for a true “personal” cytometer should be under $5,000. Since such an apparatus could find a wide range of applications in cytomics in both developing and developed countries, it seemed desirable to consider its technical and economic feasibility.

Methods:

Using resolution targets and a variety of fluorescent bead standards immobilized on filters and/or slides, we evaluated high-intensity LEDs as fluorescence excitation sources, relatively inexpensive CCD cameras as detectors, and 35 mm camera lenses and plastic low-power microscope optics for light collection in a simple, inexpensive low-resolution imaging cytometer.

Results:

The components tested could be combined toproduce an instrument capable of detecting fewer than 10,000 molecules of cell-associated fluorescent label, and thus applicable to a broad range of cytometric tasks.

Conclusions:

Given the requirements for light sources, detectors, optics, mechanics, electronics and data analysis hardware and software, and the components presently available, it should be easier to reach the desired $5,000 price point with an image cytometer than with a flow cytometer. © 2006 International Society for Analytical Cytology

“PERSONAL” CYTOMETERS—WHAT AND WHY?

The analytical flow cytometer was developed at IBM (Armonk, NY) in the early 1960s (1, 2). The company's principal products at the time were computers; the original cytometer was itself interfaced to an IBM 1130 computer the size of a large desk. This had only a fraction of the processing power of the then-current 7094 mainframe, which filled a large room, consumed over 100 kW of power, and cost between three and four million dollars. By 1981, IBM had introduced its first desktop personal computer, which cost less than US$5,000, consumed a few hundred watts of power, and matched or exceeded the processing and storage capabilities of the 7094. IBM's Personal Computing Division, recently acquired by Lenovo Group (Beijing, China), now makes laptops that sell for less than $1,000 and have hundreds of times the processing power and thousands of times as much storage.

The IBM flow cytometers were commercialized in the early 1970s by Bio/Physics Systems (Mahopac, NY), which made benchtop instruments with low-power air-cooled helium-neon or argon ion laser light sources, diode detectors for light scattering and extinction signals, and photomultiplier tube (PMT) detectors for fluorescence, and selling prices in the $20,000 range. Although flow cytometers have evolved considerably over the subsequent decades, measuring increasing numbers of parameters with greater speed, accuracy, sensitivity, and precision, and, in at least some cases, decreasing substantially in size and power consumption, today's benchtop systems typically cost over $50,000, and no commercial flow cytometer has yet met the $5,000 price criterion according to which it could truly be called a “personal” cytometer.

The cost and, to some extent, the complexity of current flow and static cytometers have limited the diffusion of cytometric technology. In affluent countries, patients with HIV infection are monitored with the aid of counts of CD4-positive T lymphocytes performed using fluorescence flow cytometers. The resource-poor nations of Africa, Asia, Eastern Europe, and South America, in which HIV poses a far greater public health problem, would benefit greatly if simpler and less expensive apparatus with equivalent performance were available (3). Such systems could be useful for a relatively wide range of other cell analysis and fluorescence measurement tasks in cytomics, genomics, and proteomics.

For example, inexpensive “personal” cytometers should be readily adaptable for rapid detection and characterization of relatively low concentrations of microorganisms, a problem faced in common by environmental microbiologists studying microbial ecology, sanitary microbiologists analyzing food and drinking water, clinicians examining body fluids, swabs, etc., and military and security personnel searching for biowarfare agents.

It was recently suggested (4) that simple, low-magnification static imaging cytometers with selling prices in the $5,000–$10,000 range, using high-intensity light-emitting diodes (LEDs) as light sources and charge-coupled devices (CCDs) as photodetectors, could provide performance comparable to that of many benchtop cytometers costing 10 times as much. This paper will provide further evidence that this is the case, and that it is unlikely that a flow cytometer could provide similar performance at a similar price.

Cytometers: Function, Structure, and Cost

The present discussion will be restricted to optical cytometers, which make measurements of light scattered or absorbed by or fluorescence emitted from individual cells or other particles in the size range from tens of micrometers to tens of nanometers. A cytometer must incorporate one or more light sources, optical and mechanical arrangements for illuminating and collecting light from individual cells, one or more photodetectors, and the hardware and software necessary to quantify, record, analyze, and display measurement values associated with each cell.

Both the structure and the cost of a cytometer are determined by what is available in the way of components at the time the instrument is designed. It is generally accepted within the instrument industry that the component costs of a commercially competitive apparatus can be no more than one-third of the selling price, meaning that the components of a “personal” cytometer could cost no more than US$1,700. Almost all cytometers now available are designed to collect light from one cell at a time, and a brief review should make it clear that this modus operandi necessitates the incorporation of one or more components that raise costs above this target level.

Microspectrophotometers, the first cytometers (5, 6), are optical microscopes using field stops to direct light from the region of a single cell to one or more photodetectors. Smaller field stops permit measurement of small areas within a cell, providing a medium- to high-resolution image, but increasing analysis time. Comparatively slow and expensive electromechanical hardware is required for movement in the specimen plane, and also for focus control, since resolution of subcellular detail demands high numerical aperture (N.A.), which goes hand in hand with low depth of field. Using galvanometer mirror-based scanners or video cameras as photodetectors increases the rate at which cell images can be acquired, but further increases cost.

Flow cytometry (7), in which mechanical means drive cells in single file through the field of view of the measuring objective, allows whole-cell measurements to be made at a much more rapid rate than can be achieved in a microspectrophotometer. Although flow cytometer fluidics are relatively inexpensive, they introduce two classes of problems. Clogs and bubbles may interrupt flow, and minimizing coincidences requires a low duty cycle, limiting measurement time to a few microseconds/cell if thousands of cells are to be analyzed per second. Multiparameter measurements require multiple detectors, with the response of each restricted by lenses, mirrors, and optical filters.

Relatively inexpensive detectors such as photodiodes, and, in some cases, inexpensive light sources such as filament lamps and low-power red or infrared diode lasers, are usable for flow cytometric measurements of light scattering, absorption, and extinction, which are usually not photon limited. Measurements of fluorescence, particularly of low-intensity signals such as those from fluorescently labeled antibodies or nucleic acid probes, are typically photon limited, and require photomultiplier tubes (PMTs) as detectors. Even at emission wavelengths as low as 500 nm, it is desirable to use fairly expensive “red-sensitive” PMTs because of their higher quantum efficiencies. Also, since most fluorescent probes and labels require excitation below 500 nm, relatively expensive lasers or arc lamps are needed as light sources.

Modern flow cytometer optics are close to their theoretical limit of efficiency, and further improvements in performance would require increasing the quantum efficiency of detectors. Although avalanche photodiodes (APDs) offer higher quantum efficiencies than PMTs, they lack the relatively noise-free gain mechanism of PMTs, and often require temperature control to stabilize gain; these and other characteristics have made it difficult in practice to use APDs as fluorescence detectors in conventional flow cytometers.

The information collected by a flow cytometer comes in the form of pulses in the detector output current. The processing electronics must capture and digitize the peak intensity, or height, the integral, or area, the duration, or width, and/or other characteristics of the pulse shape. In some newer instruments, signals are digitized at high enough frequencies (1.25–10 MHz) to provide eight or more samples of each pulse, and all of the relevant measures are derived by digital computation; in other systems, analog and hybrid circuitry produce held signal voltages representing pulse height, area, etc., and these voltages are then digitized. It is now standard practice to use a separate high-resolution (14- to 16-bit) analog-to-digital converter (ADC) for each detector channel, to accurately represent signal intensities that may vary over a four-decade range. Advances in electronics over the three and a half decades during which commercial flow cytometers have been available have substantially increased the speed and complexity of flow cytometer electronics while reducing their size, cost, and power consumption. Nonetheless, although it has been feasible since the 1980s to use personal computers for storage, analysis, and display of flow cytometric data, the costs of the processing and control electronics in even a relatively simple flow cytometer exceed the cost of a personal computer.

All other things being equal, slowing the flow rate in a flow cytometer increases the transit time of particles through the illuminating beam(s); more photons are then collected from each particle, increasing the precision ofmeasurements. Discrimination of two groups of particles with relatively low fluorescence intensities becomes easier because the variances of the fluorescence intensity distributions of both groups are decreased, reducing overlap.

Slow flow cytometers with beam transit times in the millisecond range have been used to detect and characterize single molecules. Nguyen et al. (8) and Peck et al. (9) detected single molecules of phycoerythrin; others (10–13) describe detection and sizing of DNA fragments labeled with highly fluorescent cyanine dyes. The detectors favored for such “molecular cytometry” are APD-based modules operated in a nonlinear mode as single photon counters; they detect bursts of photons emitted as labeled fragments traverse the illuminating beam. The modules currently cost at least several times as much as a red-sensitive PMT.

As noted above, the duty cycle of a flow cytometer is kept low to minimize coincidences. An instrument that measures 5,000 particles/s with a 5 μs beam transit time has a 2.5% duty cycle. Increasing the transit time to 1 ms while maintaining the duty cycle would limit analysis to 25 particles/s. Since the typical sample volume flow rate through a DNA fragment-sizing apparatus is 33.3 pl/s, obtaining an analysis rate of 25 particles/s in practice requires that there be 7.5 × 10⁸ particles/ml in the input sample. For molecules, this concentration, which corresponds to 1.25 × 10⁻¹² M, is readily attainable. Moreover, fragment size distributions of restriction digests of microbial DNA can be accurately characterized by measuring only a few hundred fragments (14), allowing analyses to be done in less than a minute.

Although slow-flow measurements could also increase precision of measurement of weak fluorescence signals from viruses, bacteria, cellular organelles, or even whole eukaryotic cells, the range of potential applications is restricted by practical considerations. It is unusual to find microorganisms in natural samples at concentrations anywhere near 7.5 × 10⁸/ml, although most rapidly growing bacteria can reach this level in culture within a few hours. Also, whereas bacteria, with volumes on the order of 1 fl (10⁻¹⁵ l), make up a very small fraction of the volume of a solution containing 7.5 × 10⁸ organisms/ml, at the same concentration, eukaryotic cells with a volume of 200 fl each would occupy 15% of the volume of the sample, with significant effects on its rheologic properties. The typical upper limit of cell concentrations used in high-speed cell sorting, 5 × 10⁷/ml, would yield an analysis rate of only 1.7 cells/s in an instrument with a 1 ms dwell time and a 2.5% duty cycle. Slow flow analysis becomes even less practical when particles of interest comprise a relatively small fraction of the sample; at 25 particles/s, collecting data from 10,000 particles takes 6 min and 40 s, and analysis of only 10 particles present at a frequency of 1 in 10⁴ cells therefore requires over an hour.

The above considerations limit the extent to which designing the instrument with a slower flow rate can be used to lower the cost of building a flow cytometer. Keeping beam transit time at or below 100 μs would allow analysis of a few hundred particles/s. This could permit the use of a high-intensity light-emitting diode (LED), costing only a few dollars, as a light source (15), in place of a laser or an arc lamp costing hundreds or thousands, but individual detectors would still be required for each measurement channel, and relatively expensive APDs or PMTs would likely be needed to maintain reasonable fluorescence sensitivity and precision.

Scanning laser cytometers make measurements similar to those made in flow cytometers, but bring the beam to the cells instead of bringing the cells to the beam. They require a separate detector for each measurement channel, and use PMTs as fluorescence detectors. Cells may be measured on slides, as is done in the LSC®, iCyte™, and iCys™ instruments (CompuCyte Corporation, Cambridge, MA (16–18)), in a capillary of defined volume, as is done in the IMAGN™ system (BD Biosciences, San Jose, CA (19)), or on membrane filters, as is done in the ChemScan RDI apparatus (Chemunex SA, Ivry-sur-Seine, France (20)). In these instruments, stage motion is used to scan in one axis of the specimen plane, typically the longer axis of the scanned area, while a galvanometer mirror moves the laser beam along the other axis. The spatial resolution of a scanning laser cytometer is determined by laser spot size and stage step size; the instruments are capable of resolving subcellular detail to varying degrees. Commercial scanning laser cytometers typically incorporate microscope components, and most have selling prices near those of flow cytometers; although cell analysis rates are typically lower by an order of magnitude, sample throughput may be comparable to that of a flow cytometer in some applications. The “Cell Tracks” instrument described by Tibbe et al. (21) uses a relatively large (5 × 14 μm²) focal spot to make whole cell measurements, and can discriminate unstained and weakly fluorescent particles somewhat better than a conventional benchtop cytometer. It incorporates optical and mechanical components developed for compact disc readers, and could probably be produced to sell for a relatively low price.

Quantitative fluorescence cytometry can be done using a fluorescence microscope with a CCD camera attached; a recent report by Varga et al. (22) provides a good example. They used an instrument with a scanning stage and automatic focusing and captured multiple fields through a 20×, 0.5 N.A. objective with epiillumination from a mercury arc lamp. A “black”, or “dark frame” image made with the camera shutter closed was used to correct for dark current error in the camera; a “white”, or “flat field” image of a uniform thickness of a fluorescein solution (23) was used for shading correction, i.e., compensation for nonuniform illumination. This reduced the coefficient of variation (CV) of measurements of uniform fluorescent calibration beads from 24.3% to 3.9%. Antifading reagents were added to both standard and specimen slides. The calculated depth of focus for the objective is 1.9 μm; however, the CV of bead fluorescence remained below 4% for images taken within 5 μm of correct focus. Remarkably, data quality was not affected significantly by moderate lossy compression of image data. These results are impressive, but, although a fluorescence microscope with computer-controlled stage motion and focus and an arc lamp is less expensive than most cytometers, its price remains well above the target range for a “personal” cytometer.

Given the technological options available at present, it thus appears that a cytometer that measures one cell at a time must incorporate a component or components that push its selling price above US$5,000. This means that the first “personal” cytometer will almost certainly have to incorporate a static imaging system that measures cells in parallel.

Minimalist Imaging Cytometry—How Much For How Little?

A wide range of specific reagents and fluorescent labels now make it possible to detect, identify, and characterize many prokaryotic and eukaryotic cell types without collecting morphologic information. In a conventional flow cytometer, every cell is a single pixel (or voxel), and intensities of light scattering and fluorescence in different wavelength regions provide all the data necessary to accomplish the task at hand.

As was previously noted (4), because structural details of individual stars are not optically resolvable, astronomers are forced to characterize stars in a fashion similar to that now used for cells, i.e., by measuring emission intensities at different wavelengths. They typically do so by making CCD images of very large regions of space containing many stars and galaxies, changing optical filters as required between exposures, and segmenting those images to collect the needed data. The software needed to register multiple images and determine the intensities of multiple objects in a field is now widely available, much of it as shareware or freeware.

Although the term “cellular astronomy” was coined in 2004 to describe the use of astronomers' methods for cytometric purposes (4), a publication by Wittrup et al. (24) 10 years earlier described the actual implementation of a cytometer operating in this fashion. Camera lenses formed a 1:1 image of a 1 × 1 cm² field of view on a cooled 512 × 512 pixel CCD detector with 20 μm square pixels; each pixel thus collected light from an area larger than the area of a typical cell. An expanded 488 nm argon laser beam provided low-intensity (1 mW/cm²) illumination, intercepting the surface of the specimen at Brewster's angle to minimize light scattering. The instrument's only moving part was a focusing stage. Although sensitivity was impressive, with noise equivalent to only a few hundred fluorescein MESF (molecules of equivalent soluble fluorochrome)/pixel, precision was poor because software could not fully correct for the uneven illumination from the laser.

Although the laser, lenses, and CCD camera used by Wittrup et al. (24) in their 1994 work together cost tens of thousands of dollars, it is at present possible to implement similar instrument designs using far less expensive components. In an attempt to determine what the least expensive of these might be, we examined the performance of high-intensity LEDs as light sources for fluorescence excitation, SLR camera lenses and plastic microscope optics for light collection, and CCD and CMOS cameras costing between US$150 and US$1,600 for detection of fluorescence signals at intensity levels typically encountered in flow cytometry.

MATERIALS AND METHODS

Light Sources and Illumination Optics

We evaluated Luxeon III Star (3 W) (quantity one price US$8.75) and Luxeon V Star (5 W) ($20.01) devices (LumiLEDs, San Jose, CA; http://www.lumileds.com) with “royal blue” (center wavelength ∼450 nm), “blue” (∼470 nm) and “green” (∼530 nm) emission, all with Lambertian emission profiles. The 3-W device dies are comprised of four emissive strips with a total area ∼ 1 mm²; they are selected from devices intended to operate with a power input of 1 W. The 5-W devices incorporate four “1-W” dies and typically emit nearly twice as much radiant power as 3-W devices, but have four times the area. LEDs were powered by current-regulated power supplies (700 mA–1 A) operating on 110 VAC, 60 Hz line power or batteries. We estimated the power output of LEDs and illumination optics configurations using a Nova power meter with a 1 cm² photodiode detector head (Ophir Optronics, Wilmington, MA; http://www.ophiropt.com/).

The basic illuminator configuration is shown in Figure 1. Narrow-beam FHS ($4.35) and FLP ($3.20) series molded plastic lenses (Fraen S. r. l., Cusago, Italy, a division of Fraen Corp., Reading, MA; http://www.fraen.com) (25), which incorporate reflective and refractive optics, capture approximately 80% of the total power emitted by an LED. These lenses are mounted in plastic holders that fit directly over the LED circuit board, which is mounted on a heat sink (Model 500400B00000 ($1.20), Aavid Thermalloy, Concord, NH). A 445 nm, 50 nm bandpass filter (D445/50, Chroma Technology Corp., Rockingham, VT; http://www.chroma.com) was used to restrict wavelength in illuminators incorporating royal blue and blue LEDs. This filter is blocked to optical density (O.D.) 5.0 or higher between 350 and 410 nm and between 490 and 850 nm; measurement with a spectrophotometer (USB2000, Ocean Optics, Dunedin, FL; www.oceanoptics.com) confirmed that at least 99% of output power was confined to the range 410–490 nm, with the center wavelength 450 nm.

Figure 1. Components of an LED illuminator. The spaces shown between optical elements are not present in the actual illuminator.

To restrict output wavelengths from green LEDs, we used 530 nm, 30 nm bandpass interference filters, blocked to O.D. 5.5 or higher outside the passband. The optical filters were placed in front of the Fraen lenses. A 25-mm circle cut from a sheet of 5° LSD® holographic diffuser material (Physical Optics, Torrance, CA; http://www.poc.com/) was placed in front of the filter to provide more homogeneous illumination (a 10″ × 10″ sheet from which approximately 100 such diffusers can be made costs $150), and a singlet lens with focal length between 12.7 mm and 50 mm was placed in front of the diffuser as a condenser to focus the beam to an appropriate spot size in the specimen plane of the instrument. A 17-mm molded plastic aspheric singlet came from Fraen; the glass singlet lenses, costing no more than $30, were from Edmund Industrial Optics, Barrington, NJ.

The lenses, holographic diffuser, and filter(s) could conveniently be held together using heat shrink tubing with an initial diameter of 1 1/2″. Figure 2 shows an illuminator assembled in this fashion mounted to provide oblique illumination in a fluorescence measurement apparatus built following the general plan used by Wittrup et al. (24). Figure 3 shows this illumination configuration (middle panel) and two alternatives; the top panel shows the illuminator set up to provide “transmitted light” fluorescence excitation, and the bottom panel illustrates an epiillumination scheme which differs from that classically described by Ploem (26) (and now widely used in fluorescence microscopy) in that the dichroic is placed between the fluorescence collection lens and the specimen, rather than between the fluorescence collection lens and the image forming optics. The advantages and disadvantages of these configurations are discussed in more detail below.

Figure 2. A low-power fluorescence imaging cytometer with an oblique illuminator. The base plate measures 8″ × 8″.

Figure 3. Alternative configurations for LED illumination. The top panel shows the illuminator set up to provide “transmitted light” fluorescence excitation, the middle panel shows oblique illumination, the bottom panel illustrates epiillumination.

Fluorescence Collection Optics

Our “standard” collection optics setup incorporated two manual focus bayonet mount 50 mm f/2 SLR camera lenses (Pentax Corporation, Tokyo, Japan) ($49.95), used at full aperture with focus set at infinity. The lenses were coupled front-to-front with a macro reversing ring, providing a nominal 1:1 magnification. With royal blue excitation, we initially used a 49-mm Wratten 15 (530 nm long pass plastic) filter and a 49-mm reversing ring, and later mounted 2″ or 50 mm diameter interference filters between the lenses, using 49–55 mm thread adapters and a 55-mm reversing ring, which could accommodate and retain the filter(s). The ring adapters cost <$10 each. With green excitation, we used only interference filters. In order to work at higher and lower magnifications than 1:1, we used 50 mm f/1.4 and 100 mm f/2.8 camera lenses from Olympus Corporation (Tokyo, Japan) (no longer available), coupled front-to-front, providing a 2:1 (2×) image when the 100-mm lens was mounted on the camera and a 1:2 image when the 50-mm lens was mounted on the camera.

CCD Cameras

At the “high end”, we investigated two cooled CCD cameras. The ST-7XMEI camera (Santa Barbara Instrument Group (SBIG), Santa Barbara, CA; http://www.sbig.com/) ($1,495) incorporates a Kodak KAF-0402ME CCD, a microlensed full frame device with 768 × 512 pixels, each 9 μm square, and a peak quantum efficiency near 80%. The chip's capacity of 100,000 electrons/well and the camera's 16-bit analog-to-digital converter (ADC) provide a measurement dynamic range of just under 4 decades. The ST-7XMEI incorporates a Peltier cooler that can maintain chip temperatures as much as 30°C below ambient, and interfaces to a computer through a USB port. Unger et al. (27), using an earlier version of the camera with a less efficient CCD chip on a fluorescence microscope with oil immersion optics and mercury arc lamp illumination, demonstrated that fluorescence from single molecules of dyes and fluorescent proteins could be detected in a 100 ms exposure.

The SXV-M7 camera (Starlight Xpress Ltd, Holyport, Berks, England; http://www.starlight-xpress.co.uk) ($1,595) incorporates a Sony (Tokyo, Japan) ICX429AL microlensed interline CCD with 752 × 580 pixels, each 8.2 × 8.4 μm², with a capacity of ∼70,000 electrons/well. The quantum efficiency of this chip between 450 and 800 nm is comparable to that of the Kodak chip; the readout noise of the Sony CCD is claimed to be lower. Peltier cooling maintains CCD temperature at 30°C below ambient; 16-bit data reach the computer through a USB 2.0 port. The SXV-M7 is more compact than the SBIG camera, and, because it uses an interline rather than a full-frame CCD, does not require a mechanical shutter. Both cameras were connected to our reversed lens optical setups using C-mount to SLR lens adapters.

We also examined two inexpensive cameras; the Philips ToUcam II Pro 840, a WebCam popular with amateur astronomers (∼US$150), and the Unibrain (Athens, Greece; http://www.unibrain.com/) Fire-i board camera (also ∼US$150). Each of these cameras incorporates a single Sony ICX098 CCD chip, which has a capacity of ∼10,000 electrons/well and produces images from 640 × 480 5.6-μm square pixels. The ToUCam (ICX098BQ CCD, with a Bayer filter mask) transmits a 24-bit color image to a computer through a USB port; the Fire-i board camera (ICX098BL CCD, without a Bayer filter mask) transmits an 8-bit monochrome image via an IEEE 1394 (FireWire) connection. The quantum efficiency of the ICX098 chip between 500 and 800 nm is on the order of one-third to one-half that of the ICX429 used in the SXV-M7. The ToUcam and Fire-i cameras were connected to our optical setups using WebCam lens adapters (US$44) made by Steven Mogg (http://webcaddy.com.au/astro/adapter. htm).

Measurement Standards

We determined the field of view of optical setups with an England finder (Structure Probe, Inc./SPI Supplies, West Chester, PA; http://www.2spi.com/spihome.html) (Fig. 4A), and tested resolution with a USAF resolution target (Edmund), using the monochrome Fire-i board camera (Fig. 4B). Illumination for these tasks came from a pocket slide viewer (KLV-M35 Handyview, Hakuba USA, Los Alamitos, CA; http://www.hakubausa.com) placed under the stage; an Edmund positioner provided focus adjustment.

Figure 4. A: Image of an England Finder made using the Fire-i camera. Each large square on the finder is 1 × 1 mm² in area. B: Image of a negative resolution target made using the Fire-i camera. The inset is a magnified view showing the region of the target in which each pair of alternating light and dark lines is approximately 16 μm wide.

Images of a variety of fluorescent particles were made using oblique and substage illumination. These included 2 μm Fluoresbrite® yellow-green fluorescent hard dyed beads (Polysciences, Warrington, PA; http://www.polysciences.com/), 3.0–3.4 μm SPHERO™ Rainbow hard dyed beads with single (RCP-30-5) and multiple (RFP-30-5) intensity peaks (Spherotech, Libertyville, IL; http://www. spherotech.com/), and 6 μm QuantiBRITE™ beads bearing mean numbers of 863, 8612, 31779, and 66408 molecules of phycoerythrin (BD Biosciences, San Jose, CA; http://www.bdbiosciences.com/).

Intel and Digital Blue Computer Microscopes

In 1998, the microprocessor maker Intel (Santa Clara, CA) and the toy maker Mattel (El Segundo, CA) joined forces to develop the Intel Play line of high-technology toys, the flagship product of which was the QX3™ Computer Microscope (28). This device reached the market in 1999 with a selling price of <$100. It included a microscope body with an incident illuminator, a plastic focusing stage with a substage illuminator, and three plastic lenses that, in combination with a 320 × 240 pixel CMOS camera chip, could produce 10×, 60×, and 200× magnified images on a 15 (computer monitor with 800 × 600 pixel display resolution. The sole connection between the QX3 and a computer is via a USB port, which provides power to the incandescent lamps used in both illuminators. In 2002, Prime Entertainment (Marietta, GA) acquired the assets and licensed the technology of the Intel Play line; they now offer the Digital Blue QX5 ($79.95), which uses LEDs for illumination and replaces the camera chip used in the QX3 with a 640 × 480 pixel CMOS chip. We tested resolution of both the “60×” and “200×” lenses in a QX3 and a QX5 using the built-in camera in each case, and made images of the fluorescence of Polysciences and Spherotech particles using blue substage illumination.

RESULTS

LED Illumination

Because the Luxeon LEDs emit over a large solid angle, accurate measurement of their total power output would require an integrating sphere, which we did not have available. We estimated each device's output power by mounting it on a heat sink without auxiliary lenses or other optics, applying 700 mA input current, and placing the 1 cm² detector head of the power meter against the front of the LED package. The power meter wavelength was set at 460 nm for royal blue LEDs, at 470 nm for blue LEDs, and at 530 nm for green LEDs; measurements are approximate, because the devices emit over a spectral range of many tens of nanometers. The 1 cm² area of the detector head makes it convenient to express illumination intensity in mW/cm², but this is also an approximation, because, even with a relatively short focal length condenser lens, a considerable amount of light falls outside the detector region, and illumination is not completely uniform over this region.

The raw power output collectible by the detector from a typical 5-W blue LED was only about one-third that of a royal blue 5-W LED (∼70 mW vs. >200 mW). When the devices were built into illuminators, we saw that fluorescence from dilute solutions of both fluorescein- and phycoerythrin-labeled antibodies was brighter using the royal blue LEDs, and used these in preference to the blue ones. With a 50 mm condenser lens, we could obtain >50 mW/cm² for oblique illumination from a 5-W royal blue LED; levels of 35 mW/cm² were attained with a 37 mm condenser lens and a 3-W LED. A transmitted light substage illuminator using a 12.7 mm lens and a 5-W LED delivered over 100 mW/cm² to a slide mounted on the specimen stage. We measured over 100 mW raw emission from a 5-W green LED, with the power meter set at 532 nm; we were able to deliver >20 mW/cm² to specimens with the green LED in an oblique illuminator, and >45 mw/cm² using the green LED as a substage illuminator. Power levels from both royal blue and green LEDs in all configurations were dramatically higher than the 1 mW/cm² described by Wittrup et al. Although light intensity falls off across the field when oblique illumination is used, it is relatively uniform when substage illumination is used; the holographic diffuser substantially increases uniformity of both types of illumination.

Lens Performance

Estimates of the resolution of lens systems were strongly dependent on the characteristics of the digital camera used. We initially examined resolution of camera lens setups using the ToUCam II Pro camera. This, like other single-chip color CCD cameras, is not usable for photometry because a Bayer mask of color filters is placed in front of the pixels, restricting response of alternate pixels to different spectral regions. The ToUCam II Pro does have an advantage over the SBIG and Starlight Xpress cameras in that it produces 30 frames/s video output, making it useful for visualization and for setting up optics. Although it was far easier to focus on the England finder and resolution target using the ToUCam camera than it was using either the SBIG or Starlight Xpress cameras, the color output led us to underestimate resolution. The Fire-i board camera, which we acquired later in the project, also produces video output, and can be used for photometry, although its dynamic range is somewhat restricted; examining monochrome images from this camera saved in an uncompressed Tagged Image File Format (TIFF) allowed us to refine our estimates of resolution.

Our standard lens setup using two 50-mm lenses is intended to produce a 1:1 image of a specimen; we found the field of view captured by the Fire-i camera (Fig. 4A) to be 3.45 × 2.59 mm², meaning that each 5.6 × 5.6 μm² pixel of the Fire-i camera CCD represents a 5.4 × 5.4 μm² area of the specimen. In the section of the resolution target in which there are 64 line pairs/mm, line pairs in the camera image were resolvable and were, as would be predicted from the above, three pixels wide (Fig. 4B).

The lens setup with a 100-mm lens attached to the camera and a reversed 50-mm lens nearest the specimen should produce a 2× magnified image; we measured the field of view as 1.80 × 1.36 mm², with each pixel in the camera representing a 2.8 × 2.8 μm² area of the specimen. In the target area containing 80.6 line pairs/mm, line pair widths in the image were, as predicted, four pixels.

Based on measurements made with the 640 × 480 pixel chip in the QX5, the plastic “60×” lens used in the Intel/Digital Blue QX3 and QX5 cameras provides a 3 × 2.25 mm² field of view, with resolution of at least 57 line pairs/mm. The field of view of the “200×” lens is 1 × 0.75 mm²; resolution is at least 114 line pairs/mm.

Camera Performance

In evaluating camera performance, our primary objective was to determine whether an inexpensive camera such as the Unibrain Fire-i was sensitive enough to detect fluorescence signals from approximately 10,000 molecules of label bound to a cell, which would allow an instrument incorporating the camera to be used for common cytometric assays such as the CD4+ T lymphocyte count.

Having established that 2 μm Polysciences yellow-green beads and single peak Spherotech Rainbow beads provided strong signals using the standard royal blue oblique illuminator and a 530 nm long pass color filter, we examined multipeak (five intensities of dye loading plus a blank) Rainbow beads and were able to detect and discriminate all five intensities of dyed beads using exposure times no longer than 33 ms. It was also possible to detect all five intensities of dye-loaded Rainbow beads using the green oblique illuminator with a 570 nm long pass interference filter (Chroma) substituted for the color glass filter used with the royal blue illuminator. Using a substage green illuminator with the 570 nm long pass filter, we could discriminate signals from BD Biosciences QuantiBRITE beads nominally bearing 8,612, 31,779, and 66,408 molecules of phycoerythrin, with exposure times no longer than 133 ms. Fluorescence from beads bearing 863 molecules of phycoerythrin was not detectable above background; we were also unable to visualize this level of fluorescence in a fluorescence microscope with either arc lamp or LED illumination.

The Intel/Digital Blue QX3 and QX5 microscopes, fitted with 530 nm dichroic and long pass color filters and a royal blue substage illuminator, could readily detect 2 μm Polysciences yellow-green beads and single peak Spherotech Rainbow beads; only the highest two or three intensities of multipeak beads were detectable above noise.

It was difficult to focus the apparatus when either the SBIG or the Starlight Xpress camera was mounted in place of the Unibrain Fire-i, because neither of the former cameras provides as high a frame rate in preview or focus mode as does the Fire-I; our success with the latter led us to deemphasize experiments with the two more expensive cameras.

We found that, even at what appeared by eye to be best focus, the image of a 2 μm yellow–green bead would encompass substantially more pixels than our resolution target results initially led us to expect. Figure 5 illustrates one such image; portions of the signal from a single bead are detectable above threshold in 30 contiguous pixels of the Fire-i camera. The implications of this finding will be discussed below.

Figure 5. Screen shot using the Matlab Image Processing Toolbox, showing the area of the image of a single 2-μm yellow–green bead in a frame captured by the Fire-i camera.

Using Matlab 7 software (The MathWorks, Natick, MA), and correcting Fire-i camera images of a slide of 2 μm Polysciences beads using an LED-illuminated uranium glass slide (Newport Industrial Glass, Stanton, CA) to provide a “flat field,” we found the coefficient of variation of the integrated intensity measurements of 221 2-μm beads to be less than 10%.

DISCUSSION

Our results strengthen our belief that it should be possible to develop inexpensive “personal” imaging cytometers using LEDs as light sources and CCDs as detectors. Both classes of devices continue to improve in performance while decreasing in price, and their characteristics and those of the imaging cytometers make it possible to adapt strategies for sample preparation and labeling different from those now commonly used for flow cytometry.

LEDs as Light Sources

LumiLEDs now offers 1-W and 3-W Luxeon LEDs with “amber” (∼590 nm), “red–orange” (∼617 nm), and “red” (∼627 nm) emission, and 1-W, 3-W, and 5-W “cyan” (∼505 nm) LEDs, in addition to the 1-W, 3-W, and 5-W royal blue, blue, green devices described here. All of the devices in a given power class have the same price. Other companies have also produced high-intensity LEDs; for example, Cree (Durham, NC) has chips that emit over 250 mW at ∼405 nm, and Nichia (Anan, Tokushima, Japan) is producing a 365-nm LED that emits over 150 mW. The latter is considerably more expensive (small quantity price $100) than the Luxeon LEDs, but still substantially cheaper than any other source at this wavelength usable for cytometry. Adding additional illumination wavelengths to an inexpensive imaging cytometer need not raise its price out of the “personal” range.

CCDs and CMOS Image Sensors

Both CCD and CMOS image sensors are now being incorporated into consumer products such as cellular telephones, and economies of scale are dropping the prices of these components. The newest CMOS sensors are much less noisy than their predecessors, and may become competitive with CCDs for some instrumentation applications. We have recently examined a camera incorporating a Micron (Boise, ID) MT9M001 CMOS sensor, a 1.3 megapixel device with 1280 × 1024 5.2-μm square pixels, 40,000 electron well capacity, and 10-bit output; this device could easily detect BD QuantiBRITE beads, and industrial grade boards incorporating the chip should shortly be available in quantity for less than $300.

Major drawbacks of the Unibrain Fire-i camera are its low well capacity (10,000 electrons) and restriction to 8-bit monochrome output, which make it difficult to use the camera for measurement of signals with a wide dynamic range. Most of the 16-bit cameras now available are either much more expensive, or restricted to much lower frame rates, or both, but SAC Imaging (Melbourne, FL; http://www.sac-imaging.com) has recently announced the $750 SAC-10 camera, a cooled unit providing 16-bit monochrome output from the same Sony ICX429AL CCD used in the Starlight Xpress SXV-M7, and Meade Instruments (Irvine, CA; http://www.meade.com) now sells the Deep Sky Imager Pro, a 16-bit monochrome camera, for $399. The current availability and price of digital SLR cameras with 6.1 megapixel CCD and CMOS sensors suggest that it may become feasible to use even chips such as these in relatively inexpensive imaging cytometers.

Cytometer Optics

As was mentioned above, we noted that the image of a 2-μm bead on our cameras was substantially larger than 2 μm; the bounding rectangle of the bead image shown in Figure 5 is approximately 30 × 40 μm². Wittrup et al. (24) reported a similar result; at best focus, images of 6-μm beads occupied an average of 6 20 × 20-μm² pixels. Thus, the effective resolution of the camera chip in both cases is/was higher than the effective resolution of the optics. Using larger-area camera chips and higher sample dilutions could minimize some problems associated with poor optical resolution, but it would be preferable to increase resolution to reduce the areas of particle images and thus improve discrimination of multiplets.

Our camera lens setups, using f/2 lenses for primary light collection, have an N.A. of 0.25; it might be desirable to work at a somewhat lower N.A. to eliminate any need for focus adjustment in a production instrument. Although low-power (2–4×, N.A. 0.1) microscope lenses appear to offer better optical resolution than the camera lenses, and are available at similar prices, their lower light gathering power necessitates the use of longer exposure times. Our industrial colleagues are now investigating alternative lens designs.

Wavelength selection for fluorescence measurement could readily be accomplished by substituting a motorized filter wheel for the single fixed filter used in our experiments; it should also be possible to build two or three cameras into an apparatus, using dichroics to separate signals in different spectral regions. This would increase the cost of the instrument, but reduce the number of moving parts.

We were pleased to find that modern interference filters, which can be blocked for very low transmission (O.D. >6) outside the desired passband(s), make it possible to do low-intensity fluorescence measurements with transmitted-light illumination, shown in the top panel of Figure 3. This delivers more power to the specimen and provides more even illumination than does oblique illumination, shown in Figure 2 and the middle panel of Figure 3, but, in the transmitted light configuration, the filters do not eliminate background due to fluorescence in the condenser and/or collection lenses. External epiillumination, as shown in the bottom panel of Figure 3, deals with these background sources and facilitates examination of specimens on opaque substrates or in multiwell plates. Although effective, interference filters, which may cost hundreds of dollars, are likely to account for a substantial fraction of the production cost of a simple imaging cytometer.

Two relatively new technologies may prove useful in this regard. Polymers have been used by 3M Corporation (St. Paul, MN; http://www.3m.com/lightmanagement/) to make interference filters with passbands much less dependent on the polarization and angle of incidence of light than are those of conventional coated filters (29, 30). NanoOpto (Somerset, NJ; http://www.nanoopto.com) has produced filters with similar characteristics by nanofabrication of subwavelength structures in silicon (31, 32). Either class of filters should be less expensive to manufacture in quantity, less susceptible to extreme environmental conditions, and, potentially, easier to design into optical systems than are conventional interference filters, but it is not yet clear whether the newer filter types can achieve the high out-of-passband O.D. levels needed for fluorescence cytometry.

Software and Hardware Requirements for Production Instruments

A bewildering profusion of freeware, shareware, and commercially available programs, many developed primarily for use by astronomers, makes it possible to capture low-magnification images of cells and extract integrated fluorescence intensity values that can readily be imported into and analyzed by other programs now used for flow cytometric data analysis. We are aware that a “personal” cytometer will require an integrated software package that makes minimal demands of the user. As the building blocks exist, many of them in open-source formats, we do not foresee insuperable problems in developing the needed software.

The simplest instruments we envision will require neither focus adjustment nor movement of the specimen in the specimen plane. The present optical format, with a field of view anywhere from 6 mm² to 100 mm², should be well-adapted for work with multiwell plates, especially filter bottom plates, which allow all of the particles in a predetermined volume of sample to be collected and effectively immobilized within a small area. The mechanical precision needed to move individual wells of a multiwell plate into the field of view of the cytometer is no higher than that required of the motion components of inkjet printers and scanners, and the low prices of these devices suggest that the hardware and firmware required to handle plates will not drastically increase the cost of a production instrument.

Getting an Edge for Imaging: New Labeling Strategies

We have demonstrated that an instrument with inexpensive components can detect fluorescence signals from particles bearing fewer than 10,000 molecules of phycoerythrin, an antibody label widely used in flow cytometry, and expect that at least some phycoerythrin tandem conjugates, e.g., PE-Cy5, will also prove well suited for use with the apparatus. In addition, some other labeling techniques that have been used in flow cytometry may be even better suited to imaging devices with relatively long-duration LED excitation.

Semiconductor nanocrystals, or quantum dots (33, 34), have narrower emission spectral peaks than organic dyes and fluorescent proteins, and are more photostable. The emission wavelength of a quantum dot is dependent on its size; for quantum dots of any size, excitation cross section increases at lower excitation wavelengths, allowing a single deep blue (<460 nm), violet, or UV source to provide efficient excitation for materials with emission wavelengths ranging from blue-green to infrared. Moreover, whereas blue-excited conventional labels that fluoresce at long wavelengths, e.g., PE-Cy7, typically yield much less fluorescence than those that fluoresce at shorter wavelengths, the longer wavelength fluorescence from larger quantum dots is typically more intense than fluorescence at shorter wavelengths from smaller crystals, because excitation cross section increases with emission wavelength (35). Moreover, relatively inexpensive CCDs are well suited for detection of far red and near-infrared fluorescence from quantum dots.

Six sizes of quantum dots were used by Perfetto et al. (36), in combination with organic dyes, biliproteins, and tandem conjugates, in 17-color flow cytometry, with a violet laser providing excitation for the quantum dots. Because the lifetimes of quantum dots are on the order of 10 times those of organic fluorophores, fewer photons are collected from quantum dots during the few microseconds of observation common in flow cytometry; however, in an imaging system with a longer measurement time, quantum dots continue to undergo excitation and emission cycles, whereas fluorescent dyes and proteins bleach at a fairly rapid rate.

Although they have broader emission spectra than quantum dots, 60–70 nm silica particles doped with metal-chelate dyes have been shown to produce extremely bright surface labeling of biological materials (37, 38). Multicolor measurements should be possible using dyes incorporating different metals; several excite well at the 450 nm LED wavelength. The long fluorescence lifetimes of these dyes, hundreds of nanoseconds in some cases, make them impractical for flow cytometry, but eminently practical for imaging, and, like the UV-excited lanthanide chelates (39), they should be usable for time-resolved fluorescence measurements, for which pulsed LEDs can provide effective excitation (40–42).

Recent Commercial Developments

Several companies now offer or have announced products incorporating image cytometers using LEDs for fluorescence excitation and camera chips for detection.

Since 2002, Chemometec (Allerød, Denmark; http://www.chemometec.com) has produced the NucleoCounter, which counts cells and determines “viability” (membrane integrity), based on propidium iodide fluorescence excited by green LEDs. Specialized models are available for analysis of sperm, yeast cells, and somatic cells in milk; the instrument price in the U. S. is approximately $8,000.

More recently, Trophos (Marseilles, France; http://www. trophos.com) has developed the Flash Cytometer, designed for characterization of cells in 96-well plates (43); this offers multiple-wavelength illumination and detection, using UV, blue, and green incident illuminators incorporating Luxeon LEDs. The limited amount of detail available from the manufacturer's description suggests that this system would cost substantially more than $5,000.

LabNow (Austin, TX; http://www.labnow.com) is poised to introduce an instrument for CD4 T lymphocyte counting incorporating an LED-illuminated low-magnification imaging cytometer; the target price for this apparatus is reported to be under $5,000 (44).

CONCLUSIONS

It has recently been shown (45) that high-intensity LEDs can provide sufficient excitation to permit detection of single dye molecules using a cooled CCD camera. Our results and others' indicate that simple, widely applicable LED and CCD-based imaging cytometers, with the more modest purpose of detecting thousands of molecules of cell- or particle-associated dyes and labels, are within reach. The task of getting the most accessible and least expensive suitable instruments to those areas of the world in which they are most needed may require a combination of large companies' good corporate citizenship and smaller organizations' and individuals' audacity, but the benefits of making cytometry and cytomics technology affordable and available worldwide should far outweigh the costs.

Acknowledgements

We thank Marco Angelini (Fraen S. r. l.) for shedding new light on cells with LED illuminators, Michael Stanley (Chroma Technology Corp.) for clarifying filter specifications to make us less optically dense, and Rob Webb for his continuing guidance in optical and experimental design.