Calibration and standardization of the emission light path of confocal microscopes

Authors


Dr Stephen J. Lockett. Tel: +301 846 5515; fax: + 301 846 6552; e-mail: slockett@ncifcrf.gov

Summary

Recently, there has been a large expansion in the usage of optical microscopes for obtaining quantitative information from biological samples in order to determine fundamental biological information such as molecular kinetics and interaction, and heterogeneity within cell populations. Consequently, we built a highly stable, uniform, isotropically emitting and convenient-to-use light source, and designed image analysis procedures for calibrating the emission light path of optical microscopes. We used the source and procedures to analyse the quantitative imaging properties of a widely used model of laser scanning confocal microscope. Results showed that the overall performance was as high as could be expected given the inherent limitations of the optical components and photomultiplier tubes. We observed that the photon detection efficiency did not vary with photomultiplier tube gain and that the highest dynamic range was achieved with relatively low gain and 12-bit digitization. Practical applications of the light source for checking the transmission of optical components in the emission light path are presented.

Introduction

Optical microscopes are widely used tools for understanding the underlying molecular mechanisms in live cells and tissues because of their capabilities for detecting specific, fluorescence-labelled molecules. Laser scanning microscopes, in particular, are used for such applications because of their ability to directly acquire three-dimensional images at high spatial resolution and their versatility for a wide range of biological studies including quantitative analysis of molecular interactions. Examples of such applications include fluorescence recovery after photobleaching for measuring the translational movement of proteins as well as interaction kinetics (Phair & Misteli, 2000; Sprague et al., 2004) and image correlation spectroscopy for measuring fast protein dynamics (Digman et al., 2005). Furthermore, most confocal microscopes can be readily extended to include two-photon excitation for imaging deep into tissue (Denk et al., 1990) and for fluorescence life-time imaging to measure binding and conformational changes of fluorescence-labelled macromolecules (Wang et al., 1992). In addition, confocal microscopes are used for imaging live samples for durations of several days when studying cell migration and tissue development processes. For optimum performance of all of these types of studies, the excitation light intensity at the sample must be minimal in order to reduce sample damage and photobleaching of fluorescence labels. This, in turn, requires optimal performance of the microscope in terms of efficiency in detecting emitted photons, as well as other performance parameters. However, there are frequently considerable differences in performance between optical microscopes, between different configurations of the same microscope and from one imaging session to another for no apparent reasons. Consequently, we have developed a convenient-to-use standard light source and analysis methods for evaluating the absolute performance of optical microscopes for quantitative imaging. Although the assessment of quantitative performance of optical microscopes for fluorescence detection requires the evaluation of multiple parameters, we believe that the fraction of emitted photons recorded in the image, which we call ‘efficiency’, is the most relevant. Furthermore, being an absolute parameter it readily enables comparisons between different instruments and reports lower values for suboptimal configuration of the same instrument or for degradation due to misalignment, the presence of dirt, etc. The light source also measures several other parameters that are important for assessing quantitative imaging.

To date there have been relatively few published reports about quantitatively characterizing the quantitative performance of confocal microscopes or other optical microscopes. One of the earliest analyses of the quantitative performance of confocal microscopes was undertaken by Wells et al. (1990). They estimated that only two fluorescence photons are detected of 1000 emitted from the sample, based on estimates of the transmission efficiencies of the optical components and with the detector aperture maximally open. However, their approach is not practical for checking microscope performance on a routine basis and is not even possible for most models of confocal microscopes. Subsequently, Murray (1998) devised a more convenient method of comparing the performance of different optical microscopes. The essence of this method was measurement of the signal-to-noise ratio of the image as a function of the photobleaching rate of a fluorescent test sample. This had the advantage of accounting for differences in the excitation intensity of different instruments, which would be manifested by changes in the bleaching rate. However, this method is somewhat time consuming as it requires the user to prepare the sample and to record multiple images using different excitation intensities for each instrument and then to analyse the images. Since then, Zucker & Price (1999, 2001a,b) have developed other methods that included laser stability, objective lens transmission, uniformity of field illumination, axial resolution and sensitivity. Sensitivity was defined as the variation of pixel intensity from within standard fluorescence beads relative to the mean intensity. However, this definition does not evaluate the performance of the emission light path independently of the excitation light path. Consequently, it is only convenient for comparing the same instrument when components in the excitation light path and the excitation light intensity are kept constant.

Many other studies have assessed the performance of specific microscope components (Tsay et al., 1990), assessed specific parameters relevant to quantitative accuracy (Wilson & Tan, 1996; Model & Burkhardt, 2001; Lerner & Zucker, 2004; Zucker, 2004), assessed performance relative to a standard for the purposes of calibration (Biju et al., 2002; Dundr et al., 2002; Zwier et al., 2004; Wu & Pollard, 2005) or presented methods to correct for certain distortions introduced during microscope acquisition (de Monvel et al., 2003; Kervrann et al., 2004; Rooms et al., 2005).

In summary, the methods developed to date for the evaluation of confocal microscopes are generally rather time consuming to implement and it is difficult to compare different instruments because they do not yield measurements in absolute terms.

In this work we first provide a mathematical model of the emission light path of a confocal microscope and detection electronics. Later we use the model to determine the efficiency of the emission light path and other performance parameters. We then describe the design and characteristics of the light source and its operation. In addition to measuring efficiency, we show how the source is used to measure several other instrument parameters essential for quantitative imaging, i.e. dynamic range, linearity, uniformity over the detection area, amplification noise and background noise. We use the source to characterize the performance of LSM510 confocal microscopes (Carl Zeiss, Jena, Germany), and deduce and confirm several significant properties of these confocal microscopes. Finally, we demonstrate the application of the source for conveniently checking and optimizing the performance of a confocal microscope.

Mathematical model of photon detection efficiency and noise of a confocal microscope

Figure 1 shows schematically the emission light path of a confocal microscope. Photons emitted from the source are recorded as signal in the image but most photons are lost during passage through the objective lens, mirrors, interference filters, secondary lenses, etc. and at the photomultiplier tube (PMT) faceplate. Upon detection of a photon, the PMT produces a pulse of electrical charge. As we are primarily interested in the performance of confocal microscopes under low light intensity conditions and the response time of PMTs is typically less than 20 ns (Hamamatsu Corporation, Bridgewater, NJ, U.S.A.), we will assume that each pulse corresponds to at most one detected photon. However, generally in laser scanning confocal microscopes the detection electronics are operated in ‘integration mode’, meaning that the signal in each pixel is the integral of the output pulses from the PMT over the duration of time that a pixel's corresponding point in the sample is exposed to laser excitation light (dwell time). As dwell times are relatively long (> 1 µs), the number of detected photons recorded in each pixel is generally more than 1, even for dim samples.

Figure 1.

Schematic of the emission light path and detection electronics of a confocal microscope. (A) The sample emits an average of a photons per pixel dwell time. (B) Optical elements [objective lens, interference filters, mirrors, secondary lenses and photomultiplier tube (PMT) face plate] have a combined transmission efficiency of s. (C) Therefore, the PMT detects as photons from the sample, plus an additional b counts from background photons and dark noise. (D) Amplification by the PMT and an electronic pre-amplifier results in each detected photon producing an electric charge of g + o ± n, where g is proportional to the electronic gain, o is the fixed amplifier offset and n is the random amplification noise. (E) An integration circuit integrates the electronic signal over the duration of the dwell time for each pixel and the integrated signal is converted to a digital number by the analogue-to-digital converter (ADC).

We use the Poisson distribution to model the loss of photons in the emission part of the microscope

image(1)

where P(x) is the probability of detecting x photons during the dwell time, a is the average number of photons emitted from the light source during the dwell time, which is known from calibration of the light source, s is the probability that an emitted photon from the sample is detected (efficiency) and b is the average number of ‘background’ photons that are detected during the dwell time.

The magnitude of each pulse from the PMT is not the same for each detected photon but varies significantly due to variations in electron amplification inside the PMT. This variation is known as the ‘multiplication noise factor of secondary emission’ or ‘amplification noise.’ We use a Gaussian distribution to model the effect of amplification noise

image

where G1(y) is the distribution of pixel intensities, y when exactly one photon has been detected in each pixel. g is the average pixel intensity resulting from the detection of one photon, which is proportional to the gain of the PMT and subsequent electronics. σ parameterizes the amplification noise and the offset voltage of the electronics is assumed to be zero.

When exactly x photons are detected in each pixel, the distribution of pixel intensities broadens to

image(2)

where Gx(y) is the distribution of pixel intensities, y when exactly x photons have been detected in each pixel. Gx(y) is the convolution of G1(y) with itself x − 1 times and shifted to a mean of xg.

Convolving Eqs (1) and (2) gives

image(3)

where I(y) is the distribution of pixel intensities for the distribution P(x) of detected photons. Thus, I(y) is the expected pixel intensity distribution from an image of a uniform source emitting on average a photons per pixel dwell time.

Materials and methods

Design of the light source

The calibrated light source (Fig. 2) was fabricated from a 35-mm glass-bottom dish (MatTek Corp., Ashland, MA, U.S.A.); a light-emitting diode (LED) (green LED 276–304, RadioShack, Fort Worth, TX, U.S.A.) emitting at 565 ± 30 or 587 nm; an opal glass, near Lambertian diffuser (Edmund Industrial Optics, Barrington, NJ, U.S.A.); resistors; two AA alkaline batteries (1.5 V each); and a breadboard for wiring. The LED was glued into holes that were made in the lid of the 35-mm dish to hold it in place properly and was soldered to wires connecting to the breadboard. A thin layer of immersion oil (Immersol™ 518F, Carl Zeiss) was placed into the microwell between the glass cover slip and the diffuser in order to maintain a high refractive index between the cover slip and the glass diffuser. The purpose of the glass diffuser was to convert the directional light from the LED to an isotropic distribution, thus mimicking the fluorescence signal from a biological sample.

Figure 2.

Schematic of the light source using a light-emitting diode (LED). Resistors control the intensity of the light and a pair of AA batteries provide power. The diffuser ensures that the light from the LED is isotropic, thus mimicking a real fluorescent sample. Oil was used between the cover slip and the diffuser to keep the refractive index high and thus avoid internal reflection at the diffuser/cover slip interface.

Calibration of the light source

The absolute intensity from the light source was calibrated by covering the cover slip with opaque black tape except for a small square opening of 1.08 mm2 (measured with a 10 × objective lens) at the most intense part. The intensity of light was measured with a light meter (LM-2/Fieldmaster-GS, Coherent, Santa Clara, CA, U.S.A.) and was converted to photons µm−2 µs−1. Measurements were repeated on several occasions to check long-term stability. The relative intensity from the source was measured as a function of resistance in the circuit without covering the cover slip. Short-term stability on the microsecond time-scale was checked by placing the source directly in front of the PMT of a Carl Zeiss LSM410 confocal microscope and acquiring a virtual image at the maximum scan rate. The source was checked for isotropic emission by coupling it to a half-ball lens (Edmund Optics, Blackwood, NJ, U.S.A.) with oil and measuring the intensity at different angles.

Image acquisition using the light source

The source was placed on the stage of an inverted LSM510 META confocal microscope (Carl Zeiss Inc., Thornwood, NY, U.S.A.), which was configured to use a 40×/1.3 numerical aperture (NA) oil objective lens, an 80/20 neutral mirror (transmits 80% of emitted light) in the main dichroic position and no other filters in the emission light path. The light source was translated in the lateral directions in order to find the brightest area and the objective lens was approximately focused on the inside of the cover slip. Images were acquired with the amplifier offset adjusted to produce positive pixel intensity when the light was off and pixel intensities were digitized to 12 bits. Images were acquired using different pinhole sizes, dwell times, number of scanned images averaged together to form the final image, PMT gain and resistance in the light source circuit. Images were acquired with a standard PMT and with the META detector. Image acquisition was performed on two different LSM510 META confocal microscopes.

Analysis of images from the light source

Initially, images of the light were checked for uniformity of response and a lack of correlation of intensities between adjacent pixels by measuring the width of the peak of the autocorrelation function. As these were found to be the case, we could assume that each pixel's intensity was an independent measure of the same intensity from the light source, and determination of quantitative accuracy could proceed. We also checked that images of the source were not a function of the distance from the source to the objective lens (‘focal’ distance), which should be the case for a uniform and isotropically emitting source.

Determination of the quantitative accuracy involved four steps. The first step assessed whether the light intensity of the source was linear with respect to the mean of the pixel intensity values of the image. The second step calculated approximate values of the average detected photons per pixel (as + b) and gain g using the following equations and assuming the amplification noise σ = 0 (van Vliet et al., 1998)

image(4)

and

image(5)

where M is the mean of the image and V is the variance. The offset o was the pixel intensity when the source was off and was due to the amplifier offset. It was taken as the modal value of the intensity distribution of an image recorded with no illumination from the source (background image). The third step of the analysis calculated the SD of the amplification noise σ by fitting the distribution of intensities in an image to the model (Eq. 3) using the function ‘nlinfit’ in version 7 of matlab (The Mathworks, Natick, MA, U.S.A.) and using the average detected photons per pixel (as + b), gain and offset calculated in step two as fixed values. The gain divided by two was used as an initial estimate for σ. In the fourth step, the probability that an emitted photon from the source was detected (s) was calculated from the average number of photons emitted from the source (as + b), where a was determined from calibration of the source and b was determined from a background image.

Uniformity over the imaged area

As in actual confocal microscopy uniformity over the imaged area is affected by non-uniform excitation as well as non-uniform collection of the emitted fluorescence, we also assessed uniformity by imaging a solid red fluorescence reference slide (Fluor-Ref™, Microscopy/Marketing and Education, Allen, TX, U.S.A.) as described previously (Zucker & Price, 2001a). Images were acquired with a 40×/1.3 NA oil immersion objective lens. Pinholes were aligned prior to image acquisition in order to prevent any artefact in uniformity from pinhole misalignment. The fluorescence reference slide was excited using a 488-nm Ar+ laser and emission light was collected with a 560-nm long pass emission filter. The plane of focus for the acquired images was taken approximately 20–30 µm into the fluorescent slide to ensure that a plane of even fluorescence in the middle of the slide was chosen.

Preparation and imaging of green fluorescent protein (GFP)-tubulin sample

Our results from the light source indicated that image quality was independent of PMT gain and that the highest dynamic range was achieved with lower PMT gains. We illustrated these points using images of GFP-tubulin samples. MCF-7 cells with GFP-tubulin were fixed in pre-warmed (37 °C) 0.068 m Pipes, 0.025 m HEPES, 0.015 m EGTA, 0.003 m MgCl2, 10% dimethylsulphoxide, 0.5% Triton-X-100, with 3.7% formaldehyde and 0.05% glutaraldehyde fixative. Fixation proceeded at room temperature for 30 min, followed by two phosphate-buffered saline washes (Pipes, EGTA, MgCl2 and dimethylsulphoxide provided microtubule stabilization). Chromatin was visualized with 4′,6-diamidino-2-phenylindole (DAPI) (0.25 µg mL−1). All images were acquired using 12-bit digitization and averaging together 16 scanned images in order to form the final images.

Application of the light source for measuring the performance of confocal microscope configurations

We used the light source to measure the transmission of emission light through different objective lenses and dichroic mirrors in order to demonstrate its application for checking and optimizing confocal microscope configurations. Although these tests can be performed in other ways, we believe that using the source is by far the most convenient approach. Table 1 shows the combinations of components that were compared. Results are expressed in terms of relative efficiencies calculated as the mean image intensity after background subtraction divided by the mean intensity of the image acquired with the 40×/1.3 NA oil objective, 80/20 main dichroic mirror and 530–600-nm emission filter.

Table 1.  Assessment of the relative efficiency of a confocal microscope for various configurations of objective lens, main dichroic mirror and emission filter.
ObjectivelensMain dichroic mirrorRelative efficiency
  • An emission filter that transmitted from 530 to 600 nm was inserted in front of the photomultiplier tube.

  • a

    The value was corrected for the difference in magnification of the objective lenses.

40×/1.3 NA80/201
40×/1.3 NAHFT 4581.16
40×/1.3 NAHFT 458/5610.93
40×/1.3 NAHFT 405/458/5610.74
63×/1.4 NA80/200.84a

Results

Emission from the light source is equivalent to a fluorescence sample

Using the half-ball lens oil coupled to the light source, the intensity of light exiting at 45° to the plane of the cover slip was 80% of the intensity of the light exiting perpendicular to the cover slip. Thus, we conclude that source emission was approximately isotropic within the numerical aperture of microscope objective lenses. By imaging the source directly in front of a PMT and recording an ‘image’, we observed that there were no significant fluctuations in pixel intensities, indicating that the intensity of the source was stable on timescales from microseconds to seconds. By calibrating the light source with a power meter, we calculated the absolute intensity emitted from the source as 32.5 photons µm−2 µs−1, which is equivalent to a dim fluorescence sample. These results show that the source mimics a fluorescence sample imaged under low light conditions and thus the source is well suited for assessing confocal microscopes under practical conditions.

Images of the light source are uniform, invariant to focal distance and pixel intensities are uncorrelated

Images of the light source demonstrated an extremely high degree of uniformity (Fig. 3A, upper line) with an average relative deviation of pixel intensities from the mean intensity of 1.74%. This high uniformity is attributed both to the uniform response of the emission half of the confocal microscope and the uniformity of emission from the light source itself. In contrast, the uniformity measured from the fluorescence reference slide (Fig. 3A, lower line) was much poorer and had an average relative deviation of pixel intensities from the mean intensity of 7.04%. We speculate that the difference is due to chromatic aberration in the objective lens and other optical components leading to a slight lateral shift in the positions of the focal points of the excitation and emission light when imaging off axis. For quantitatively accurate imaging it is essential to correct images for non-uniformity measured with the fluorescence reference slide.

Figure 3.

Properties of images of the light source. (A) Relative intensities along lines through images of the light source (upper line) and the fluorescence reference slide (lower line) after smoothing the images with a Gaussian filter of SD equal to 32 pixels to remove local variations in intensity due to Poisson noise. (B) Mean intensity of images of the source as a function of distance from the source to objective lens and showing insignificant variation. (C) The autocorrelation function from a horizontal intensity line through an image of the source (red) and random data (blue). The similarity of the two curves shows that the intensities of neighbouring pixels from the source are independent.

The image intensity from the source did not vary significantly as a function of the distance from the source to the objective lens (Fig. 3B), as expected for a uniform and isotropic source. This has the major practical advantage that the same conditions are obtained without any precise adjustment when the source is removed from and replaced on the microscope.

The intensity in each pixel in the image of the source is independent of the intensities of neighbouring pixels based on the close similarities of the normalized auto-correlation function along a line through the image (Fig. 3C, red line) and the autocorrelation function of random intensities (Fig. 3C, blue line).

Taking these three results together we can conclude that the intensity in each pixel in the image of the source is an equivalent and independent measure of the same intensity from the light source and that the same intensities are reproduced when the source is removed from and replaced on the microscope. Hence we could proceed with modelling the intensity distribution from the source using Eq. (3).

Image intensity is proportional to input intensity, the efficiency of the confocal microscope is independent of photomultiplier tube gain and amplification noise is proportional to gain

We observed a very high degree of linearity between the light flux and the mean pixel intensity of the images at PMT voltages of 650, 800 (Fig. 4A) and 950 V. The light flux was controlled by varying the pinhole size (from 1, 3 or 5 Airy units as described in Fig. 4A) and dwell time by the number of images averaged together to form the final image. Images were acquired as described in Materials and methods using a 40×/1.3 NA oil objective lens. We obtained the same results when the dwell time was varied by changing the laser scan speed instead of the number of images averaged. We note that the intercepts with the y axis in Fig. 4(A) are very close to the origin for both lines, indicating that background signals from PMT dark noise and photons not from the light source were very low, and that using the modal value of the background images as the offset intensity (o) was valid. By subtracting this offset intensity, the mean pixel intensity of the image became proportional to source intensity. Also, the high proportionality showed that the pinhole size was calibrated accurately and the effect of averaging was exactly as expected.

Figure 4.

Quantitative properties of the emission light path of a confocal microscope. (A) Graph of relative light flux vs. mean image intensity after subtracting the offset (o) for photomultiplier tube (PMT) gain set to 800 V (solid line) and 650 V (dotted line) and overlaid with the best fit straight lines. The data point at 0 relative light flux (RLF) was recorded with the light source turned off, and data points at 1, 9 and 25 were recorded with the pinhole set to 1, 3 and 5 Airy units (AU). The data point at 36 was recorded with the pinhole set to 3 AU and four images were averaged together. (The RLF at this data point was multiplied by 4 in order to counteract the automatic division by the LSM510 software of the signal by the number of images averaged together.) Correlation coefficients at 800 and 650 V were, respectively, 0.9998 and 0.998. (Similar results were obtained at 950 V and for the META detector.) (B) RLF from the light source plotted against the average number of photons detected per pixel. The points at RLF = 5.75 and 23 were recorded with additional resistance in the light source circuit that reduced intensity 4.4-fold. (C) Total gain of the system as a function of RLF. (Gain decreased 4-fold at RLF = 23 and 36 because when four images are averaged together the LSM510 software divides images by the number of images averaged together.) Similar results were obtained for the PMT at different voltages and for the META detector. (D) Histogram of the pixel intensity distribution from the light source with the pinhole at 5 AU diameter, no averaging of images and PMT voltage set to 800 V (green) overlaid with the Poisson distribution model (red) and the full model including amplification noise (blue). The best fit to the model was restricted to the pixel intensity range 215–1500. (E) Histogram of the pixel intensity distribution under the same conditions as (D) except that the PMT voltage was 650 V and the fit was restricted to the range 215–530.

Given the proportionality of the microscope's response and the properties of the source, we had validated the assumptions of the model expressed in Eq. (3) and therefore we proceeded to calculate the total average detected photons per pixel (as + b) and the overall gain of the detection process (g) using Eqs (4) and (5) but initially making the approximation that the SD of the amplification noise (σ) was zero. We observed that the average photons per pixel increased in proportion to light flux (Fig. 4B) and the gain was approximately invariant to the intensity of light impinging on the PMT (Fig. 4C). Furthermore, we observed that the average detected photons per pixel was not a function of PMT gain. This was the case for both the standard PMT and the META detector. Although this is a known property of PMTs operating at relatively high voltages, it is widely ignored when operating confocal microscopes.

In order to validate the model used above, we compared the empirical pixel intensity distributions (green curves in Fig. 4D and E) with the Poisson distribution model determined from the average detected photons per pixel and the gain calculated above (red curves in Fig. 4D and E) for the PMT operating at 800 and 650 V, respectively. The predominant differences between this approximate model that assumed that amplification noise was zero and empirical distributions are the quantized nature of the former vs. the continuity of the latter. We attributed this difference to the amplification noise, which causes variance in the magnitude of the signal from the PMT corresponding to each detected photon. Thus, we calculated the SD of the amplification noise (σ) by fitting the empirical distributions to the full model of photon detection efficiency and noise expressed in Eq. (3). The best fits (blue curves in Fig. 4D and E) were obtained for SDs of 27 and 7 pixel intensity units for PMT gains of 800 and 650 V, respectively. Based on Eqs (4) and (5) above, the previously estimated values of average detected photons per pixel and gain under the assumption of zero amplification noise should change when amplification noise is taken into account. However, for the above values of SD, the changes are slight; the average detected photons per pixel increased by 7% and gain decreased by 7%. Interestingly, we observed that the ratio of the SD of the amplification noise to the gain was constant (≈ 0.25) for different PMT voltages. This was not anticipated as the signal-to-noise ratio of PMTs is known to be a function of dynode gain.

From knowing the average detected photons per pixel (as + b), background photons per second (b) and intensity of the light source (a) from its independent calibration, we calculated the fraction of photons emitted from the light source that were detected in the final image (s) to be 3.3 × 10−3. A very similar result was obtained from a different LSM510 a year earlier. This is significantly higher than early estimates based on the transmission of individual components (Wells et al., 1990) but lower than values published for specialist instruments using avalanche photodiodes (Garcia-Parajo et al., 2000). However, the META detector was approximately 5-fold less efficient under the conditions used in this study, which was attributed to loss of light in the diffraction grating and the narrow spectral range of 11 nm of a META anode relative to the 30-nm wavelength range of the light source.

Implications of the invariance of photon detection efficiency and amplification noise-to-gain ratio for confocal image acquisition

The observed invariance of photon detection efficiency and amplification noise-to-gain ratio with respect to PMT gain has two important implications for confocal microscope image acquisition. Firstly, the invariance of efficiency indicates that sensitivity is not improved by increasing PMT gain and secondly, the signal-to-noise ratio, which is proportional to the SD of the amplification noise (σ)-to-gain ratio for a given light intensity, is the same at different PMT gains. (This is assuming that PMT gains are in the normal range used in confocal microscopy and that σ is more than one grey level intensity, which is normally the case for 12-bit digitization.) Taking these two properties together indicates that there will be virtually no difference between the qualities of two images of the same sample taken under the same conditions except with different PMT gains, with the exception of overall brightness of the images, which can be compensated for by offline contrast stretching of the display image. This is shown by the similarity of the distributions in Fig. 4(D and E) where the only difference is in the x axis scaling. We further illustrate this result using images of GFP-tubulin taken under the same acquisition conditions except that the PMT gains were different. Figure 5(A) shows the image acquired using a PMT voltage of 950 V, whereas Fig. 5(B) shows the equivalent image acquired using a voltage of 650 V and linearly contrast stretching the display of the image offline so that the brightest point is white. Clearly there is no significant visual difference between the images.

Figure 5.

Images of GFP-tubulin acquired using different photomultiplier tube (PMT) and laser power settings. (A) Image acquired with a PMT voltage of 950 V. (B) Same scene as in A but imaged with a PMT voltage of 650 V, followed by offline contrast stretching of the display of the image so that the brightest point is white. (C) Image of a very dim part of the specimen acquired using a PMT voltage of 950 V and then offline contrast stretching the display so that the brightest point in this part is white. (D) Image of the same part of the specimen acquired using a PMT voltage of 650 V, increasing the excitation light intensity so that the brightest point in the entire image is just below saturation and then contrast stretching the display so that the brightest part is white.

The second implication of these properties of the confocal microscopes that we tested was that a higher dynamic range can be achieved by reducing PMT voltage. This is because the maximum number of resolved levels of photon counts per pixel is approximately the intensity range (4095 for 12-bit digitization) divided by the SD of the amplification noise associated with the detection of one photon (i.e. 4095/σ). Clearly, this increases with decreasing PMT gain. The benefits of this higher dynamic range are readily observed when imaging very dim areas of a sample without saturating the bright areas. This is illustrated in Fig. 5(C and D), which shows a very dim part of the GFP-tubulin sample acquired using PMT voltages of 950 and 650 V, respectively. At 950 V there is clearly less detail of the microtubules compared with 650 V.

Application of the light source for checking confocal microscope performance

We demonstrated the utility and convenience of our light source (emitting at 587 nm) for verifying the performance of a confocal microscope by using it to check the transmission of different main dichroic mirrors and objective lenses using the parameters described in the Materials and methods and Table 1 because neither can be conveniently checked with light sources that are standard parts of confocal microscopes. These mirrors cannot be checked with the epi-illumination sources (arc lamp or lasers) because, with the exception of the 80/20 mirror, they, by design, do not reflect the same wavelengths that they transmit and so ideally should not transmit any light to the detector. The transmitted light could be used but an interference filter would need to be inserted into the light path for selection of the wavelengths of interest. The results that we obtained for the transmission of dichroic mirrors relative to the 80/20 mirror and using an emission filter that transmits from 530 to 600 nm are reported in Table 1. As expected, the HFT 458 had improved transmission, whereas the HFT 458/561 and HFT 405/458/561 had poorer transmission because they reflect some of the light from the source back towards the laser. Unexpectedly, the HFT 405/458/561 had significantly poorer transmission (74%) compared with the HFP 458/561 (93%). At present we can only speculate on this difference as either being actual properties of the mirrors (double dichroic versus triple dichroic) or caused by dirt on the HFT 405/458/561.

Measurements of objective lens transmission are very difficult to perform with epi-illumination sources on confocal microscopes because the positioning of a mirror at exactly the focal distance is required in order to efficiently reflect light to the detector. In practice this is very difficult to do consistently. Although the transmitted light source can be used to measure lens transmission, the numerical aperture of most condensers is much less than oil objective lenses. Thus, the objective lens will not be tested at its full range of acceptance angles. In addition, an interference filter must be placed in the light path to select wavelengths of interest. We compared the transmission of a 40× oil Plan Neofluar objective lens and a 63× oil Plan Apochromat lens, and found that the transmission of the latter was 86% of the former after taking into account differences in magnification. Qualitatively this difference was to be expected as the Apochromat contains more optical elements in order to achieve improved chromatic and flat-field correction.

Discussion

The goal of this study was to develop methods for conveniently calibrating the quantitative performance of confocal microscopes. We focused our study on assessing the emission side of the instrument, i.e. between the sample and digital image, because this part is critical for optimal performance and previous studies have not reported easy-to-use methods. We solved this problem by inventing a stable, uniform and isotropic light source inside a standard 35-mm cover slip bottomed dish normally used for live cell imaging. Using the source, it was possible to determine the background count rate, uniformity, amplification noise and photon detection efficiency by mathematical modelling of the intensity distributions from single images of the source, as well as to determine linearity and dynamic range from a series of images. Thus, it is particularly easy to compare the performance of an instrument under different configurations (objective lens, emission filters, etc.) and settings (pinhole size, detector gain, pixel dwell time, etc.). As measurements are in absolute terms, our approach is well-suited for monitoring the day-to-day performance of instruments and comparing different instruments.

Our calculations showed that the probability of a photon emitted from the light source being counted in the image was 0.33%, This is clearly sufficient for detecting single fluorescence molecules as a fluorescein molecule typically emits 10 000 photons (Murray, 1998) and an enhanced green fluorescence protein molecule typically emits 100 000 photons before photobleaching (Chirico et al., 2001; Thompson et al., 2002). However, further studies are necessary to discover whether detecting a molecule on average 300 times is sufficient for measurements of single molecule dynamics, such as in fluorescence correlation spectroscopy (Digman et al., 2005) and fluorescence anisotropy imaging (Velez & Axelrod, 1988).

Our calibration method enabled the measurement of the amplification noise, which was parameterized by the SD (σ) of the signal variation from detecting individual photons. The ratio (σ/g) of this value to the gain of the system g was approximately 25%. Usually amplification noise is expressed as a constant of proportionality (se) in the equation for the variance of the signal from a PMT (Art, 1990): variance =se2 × N, where N is the number of detected photons. Using Eq. (5), inline image, which as expected, is close to for a PMT operating at high gain. However, it still has a pronounced effect on the intensity distributions (Fig. 4D and E), resulting in a loss of resolution of intensity peaks corresponding to different numbers of detected photons per pixel. In future studies, it would be interesting to investigate whether circumventing the effects of amplification noise by operating the PMT in photon-counting mode will resolve these peaks. However, we point out that PMT operation in integration and photon-counting modes is equivalent when the probability of detecting a photon is significantly less than 1 for each pixel.

Taking together the observations that both the photon detection efficiency and the ratio of amplification noise to gain are invariant to changes in PMT gain led to the interesting conclusion that overall image quality is invariant to PMT gain (compare Fig. 5A and B). Furthermore, these observations led to the result that the highest dynamic range is achieved by lowering the PMT gain and increasing excitation light intensity (Fig. 5C and D), while using 12-bit digitization and long pixel dwell times (either by averaging or slow scanning). Although these features of laser scanning microscopy have been known for years, we believe that they are widely ignored by most users who instead assume that the highest PMT voltage will give the most sensitivity and highest image quality. An important practical implication of these properties of confocal microscopes is in the common situation when imaging a set of samples where it is not known a priori which samples are the brightest. In order to minimize the risk of saturation, image acquisition should be performed with 12-bit digitization, amplifier offset set such that there is a slight positive signal in background regions and a relatively low PMT gain so that bright signals are barely visible without contrast enhancement of the display image. Generally, this will enable samples that are two to four times brighter to be imaged using the same settings and without saturation.

The complete assessment of confocal microscopes may require the measurement of other parameters in addition to those calculated from images of the light source. One parameter that is essential for quantitative imaging is the uniformity of response over the detection area, which depends on the performance of the excitation as well as emission light path. Consequently, we included results from the measurement of uniformity using a fluorescence sample and showed that non-uniformity from this sample is significantly worse than non-uniformity from the light source. We attribute the difference to slight chromatic aberration. Recently, methods to measure axial chromatic shift, as well as non-uniformity over the field of view and axial resolution as a function of position, have been reported (Brakenhoff et al., 2005). We also point out that measurement of spatial resolution and laser stability are important performance parameters but methods to assess them have been published previously (Zucker, 2004).

Currently we are using only one LED in the light source. However, given the availability of inexpensive LEDs that emit from blue to red as well as infrared, it would be easy to construct a light source containing multiple LEDs.

Characterizing optical microscopy systems for the absolute quantification of image data opens radical new doors for quantitative biology. The methods proposed in the present study allow for the true quantitative measurement of biological samples as an alternative to the relative measurements that are common today, making it an immensely valuable tool for gathering significant information for many biological studies. With the advances in high-throughput imaging, it has become increasingly imperative to be able to gather calibrated, quantitative data from large data sets. The study of protein dynamics with fluorescence microscopy has also become a fundamental goal for many cell biologists in order to discern the interactions of proteins and their molecular partners in vivo. Recent work on the ability to study protein–protein interaction through quantitative colocalization image analysis algorithms (Costes et al., 2004; Daelemans et al., 2004) and to study protein interactions through fluorescence energy resonance transfer (Gu et al., 2004) has proven the need for quantitative tools for fluorescence imaging. The ability to very conveniently determine the number of actual photons that are detected by the detector gives researchers a method of acquiring truly quantitative data.

Acknowledgements

The cells were a kind gift of Drs Mary Ann Jordan and Kathy Kamath (Department of Biological Science, University of California at Santa Barbara). Labeling of the GFP-tubulin sample was kindly performed by Dr M. Katherine Jung (Toxicology and Pharmacology Branch of the Developmental Therapeutics Program, National Cancer Institute at Frederick/Scientific Applications International Corporation, Frederick, MD, U.S.A.). The authors are also grateful to Dr James McNally for insightful discussions and critical review of the manuscript. This project was funded in whole or in part with federal funds from the National Cancer Institute, National Institutes of Health under contract N01-CO-12400. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services and nor does mention of trade names, commercial products or organizations imply endorsement by the U.S. Government.

Ancillary