The task of a biochip reader is to measure the intensity of light emitted from fluorescence-labeled biochemical species in the spots of a biochip that have been printed on a chemically surface-treated substrate.1, 2 A number of techniques have been developed during the last several years for this purpose. 1, 2, 3, 4, 5, 6, 7, 8, 9 These experimental approaches differ with respect to excitation mechanism (direct irradiation or via evanescent waves; single-spot, multispot, or full-area excitation) and detection technique (type of sensor; temporally, spatially, and/or spectrally resolved signals).
To obtain a high biochip throughput, full-area flood illumination of the fluorescence samples is often employed. Equipped with a CCD-based detection unit, diagnostic systems using this approach are rapid and cost-efficient.3, 4 Parallel readout of the biochip enables screening of many fluorescence samples simultaneously in a short time. Also, these systems are versatile with respect to the printing format of the biochips, i.e., they do not require a well-defined pitch of the spots of the microarray. On the other hand, a comparatively high level of background fluorescence significantly reduces their sensitivity.
Alternatively, highly sensitive fluorescence detection can be accomplished by excitation with a focused laser beam.5 The fluorescence light from the spots is then detected by a photomultiplier tube (PMT), where the background fluorescence is reduced by a confocally mounted pinhole in front of the PMT. Here, an increased SNR enables an improved detection limit of the fluorescence light. Combined with fluorescence excitation by means of evanescent waves6, 7 (sometimes referred to as the total internal reflection fluorescence or TIRF), this method is capable of detecting even single fluorescent molecules.
However, to read out a complete microarray using the laser excitation approach, a mechanical scanning system is required because only one spot can be excited and read out at a time. Due to this sequential readout, the throughput of these systems is low.
Mehta 8 and Blom 9 demonstrated parallel fluorescence detection through multispot excitation using a laser-illuminated diffractive optical element. In this way, two essential advantages were realized: highly parallel excitation and an excitation only at the position of the spots of the biochip. Highly parallel multispot excitation increases the throughput of samples; excitation only at the position of the spots reduces background fluorescence from unbound marker molecules in the analyte surrounding the spots on the microarray, which means an increased excitation efficiency and improved SNR. However, when using static diffractive optical elements for holographic excitation, matching the diffraction pattern to the spot pattern requires additional imaging optics and mechanical translation stages. Therefore, the adaptation of the excitation unit to new spot patterns requires modification in the hardware of the setup.
Due to the availability of dynamic diffractive optical elements (D-DOEs), this limitation can now be overcome, such that arbitrarily pitched spot arrays can be excited without need for any further optical components. We present here a biochip reader that provides a means for dynamic holographic excitation in combination with a CCD-based hyperspectral detection unit for simultaneous fluorescence analysis at several wavelengths. The combination of these new technologies makes it possible to realize a high-throughput biochip reader with a high sensitivity and great versatility.
The basis of holographic excitation is a diffractive optical element (DOE), which is able to generate an arbitrary spatial phase modulation of an incoming lightwave. The appropriate phase modulation can generate a well-defined, 3-D modulation of the light intensity after reflection from or transmission through10 the DOE.
The calculation of the required phase modulation for the DOE to achieve a desired arbitrary intensity profile is based on the reversibility of the light paths, i.e., light passes through an optical system backward in the same way as it does forward. Tracing back the light paths from the desired intensity distribution on the image plane (i.e., at the biochip) to the phase-shifting DOE enables the determination of the required phase shift to yield this intensity distribution.11, 12
For adjustment of the pitch of the spot intensity pattern without replacement or mechanical movement of components, reflective liquid crystal displays [for example, a Hitatchi liquid crystal on silicon (LCoS) chip with ] can be used as a D-DOE. The liquid crystal has an electric-field-dependent refractive index such that the refractive index of each cell of the LCoS display can be individually modified. The resulting spatial distribution of the refractive index causes a spatial phase modulation of the lightwave reflected from the LCoS display. Thus, for adaptation of the biochip reader to a new fluorescence spot pitch, only the calculation of the new phase modulation pattern of the D-DOE is required. The phase modulation pattern of the LCoS displays can be refreshed dynamically at a rate of , thereby enabling a linear movement or rotation or pitch adjustment of the intensity pattern at that rate.
Figure 1 shows the optical setup for parallel holographic excitation. Two lasers illuminate two reflective liquid crystal displays (LCDs). The first, a laser diode, has a maximum power of (Chromalase 405, Blue Sky Research) and the second, a diode-pumped solid-state green laser, at (GLMC1-30). The intensity of both sources can be individually adjusted by the laser current. The two laser beams are expanded by two-beam expanders with confocal pinholes to fully illuminate the LCDs, which then modulate the phase of the incident light in a 2-D pattern as already described. The two separate light paths are joined together by mirror M1 and the dichroic mirror. A further mirror (M2) directs the modulated laser light through the biochip to excite the sample spots.
Generally, the rectangular aspect ratio of the LCDs leads to an asymmetric rectangular beam profile of the reflected light. To correct this asymmetry, the LCDs are illuminated under an oblique angle of incidence, which is given by the arccosine of the aspect ratio of the LCoS display. Then, the rectangular LCoS display appears quadratic to the incident beam, and the reflected beam has a square truncated Gaussian beam profile. At the same time, reflecting the circular expanded laser beam to a square beam profile is more efficient than reflecting it to a rectangular profile. Oblique illumination of the D-DOE also simplifies the optical setup, because no additional beamsplitters are required. On the other hand, oblique illumination requires a higher effort in the calculation of the holograms. First, the hologram must be stretched in one direction because of the projection onto the inclined plane of the LCD. Otherwise the focal spots would become astigmatic. Second, the varying distance from the LCD to the focal plane (i.e., the position of the biochip) must be taken into account because the LCD plane and the focal plane are not parallel. Based on these considerations, holograms were calculated with different spot distributions and pitches.13 Figure 2a shows a calculated hologram for generation of 25 light spots with a pitch of at a wavelength of . Figure 2b (left and middle image) shows the type of aberrations, such as coma and astigmatism, which occur if the already-mentioned compensation for the oblique illumination of the D-DOE is not sufficiently considered in the hologram calculation. Finally, software-based full compensation of all effects resulting from oblique incidence leads to an optimized intensity profile of the spot [Fig. 2b, right image].
By focusing the spot intensity pattern generated by the dynamic hologram onto a liquid fluorochrome film (rhodamine B, purchased from Fluka, dissolved in distilled water) sandwiched between a glass slide and a cover slide, the quality of the excitation unit was measured directly in the experimental setup. Hyperspectral detection of the fluorescence light does not provide an image of the spots, but rather an exact determination of the intensity of each spot. Figure 3a shows an image plane cross section of 21 spots generated with a D-DOE at a wavelength . The contrast between bright and dark zones is 0.93, and the standard deviation of the intensity maximum over the 21 foci is 10%. The truncated Gaussian intensity distribution of the incident light is the reason for the smaller intensity in the last two spots on the left side. Under consideration of the light intensities of each spot, the evaluation software can determine the dye concentration from the measured fluorescence intensity of each fluorescence sample. Figure 3b shows a picture of 10 excitation spots imaged directly onto a CCD sensor in the focal plane, i.e., at the position of the biochip.
The fluorescence light emitted from a thus excited line of spots is imaged onto a CCD sensor (a 2-D object), with one coordinate on the CCD for the spatial information (i.e., which spot in the row) and the other coordinate for the spectral information from the fluorescence light of that spot.14 The essential elements of this hyperspectral detection unit, as seen in Fig. 1, are a dispersion prism (P) for spectral separation, a highly sensitive CCD-camera (SamBa SE-34; Sensovation AG; square pixels with resolution), a special filter set (see later in the paper), and an objective lens with high numerical aperture (Schneider-Kreuznach Xenon 25/0.95).
Lateral magnification, i.e., the size of the imaged spot on the CCD sensor is an important issue in hyperspectral detection. A high lateral magnification leads to a high spatial resolution of detected signals and the number of illuminated pixels for a spot increases quadratically with as can be seen fromis the number of illuminated pixels, is the diameter of the spot, is the width of the CCD pixel, and is given by is the -number of the objective lens. Dividing Eqs. 1, 3 gives the intensity collected for :4, the intensity per pixel decreases with increasing magnification . At small the right part of Eq. 4 converges toward unity, but the number of illuminated pixels also decreases rapidly.
For these reasons, the lateral magnification has to be optimized such that a reasonable number of pixels (10 to 100) is illuminated and the intensity per pixel is sufficiently high for reliable detection.
The typical spectral width of a fluorochrome emission spectrum is15 about . Therefore, to distinguish five different fluorochromes in the range of with this detection unit, a spectral resolution of is sufficient. Furthermore, a spatial resolution of about is sufficient for measurement of a microarray with a pitch of . Using these values, and with a pixel size of the used CCD sensor of , we calculated an optimal magnification .
Although they are frequently used in spectrometers, the overall performance of diffraction gratings turned out to be inferior when compared to the dispersive prism used here, due to their strongly wavelength-dependent diffraction efficiency. An antireflection-coated, BK7 prism proved to have superior performance and was thus used for spectral separation. Depending on the dispersive power of the prism, its position in the optical path, the pixel pitch of the CCD detector and the objective lens, the detection unit achieved a spectral resolution of in the red, in the green, and in the blue regions of the spectrum.
The spatial resolution of the system is , sufficient to resolve the printed DNA spots of the microarray. A printed spot with a diameter of is imaged to in the spatial axis of the CCD sensor, and to in the wavelength axis, the latter depending on the wavelength range. As we can see from Fig. 7 in Sec. 5 this pixel number is sufficient to resolve the spectra of all fluorochromes used. As a further advantage of hyperspectral detection over detection in a small spectral range (as is done using interference filters), detection of the full emission spectra of the spots allows the discrimination of the fluorescence signals of the spots and that of the background (mostly emitted from the substrate).
The fluorescence emission peak is shifted towards longer wavelengths (“Stokes shift”) with respect to the excitation wavelength;15 this shift is typically about . To detect weak fluorescence signals, the excitation wavelength must thus not reach the detection unit, such that efficient filters for blocking the excitation light are required.
In the setup of Fig. 1 a special filter combination is used to measure the fluorescence of different sets of fluorochromes. A long-pass filter (Schott Glas; GG435) blocks the shortest excitation laser wavelength , and a notch-filter (Semrock StopLine™ 532.0) with an optical density higher than at blocks the second excitation laser line at . With this combination of filters, most fluorochromes in the visual spectrum can be detected. The blue and green fluorochromes can be stimulated with the near-UV light of the blue laser. To stimulate green and red fluorochromes, the laser at can be used. Figure 4 illustrates the transmission spectra of the filters employed in relation to the two laser lines as well as the emission and excitation spectra of the three fluorochromes, Dy415, Dy550, and Dy631.
To measure complete microarrays, the entire system was computer controlled. Several parameters, such as the spot grid, the integration time of the CCD, or the spectral and spatial range, can be dynamically altered. Thus, the use of different biochips with differing array pitches does not require hardware changes since the holographic excitation pattern is defined purely by software.
In a typical measurement, a manual visual alignment of the microarray via the CCD image is necessary before a readout sequence can be started. During the microarray scan, a hardware-based handshake between the motorized stepper, the CCD camera, and the lasers controls the readout. After moving the microarray to the appropriate line of spots to be measured, the motion controller sends a signal to the camera to start the data acquisition. The two lasers are switched on by the camera only during data acquisition such that the photobleaching of the organic fluorochromes is kept to a minimum. After the measured data from the camera are received, the program starts the motion controller again to read out the next line of the microarray and this procedure is repeated until the full microarray is read out.
For most measurements, an integration time of is sufficient but for more sensitive detection, a longer integration time can be chosen. Integration times longer than are unsuitable, however, since the background noise due to the dark current of the CCD begins to dominate the signal. Consequently, the readout of a standard microarray slide of with an array pitch of takes about for about 3000 single spots.
Quantitative Concentration Measurement
To quantify the performance of the system, a DNA-based test chip was employed. On an amino-silane-coated glass slide (GAPS™ II, Corning), a microarray with different oligonucleotide concentrations was printed. The format of the chip was and the distance between the spots ; a standard VHL sense-strand oligonucleotide (SVSO) was used. Concentrations from were printed with a high precision needle printer on the slides. After washing, UV crosslinking and drying, the slides were hybridized for at with SVSO-FL-Cy3 fluorescence labeled oligonucleotide solution. The slides were subsequently washed and dried again.
To verify the hybridized oligonucleotide concentrations, the chip was first examined with two commercial biochip readers, an Axon GenePix® 4000 and a GeneScan Biodetect 645/4. The Axon reader, with photomultiplier detection, was able to detect all concentrations with a resolution of in . The CCD-based reader (GeneScan) detected the concentrations from 1000 down to with an integration time of .
Figure 5 (a, top) shows a picture of the area of the test chip measured with the system developed here. With an illumination power of per spot, the chip was read out line by line using holographic excitation. The exposure time of each line was . The readout of the subareas shown in Fig. 5 took in total ( for the whole readout of one line). After integration of the fluorescence signal of each spot the signal shows the correspondence between the measured fluorescence intensity and the oligonucleotide concentration. As seen in Fig. 5c, the system was able to measure oligonucleotide concentrations from 1000 down to with a linear dependence between the dye concentration and the fluorescence intensity. In spite of the hyperspectral evaluation, which decreases the sensitivity of the detection by the distribution of the fluorescent light on more camera pixels, the same sensitivity as the commercial CCD-based microarray reader is attained,16 showing that implementation of hyperspectral detection in microarray biochip readers may not necessarily adversely affect performance.
Both CCD-based systems showed a reduced SNR at longer integration times due to a strong increase of the background noise caused by the autofluorescence of the slide and the dark current noise of the CCD. To minimize the autofluorescence of the slide, we used amino-silane-coated glass slides, which showed a much lower autofluorescence compared to TOPAS® and PMMA plastic slides. The dark current noise of the CCD camera can be reduced by cooling the CCD chip. To detect lower concentrations than those demonstrated here, the use of a cooled CCD camera is reasonable and with this technology the system could lead to sensitivities comparable to a photomultiplier-based system.
Detection of Emission Spectra
Modern biochips are able to assay complex biochemical reactions. To obtain more information from a single spot, several fluorochromes are frequently used simultaneously.1 This requires a biochip reader capable of a detailed spectral analysis of the fluorescence signal to discriminate different fluorochromes, one motivation for the hyperspectral detection unit presented here.
To evaluate the performance of the microarray scanner, a test system with the fluorochromes Dy631 (red), Dy550 (green), and Dy415 (blue) (Dyomics GmbH) was made. Different mixtures of these dyes were spotted onto a glass slide using a commercial needle printer. As we can see in Fig. 6 , the hyperspectral detection unit can measure the fluorescence intensity of the three different dyes simultaneously. The color picture illustrates the spectral distribution of each spot of the microarray biochip and a qualitative correlation between the concentrations and the intensity of the fluorescence signal can be observed. Figure 7 shows the measured fluorescence intensities versus the wavelength for different dye mixtures; in each case, the spectrum of the individual dyes is also shown for comparison. By comparing the fluorescence spectra of the spots with the known spectra of the fluorochromes employed, evaluation of the spectra may be used to determine the dye concentrations. It is clear in the figure that the fluorescence intensity peaks correspond to the dye concentrations.
In addition, the different fluorescence intensities measured for identical concentrations of different dyes is due to the different quantum efficiencies of the dyes. As we can see in Figs. 7a and 7b, the fluorescence intensity of the red fluorochrome Dy631 is nearly the same as that of the Dy550 (green), even though the excitation with the green laser at is not optimal for Dy631. The results also show that excitation with the two lasers at 405 and is sufficient to excite all three fluorochromes at the same time. However, if red dyes are extensively employed, the use of a third red laser source would be preferable.
Figure 7c shows the spectral distribution of two spots with three different dyes. In this case, the fluorescence intensity of the green dye is less than the intensity of the red dye. This dimming of the green fluorescence intensity is caused by the fluorescence resonance energy transfer (FRET) from the green to the red fluorochrome, as described by Förster17 in 1948.
A new approach for highly parallel hyperspectral fluorescence spectroscopy using holographic excitation of a microarray biochip was presented. A key feature of the instrument is the flexible, multipoint excitation of fluorescence spots using spatial light modulators. This new approach has four notable advantages. First, due to the holographic multipoint illumination, excitation light is focused only onto the spots under investigation; background fluorescence is thereby reduced and the measurement shows a higher contrast when compared to flood illumination. Second, an adaptation to arbitrary spot configurations can be achieved by changing the holograms by software modification only; modifications of the hardware are not required. Third, parallel illumination and readout of 1-D spot arrays increase the throughput compared to single-point scanning techniques. Finally, continuous spectral fluorescence detection enables discrimination of various dyes without modification of the optical setup, providing additional information on the chemical reactions that take place on the biochip and increasing the sensitivity by eliminating significant background fluorescence. The presented system thus represents a fast, universal, and reliable prototype biochip readout unit and enables high throughput complex DNA detection.
This work was funded by MOBA (Micro-Optical Biochip Analysis) as part of the research program Biophotonik 1 of the German Ministry of Education and Research (BMBF). We gratefully acknowledge cooperation and support from Sensovation AG (Stockach, Germany) and the assistance of Dipl.-Chem. Thorsten Neumann (IMTEK, Laboratory for Chemistry and Physics of Interfaces).