The programmable array microscopes (PAMs) are a family of instruments incorporating arbitrary control of the patterns of illumination and/or detection. The PAM can be used in sectioning and nonsectioning modes, thereby constituting a useful platform for fluorescence lifetime imaging.
Methods and Results
We used a PAM for acquisition of optically sectioned and widefield fluorescence lifetime images, in which contrast was increased predominantly by suppressing out-of-focus light contributions. We simulate, display, and discuss the effects of blurring and fluorophore heterogeneity on lifetime imaging in widefield and confocal configurations.
The burgeoning use of fluorescent lifetime imaging microscopy (FLIM) in cell biology (1) has been prompted by the need to discriminate between multiple fluorophores with overlapping emission spectra and to assess molecular microenvironment, conformational states, and dynamics. Molecular association is most generally probed via the phenomenon of Förster resonance energy transfer (FRET). Among the many manifestations of FRET exploitable in imaging systems (2), the reduction of the fluorescence lifetime of a donor in the presence of a suitable acceptor is one of the most fundamental. Lifetime determinations are well suited for cell biological applications because they are relatively independent of optical pathlength and illumination homogeneity. However, as with many spectroscopic techniques, lifetimes measured in a conventional widefield microscope are subject to distortions arising from contributions of out-of-focus light originating above and below the plane of best focus. This realization has led to the development of optically sectioned lifetime imaging techniques implemented in single (3, 4, this publication) and multiphoton configurations (1, 5–8). Image reconstruction procedures applied to widefield and structured illumination data have also been used to generate sectioned fluorescence lifetime images. Different optimized lifetime imaging systems for the time domain (9, 10) and frequency domain (11, 12) have been reported; see Suhling et al. (1) and Bunt and Wouters (13) for recent reviews.
Programmable array microscopes (PAMs), including those extensively developed in our laboratory, provide a unique opportunity for studying the effects of varying degrees of optical sectioning on fluorescence lifetime measurements. In the PAM, a spatial light modulator (SLM) is placed in an image plane of the microscope and used to generate patterns of critical illumination and/or detection. PAMs have been implemented with SLMs based on micromirror (14–18) and liquid crystal (19, 20) technologies. In the fluorescence mode, white light and laser sources have been employed (14, 15, 18, 21). PAMs incorporate no macroscopic moving parts and feature software control of the patterns of illumination and detection that can be restricted to one or many regions of interest, thus enabling a large variety of technologies and applications (see 22 and references therein). In optical sectioning arrangements, the sectioning strategy can be matched to specific experimental needs.
The theory of fluorescence image formation in the PAM has been treated in detail (15, 18, 23) and comparisons between the theoretical predictions and measured responses have been reported. PAMs have also been used in nonsectioning configurations for spatial encoding in two- and three-dimensional imaging spectroscopic applications (24), optically sectioning imaging spectroscopy (21, 25), and a widefield system for measuring spectrally resolved FLIM (26).
In this report, we describe a PAM adapted for the measurement of fluorescence lifetimes in widefield and in optical sectioning configurations and explore the influence of fluorescence heterogeneity and diffraction-limited effects under these conditions.
MATERIALS AND METHODS
Theory and Computation
Frequency domain lifetime imaging has been treated extensively in the literature. We restrict the coverage here to that required for understanding the effects of sectioning on the lifetime images. Additional information can be found in the numerous publications cited above and in monographs (27–30).
Two apparent single-frequency fluorescence lifetimes are computed from the phase difference and relative modulation depth of the fluorescence emission compared with the excitation light source.
In equation 1, τm is the modulation lifetime, ω is the radial frequency (2πFo, where Fo is the excitation intensity modulation frequency), m is the relative (emission/excitation) modulation depth, τϕ is the phase lifetime, ϕ is the phase shift between the sinusoidal excitation and emission, and the index k designates a particular location (xk, yk, zk spatial coordinates) in the sample. In the context of this study, we were concerned with the effects of blur and its removal on the determination of these spectroscopic parameters. In heterogeneous systems, comprising a mixture of n individual exponentially decaying components, which may or may not originate from the same chemical entity, blur decreases the range of lifetimes observed in an image due to mixing of out-of-focus light from regions above and below the plane of focus. The modulation depth and phase in such heterogeneous systems is dictated by the characteristics of the sample in the detection region, i.e., the particular linear combination of fractional contributions from the n components. The measured values of mk and ϕk depend on the lifetimes (τi) and normalized pre-exponential factors (αi) of the i = 1 − n decay components and on ω according to equations 2 to 4.
where the dimensionless quantities Ak and Bk are conveniently expressed in terms of the ith fractional contribution fi,k to the steady-state fluorescence and the dimensionless quantity Ti, the ith lifetime multiplied by ω.
In an optically sectioned FLIM image of a specimen with a distribution of components varying in space (over x, y, z), the dynamic range of operative αi and fi values will be greater than in the corresponding nonconfocal widefield image, thereby generating improved discrimination. For example, the signal from a sample consisting of two overlying (along the optic axis) “slabs” containing fluorophores with different lifetimes but equal intensities (f1 = α1τ1 = f2 = α2τ2) appears constant at different focus levels in a widefield system. However, as optical sectioning is applied, a range of f1 and f2 values is observed (Fig. 1). Although the treatment presented here applies to the frequency domain, a related set of issues, including photon statistics and optimal selection of experimental parameters (31), arises in time domain measurements, particularly those based on the collection of a small number of time windows (9, 10). Lifetime determinations of specimens with only a single type of unperturbed fluorophore (a condition we will refer to as fluorophore homogeneity) or thin specimens with multiple fluorophores that are spatially separated but lie within the diffraction limit will not be improved by optical sectioning in general.
Frequency domain imaging data obtained from heterogeneous systems of fluorophores have been analyzed according to a variety of formalisms (32–35), including the LiMA treatment of A. Esposito and F. Wouters (personal communication). A convenient graphical representation of the mixing behavior described by equations 1 to 4 is a two-dimensional display of τm and τϕ (36–39). In this representation, all possible values of fi for a system of two fluorophores lie along an arc below the line of equivalence (examples are given in Fig. 5D–F). Recently, an alternative graphical representation has been proposed, the plot of Ak versus Bk (AB plot) (40, 41), which offers certain advantages. (i) Ak and Bk are well defined for all values of mk and ϕk, whereas τm and τϕ are ill behaved as m → 1 and ϕ → π/2, respectively. The former condition is a concern in noisy measurements of very short lifetimes and the latter condition can arise with long lifetimes and noisy data or when monitoring the acceptor in a FRET pair. (ii) For a binary system of fluorophores with distinct and invariant lifetimes (τ1, τ2) but mixed in arbitrary, spatially varying stoichiometries, the AB plot transforms a somewhat difficult global analysis problem to a simple linear relation between the vectors of experimental Ak and Bk values (functions of mk and ϕk, equation 4; Fig. 5G–I), the solution of which leads to recovery of the two constituent fluorophore lifetimes.
Attention to statistical issues is important in the application of equation 5. For example, because Ak and Bk are subject to experimental error, an appropriate fitting method such as major axis regression is required (42).
Programmable Array Microscope
The digital micromirror device (DMD)-based PAM and the fluorescence lifetime modules have been described previously (15, 18, 43). The PAM system consisted of a DMD mounted in the primary image plane of a Nikon E-600 microscope, where it served to define patterns of illumination and detection (Fig. 2). The frequency Fo and phase of modulation were established with a computer-controlled signal generator (Marconi Instruments 2030). A second signal generator (Marconi 2023), phase locked to the intensifier signal generator and amplified with a linear power amplifier (Electronic Navigation Industries 403LA), drove an acousto-optic modulator (AOM; IntraAction SWM-10044) at frequency Fo/2. An Ar+ ion laser (Coherent Inova 90-5) was used to illuminate the microscope field. The laser beam traversed the AOM and the zero order light, modulated at Fo, was selected and relayed through a multimode optical fiber to the PAM illumination port. The optical fiber was mechanically shaken to scramble the modes and further homogenized using a 10° holographic light-shaping diffuser (Physical Optics Corp.). The emission light captured by and reflected from “on” elements of the DMD was relayed to a gain-modulated image intensifier (Hamamatsu C5825). The light from the output phosphor plate of the intensifier was relayed with a pair of f/1.8 (Nikon AF Nikkor) camera lenses to a charge-coupled device camera (Apogee Instruments KX-2). Lifetime images were calibrated relative to rhodamine 6G in water (4.1 ns, ).
A specimen exhibiting a homogeneous lifetime was prepared from a Drosophila salivary gland expressing an enhanced green fluorescent protein posterior sex comb (EGFP-Psc) fusion protein in cellular nuclei. Whole in vivo salivary glands were observed using a 60×/1.2 numerical aperture (NA) water-immersion lens. A second specimen with a heterogeneous lifetime distribution was prepared using 2.47-μm fluorescent microspheres (Molecular Probes FluoSpheres) suspended in a ∼5 μM rhodamine 6G solution. The solution included ∼1% agarose to reduce motion of the beads; the agarose did not increase the background fluorescence.
Simulations of the confocal and widefield responses to model fluorescent objects were implemented with full numerical computations based on wave vector theory (23, 44). The object consisted of a 2.5-μm-diameter sphere with a fluorescence lifetime of 2.5 ns inside a 10-μm box with a lifetime of 4.1 ns. The simulated object was represented using 40-nm voxels. The simulations consisted of the convolution of the object with the point spread functions (PSFs) computed assuming 488-nm excitation and 520-nm emission. Three PSFs were used, corresponding to (a) a widefield measurement with a 40×/0.6 NA objective, (b) a confocal measurement with a 40×/0.6 NA objective with a 41-μm pinhole, and (c) a confocal measurement with a 60×/1.2 NA objective with a 31-μm pinhole. No noise was applied to the data. The results from the simulations were introduced into equations 1 to 4 to compute mk, ϕk, τm, and τϕ.
Imaging Quality in a Specimen With a Homogeneous Distribution of Fluorophores
A comparison of sectioned and widefield approaches to imaging Drosophila embryos demonstrated that the sectioned approach produced superior results. The intensity images produced during the acquisition of the frequency domain lifetime series exhibited reduced blur, resulting in clearer identification of the nuclear localization of the EGFP-Psc construct by intensity (Fig. 3A,B) and lifetime (Fig. 3C,D) due to suppression of the signal in the regions outside the nuclei and segmentation by simple thresholding. However, the sectioned and widefield imaging approaches yielded similar mean lifetimes within the error margins of the measurement. The sectioned lifetime image had lower intensities and thus fewer pixels above the threshold, resulting in a broader, less defined distribution. Both effects are to be expected from the considerations discussed above.
Imaging Quality and Lifetime Distribution in a Heterogeneous Sample
A heterogeneous specimen observed under sectioning and widefield conditions illustrated the contrast advantage of optically sectioned lifetime imaging for heterogeneous mixtures. The images of a sample of 2.47-μm beads embedded in an agarose-containing solution of rhodamine 6G showed dramatic improvement by introduction of optical sectioning (Fig. 4). The effects were observed in both the τm (Fig. 4A,D) and τϕ (Fig. 4B,E) images, as well as in the corresponding two-dimensional τm vs. τϕ histograms (Fig. 4C,F; lifetimes increase from the lower left corner). The tighter distribution of panel F distinguishes clearly between the fewer, lower lifetime pixels and the more numerous, higher lifetime pixels.
Simulating Effects of Diffraction Limit on Lifetimes of a Heterogeneous Sample
Simulations of a specimen consisting of a 10 μm cube with a 2.5 μm sphere in the center closely reproduced the sample of beads and R6G of Figure 4. In both cases, the fluorescence from the surrounding medium significantly affected the lifetimes in the region of the spheres (Fig. 5D–F). The out-of-focus blur diminished from ∼80% at the center of the sphere in the widefield situation to ∼34% using optical sectioning (NA 0.6). Increasing the NA to 1.2 reduced the contaminating signal further to ∼10%. Blur also restricted the dynamic range of the measured lifetimes, with the widefield image being most affected. A greater range of lifetimes was observed in the sectioned case, but only the high NA confocal case was able to approach the pure lifetime of the sphere (2.5 ns) by suppression of contamination from the long (4.1 ns) surrounding component in the cube.
These results indicate that the relation changes in donor lifetimes arising from FRET and the corresponding transfer efficiencies measured in FLIM systems, particularly widefield, will be underestimated in the presence of significant blur.
Optically sectioned FLIM images are required in many research and biotechnological applications. We have demonstrated a simple implementation of the method in a PAM. The optically sectioned measurements provide improved contrast in the lifetime images of heterogeneous systems of fluorophores, an important feature that should assist subsequent global analysis procedures and obviate the need for recording multiple image planes. The modular approach of the FLIM-PAM should be well suited to other hybrid techniques implementing other forms of patterned illumination and to measurements combined with additional fluorescence modalities such as emission anisotropy (45).
In a previous publication on FLIM from our laboratory, we noted that detectable intra-image variations in lifetime are about an order of magnitude smaller than detectable inter-image variations (43). Thus, methods exploiting the high sensitivity of intra-image lifetime variations should provide improved discrimination of processes leading to FRET. In this context a few rules of thumb can be applied. In the case of experimental systems based on a single fluorophore or a discrete FRET signal (uniform transfer efficiency to an acceptor throughout the sample), sectioning is not required and in fact is likely to degrade the lifetime measurements (although it will improve the intensity image quality). However, in the presence of multiple fluorophores or localized FRET, sectioning is highly desirable. All imaging modes will underestimate the extent of FRET due to blur from a free donor unless this feature is specifically considered in the model (46).