Download the PDF version

 

1. Introduction

Multicolor fluorescence imaging is ubiquitous in research and clinical applications. When the
fluorophores used in such applications have low spectral overlap, standard filter cubes (with fixed spectral
edges) are adequate. However, when the crosstalk is high, spectral imaging [1-4] becomes necessary.
Spectral imaging systems provide flexibility in wavelength selection. This flexibility is also useful when
experiments need to be designed with new fluorophore combinations because optical filters need not be
changed in these systems. However, despite their flexibility, conventional spectral imaging systems are
rarely able to offer the key advantages of thin-film interference filters, i.e., high transmission combined
with steep spectral edges and high out-of-band blocking. This article outlines a novel approach to spectral
imaging based on recently introduced tunable thin-film optical filters.
2. Technology

At the heart of this spectral imaging technology is Semrock’s VersaChrome® filters technology.
These are the first widely tunable thin-film optical filters [5]. Unlike standard thin-film interference filters,
the spectra of VersaChrome® filters can be angle-tuned – the filter spectrum changes as a function of
angle of incidence – without exhibiting any appreciable change in the shape of the spectrum.

Figure 1: (Left) Transmission spectra at several very high angles for the TBP01-620/15 filter. (Right)
Shift of center wavelength with increasing angle of incidence for this filter. The FWHM bandwidth remains
fixed at 20 nm (corresponding to 15 nm guaranteed minimum bandwidth). Note that the filter spectrum is
continuously tunable over 0 to 60º AOI.

Figure 1 shows snapshots of spectral profiles of one of Semrock’s VersaChrome® filters, TBP01-
620/15, at different angles of incidence (AOI). This filter has a guaranteed minimum bandwidth of 15 nm
and a full-width-at-half-maximum bandwidth of 20 nm. As seen in this figure, the filter not only retains its
bandwidth at higher AOIs, but also the high transmission, steep spectral edges, and high out-of-band
blocking (not seen in this plot) remain virtually unchanged.

The center wavelength of this tunable filter (in fact for all VersaChrome® filters) is dictated by the
following equation, where neff is the effective refractive index of the thin-film coatings:

Here neff is approximately 1.85. Note that the spectrum of this tunable filter is continuously tunable. With
a tuning range of greater than 12% of the normal-incidence wavelength (by varying the angle of incidence
from 0 to 60º) only five filters are needed to cover the full visible range.
3. Materials and Methods

Microscopy: BPAE cells (sample courtesy of Mike Davidson, Molecular Expressions™) labeled with
MitoTracker® red (Mitochondria), Alexa Fluor® 568 (F-actin), and SYTOX® Orange (Nucleus) were imaged
with an Olympus BX41 microscope equipped with a Hamamatsu ORCA C8484 camera. The emission
pathway of this microscope was modified to include a computer-controlled tunable filter module. The
principle of operation of this microscope is shown in Figure 2. In order to simultaneously excite all the
fluorophores in the sample, a single excitation filter (FF01-543/22-25) and a single-edge dichroic mirror
(FF562-Di02-25×36) were installed in a filter cube in the standard filter turret of the microscope. The
tunable emission filter, TBP01-620/15-25×36, was placed in the tunable filter module, and a sequence of
images were acquired (called a lambda stack of images) by varying the angle of incidence of the filter
with respect to the emission beam in 1º increments.

Figure 2: (Left) Principle of operation of the microscope used in this study. (Right) The emission path of
a standard Olympus BX41 microscope was modified to include a (stepper) motor controlled tunable filter
module. The angle of incidence of the tunable filter (TBP01-620/15-25×36) was adjusted in 1º increments
to acquire images of the sample corresponding to different wavelengths. The location of the tunable
emission filter is marked with a dotted rectangle.

Linear unmixing: In order to spectrally deconvolve the data (called linear unmixing, [1-4]), the pixel
intensity values were extracted from the lambda stack images and arranged in matrices. MATLAB was
used to solve a linear least squares problem of min spec_img_versachrome_fig_9 with a nonnegativity constraint on x, where
is the matrix of deconvolved spectral contributions from each fluorophore at a given pixel. The matrix A
contains the reference fluorophore spectra corresponding to each of the fluorophores (Figure 4), and the
matrix b comprises the intensity values for a given pixel from the lambda stack images.
4. Results

In this study, spectral imaging using tunable filters is illustrated with an example. Lambda stack
images of a sample labeled with MitoTracker® red, Alexa Fluor® 568, and SYTOX® Orange were acquired
using a Semrock VersaChrome® tunable filter (see Materials and Methods for details). Figure 3 presents
images acquired at about 5 nm intervals.

Figure 3: Lambda-stack mages of the sample acquired using a tunable emission filter. Images are
presented at about 5 nm intervals (refer to Fig.1 for corresponding filter spectra). Cellular components
labeled with fluorophores of distinct emission spectra can easily be resolved even in the raw data,
however, fluorophores with similar spectra benefit from linear unmixing algorithms. Individual frames
represent an object size of about 47µm x 38 µm.

It is evident from these images that the nucleus stained with SYTOX® Orange can be easily
discriminated from the other cellular structures – by merely utilizing a single tunable emission filter used to
visualize all the fluorophores. However since F-actin and mitochondria are labeled with fluorophores that
have a high degree of spectral overlap (Alexa Fluor® 568 and MitoTracker® red, respectively), linear
unmixing was necessary to discern the corresponding cellular constituents.

Figure 4: Reference spectra for each fluorophore are plotted as normalized intensity values of the selected
regions corresponding to each of the fluorophores in the sample. Note that the measured spectrum of a fluorophore
can be different from its ideal spectrum. This can happen due to a change in the environmental conditions or due to
limitations in experimental protocols. The background signal is also plotted as a function of the wavelength.

Regions of interest representing the pure spectral contribution of each fluorophore were selected from
the lambda-stack images shown in Figure 3. The normalized intensity values (following background
adjustment) corresponding to each of these fluorophores are plotted in Figure 4. These represent the
reference spectra of the fluorophores used in the unmixing algorithm. The lambda-stack images were
then used together with a linear unmixing algorithm (see Materials and Methods) to arrive at spectrally
deconvolved images for all the fluorophores (Figure 5).

Figure 5: Spectrally unmixed data. The grayscale images correspond to cellular components labeled
with specific fluorophores: nucleus (top left), F-actin (top right) and Mitochondria (bottom left). Bottom
right is a composite image. Individual frames represent an object size of about 47µm x 38 µm.

5. Conclusions

VersaChrome® filters can be placed in the excitation or emission path of a fluorescence instrument and
by merely changing the angle of incidence of the filter with respect to the beam, different spectral features
can be obtained. Also, it is worth pointing out that the spectral properties of these tunable filters are
almost identical for both s and p polarizations of light – a feature that cannot be easily obtained using
liquid-crystal and acousto-optic tunable filters [2-5]. Polarization independence is highly desirable for
spectral imaging systems, and yet polarization limitations of current tunable filters can account for a loss
of half of the signal in many spectral scanning instruments. VersaChrome® filters, on the other hand, do
not exhibit such a loss of signal. Therefore these filters can not only enhance the throughput in spectral
imaging but they can also greatly simplify the complexity of instrumentation [5].
6. References

[1] Visualization of Microscopy-Based Spectral Imaging Data from Multi-Label Tissue Sections,
Mansfield JR, Hoyt C, Richard, RM, Current Protocols in Molecular Biology, 84:14.19.1-14.19.15
2008.
[2] Spectral Imaging: Principles and Applications. Garini Y, Young IT, McNamara G., Cytometry A,
69(8):735-47, 2006.
[3] Multispectral Imaging Fluorescence Microscopy for Living Cells, Hiraoka Y, Shimi T., and Haraguchi
T. Cell Structure and Function, 27: 367-374, 2002.
[4] http://zeiss-campus.magnet.fsu.edu/

Authors

Prashant Prabhat, Ph.D., is Applications Scientist, Neil Anderson, Ph.D., is Technology Development
Analyst, and Turan Erdogan, Ph.D., is co-founder and CTO of Semrock, Inc., a Unit of IDEX Corporation.
E-mail: pprabhat@idexcorp.com, Tel: (585) 594-7064; Fax: (585) 594-7095.
Acknowledgements

The authors would like to acknowledge the support of Professor Mike Davidson, at Florida State
University, for this study.