Abstract
Tumor spheroids more faithfully mimic tumor biology than monolayer cultures and require three-dimensional microscopy. Our goal in this study was to overcome the limitations of signal to noise ratio that have traditionally limited three-dimensional imaging to depths of 100 μm or less. We studied the expression of hypoxia-inducible factor 1α (HIF-1α), the main regulator of cellular hypoxic response in C6 glioma spheroids. In our spheroids, red fluorescent protein is expressed constitutively and green fluorescent protein is expressed conditionally under control of a HIF-1α promoter. In this article, we show a series of optimizations that allowed us to obtain excellent quality confocal microscopy images at imaging depths of up to 320 μm. The combined use of special objectives, glass-bottomed culture dishes, and depth-dependent laser output modulation extended our depth range beyond previously accepted limits. This allowed us to image up to the equator of spheroids of 650 μm diameter, allowing interrogation of HIF-1α expression from the spheroid periphery to its hypoxic center.
TUMOR SPHEROIDS are an ex vivo tumor model of great importance because they more faithfully mimic tumor biology than monolayer cultures. Spheroid tumors can be thought of as midway in complexity between monolayer cultures and in vivo tumors and are an excellent ex vivo model of cancer. Specifically, tumor spheroids are characterized by high cell density and have volumes of central hypoxia similar to those of real tumors. Hypoxia-inducible factor 1α (HIF-1α) is an important regulator of cellular responses to hypoxia, playing a role in tumor progression, angiogenesis, and therapy resistance. The tumor spheroid cells used in this study express two fluorescent markers: red fluorescent protein (RFP) is constitutively expressed by all cells, whereas green fluorescent protein (GFP) is expressed under the control of a HIF-1α promoter under hypoxic conditions. 1
There are significant technical challenges, however, to the microscopic study of these spheroids. Microscopic imaging becomes progressively more challenging as one moves from two-dimensional monolayer cultures to three-dimensional tumor spheroids. Laser scanning confocal microscopy presents one methodology that allows three-dimensional microscopy of these structures, but there are significant technical challenges in dealing with loss of image quality (signal to noise loss) at increasing imaging depths.2–5
The goal of this study was to optimize confocal microscopy imaging of our model three-dimensional tumor spheroids to allow imaging at greater imaging depths. Specifically, we describe the methods used to deal with the problem of signal to noise loss at increasing optical depths. A combination of optimizations allowed us to visualize depths of up to 320 μm. In our model, this covers the spheroid equator where oxygen concentrations would be expected to be at a minimum. Imaging at sufficient depth allows the study of biologically relevant tumor spheroid hypoxia.
Materials and Methods
Spheroid Manufacture
Spheroids were made using a well-characterized C6 glioma cell line, stably transfected with a genetic reporter consisting of constitutively expressed RFP under the control of a cytomegalovirus promoter and GFP under transcriptional control of the hypoxia response element HIF-1α. 1 We manufactured three-dimensional tumor spheroids with an average size of 500 μm (containing ~ 30,000 cells/spheroid) by mixing a C6 #4 glioma cell pellet on ice with Matrigel (#354248, BD Biosciences, San Jose, CA), followed by immersion of the cell-Matrigel mixture into room-temperature culture media (Dulbecco's Modified Eagle's Medium/F12 #11330-057; 10% fetal bovine serum #26140-079, and antibiotic-antimycotic #15240104, all from Invitrogen, Carlsbad, CA). All imaging was performed on 24- to 48-hour-old spheroids that were rendered chemically hypoxic by the addition of 150 μM CoCl2 to the culture media.
Optimizing Image Quality
In this study, we used a multilevel approach to maximize image quality, specifically by addressing the problems of imaging at increasing optical depths in tumor spheroids.
Culture Dishes
We imaged three-dimensional spheroids in two alternative culture dish systems: (a) optically clear, flat well-bottomed, polystrene plates designed for direct microscopic viewing (Corning Costar #3603, Corning Costar Corp, Corning, NY) and (b) chambered slides using borosilicate cover glass as the growth surface (Labtech #1554 11, Labtech International Ltd., East Sussex, UK). The spheroids attached to the surfaces of both the polystyrene plates and the glass-bottomed slides prior to imaging (Table 1).
Confocal Microscopy
The spheroids were imaged live with an inverted Olympus FV1000 (Olympus Inc, Center Valley, PA) laser scanning confocal microscope using standard dry and wet Universal Plan Apochromat Lens (UPlanApo) objectives (Table 2).
In this study, we used both the argon ion 488 nm and green HeNe 543.5 nm lasers (Melles Griot, Albuquerque, NM) and suitable filter sets (DM405/488/543 nm). We analyzed the constitutively expressed RFP for which the excitation/emission spectra were 543 nm/591 nm. The laser output for initial experimentation with both the dry and wet 10× UPlanApo objectives was at 11% for the 543 nm laser, with no correction for light attenuation.
The imaging was performed by taking a series of horizontal scans through the spheroids, arranged as a vertical stack/z-stack by means of image analysis software (FV1000 version 1.6, Olympus Inc). Optical sections were acquired at 20 μm intervals, with a typical imaging stack consisting of 12 to 22 optical sections per spheroid. Image size was an 800 × 800 pixel matrix/image (1,272 × 1,272 μm) with a color depth of 12 bits/pixel.
To determine at which z-level optical sections should be sampled, the spheroid size was determined by measuring the spheroid at the widest possible diameter on both the bright field and red fluorescent channels. We simplified calculations for the purpose of this study by assuming that the spheroids were spherical. The spheroid equator was determined by calculating the radius (diameter/2) and then imaging at least to the equator, preferably beyond it by some margin. A typical spheroid of 500 μm should thus be imaged to at least 250 μm from the plate bottom to study at least 50% of the total spheroid volume.
Specifications of Culture Dishes Used in Spheroid Imaging
Well bottom elevation signifies the distance from the microscopy stage surface to the bottom of the plastic well bottom.
Specifications of the Olympus FV-1000 Confocal Microscope Objectives
NA = numerical aperture; N/A = not applicable; UPlanApo = Universal Plan Apochromat Lens.
Light Attenuation Correction
Imaging signal to noise loss decreases as a function of increasing imaging depth, and one of the ways to counter this loss is to increase laser output gradually to compensate for the loss. Laser output was scaled linearly from an initial 1% near the plate bottom to 10% at maximum imaging depth near the spheroid equator.
Image Analysis
The three-dimensional image stacks acquired were analyzed using custom software written as a script for ImageJ (National Institutes of Health, Bethesda, MD). 6 This software performed radial profile image analysis on user-specified image slices through the spheroid. The RFP channel was used to threshold the data and to determine the center of the spheroid. We swept a radial arc through 360° at 1° increments around this center while plotting an RFP and GFP expression plot profile along each radius (plot line thickness = 1 pixel). Numerical plot profile data thus obtained were further analyzed using a spreadsheet (Excel, Microsoft, Redmond, WA).
Results
Spheroid Manufacture
We could reliably make spheroids of 500 μm (± 50 μm) diameter. Spheroids were rounded near-spheres to microscopic inspection and retained their shape through the course of the experiment. Spheroids retained their firm spontaneous attachment to culture surfaces despite movement and manipulations during culture and imaging.
Optimizing Image Quality
The image quality required for this study was defined pragmatically: our custom-written software needed to be able to use the images as input for its algorithms. Software image analysis depends critically on the signal to noise ratio of the imaging data and correlates well with human-defined measures of image quality, such as object brightness, background darkness, focus, clarity, and discrimination of features. Analyzable images were of good quality to human visual inspection and have a dark background, bright nuclear green fluorescence, bright cytoplasmic red fluorescence, a clearly defined spheroid border, and relatively constant brightness across the visual field covering the spheroid (no or little central loss of signal). See Figure 1 and Figure 2 for examples of imaging stacks, with the last analyzable image marked with an asterisk.
Comparison of the effect of 10× Universal Plan Apochromat Lens (UPlanApo) standard (dry) versus 10× UPlanApo W3 water immersion objectives on tumor spheroid image quality at increasing optical depths in polystyrene-bottomed (0.5 mm thick) 96-well plates and in glass-bottomed (0.13-0.16 mm thick) chambered slides as a function of imaging depth. A, Tumor spheroids imaged with a 10× standard objective in 96-well polystyrene-bottomed dishes had an analyzable depth of 120 μm. B, Tumor spheroids imaged with a 10× water immersion objective in 96-well polystyrene-bottomed dishes had an analyzable depth of 160 μm. C, Tumor spheroids imaged with a 10× standard objective in glass-bottomed slides had an analyzable depth of 160 μm. D, Tumor spheroids imaged with a 10× water immersion objective in glass-bottomed slides had an analyzable depth of 220 μm. Scale bars: x-axis = 652 μm; y-axis = 308 μm. *Last analyzable image.
A maximum intensity projection reconstruction of a three-dimensional spheroid optical data stack showing a side projection of tumor spheroids imaged with 10× Universal Plan Apochromat Lens (UPlanApo) standard versus 10× UPlanApo W3 water immersion objectives in polystyrene-bottomed (0.5 mm thick) 96-well plates and glass-bottomed (0.13-0.16 mm thick) chambered slides. A, 10× standard objective in 96-well plates. Notice that the rounded shape of the spheroid is only partially visualized; we cannot resolve images to the spheroid equator. B, 10× water immersion objective in 96-well plates. Note that at 180 μm, the images start repeating owing to mechanical interference between the objective and the 96-well imaging plate and stage (see the text for details). C, 10× standard objective in glass-bottomed chambered slides. D, 10× water immersion objective in glass-bottomed chambered slides. Notice that slightly more than half of the circumference of the round shape of the spheroid can be perceived. We are able to image to the spheroid equator. E, 10× water immersion objective in a glass-bottomed chambered slides with correction for light attenuation. Notice that a high signal to noise ratio was achieved in spheroids with a diameter of 650 μm, which equates to the spheroid equator. Scale bars: x-axis = 652 μm; y-axis = 398 μm.
Culture Dishes
Image quality, especially at deeper optical levels, was improved when imaging in thin borosilicate glass-bottomed (0.13-0.16 mm thick) chambered slides versus polystyrene flat-bottomed (0.5 mm thick) culture plates. Initial imaging (with the standard objective) yielded analyzable images up to a depth of 120 μm in polystyrene bottoms, which improved to 160 μm with use of the glass-bottomed culture chambers. Figure 1A demonstrates an imaging stack obtained using polystyrene-bottomed plates, whereas Figure 1C demonstrates the results for a glass-bottomed plate, imaged with the standard 10× objective. (See Table 3 for a summary of analyzable depths.) Figure 4, A and B, demonstrates the average decline in fluorescent intensity with increasing imaging depth as determined by the custom-written software.
Analyzable Imaging Depth in Spheroid Imaging for Various Optimizations as Determined by the Use of Software and Scientific Inspection
Confocal Microscopy Objectives
Using a water immersion objective allowed improved imaging depths compared with the standard objective. The analyzable imaging depth improved from 120 μm with the standard objective to 160 μm with the water immersion objective using the 96-well polystyrene-bottomed plates (see Figure 1, A and B). For glass-bottomed culture chambers, imaging depth improved from 160 μm with the standard objective to 220 μm for the water immersion objective (see Figure 1, C and D). The water immersion objective allowed analyzable optical sections through the spheroid equator when imaging on thin borosilicate glass-bottomed slides, which is demonstrated by the spheroid maximum intensity projection (MIP) image (see Figure 2D).
At an imaging depth of 180 μm, images taken with the water objective in the 96-well polystyrene-bottomed plates were repetitively imaged (see Figure 1B) owing to mechanical interference between the plate and the objective. The MIP images of a horizontal side projection of spheroids showed these repeating images (see Figure 2B) in comparison with the standard objective, which showed a marked loss in signal to noise ratio at an optical depth of 140 μm and above (see Figure 2A).
Light Attenuation Correction
Light attenuation correction by laser modulation was done only for thin borosilicate glass-bottomed slides with the water immersion objective and was able to improve the analyzable imaging depth to 320 μm. Figure 3 demonstrates the use of light attenuation correction on a glass-bottomed plate with the water immersion objective. We performed simple linear laser output modulation from 1 to 10% and found a more even distribution of fluorescence intensity with depth (Figure 4C) and improved image quality at depth (see Figures 2E and 3).
Optimized confocal imaging of tumor spheroids using glass-bottomed (0.13-0.16 mm thick) chambered slides with a 10× Universal Plan Apochromat Lens (UPlanApo) W3 water immersion (wet) objective and correcting for the effect of light attenuation on imaging quality at increasing optical depths by a modulated increase in laser output. Maximally optimized images have an analyzable depth of 320 μm. Scale bars: x-axis = 652 μm; y-axis = 398 μm. *Last analyzable image.
Note that this combination of optimizations allows us to comfortably exceed the equatorial plane of a 500 μm spheroid; in theory, we would be able to attempt spheroids up to about 650 μm with this set of optimizations (see Figure 2E).
Discussion
Tumor spheroids in cultures closely simulate conditions in actively growing tumors, in which central regions within the tumor develop low pH as well as oxygen and nutrient deprivation. Larger spheroids (650 μm in diameter) mimic in vivo conditions more closely regarding central hypoxia. 7 However, these larger spheroids are also more difficult to image. It is generally accepted that confocal microscopy becomes challenging at depths exceeding 100 μm owing to signal to noise loss. Recent findings indicate that it is not possible to examine the central regions of the spheroids with scanning confocal microscopy and that two-photon exciting microscopy is essential. 8 Some of the reasons for impaired imaging performance are intrinsic to the imaging of living systems, such as (a) light scattering, (b) photon attenuation, and (c) local refractive index differences. These factors, for the most part, cannot be changed by the microscopist. Refraction differences extrinsic to the sample, however, are a major cause of spherical aberration-related loss of signal, and some of these differences can be reduced or eliminated.
Graphs plotting the average fluorescence intensity values (y-axis) of spheroid optical sections at increasing optical depths (x-axis) as measured in μm. A, Light attenuation demonstrated in tumor spheroids imaged on polystyrene bottom plates (0.5 mm thick) with both standard and water immersion objectives with no correction for light attenuation. B, Light attenuation demonstrated in tumor spheroids imaged on borosilicate glass bottom slides (0.13-0.16 mm thick) plates with both standard and water immersion objectives with no correction for light attenuation. C, Imaging tumor spheroids with a water immersion objective on borosilicate glass bottom slides (0.13-0.16 mm thick) plates with correction for light attenuation. Standard deviation bars indicated
Intrinsic factors degrading imaging performance are, for the most part, difficult to change. Local refractive differences in the three-dimensional matrix of the sample can potentially be corrected for by means of sophisticated mathematical modeling, 2 but this technique falls outside the scope of our article. Fortunately, light scattering and absorption in the sample itself are minor players in a transparent biologic system such as ours. 9
Refractive errors are the most important light interactions in the formation of aberrations. 4 Spherical aberration is induced at interfaces with refractive index mismatches and results in an optical wave front that can no longer be focused to a single point. Fortunately, many of the effects of spherical aberration can be minimized by the microscopist, as shown in the following optimizations.
Culture Dishes
The culture dishes used during imaging had a significant impact on image quality, especially at increasing optical depths. Most objectives (including ours) are designed for use with glass coverslips of 0.17 mm thickness and a refractive index of 1.525. 4 In our system, the borosilicate culture chamber bottoms approached this ideal much more closely than the polystyrene 96-well plates. The culture chamber bottoms had the physical properties of cover glass (0.13-0.17 mm, refractive index = 1.523) as opposed to the polystyrene bottoms (0.5 mm, refractive index = 1.57-1.60). High-throughput applications will likely require the use of 96-well plates with thin glass bottoms.
The well bottom elevation of 3.57 mm in our 96-well polystyrene plates is not well suited to use with a water immersion objective as it does not allow the close physical proximity required for this objective (water immersion requires the objective to nearly touch the plate bottom). We noted problems with mechanical interference between the water immersion objective and these plates (see Figure 2B). At present, we are not aware of any polystyrene 96-well plate manufactured with a well elevation of less than 2.28 mm, but special plates may become available in future. High-throughput applications will likely require the use of recessed plate holders to minimize the effect of well elevation observed while imaging with 96-well plates.
Objectives
We observed a dramatic improvement in imaging quality at depth by using the water immersion objective. Using water rather than air reduces the mismatching of refractive indices, thus minimizing spherical aberration, maintaining the point spread function with increasing depths, and improving imaging performance, especially for high numerical aperture (NA) objectives. 10 , 11 The refractive index of a spheroid is 1.3406 to 1.341 at 20°C and 1.338 to 1.339 at 37°C (Frank Mannuzza, BD Biosciences, personal communication, 2008). Thus, the water immersion objective provides better images than the dry objective at increasing imaging depths because the index mismatch between air (nd = 1.00) and spheroid (nd = 1.34) is greater than the combination of water (nd = 1.33) and spheroid (nd = 1.34). For this reason, many manufacturers now offer water immersion objectives such as the one used in this study as water matches the refractive index of most living biologic systems even better than oil immersion. 3
Spherical aberration results from the inability of spherically ground lenses to bring light to a single point focus. The vulnerability of an objective to spherical aberration is correlated to NA, with large NA objectives being more vulnerable. 11 This vulnerability to spherical aberration is addressed by means of the water immersion objective. In apochromatic lenses, such as in our system, the effects of chromatic aberration are small to negligible.
Confocal systems suffer doubly from spherical aberration, first, because it reduces the photon flux focused on the voxel and thus reduces the fluorescence emitted, and, second, the fluorescent light harvested at the gathering confocal aperture is deprived of many of the photons by the same process.
Laser Output Modulation
Some of the light losses described so far are not normally a serious consideration in two-dimensional microscopy. The issue only comes up when imaging at depth in three dimensions, such as our application in the study of tumor spheroids. When photons are lost from the interrogating beam for whatever reason, compensation (within limits imposed by the biology and/or laser output available) can be made by simply increasing photon flux. We modulated laser output output as a function of depth and had good success in increasing imaging depth and quality. Doubtless, further improvements could be expected with exponential modulation of laser output based on actual system imaging characteristics with feedback loops.
Image Analysis
We demonstrated the use of a software computer vision algorithm, discussed in greater depth elsewhere. 12
Other Technologies
Although our article deals with confocal microscopy, many alternative imaging modalities are under development. Two- or multiphoton imaging shows great promise for imaging at depth 13 but has not yet achieved the penetration in the community that confocal microscopy has. Many scientists would like to image living systems at depth using their existing confocal microscopes and may not have access to multiphoton imaging. In addition, fascinating methodologies for imaging spheroids or similar biologic systems with light sheets, 14 magnetic resonance imaging, 7 and ultrasonography 15 are evolving, each with unique strengths and abilities.
Conclusion
Confocal microscopy imaging in three-dimensional applications is a complex optimization of many variables. The most important variable is imaging depth, which correlates with increases in spherical aberration and photon loss for other reasons, leading to a reduced signal-to-noise ratio. Users can optimize imaging at depth by choosing appropriate parameters for lens NA, dry versus wet objectives, 96-well plates versus glass-bottomed culture chambers, thickness of plate bottom, and modulation of laser output. The complex interplay between these variables will determine the success of three-dimensional confocal imaging at depth.
We describe a series of optimizations that allow greatly increased imaging depth in the confocal microscopy of three-dimensional tumor spheroids. Thin glass-bottomed culture chambers, water immersion objectives, and laser output increase as a function of depth were key parameters that we optimized to increase analyzable imaging depth from about 120 to 320 μm. This methodology will allow investigators with access to confocal microscopy to extend their depth of imaging, allowing the study of complex three-dimensional structures such as tumor spheroids.
Footnotes
Acknowledgments
We would like to thank Laura Driscoll, Angela Goodacre, and Mark Blaylock from Olympus America and Jeff Lovett and Al Butzer from Leeds Instruments for their expertise and assistance during this project. Thanks to Jessica Hunter for help in preparing the manuscript.