SEARCH

SEARCH BY CITATION

Keywords:

  • 3D image cytometry;
  • tissue cytometry;
  • rare cell detection;
  • multiphoton microscopy

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. PROCEDURES AND RESULTS
  5. DISCUSSION
  6. CONCLUSIONS
  7. LITERATURE CITED
  8. Supporting Information

Image cytometry technology has been extended to 3D based on high-speed multiphoton microscopy. This technique allows in situ study of tissue specimens preserving important cell–cell and cell–extracellular matrix interactions. The imaging system was based on high-speed multiphoton microscopy (HSMPM) for 3D deep tissue imaging with minimal photodamage. Using appropriate fluorescent labels and a specimen translation stage, we could quantify cellular and biochemical states of tissues in a high throughput manner. This approach could assay tissue structures with subcellular resolution down to a few hundred micrometers deep. Its throughput could be quantified by the rate of volume imaging: 1.45 mm3/h with high resolution. For a tissue containing tightly packed, stratified cellular layers, this rate corresponded to sampling about 200 cells/s. We characterized the performance of 3D tissue cytometer by quantifying rare cell populations in 2D and 3D specimens in vitro. The measured population ratios, which were obtained by image analysis, agreed well with the expected ratios down to the ratio of 1/105. This technology was also applied to the detection of rare skin structures based on endogenous fluorophores. Sebaceous glands and a cell cluster at the base of a hair follicle were identified. Finally, the 3D tissue cytometer was applied to detect rare cells that had undergone homologous mitotic recombination in a novel transgenic mouse model, where recombination events could result in the expression of enhanced yellow fluorescent protein in the cells. 3D tissue cytometry based on HSMPM demonstrated its screening capability with high sensitivity and showed the possibility of studying cellular and biochemical states in tissues in situ. This technique will significantly expand the scope of cytometric studies to the biomedical problems where spatial and chemical relationships between cells and their tissue environments are important. © 2007 International Society for Analytical Cytology.

Cytometry is the quantitative measurement of the physical and biochemical states of cell populations. Cytometry provides information of the cell population on a cell-by-cell basis rather than the population average. Many cytometric approaches are high-throughput and allow for categorizing large cell populations into subgroups revealing rare subpopulations. Flow cytometry is a widely used technique in which cellular specimens are prepared in fluid suspensions and the properties of individual cells are measured in a narrow fluid stream (1–4). The properties of cells are assayed based on optical characteristics, such as fluorescence, light scattering, and light absorption. Flow cytometry has very high throughput reaching a rate up to 10,000 cells/s (3). In combination with a cell sorting apparatus, precise physical separation of the cellular subpopulations is routinely achieved. Flow cytometry is an indispensable tool in immunology, molecular and cell biology, cytogenesis, and the human genome project.

Image cytometry is a complementary method to flow cytometry, overcoming some of its limitations such as loss of cellular morphological information (5–10). In image cytometry, cell specimens are prepared as 2D tissue cultures and are imaged by either wide field or laser scanning microscopy. Quantitative image analysis is performed to segment and categorize cellular subpopulations. Although throughput of this method is relatively low—hundreds to thousands cells per second—it has several advantages: (1) Detailed cellular morphological and structural information are retained. (2) The cells of interest can be identified and relocated for further analysis. One key example is the ability to monitor temporal evolution in live specimens. (3) Image cytometry also provides the spatial information of protein distributions within cells such as receptor distribution in the membrane versus cytosol. Quantitative image analysis tools have been further developed based on intensity, morphology, and spectral information of the image. Image cytometry often utilizes multiple excitation light sources and detection channels to distinguish multiple labels. More advanced systems allow hyperspectral analysis of higher content images acquired using devices such as liquid-crystal tunable filters (11).

Cells in tissues behave differently from cells in cultures (12, 13). In tissues, cells become specialized, acquiring specific functions. Cellular behaviors such as proliferation and differentiation are regulated by adjacent cells and their tissue locale, and cells interact either directly, such as via cell adhesion molecules (CAM), or indirectly by chemical signaling. Cells are connected to the extracellular matrix (ECM) via their membrane receptors such as integrin. ECM is composed of fibrous proteins, such as collagen, and adhesion molecules, such as fibrin and fibronectin. ECM not only provides physical support but cell–ECM interactions are keys to the collective behaviors in tissues and play a critical role in processes such as mechanotransduction (14). Therefore, it is important to study cellular behaviors in the native tissue environment. Those interactions are either lost or altered when cells are grown in culture or dissociated from organs (15). The physiology of pancreatic islets is an example showing different cellular behaviors depending on environments (16, 17). Bennett et al. (16) used multiphoton microscopy (MPM) to monitor redox activity inside intact pancreatic islets based on their NAD(P)H level. Previous research on the cells grown as 2D cultures showed significant cell–cell variations in NAD(P)H level as response to external glucose levels. This 2D culture experiment resulted in metabolic models of glucose metabolism based on step-wise recruitment of individual cells. However, when Bennett et al. (16) imaged cells within intact islets, they found significantly more homogeneous glucose response of cells in the islet suggesting that the step-wise model based on 2D culture results might not be relevant to the actual physiological insulin response in the pancreas. This difference is clearly important for the design of pharmaceuticals for diabetes treatment. Furthermore, in traditional histological tissue examination with image cytometry, tissue specimens need to be sliced very thin in order to be imaged. Slicing generates distortion in the structure of tissue specimens (7). 3D image cytometry can avoid this artifact by imaging directly inside tissues down to hundreds of micrometers deep. Tissue structures and gene expression distributions can be studied based on various labeling techniques such as fluorescence in situ hybridization.

Recently, image cytometry has been extended to 3D for the quantitative study of histological tissue specimens (8–10, 18–21). It is based on confocal laser scanning microscopy (CLSM) with multiple excitation light sources. CLSM obtains 3D resolution by rejecting light from out-of-focus regions in the specimen with a pinhole in front of a detector. Its maximum imaging depth is approximately 100 μm in typical tissues. Quantitative image analysis to identify individual cell types can be a challenge, because cells are tightly packed in some tissues. However, various studies have showed promising results. In the lymph node, the spatial distribution of leukocytes expressing different antibodies was mapped in thin tissue slices (9). The phenotypes and quantities of tissue-infiltrating leukocytes were further characterized to monitor immune system response to transplants and therapies in renal tissues. Confocal 3D image cytometer demonstrated in situ quantification of proliferation markers and tumor suppressors, and in situ quantification of apoptosis in tissues (8, 20). In the study of Alzheimer's disease, neuron subsets with and without cyclin B1 expression were identified and mapped throughout a 120 μm thick tissue (21). However, image cytometry based on CLSM is limited to fairly thin specimens due to tissue turbidity.

We introduce a new 3D tissue cytometric approach based on high-speed multiphoton microscopy (HSMPM). MPM is a 3D imaging technique, similar to CLSM. However with its features of higher imaging depth in tissues and minimal phototoxicity compared with CLSM, MPM is appropriate for in vivo tissue studies (22–24). The main obstacle in applying MPM for tissue cytometry was the limited imaging speed of conventional MPM systems, but this difficulty has been circumvented by the introduction of HSMPM. We developed instruments that can assay the tissue volume of 1.45 mm3/h at <1 μm3 resolution. When the tissue volume contains tightly packed, stratified cellular layers, these instruments can sample approximate 100–200 cells/s. Therefore, the 3D tissue cytometer based on HSMPM can be used to screen a large cell population inside intact tissues where 1 million cells can be quantified within about 3 h.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. PROCEDURES AND RESULTS
  5. DISCUSSION
  6. CONCLUSIONS
  7. LITERATURE CITED
  8. Supporting Information

Multiphoton Microscopy

MPM is based on nonlinear excitation of fluorophores (22, 25) with two-photon excitation being the most important and practical case. Nonlinear excitation results in fluorescence generation only at the focus of excitation light. This localization of excitation volume further results in reduced specimen photodamage and photobleaching. Most importantly, MPM allows deeper imaging into tissue specimens than other microscopy technique such as CLSM, because of the reduced scattering and absorption of excitation light and high collection efficiency of emission light in turbid tissues. Many studies have demonstrated that MPM is very suitable for in-vivo deep tissue imaging. MPM has become an inevitable tool for biomedical studies, such as neuronal plasticity (26–28), angiogenesis in solid tumors (29), and noninvasive optical biopsy (30).

3D Tissue Cytometry Based on HSMPM

We developed two HSMPM systems that could sustain a frame rate of about 10 times faster than the conventional MPM. These HSMPM systems have been reported elsewhere (31, 32) and are explained briefly here. The first method adapted a polygonal mirror scanner to increase scanning speed rather than galvanometric scanners, which have a limited bandwidth (31). The polygonal mirror scanner was a light weight metal cylinder with mirror facets machined around its perimeter. Rotation of the cylinder swept the facets and generated line-scans along one axis in the sample plane. It achieved a higher scanning speed, since the polygon rotated at a constant speed, rather than moving back-and-forth as the galvanometric scanner that requires repeated start-stop. The schematic of the current system is presented in Figure 1 achieving a frame rate of 13 fps.

thumbnail image

Figure 1. Schematic of the high-speed multiphoton microscope based on a polygonal mirror scanner. Excitation beam is drawn in red and emission beam is in green. Excitation light coming from a titanium-sapphire laser (Mira 900, Coherent, Palo Alto, CA) was reflected on the polygonal mirror (Lincoln laser, Phoenix, AZ) and was relayed to a galvanometric mirror scanner (Cambridge technology, Cambridge, MA). The polygonal mirror scanned the excitation beam along the fast axis of the sample plane and the galvanometric mirror scanned along the slow axis. After the scanners, the excitation beam was directed into an upright microscope (Axioscope, Zeiss, Thornwood, NY) via a modified epiluminescence light path. The excitation beam was expanded by a lens pair and reflected on a dichroic mirror (short pass, DC700SP, Chroma technology, Brattleboro, VT) toward an objective. The objective focused excitation light into the specimen generating multiphoton excitation. The excitation focus scanned in a raster pattern via the polygonal mirror and galvanometric scanner. Emission light was generated at the focus and was collected by the objective. It was transmitted by the dichroic mirror and collected by a photomultiplier tube (PMT, R3896, Hamamatsu, Bridgewater, NJ). The signal from the PMT was low-pass filtered and amplified via a transimpedance amplifier and measured by an analog-to-digital converter (AD9220EB, Analog Device, Norwood, MA). A separate laser diode and photodiode are used to sense the position of facets of the polygon and to synchronize the other scanner and the detectors. The schematic shows a single detector for simplicity, but two detectors with a dichroic mirror were used for two channel imaging. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Download figure to PowerPoint

The second method increased imaging speed by parallelizing the multiphoton imaging process and it has been called as multifocal multiphoton microscopy (MMM) (33, 34). It scans a specimen with multiple foci of the excitation light instead of a single focus and collects emission light from the multiple foci simultaneously with a spatially-resolved multi-channel detector. The imaging speed increases proportional to the number of excitation foci scanning together. In our implementation, we used 6 × 6 excitation foci for scanning and multianode photomultiplier tubes (MAPMTs), which have 6 × 6 pixels, for simultaneous signal collection (32) (Fig. 2). Each MAPMT pixel collected signal from the corresponding excitation focus in the sample, and the signal collection was synchronized with scanning of the excitation foci. This MAPMT-based MMM design was to achieve equivalent imaging depth as the conventional MPM, with higher imaging speed. The imaging speed increased approximately 30 times.

thumbnail image

Figure 2. Schematic of the multifocal multiphoton microscope based on MAPMT. Excitation light from Titanium-Sapphire laser source (Tsunami, Spectra-physics, Mountain View, CA) is expanded and illuminated a microlens array (1000-17-S-A, Adaptive Optics, Cambridge, MA). The microlens array comprise of an array of small lenses of 1 mm × 1 mm square in size and 17 mm in focal length. The microlens array generates multiple foci at its focal plane and another lens (L1) collimated these beams from the foci creating 6 × 6 beamlets. Two excitation beam-lets are traced in the schematic. The beam-lets pass through a dichroic mirror and are reflected on x-y galvanometric mirror scanners (6220, Cambridge Technology, Cambridge, MA). The beam-lets after the scanners are expanded by a combination of an eyepiece (L2) and a tube lens (L3) to fill the back aperture of an objective. The objective focuses the excitation beam-lets to generate an array of excitation foci in the sample plane. The x-y scanners raster scan the array of excitation foci in sample plane. The excitation foci are separated from each other by 45 μm in the current instrument. Emission beam-lets from the sample are collected by the objective and traced in green toward the MAPMT. After reflected on the scanners, emission beamlets are de-scanned and become stationary regardless scanning. The dichroic mirror reflects the beam-lets toward a MAPMT (H7546, Hamamatsu, Bridgewater, NJ). The MAPMT is a spatially resolved multichannel PMT with a rectilinearly divided anode and had 8 × 8 channels. We use 6 × 6 channels in the middle of the MAPMT for simultaneous signal collection. The outputs from the MAPMT channels are read by a custom built single photon counting circuitry that convert the photon pulse signals from these channels into digital counts in parallel. The signals are transferred to an acquisition computer via a 32 bit parallel bus. The schematic shows a single detector for simplicity, but two channel detections were implemented with two MAPMTs and a dichroic mirror. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Download figure to PowerPoint

Since the imaging area was limited by the view field of microscope objectives on the order of a few hundred micrometers on a side, an automated translation stage was needed to image a large sample region. A computer-controlled specimen translation stage (H101, Prior Scientific, Rockland, MA) was used to translate the specimen rapidly and precisely. The stage was driven by step motors in three axes and its resolution was ±3 μm. The stage communicated with a control computer via serial port. Its response time was approximately 0.5 s and was limited by the computer communication speed and its traveling distances. A significant improvement of the imaging speed will be possible with a higher bandwidth stage.

The microscope field of view was an important factor that determined instrument throughput. Since raster scanning of the excitation light was always faster than the mechanical translation of the specimen, the throughput was higher with microscope objectives which had larger view fields. Therefore, microscope objectives with lower magnifications were preferred. On the other hand, it was also critical to use the objectives of high NAs to maximize collection of emission light. For the image cytometer based on the HSMPM using polygonal scanner, a 25× water immersion, NA 0.8 (LCI Plan-Neofluar 25×, NA 0.8, Zeiss, Thornwood, NY) was used and the field of view was approximately 200 μm × 200 μm. For the one based on the multifocal multiphoton microscope, a 20× water immersion, NA 0.95 objective (XLUMPLFL20XW, Olympus, Melville, NY) was used and the field of view was 270 μm × 270 μm.

PROCEDURES AND RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. PROCEDURES AND RESULTS
  5. DISCUSSION
  6. CONCLUSIONS
  7. LITERATURE CITED
  8. Supporting Information

Rare Cell Detection in 2D and 3D Cell Cultures

The performance of the 3D tissue cytometer was characterized by quantifying rare cell subpopulations in 2D and 3D specimens in vitro. Two cell populations of different fluorescent labels were mixed at various ratios from 1/1 down to 1/105. The cell specimens were prepared either as 2D tissue cultures on glass slides or as 3D in collagen scaffolds. The population ratios were measured by 3D HSMPM imaging and subsequent image analyses.

2D and 3D cell culture preparation.

Preparation of the 2D cultures was as described in the literature (35). In brief, mouse 3T3 fibroblast cells were transfected with plasmids containing the coding region for enhanced yellow fluorescent protein (EYFP) or enhanced cyan fluorescent protein (ECFP) (Clontech, Palo Alto, CA), each under the control of the pCX (chicken beta actin) promoter (pCX Yellow and pCX Cyan, respectively). These proteins were chosen with the consideration that both colors could be simultaneously excited at a single excitation wavelength and that the emission spectra of the two fluorescent proteins could be easily distinguished. 3T3 cells were lipofected with the plasmids using LipofectAMINE-Plus (Invitrogen, Carlsbad, CA). EYFP and ECFP cells were grown separately. EYFP and ECFP cell stocks were trypsinized, counted, and mixed at various ratios from 1/1 down to 1/105. The mixed cells were cultured on glass slides for 12 h. The specimens were rinsed with PBS and then sealed with coverslips before imaging. The cell specimens were imaged live.

For 3D culture, NIH 3T3 mouse fibroblast cells were grown in collagen scaffolds to mimic tissue environment. Collagen scaffolds have been used extensively as a biomaterial in tissue engineering (36). Collagen is a natural constituent of ECM and the scaffolds have porous structures allowing cell seeding and nutrient diffusion. The collagen scaffolds can serve as analogs of ECM by providing physical support and working as insoluble regulators of biological activities that direct cellular processes such as migration, contraction, and division. The preparation of the collagen-glycosaminoglycan copolymer scaffolds using a freeze-drying method has been previously described in detail (36–38). Two groups of cells were prepared, one group labeled with green CellTracker (C2925, Molecular probes, Eugene, OR) and the other group not labeled, and both cell populations were nucleus-labeled with Hoechst. The CellTracker label was chosen because it was permanent in live cells through multiple cell divisions, and also because both labels were excited efficiently with single wavelength excitation light. In the labeling procedure, CellTracker dye was diluted to 10 μM in serum free medium. Culture medium in the culture dishes was replaced with dye solution and the dishes were incubated at 37°C for 45 min. Then, the dye solution was replaced with culture medium and the dishes were incubated for another 30 min. Afterwards, both labeled and unlabeled cells were trypsinized from the dishes and cell densities were measured using a hemocytometer (1483, Hausser scientific, Horsham, PA). The stained cells were mixed with unstained (control) cells at various ratios from 1/1 down to 1/105. The final concentration of the cell mixtures were adjusted to 2 million cells/ml, and 25 μl of cell mixtures was seeded onto each collagen scaffold of 15 mm × 15 mm. Prior to seeding, collagen scaffolds were incubated in culture medium for 6 h so that the medium was in equilibrium within the scaffolds. The collagen scaffolds were then dried out slightly for 0.5 h in order to absorb the additional 25 μl of cell mixture solution. One hour after seeding, culture medium was added around the collagen scaffolds and followed by 12 h incubation allowing the seeded cells to migrate inside scaffolds. The collagen scaffolds containing the cell mixtures were subsequently fixed with buffered zinc formalin (Z-fix, Anatech LTD, Battle Creek, MI). The cells were fixed in this experiment to prevent cell division that might occur at different rates among stained and un-stained cells. After fixation, all the cells were nucleus-labeled with Hoechst (33342, Molecular probes, Eugene, OR). The Hoechst stock (10 mg/ml) was diluted 1/2,000 times in PBS to a final concentration of 5 μg/ml. The Hoechst concentration was chosen to approximately equalize the fluorescent intensities of Hoechst and CellTracker at the photodetectors. The scaffolds were incubated at 37°C for 10 min with Hoechst solution and washed with PBS. The scaffolds were placed on glass slides and sealed with silicone isolators (JTR20-A2-1.0, Grace Bio-Labs, Bend, OR).

Data acquisition and analysis.

For the study of 2D specimens, the 3D tissue cytometer based on the polygon scanning HSMPM was used. Excitation wavelength was set at 910 nm to excite both ECFP and EYFP. Although the specimens were monolayer cells, we imaged 10 layers along axial direction with 2 μm step size. Since the high NA objective had a narrow depth of focus on the order of microns, imaging was very sensitive to leveling of the specimen. Imaging a small volume stack ensured that the cell layer was captured in focus. To achieve reasonable statistical accuracy, we aimed to detect at least 10 rare cells at each mixture ratio. Since these specimens were composed of monolayer cells, the throughput rate was much lower as compared to that of 3D samples in which multiple cell layers were present.

To distinguish two emission colors, a dichroic mirror at 495 nm wavelength (495DCXR, Chroma Technology, Brattleboro, NH), a long pass filter at 500 nm (E500LP, Chroma Technology, Brattleboro, NH), and a short pass filter at 490 nm (E490SP, Chroma Technology, Brattleboro, NH) were used, dividing emission light into two channels, referred to as the yellow channel (YC) for EYFP, and the cyan channel (CC) for ECFP. Figures 3a and 3b are representative images of a 1/10 mixture ratio (EYFP/ECFP) in YC and CC, respectively. Both cell types appeared in the YC image (Fig. 3a), since there was significant bleed through of the ECFP emission spectrum into YC. On the other hand, only ECFP expressing cells appeared in the CC image (Fig. 3b). For clearer distinction, ratio images were constructed by combining two channel images according to the following equation.

  • equation image(1)

IR represents an intensity value of the ratio image pixels. ICC and IYC denote pixel values of the CC and YC images, respectively. To remove background noise from the regions not containing cells, ratio images were masked with binary images obtained by thresholding YC images. The threshold was chosen based on noise level of the images. Pixel values of the ratio images ranged between −1 and +1 by the definition of Eq. (1). Given the spectral properties of fluorescent proteins, EYFP expressing cells had pixel values (IR) around 0.95 in the ratio images, and ECFP expressing cells had values around 0.5. These two differently labeled cell populations were easily distinguishable in the ratio images. Figure 3d is a ratio image of a 1/103 mixture ratio specimen where only 2 EYFP cells appear. Specimens of other mixture ratios were imaged and analyzed the same way and the results are shown in Figure 3e. Expected ratios and measured ratios are in horizontal and vertical axes respectively. The good correlation between the expected and measured ratio can be seen from the measured slope (1.08) and the goodness-of-fit (R2 = 0.9855) of the linear regression line. The linearity of the relationship demonstrates that the system can detect rare cell populations down to 1/105 ratio.

thumbnail image

Figure 3. Image analysis for 2D cell cultures. Two cell populations expressing different fluorescent proteins (EYFP, ECFP) were mixed at various ratios. (ac) are images of a cell mixture at 1/10 ratio (EYFP/ECFP) in the YC, CC and after the ratio processing respectively. EYFP expressing cells appears only in YC and these cells appear in yellow color with high ratio values in the ratio image (c). (d) is a ratio image of a 1/100 mixture ratio (EYFP/ECFP) specimen. Only two EYFP cells are detected. Images of other mixture ratios were analyzed in the same way and the measured ratios are plotted against the expected ratios down to 1/105 in (e). This figure was published in a SPIE proceeding paper (Kim KH et. al, 2001; Proceedings of SPIE Vol. 4262.) and is republished here with permission of SPIE.

Download figure to PowerPoint

Similar characterization experiments were performed for the 3D cell blocks grown in collagen scaffolds. The specimens were imaged with the 3D tissue cytometer based on the multifocal multiphoton microscope operating at 30 frames/s. The excitation wavelength was 800 nm. Input power was 18 mW per excitation focus. The size of image was approximately 270 μm × 270 μm containing 192 × 192 pixels. For 3D imaging, 50 layers were imaged with 2 μm depth increment. This depth range covered about 3–4 layers of cells in the specimens. The two cell populations were distinguished by two-color imaging: a blue channel (BC) and a green channel (GC) utilizing a dichroic mirror at 500 nm. No additional barrier filters for the two channels were used to maximize emission signal collection. The acquired data was analyzed to count the number of cells in each population. The cell counting algorithm was written in MATLAB (MathWorks, Natick, MA) and is explained in the next section. Analyzed results are presented in Figure 4. Figure 4a is a representative 3D image of a 1/1 ratio specimen. One group of cells which was labeled only with Hoechst showed blue emission in their nuclei. The other cells, which were labeled also with CellTracker green, had both blue and green emissions in their nuclei and green emission only in their cytoplasms. We focused our analysis on the intensity ratio (IR) of green to BC (IR = IGC/IBC) in the nuclei to distinguish the two subpopulations. This method was very effective for the following reasons. (1) The intensities at nuclei were high in general so that it was easy to discriminate from background noise. (2) Nuclei were generally round and could be readily discriminated from other objects based on shape analysis. (3) Nuclei were often well separated each other. Two channel images were analyzed to segment cell nuclei and to quantify each cell populations. Figure 4b is a cell scatter plot from a specimen of 1/1 population ratio. It was based on average intensity of individual nuclei in two channels (BC intensity in the horizontal axis and GC intensity in the vertical axis). Marks in the plot represent individual cells counted in the analysis. The IRs in the cells with only Hoechst label had a characteristic ratio of 3/10. The IR values from the cells labeled with both were all above the IR values from singly labeled cells, although they varied appreciably. Therefore, a line with the slope of 0.3 (IR0) from a point (100, 50) in (BC intensity, GC intensity) distinguished the two cell populations. The analysis of cell counting and discrimination was performed in the images from specimens of different population ratios from 1/1 down to 1/105 and the results are presented in Figure 4c. The good correlation between the expected and measured ratio can be seen from the measured slope (0.91) and the goodness-of-fit (R2 = 0.9892) of the linear regression line. However, one can observe that better correlation is achieved for the 2D as compared with the 3D case. Significant deviation from the linear regression line is seen for the lowest concentration data point. This may be due to image analysis limitations. We kept the cell density high in the scaffold to ensure fast imaging speed; the image analysis method employed was less reliable for closely spaced cells since it is based on global intensity thresholding.

thumbnail image

Figure 4. Image analysis for 3D cell cultures grown in collagen scaffolds. (a) is a 3D reconstructed image of a 1/1 ratio specimen. Cells with Hoechst label only appear as blue nuclei and the other cells with both Hoechst and CellTracker labels appear as blue nuclei in green cell bodies. A movie of 3D cell culture in collagen scaffold (1/10 mixture ratio) is available as supplementary material. (b) is a representative scatter plot for 1:1 ratio specimens. Two subpopulations are clearly distinguished based on the IR. (c) is the result of cell counting analysis of 3D cell cultures with various mixture ratios down to 1/105. Measured ratios (in x-axis) are plotted against the expected ratios (in y-axis).

Download figure to PowerPoint

Segmentation and cell counting algorithm.

While image processing was required in all our experiments, we present a detailed description of the algorithm used for segmenting in the 3D cell blocks as an example. There have been sophisticated image processing algorithms developed for segmenting cell nuclei in 3D within real tissue specimens (8, 39–49). Here we use a simple segmentation method, because the cell density is lower in our specimens and as a demonstration. Its procedure is presented as a flow chart in Figure 5.

thumbnail image

Figure 5. Cell counting analysis procedure for the 3D cell cultures grown in collagen scaffolds. Both BC and GC images (a,b) are converted to BW nucleus images (c,d) based on thresholding. Erosion and dilation remove noise in the images (e,f). Convolution with a cone-shaped kernel enables cell counting in both channel images (h,i). (j) is a 3D view of BC image after kernel convolution (h). [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Download figure to PowerPoint

Conversion of BC images in each layer to binary black and white (BW) images by intensity thresholding: BC images mapped spatial distribution of nuclei in both cell populations with Hoechst label. Proper thresholding selected the pixels of nuclei from background noise. The BC images were smoothed by applying a median filter and then converted to BW images (IMmath image) by intensity thresholding (Ithresh). The average intensity of the nuclei in the focal plane was approximately 160 and the background noise was less than 10. Ithresh was set at 100 so that only the nuclei in the focal plane were captured in the BC-BW images (Ithresh = 100).

  • equation image(2)

where ir and ic were pixel indices along row and column directions in the image respectively.

Conversion of GC images to BW images: Valid pixels in the GC images were selected using the binary images generated in step (1). The IR of the valid pixels was calculated as the GC intensity to BC intensity. GC images were converted to BW images (IMmath image) based on thresholding of intensity ratio IR0 (Fig. 5b). These BW images (IMmath image) contained the nuclei of the cells which were double labeled.

  • equation image(3)

Noise reduction with a series of erosion and dilation operations: Erosion operation is a logical “AND” operation among a pixel to be processed and its neighboring pixels. The value of the processed pixel becomes 1 only if the values of all the neighboring pixels including the processed pixel are 1. Therefore, erosion operation could remove small structures including noise in the images. Dilation operation is a logical “OR” operation among the processed pixel and its neighboring pixels. To remove noise in BW images, erosion was performed with a disk shape element resulting in shrinking the size of nuclei and removing spurious pixels due to noise. A subsequent dilation operation with the same structure element was performed to restore the size of the nuclei.

  • equation image(4)

Convolution with a cone shape kernel: To find the locations of nuclei, the BW images (IMmath image, IMmath image) were convoluted with a kernel (K) which had a cone shape intensity distribution with its peak at the center. It was noted that this convolution operation tended to blur the images. Here we assumed that nuclei were separated from each other. Also we chose the kernel size smaller than the average nucleus size in order to reduce the blurring effect.

  • equation image(5)

The locations of nuclei were identified by peak finding and masking: After convolution with the kernel image, the images had peak intensities at the center locations of individual nuclei. A maximum finding procedure was performed to find the highest peak location, and an area around the peak location was masked with a masking disk. The area of masking disk was initially set to be slightly bigger than the average nucleus size and was adjusted proportional to the peak intensity levels. Then the next maximum finding operation gave location of the next highest peak (location of the next nucleus). The operation of masking with a disk and maximum finding was continued until the peak value became less than a chosen threshold determined by image noise level.

The above operations (1–5) were performed in each layer image. The continuity of nuclei in the images of several layers was checked by measuring the separation distance among their locations found in the 3D image stack. If two locations were within a distance, they were assumed to be the same nucleus and the plane containing the highest intensity value was considered to be the central location of nucleus in 3D.

Counting of both cell groups was performed for each data set. Each 3D section of 270 μm × 270 μm × 100 μm contained approximately 50 cells on average.

Rare Structure Detection in Ex-Vivo Skin

A major strength of this 3D tissue cytometer was to quantify cellular and ECM states in tissue environments. As a demonstration, an ex-vivo human skin specimen was imaged based on its autofluorescence and second harmonic generation by the 3D tissue cytometer based on the polygon scanning HSMPM. The objective used in this experiment was 40× (Fluar, NA 1.3, oil; Zeiss, Thornwood, NY). Since the field of view of the microscope was limited to 120 μm × 120 μm, the large section imaging was performed by translating the specimen with the computer-controlled sample stage, once 3D imaging for each section was completed. Image sections (25 × 25) were acquired and 50 layers were imaged from surface down to 67 μm deep for each section. A stitching algorithm was applied to combine these images stacks together to form montage images of 2.5 mm × 2.5 mm. Figure 6a shows the montage image at 67 μm deep from the surface. This image shows tissue structures of collagen, elastin fibers in the dermal layer which are either autofluorescent or second-harmonic-active. In addition to the ECM components, the 3D tissue cytometry could further detect cell clusters at the base of skin hair follicles (Fig. 6b) and a sebaceous gland of the skin (Fig. 6c). This study represents one of difficult imaging conditions that 3D tissue cytometry may encounter due to the low endogenous signal and the high turbidity of the tissue. We expect most other tissue cytometric studies, which use exogenous fluorophores or fluorescent proteins, to be much easier.

thumbnail image

Figure 6. Large sectional image of ex-vivo human skin in dermal layer. (a) is a montage image of 2.5 mm × 2.5 mm in size. It is constructed by combining 25 × 25 image sections, which were acquired with the HSMPM based on the polygonal scanner. Collagen and elastin fibers are shown based on autofluorescence and second harmonic generation. A cell cluster at the bottom of a hair follicle (b) and a sebaceous gland (c) appear in marked regions.

Download figure to PowerPoint

Detection of a Rare Cell Population in Transgenic Mice

As a demonstration of rare cell detection in tissues, we applied the 3D tissue cytometer to image tissues from the fluorescent yellow direct-repeat (FYDR) mice (50, 51), a model designed to study homologous mitotic recombination. Homology directed repair (HDR) is a mechanism to repair double strand breaks but misrepair of those breaks can promote cancer. HDR repairs damaged sequences by copying sequence information from the sister chromatid or the homologous chromosome and it can be associated with exchange of DNA sequences. Cells in this mouse model could express EYFP if mitotic recombination events occurred at a specific genetic locus. Spontaneous frequency of mitotic recombination was approximate ∼1/106 for these mice. Flow cytometry was used to quantify rare fluorescent recombinant cells. Here, we attempted to apply the 3D tissue cytometer to detect these recombinant cells in tissues.

In this experiment, tissue specimens were excised from the pancreas of a selected FYDR mouse and were further stained with Hoechst. After demarcating a region containing a recombinant fluorescent focus using traditional epifluorescence microscopy, this region of the sample was then analyzed using the multiphoton imaging described here. The objective used was 40× water-immersion with 1.2 NA. Two detector channels were implemented for dual color imaging using a dichroic mirror at 500 nm. The size of each image stack was approximately 110 μm × 110 μm × 100 μm. The wavelength of excitation light was 890 nm and approximately 20 mW of input power was used for imaging. The acquisition speed was approximately 0.25 frames/s. We found that recombinant cells expressing EYFP could be identified and a 3D image of the recombinant cell and its tissue microenviroment is presented in Figure 7 (note that it is unclear how many cells are present within the fluorescent region; nuclei appear blue and cytoplasm of recombinant cells appears green in this image). This is the first demonstration that rare recombinant cells can be imaged in 3D within a tissue specimen. Further studies are underway to develop methods to both detect and to image recombinant cells in situ.

thumbnail image

Figure 7. Images of recombinant cell(s) found in a pancreatic tissue of a FYDR mouse. (a) is a representative 2D plane and (b) is a 3D reconstructed image from the image stack. Nuclei which are stained with Hoechst appear as blue in the image and the recombinant cells express EYFP (green color) in the cytoplasm.

Download figure to PowerPoint

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. PROCEDURES AND RESULTS
  5. DISCUSSION
  6. CONCLUSIONS
  7. LITERATURE CITED
  8. Supporting Information

The 3D tissue cytometer based on HSMPM demonstrated its ability to screen 2D and 3D tissue cultures containing up to 106 cells. The ex-vivo skin imaging showed that this instrument had a high sensitivity for tissue imaging based on endogenous fluorophores. The detection of rare recombinant cells showed its potential for in-vivo quantification of rare genetic events in animal models. Although the 3D tissue cytometer based on HSMPM showed some promise, it had limitations and rooms for improvement. HSMPM could excite two different color fluorophores simultaneously using a single excitation wavelength. Since signals from both fluorophores could be collected in the nonimaging configuration, two channel spectral images were guaranteed to be co-registered. While this was an advantage, this was also a limitation. Since the excitation wavelength was set to excite both fluorophores, it could not be optimized to excite both efficiently, and the imaging speed needed to be slowed down subsequently.

Performance of the 3D tissue cytometer based on the multifocal multiphoton microscope can be significantly enhanced by further instrument improvements. First its imaging speed can be increased. An important advantage of the multiple focal approach is that it can achieve higher imaging speed while keeping the pixel residence time the same as the conventional MPM. High-speed single focus systems, such as the polygon scanning system, necessarily have shorter pixel residence time to achieve higher frame rate and lower SNR subsequently. The SNR can be improved by increasing excitation power. However, fluorophore saturation and sample damage threshold ultimately limit this approach, particularly for fragile, live samples such as embryos. The multifocal approach overcomes this limitation and makes high speed 3D tissue cytometry for weakly stained or autofluorescence samples a viable possibility. Generation of multiple excitation foci requires more excitation laser power, and currently imaging speed is limited by the maximum power of available laser sources. With the availability of higher power lasers, more excitation foci can be generated and, in principle, can provide higher degree of parallelization and speed. Further, we used a microlens array to split the excitation beam into multiple beam-lets. The Gaussian intensity profile of the excitation beam led to uneven excitation intensities in the sample with the center foci producing higher signals than the surrounding foci. Additionally, since the microlens array was rectilinear and the excitation beam had a round shape so that the portion of excitation beam illuminating the outside of the square profile was not used. Therefore, there was a significantly waste of excitation light (about 30%). In the future, beam splitting mechanisms, such as ones using multiple beam splitters or ones using diffractive optical elements, can be used to generate multiple beam-lets of uniform powers and improve transmission efficiency (52). The MAPMTs used in the current system had an acceptable quantum efficiency (QE) of 20%. This was because the photocathode material in MAPMTs was not optimal. Currently, GaAsP photocathode material can achieve higher QE up to 40% and may be available for the future generations of MAPMT. The current 3D tissue cytometer based on multifocal multiphoton microscope had only two channels. A new 3D tissue cytometer with multiple channels for spectral imaging can be easily implemented by modifying the current system. Multiple color imaging can be implemented in the same configuration as the one demonstrated with a single excitation focus (53), but by generating a 1D vector of multiple foci. An improved system incorporating these improvements is under development in our laboratory.

In this work, we have demonstrated detection of rare cell populations in 2D and 3D specimens based on automated cell counting. Very simple image processing algorithms were used in this pilot study for cell population analysis and these had limitations. For the 2D specimens, both groups of cells expressed fluorescence in their cell bodies and it was difficult to do segmentation in confluent specimens. Sparse specimens were prepared for this 2D population analysis. For the 3D specimens, the processing algorithm, which was based on nucleus segmentation, also worked only for the specimens where nuclei were well separated each other with minimal variations in their size and shape. Many other cytometric analyses based on cellular morphology could be performed using data acquired by the current system, although we have not demonstrated these capabilities here. We believe that the current limitation of high throughput 3D cytometry lies less in the instrument, that can now readily provide 3D image data with sufficient SNR at high speed, but more in the availability of efficient computation approaches that can efficiently visualize, segment, and quantify 3D image data sets in large scales. Currently, many advanced algorithms for the segmentation of cell nuclei and cell bodies have been developed and offer the potential for more accurate quantification (8, 39–49). Some of these algorithms further incorporate adaptive adjustment of segmentation parameters for automatic segmentation process. For future applications, segmentation of more geometrically complex biological structures in tissues such as blood vessels and neuronal dendrites will become necessary. Some works have been done (41, 54–58), and more efficient and novel imaging processing routines need to be developed. The high throughput 3D tissue cytometry will also require high performance computational environments and resources which can efficiently handle terabyte scale data set.

While the current version of 3D tissue cytometer could detect rare recombinant cells in FYDR mouse tissues, our current instrument may not be ideal for this application. It is inefficient to image whole organs at high resolution for detecting few recombinant cells. Since recombination events are rare and fluorescent recombinant cells appear as readily distinguishable discrete clusters throughout the tissue, an improved system may be designed where these cell clusters can be first detected by low resolution wide field imaging before subsequent multiphoton imaging to provide more informative 3D high resolution assays to determine the composition of these clusters.

3D tissue cytometry is a fairly new technology and its potential for biomedical research is still far from fully explored. In the short term, we have identified a number of immediate applications. We are exploring the application of 3D tissue cytometry to study two aspects of cancer biology. The first aspect is cancer progression. The process of how cancer cells extravasate through blood vessel walls and expand to form metastatic cancer at distant organs is far from completely understood. We are exploiting 3D tissue image cytometry to explore questions such as the distribution of these cancer cells on the organ level and the spatial relationship of these metastatic cancer cells with the organ vasculature. This technology further allows us to study the time course of cancer cell clearance rate from the vascular system and provides temporal information on the growth rate of metastatic tumors. Other interesting areas of application include neural biology where we are exploring the potential application of this technology to quantify neuronal connectivity as a function of animal development, stem cell research where we are mapping stem cell distribution in organs and examining the process of adult stem cell division in animals, and tissue engineering where we are examining cellular differentiation and organ formation in situ.

Finally, with histological quantification, one may envision that tissue physiology and pathology can be better understood through modern genomic and proteomic analysis. Combining high throughput 3D tissue image cytometry with the ability to map gene and protein expression profiles, physiological models may be developed based on the underlying molecular and cellular process. These physiological models may allow us to understand how tissue structure is affected by genetic and protein expression variations. An early example of this type of analysis may be found in areas such as cancer development. Cancer is a disease which has a very strong spatial component to its etiology (59). Cancer cells can invade the stroma of the surrounding tissue and recruit nonmalignant cells to differentiate and support the growing tumor. The arrangement of normal tissue boundaries becomes pathogenic as the expression profiles of the surrounding cells are altered by cell signaling from the malignant cells (60–62). The application of 3D tissue image cytometry to simultaneously map tissue morphology, gene and protein expression patterns may allow us to better understand this important pathological process.

CONCLUSIONS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. PROCEDURES AND RESULTS
  5. DISCUSSION
  6. CONCLUSIONS
  7. LITERATURE CITED
  8. Supporting Information

The high-speed multiphoton microscope with a computer-controlled specimen stage was adapted as the 3D tissue cytometer. It still retained the advantages of a standard multiphoton microscope allowing in vivo turbid tissue imaging with subcellular resolution. Its throughput rate was up to approximately 200 cells/s. Therefore, it can be used in the studies which need to investigate a large cell population in vivo. Quantitative analysis of cell populations in 2D and 3D cultures demonstrated capability of this system for detecting rare cell populations: cell mixtures of various ratios from 1/10 to 1/105 could be accurately quantified. The wide area human skin image showed its capability of screening highly turbid tissue specimens at high speed based on endogenous fluorophores of tissue components. Using the 3D tissue cytometer, rare recombinant cells were detected in tissue specimens from FYDR mice which carry fluorescent markers for homologous mitotic recombination. 3D tissue cytometry based on HSMPM allows the study of cellular and tissue morphological and biological states in situ with subcellular resolution. This technique will significantly expand the scope of cytometric studies to the biomedical problems where cell-tissue interactions are critical.

LITERATURE CITED

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. PROCEDURES AND RESULTS
  5. DISCUSSION
  6. CONCLUSIONS
  7. LITERATURE CITED
  8. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. PROCEDURES AND RESULTS
  5. DISCUSSION
  6. CONCLUSIONS
  7. LITERATURE CITED
  8. Supporting Information

This article contains supplementary material available via the Internet at http://www.interscience.wiley.com/jpages/1552-4922/suppmat .

FilenameFormatSizeDescription
3DcellcultureinSC_1to10ratio.mpg1428KSupporting Information file 3DcellcultureinSC_1to10ratio.mpg

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.