Enumeration of leukocyte infiltration in solid tumors by confocal laser scanning microscopy
© Biggerstaff et al; licensee BioMed Central Ltd. 2006
Received: 20 October 2005
Accepted: 21 July 2006
Published: 21 July 2006
Leukocytes commonly infiltrate solid tumors, and have been implicated in the mechanism of spontaneous regression in some cancers. Conventional techniques for the quantitative estimation of leukocyte infiltrates in tumors rely on light microscopy of immunostained thin tissue sections, in which an arbitrary assessment (based on low, medium or high levels of infiltration) of antigen density is made by the pathologist. These estimates are relatively subjective and often require the opinion of a second pathologist. In addition, since thin tissue sections are cut, no data regarding the three-dimensional distribution of antigen can be obtained.
To overcome these problems, we have designed a method to enumerate leukocyte infiltration into tumors, using confocal laser scanning microscopy of fluorescently immunostained leukocytes in thick tissue sections. Using image analysis software, a threshold was applied to eliminate unstained tissue and residual noise. The total antigen volume in the scanned tissue was calculated and divided by the mean cell volume (calculated by "seeding" ten individual cells) to obtain the cell count. Using this method, we compared the calculated leukocyte counts with those obtained manually by ten laboratory personnel. There was no significant difference (P > 0.05) between the cell counts obtained by either method.
We then compared leukocyte infiltration into seven tumors and matched non-malignant tissue obtained from the periphery of the resected tissue. There was a significant increase in the infiltration of all leukocyte subsets into the tumors compared to minimal numbers in the non-malignant tissue.
From these results we conclude that this method may be of considerable use for the enumeration of cells in tissues. Furthermore, since it can be performed by laboratory technical staff, less time input is required by the pathologist in assessing the degree of leukocyte infiltration into tumors.
A variety of clinical and pathologic evidence indicates that tumors can stimulate immune responses, such as the presence of mononuclear cell infiltrates, composed of T-cells, NK cells, and macrophages in many different tumors.
In the past few decades, a number of studies have established correlations between prognosis and the degree of leukocyte (lymphocytes; dendritic cells; cytotoxic T-cells[6, 7]; gamma/delta T-cells; tumor infiltrating lymphocytes; and monocytes) infiltration in a variety of cancers. However, in all of these studies, specific staining was performed on relatively thin tissue sections. In addition, labeled leukocytes were manually counted in one or more random areas of each tumor section. Since the leukocytes are distributed in a three-dimensional volume of tissue, manual counting on thin tissue sections may not be truly representative of the actual numbers and spatial distribution of cells within the tissue.
Several microscopy and image analysis techniques have been developed for the in vitro and in vivo three-dimensional quantification of antigens in tissues, including deconvolution microscopy, stereomicroscopy[12, 13] (often used with deconvolution or Confocal microscopy (CLSM) for thicker tissue sections), or CLSM[15, 16].
In deconvolution microscopy light from all planes of focus is collected, usually via a digital or video camera, and image slices are recorded onto a computer. Since light is collected from the whole depth of the specimen at each focal plane, each image appears blurred, or convolved. Before image analysis can be performed, the images need to be deblurred, or deconvolved to render a sharp image. Depending on the thickness of the specimen this process can be very time consuming.
Stereomicroscopy, a technique mainly used for dissection and tissue manipulation has experienced a resurgence in recent years for the 3-D analysis of many proteins and gene structures in tissues. Stereomicroscopes have a long working distance and, using a dual light path, can generate 3D images over a large depth of field. However, for thicker specimens, stereomicroscopy is often coupled with deconvolution or CLSM to improve image clarity.
In terms of speed of data acquisition, CLSM is probably the most advantageous microscope technology for the production and analysis of 3D samples. CLSM refocuses fluorescent light onto a pinhole which excludes up to 95% of light from outside of the focal plane, producing clear images, even in tissue up to 100 μm thick. One apparent major disadvantage of CLSM is the initial equipment cost. However, in laboratories with a high sample throughput, this cost may be offset by its efficiency over time. This is especially important in pathology, where many specimens are analyzed and the data is required for diagnosis and prognostic determination as quickly as possible.
Therefore, this study sought to design an image based method for the calculation of leukocyte numbers in thick tissue sections (20 microns) using three-dimensional laser scanning confocal microscopy.
Indirect immunofluorescence staining for leukocyte markers
Effect of threshold value on cell count
In order to calculate the volume of tumor occupied by a fluorescent antigen, a threshold was applied to the image stack to include only antigen-specific fluorescent signal. In addition to background (black or no fluorescence) removal, a threshold can be used to remove any residual noise resulting from setting the PMT to just include control fluorescence. The higher the threshold is set, the more fluorescence data are lost. To determine the effect of threshold setting on cell counts, three image stacks containing low, moderate or high numbers of cells were taken and the cell numbers calculated at ten threshold levels. Figure 1 (middle row) shows a representative high cell infiltration (high) image (derived from an image stack (CD8 antigen) at three of the ten thresholds examined. At a threshold of 1, background noise can still be clearly seen in the image, but is not present in subsequent images. At a threshold of 5, background noise was minimized, whereas at threshold 10 significant data loss was apparent.
At any particular threshold, the cell number was determined by division of the total antigen volume by mean cell volume. To determine the mean cell volume, ten cells from each image stack were seeded to calculate the volume of all contiguous fluorescence (cell volume) above the chosen threshold. Figure 1 (bottom row) shows representative projections of low, medium and high cell distributions indicating (arrows) the ten cells chosen. Each cell was checked to be separate from other cells, and to be entirely within the volume scanned.
Computer generated volumes for 10 individual cells (rows) after thresholding of grey scales 1–10 (columns) from an image stack with a low cell infiltrate. The mean Cell Antigen Volume (CAV) for the 10 cells is shown at each threshold, as well as the standard deviation of the Cell Antigen Volume (SD CAV), the number of cells calculated for the image stack at each threshold, and the Total Antigen Volume (TAV) in the image stack at the corresponding threshold value.
Mean CAV (μm3)
SD CAV (μm3)
Number of Cells
TAV (μm3 * 10E6)
Computer generated volumes for 10 individual cells (rows) after thresholding of grey scales 1–10 (columns) from an image stack with a medium cell infiltrate. The mean Cell Antigen Volume (CAV) for the 10 cells is shown at each threshold, as well as the standard deviation of the Cell Antigen Volume (SD CAV), the number of cells calculated for the image stack at each threshold, and the Total Antigen Volume (TAV) in the image stack at the corresponding threshold value.
Mean CAV (μm3)
SD CAV (μm3)
Number of Cells
TAV (μm3 * 10E6)
Computer generated volumes for 10 individual cells (rows) after thresholding of grey scales 1–10 (columns) from an image stack with a high cell infiltrate. The mean Cell Antigen Volume (CAV) for the 10 cells is shown at each threshold, as well as the standard deviation of the Cell Antigen Volume (SD CAV), the number of cells calculated for the image stack at each threshold, and the Total Antigen Volume (TAV) in the image stack at the corresponding threshold value.
Mean CAV (μm3)
SD CAV (μm3)
Number of Cells
TAV (μm3 * 10E6)
Manual leukocyte counts (Mean cell number, SD and range) for projections containing low, medium and high cell density.
Computed leukocyte counts (Mean cell number, SD and range) for projections containing low, medium and high cell density.
Comparison between calculated and manual leukocyte counts in tumors
Comparison of leukocyte counts between tumor and non-malignant tissue
We have designed a method for the enumeration of leukocytes in three dimensions using confocal laser scanning microscopy and image analysis software. The method has several advantages over visual assessment of leukocytes in hematoxylin and eosin stained tissue sections. Specific immunostaining for the leukocyte subtypes in thick tissue sections enables the assessment of the relative numbers of several leukocytes subtypes within a single specimen. Furthermore, refinement of the method by multiple immunostaining using different fluorophore tagged secondary antibodies would show the relative distribution of leukocytes within the same three dimensional volume of tissue. Thick (20 μm) cryosections are less difficult to cut than thin tissue sections (4–5 μm) typically used for histological analysis. Using thick tissue sections, CLSM can optically image thin slices at different focal planes within the tissue, which can be reconstructed to generate a three-dimensional stack. Enumeration of leukocytes in three-dimensional volumes is likely to be more representative of the true distribution of cells in the tissue, since their distribution may vary between image planes. A thin tissue section only represents one image plane for the pathologist.
Image thresholding removes voxels representing background (black) and low level non-specific staining. By choosing an appropriate threshold, the volume of tissue occupied by specific antigen can be calculated and expressed as a percentage of the total volume scanned. Manual threshold selection is subjective and the results may vary if too much or too little threshold is applied. However, in the present application, the mean cell volume is also calculated at the same threshold as the total fluorescence volume. Thus, the calculated number of cells in each image stack remains relatively constant over a number of possible threshold choices (Figure 2). In fact, ten people who had not performed thresholding before chose thresholds between three and seven. When calculated between these thresholds, this represented a variation of only 79 cells (759 – 838) in the image stack with high CD8 cell infiltration. This variation was considerably lower than the 300 cell difference obtained when ten personnel manually counted the cells in a projection generated from the same image stack (716– 1124 cells; mean 924 ± 111). Although the two methods of counting were not significantly different (at 10 threshold levels), there appeared to be a tendency for manual overestimation of cell numbers, which became statistically significant when the calculated results were adjusted between three and seven (the range of thresholds chosen by ten personnel). These results demonstrated a tendency to manually overestimate the number of cells at higher densities in the tissues. At lower levels of cell infiltration the manual counts compared much better. For the twelve tumor samples tested, the two methods were highly correlated for all four antigens tested (Figure 4).
Using the calculated method, we then compared the leukocyte numbers in seven tumors with matched peripheral "non-malignant" tissue resected by the pathologist from the periphery of the tumor. It should be noted that although these tissues were assessed by the pathologist to be non-malignant, they may not necessarily be normal because of their proximity to the respective tumors. All non-malignant tissue sections showed minimal leukocyte numbers indicating a low level of infiltration (Figure 5). A larger variation in leukocyte numbers was observed for all cell types tested in the malignant tissues, but all tumors showed considerably increased leukocyte infiltration compared to their associated non-malignant tissue. The large variability in leukocyte infiltration into tumors may have been due to differences in the type and stage of disease in each case. Since the main purpose of this work was to compare methods of leukocyte enumeration, these factors were not taken into account.
This methodology is not restricted to the image analysis package we used to obtain the data reported in the present work. For example, we have recently adapted the method to ImagePro Plus (Media Cybernetics, Silver Spring, MD) with its plugin 3D Constructor. Many other commercially available and free packages such as NIH Image and Image J are available which will adequately perform thresholding, seeding and volume calculations without the need for special programming. In addition, three-dimensional image stacks can also be obtained from less expensive deconvolution or stereoscopic based microscopy systems which, although more time consuming, may be better suited to the budget of smaller pathology units.
These data clearly demonstrate that enumeration of leukocytes using image analysis is a robust and rapid methodology which can be performed by relatively inexperienced laboratory staff. In addition, three dimensional image analysis overcomes the tendency to overestimate the number of cells in tumors with a high degree of leukocyte infiltration. Data obtained from clinical samples may help clinicians to establish the extent and type of immune response ongoing in the tumor, and thereby assist in both diagnosis and therapeutic strategy for individual patients.
Tumors and matched non-malignant tissues
For the development of thresholding and counting methodology, twelve freshly resected tumors comprising vulvar carcinoma (1); peritoneal carcinoma (2); endometrial carcinoma (2); neuroblastoma (1) and ovarian carcinoma (6) were obtained from the histopathology department at Florida Hospital in 30 ml universal containers (Barloworld Scientific, Staffordshire, UK) containing 10 ml RPMI 1640 tissue culture medium (Gibco, Grand Island, NY). The tissues were then transferred to 15 ml cryostorage vials (Nalge Company, Rochester, NY) and frozen in liquid nitrogen until required.
For experiments comparing malignant and non-malignant leukocyte infiltration, freshly resected tissue was obtained from seven patients. Representative (as assessed by a pathologist) pieces (1–2 cm) of endometrial carcinoma (1); ovarian adenocarcinoma (5) and neuroblastoma (adrenal; 1) and matched peripheral non-malignant tissue were obtained and treated as described above.
Primary mouse anti-human monoclonal anti CD3 (IgG1), CD4 (IgG1), CD8 (IgG1), CD14 (IgG2a), and isotypic antibodies (murine IgG1 and IgG2a, were obtained from Beckman Coulter (Miami, FL). Oregon Green-labeled goat anti-mouse IgG (secondary antibody) was purchased from Invitrogen (Eugene, OR).
Sectioning and immunofluorescence staining
Tumors were removed from liquid nitrogen, placed in the cryostat, left for 30 minutes to equilibrate at -20°C, placed onto cryostat tissue holders and embedded in mounting medium (CRYOform, International Equipment Company, Needham, MA). Serial 20 μm tissue sections were cut, placed on microscope slides, and allowed to air dry. Tissue sections were fixed in acetone/methanol (1:1) for 5 min, and washed three times using phosphate buffered saline (PBS; Gibco), with 5 minutes between washes. The slides were incubated with normal goat serum (10% in PBS; 100 μl vol) for 30 minutes, and washed in PBS as described above. One hundred microliters (containing 1 μg IgG) of primary antibody (CD3, CD4, CD8 or CD14) or isotypic IgG were applied to appropriate slides for one hour at room temperature in a humidified chamber. The slides were again washed three times with PBS and 100 μl of secondary antibody (containing 1 μg IgG) added to all slides except the blank (autofluorescence control), which was left in PBS. After a further three washes in PBS, coverslips were attached to the slides using Aqua-mount (Lerner Laboratories, PA).
Confocal laser scanning microscopy
Slides were examined using a Multiprobe 2010 confocal laser scanning microscope (CLSM; Molecular Dynamics, Sunnyvale, CA), equipped with a Nikon Plan-Apo 20x (NA 0.75) air objective. Initially, slides were checked for specific antigen staining by comparing them to their isotype, second antibody, and autofluorescence controls. The photomultiplier tube (PMT) of the CLSM was set to just include light from the appropriate isotype control. Using the same PMT setting, five areas of tissue on each antibody labeled slide were serially scanned for Oregon Green fluorescence (areas were chosen using the halogen lamp alone to ensure that areas of high or low fluorescence were not inadvertently "selected"). For each field of view, a series of twenty serial optical slices, 0.6 μm apart, were scanned (i.e. 100 optical slices per slide). Data was processed on a Silicon Graphics workstation for image analysis. After image analysis, images and data were transferred to CD for storage.
Image analysis techniques
Image thresholding and masking
Calculation of antigen volume
After thresholding, the number of voxels containing data of interest (i.e. specific antibody fluorescence) can be calculated as a volume in the same way that the total volume scanned was calculated, giving the relative proportion of the total volume occupied by antigen according to the following:
(Total antigen volume/total volume scanned) * 100%
The percentage of tumor volume occupied by antigen is a semi-automated method for providing the pathologist with data regarding the antigen content in tumors, as previously described .
Seeded region-growing segmentation analysis
Having ascertained that the cell is contained within the scanned volume, a mark or seed is placed on any part of the stained cell (using a mouse pointer) and the segmentation software calculated the volume of all contiguous voxels of fluorescence above the threshold. The region-growing stops when the threshold grey scale is reached. Figure 7B (left) shows the single cell chosen in Figure 7A. After seeding the cell (right) is falsely colored to show the extent of the region growth, and the cell antigen volume is automatically calculated.
This procedure was performed on every cell seeded in this study.
Enumeration of leukocyte infiltration
To enumerate leukocytes, ten isolated cells in each field of view were seeded and the mean of their volumes recorded. The total number of cells/field was determined by: mean cell antigen volume/total antigen volume
Composite projections used for manual counting
Using Imagespace™ software (Molecular Dynamics, Sunnyvale, CA), a look-through projection (i.e. the twenty optical slices were reconstructed to produce an image which appears to have been obtained from a transparent specimen by a lens with a large depth of focus) of each image stack was generated for each field of view. Look-through projections average the voxel intensities of exactly registered (directly in-line vertically through the stack) voxels from all slices in the image, and interpolate the data from the spaces within the slices. These processes generate an image which appears to be a composite of all the slices in the stack. The projections were printed and the cells manually counted.
Comparison between calculated and manual cell counts in tumors
To compare calculated and manual cell counts in tumor tissue sections, ten cells were chosen from each slide which were entirely within the volume scanned, and the mean cell volume determined by seeding, as described above. The number of cells in each image stack was then calculated by dividing the total antigen volume by the mean cell volume. Using Imagespace™ software, the volume of tissue in each image stack was calculated. Since all scanning conditions were kept constant, a constant volume of tumor was scanned in each image stack (5.35 × 106 μm3). A threshold was set to exclude zero fluorescence (black) and any residual background staining, and the total volume of scanned tumor occupied by stained cells was determined.
A look-through projection of each image stack was also printed and the cells manually counted. The manual counts were compared to the computer generated cell numbers using Student's t-test for dependent variables.
Effect of threshold value on cell count
To determine the effect of threshold level on cell count, three image stacks were selected (two CD3 and one CD8 series) which contained low (relatively few, well separated cells), moderate (well separated cells and some cell clumps) and high (few clearly separated cells and many clumps) cellular density. Ten cells were chosen from each image stack which were entirely within the volume of tissue scanned, and their volumes determined at 10 threshold levels from grey scale 1 to grey scale 10, using ImageSpace™ software. The total tumor volume occupied by fluorescent antibody was also calculated at each threshold. The total number of cells in the image stack was then calculated by dividing the mean cell volume (of the ten cells) into the total antigen volume at each threshold.
Manual cell counts
Look-through projections were printed, and the number of cells in each projection manually counted. To assess operator variability for manual counting the printed projections were counted independently by 10 laboratory personnel. The two counting methods were then compared using a Student's t-test for dependent variables.
This work was funded by a generous donation by Florida Hospital Endowed Program in Cancer Research.
- Kuzushima K, Nakamura S, Nakamura T, Yamamura Y, Yokoyama N, Fujita M, Kiyono T, Tsuruni T: Increased frequency of antigen-specific CD8(+) cytotoxic T lymphocytes infiltrating an Epstein-Barr virus-associated gastric carcinoma. J Clin Invest. 1999, 104: 163-171.PubMed CentralView ArticlePubMedGoogle Scholar
- Hokland M, Kjaergaard J, Kuppen PJ, Nannmark U, Agger R, Hokland P, Basse P: Endogenous and adoptively transferred A-NK and T-LAK cells continuously accumulate within murine metastases up to 48 h after inoculation. In Vivo. 1999, 13: 199-204.PubMedGoogle Scholar
- Takanami I, Takeuchi K, Kodaira S: Tumor-associated macrophage infiltration in pulmonary adenocarcinoma: association with angiogenesis and poor prognosis. Oncology. 1999, 57: 138-142. 10.1159/000012021.View ArticlePubMedGoogle Scholar
- Ma D, Gu MJ: Immune effect of tumor-infiltrating lymphocytes and its relation to the survival rate of patients with ovarian malignancies. J Tongji Med Univ. 1991, 11: 235-239.View ArticlePubMedGoogle Scholar
- Eisenthal A, Polyvkin N, Bramante-Schreiber L, Misonznik F, Hassner A, Lifschitz-Mercer B: Expression of dendritic cells in ovarian tumors correlates with clinical outcome in patients with ovarian cancer. Hum Pathol. 2001, 32: 803-807. 10.1053/hupa.2001.26455.View ArticlePubMedGoogle Scholar
- Kondratiev S, Sabo E, Yakirevich E, Lavie O, Resnick MB: Intratumoral CD8+ T lymphocytes as a prognostic factor of survival in endometrial carcinoma. Clin Cancer Res. 2004, 10: 4450-4456. 10.1158/1078-0432.CCR-0732-3.View ArticlePubMedGoogle Scholar
- Matsushita N, Ghazizadeh M, Konishi H, Araki T: Association of ovarian tumor epithelium coexpressing HLA-DR and CA-125 antigens with tumor infiltrating cytotoxic T lymphocytes. J Nippon Med Sch. 2003, 70: 40-44. 10.1272/jnms.70.40.View ArticlePubMedGoogle Scholar
- Raspollini MR, Castiglione F, Rossi Degl'innocento D, Amunni G, Villanucci A, Garbini F, Baroni G, Taddei GI: Tumour-infiltrating gamma/delta T-lymphocytes are correlated with a brief disease-free interval in advanced ovarian serous carcinoma. Ann Oncol. 2005, 16: 590-596. 10.1093/annonc/mdi112.View ArticlePubMedGoogle Scholar
- Ikarashi Y, Matsumoto Y, Omata S, Fujiwara M: Recipient-derived T cells participate in autoimmune-like hepatic lesions induced by graft-versus-host reaction. Autoimmunity. 1995, 20: 121-127.View ArticlePubMedGoogle Scholar
- Haskill S, Koren H, Becker S, Fowler W, Walton L: Mononuclear-cell infiltration in ovarian cancer. III. Suppressor-cell and ADCC activity of macrophages from ascitic and solid ovarian tumours. Br J Cancer. 1982, 45: 747-753.PubMed CentralView ArticlePubMedGoogle Scholar
- Shen Z, Lipowsky HH: Image enhancement of the in vivo leukocyte-endothelium contact zone using optical sectioning microscopy. Ann Biomed Eng. 1997, 25: 521-535.View ArticlePubMedGoogle Scholar
- Distel M, Babaryka A, Koster RW: Multicolor in vivo time-lapse imaging at cellular resolution by stereomicroscopy. Dev Dyn. 2006, 235: 1100-1106. 10.1002/dvdy.20694.View ArticlePubMedGoogle Scholar
- Foschini MP, Righi A, Cucchi MC, Ragazzini T, Merelli S, Santeramo B, Eusebi V: The impact of large sections and 3D technique on the study of lobular in situ and invasive carcinoma of the breast. Virchows Arch. 2006, 448: 256-261. 10.1007/s00428-005-0116-y.View ArticlePubMedGoogle Scholar
- Zhou L, Yoshimura Y, Huang Y, Suzuki R, Yokoyama M, Okabe M, Shimamura M: Two independent pathways of maternal cell transmission to offspring: through placenta during pregnancy and by breast-feeding after birth. Immunology. 2000, 101: 570-580. 10.1046/j.1365-2567.2000.00144.x.PubMed CentralView ArticlePubMedGoogle Scholar
- Biggerstaff J, Amirkhosravi A, Francis JL: Three-dimensional visualization and quantitation of fibrin in solid tumors by confocal laser scanning microscopy. Cytometry. 1997, 29: 122-127. 10.1002/(SICI)1097-0320(19971001)29:2<122::AID-CYTO4>3.0.CO;2-F.View ArticlePubMedGoogle Scholar
- Naser SA, Shafran I, Schwartz D, El Zaatari F, Biggerstaff J: In situ identification of mycobacteria in Crohn's disease patient tissue using confocal scanning laser microscopy. Mol Cell Probes. 2002, 16: 41-48. 10.1006/mcpr.2001.0395.View ArticlePubMedGoogle Scholar
- Shaner NC, Campbell RE, Steinbach PA, Giepmans BN, Palmer AE, Tsien RY: Improved monomeric red, orange and yellow fluorescent proteins derived from Discosoma sp. red fluorescent protein. Nat Biotechnol. 2004, 22: 1567-1572. 10.1038/nbt1037.View ArticlePubMedGoogle Scholar
- Gonzalez RC, Woods RE: Image enhancement in the spatial domain. Digital Image Processing. Edited by: Gonzalez RC, Woods RE. 2002, Upper Saddle River, NJ Prentice Hall, Inc, 77-Google Scholar