Analytic studies of foam cells from breast cancer precursors

(1)

Analytic

Studies of

Foam

Cells

From

Breast

Cancer Precursors'

Eileen B. King, Lois

K. Kromhout, Karen

L. Chew, Brian

H. Mayall, Nicholas

L. Petrakis,

Ronald H. Jensen, and Ian T. Young

Departments of Pathology and of Epidemiology and International Health (E.B.K., L.K.K., K.L.C., N.L.P.), and Program for Analytical Cytology (B.H.M.), University of California, San Francisco, California 94143, Biomedical Sciences

Division, Lawrence Livermore National Laboratory, Livermore, California 94550 (B.H.M., R.H. J.), and Pattern Recognition Group, Department of Applied Physics, Technical University Delft, Delft, The Netherlands (I.T.Y.)

Received for publication January 27, 1983; accepted October 10, 1983.

A preliminary study of foam cells from nipple aspirate fluid demonstrated the ability of image analysis to discriminate categories of breast disease. Foam cell images numbering 471 were collected from nipple aspirate samples representing three to six cases of each of the four following disease categories based on breast tissue diagnosis: benign, nonproliferative; hyperplasia; atypical hyperplasia; and cancer. Twenty-two shape and density parameters were measured for each cell image. Us- ing multivariate analysis, eight nuclear and three cytoplasmic parameters showed significant differences ( P

<

0.005) when tested among cell populations from the breast disease categories. Linear stepwise discriminant analysis enabled construction of a three-pa-

rameter model that was optimal for distinguishing among cell populations from the four categories of breast disease. The means of all twenty-three parameters were then evaluated on a per-patient basis. A second three-parameter model was constructed that distin- guished, with 100% accuracy, patients with proliferative disease from those with nonproliferative disease. Grouping disease categories and comparing patients whose diagnosis was benign or hyperplasia versus atypical hyperplasia or malignant, the model placed patients in the correct group 83% of the time.

Key terms: Foam cells, breast disease, image cytometry

Breast fluid can be obtained from most nonlactating women by applying slight negative pressure to the nipple with a simple aspirating device (20,231. Using this technique, fluid can be obtained from as many as 79% of Caucasian women during their reproductive years. A somewhat lower proportion of women yield fluid when all races and ages are included. Because the cells of nipple aspirate fluid (NAF) may reflect the disease status of the breast, several authors have proposed cytologic screening of the fluid to identify malignant and premalignant breasts (3,231.

Nipple aspirate fluid contains epithelial cells, foam cells, and hematogenous cells (9,10,14). The duct epithelial cells in

NAF'

may show morphologic changes related to atypical proliferative changes in the breast that are considered to be precursors to breast cancer (22,24). We have found that abnormal duct cells in NAF are associated with increased breast cancer risk (11-13,19). Infre- quently, we have found frankly malignant cells in NAF.

Although the presence of abnormal duct cells has value

in identifying women at risk for breast cancer, most nipple aspirate fluids do not contain a s a i c i e n t number of duct cells for evaluation. A review of 2,008 nipple aspirate specimens in a n earlier study (15) showed that only 59% of specimens contained ten or more single duct cells or cell groups. In contrast, foam cells were present in 88% of the 2,008 specimens, and make up a majority of the cells in most NAF'.

Foam cells are relatively large cells with abundant foamy cytoplasm and a single central or slightly eccen- tric round or oval nucleus with a distinct nucleolus. There are occasional cells with two or more nuclei. Chro- matin is uniformly granular and the chromatinic membrane is distinct.

'Presented a t the Combined International Conference on Analytical Cytology and Cytometry I X and the VIth International Symposium on Flow Cytometry, Elmau, Bavaria, October 18-23, 1982.

Address reprint requests to Dr. E.B. King, Department of Epidemiol- ogy and International Health, 1699 HSW, University of California, San Francisco, CA 94143.

(2)

ANALYSIS O F BREAST FOAM CELLS 125 The origin of the foam cell is uncertain. There is evi- derived from ductal lobular epithelium, it might be ex- dence for epithelial as well as bone marrow derivation pected to have the same pathologic changes.

(4-6). The rationale for a duct epithelial origin is based We have been unable to distinguish morphologic dif- on the intimate association of foam cells with duct epi- ferences in foam cells from benign and malignant breasts thelium in tissue (Fig. lA,B) and on the similarity of the using conventional cytologic criteria; however, we

foam cell nuclear structure to that of duct epithelial cells thought image cytometry might be able to detect signif- by light and electron microscopy. If the foam cell is icant differences between foam cells from benign and

FIG. 1. A, B. Foam cells are seen in duct lining epithelium and form the preponderance of cells in the lumen. X40. B. Higher magnification

of foam cells in epithelium. x250. Note the intimate admixture of epithelial and foam cells within the duct lining.

(3)

malignant breasts by measuring morphologic features not appreciated by the human eye.

Other investigators have used image cytometry successfully in similar cell and disease discrimination prob- lems. Bibbo et a1 (l), found that parameters describing nuclear staining and chromatin granule densities in “benign appearing” intermediate squamous cells were useful in classification of cervical disease. A similar study by Pahlplatz et a1 (18) has recently confirmed this finding. Nuclear chromatin granularity and texture have been explored quantitatively for classification of white blood cells (2) and various types of leukemic blast cells (7).

Image cytometry applied to foam cells in NAF could provide a n objective and reproducible measure of changes in these cells. If foam cells truly reflect different types of benign and malignant breast disease, the value of NAF for cytologic diagnosis of the breast could be greatly improved with use of image cytometry.

In the present study, we examine the utility of image cytometry for the analysis of foam cells in NAF. We extract and measure a number of cellular features. These measures are able to differentiate, with encouraging accuracy, the disease status of the donor breast.

MATERIALS AND

METHODS

Nipple aspirate fluid was obtained from women who underwent breast examinations between 1974 and 1980 at breast-screening clinics a t the University of Califor- nia, San Francisco, and the Breast Screening Center of Northern California, Oakland, California. The specimens were collected in suspension and prepared on Mil- lipore@ filters, Papanicolaou-stained, and routinely screened for the presence of various cell types and cell changes. Cytologic diagnosis was based on changes in duct epithelial cells.

Breast biopsy tissue, obtained sometime later from the women who had supplied NAF, was reviewed without reference to the cytologic findings in fluid. The tissue diagnosis was based on changes in ductal lobular epithelium, and this histologic diagnosis was taken to reflect

the “true” status of breast disease. The histologic classifications were grouped as benign (nonproliferative), hyperplasia (without atypia), atypical hyperplasia (premalignant), and carcinoma. We selected 18 cases for analysis, three each from the six groups of NAF and breast tissue classifications (see Table 1). Initial analysis was done on the first four groups in which the conventional cytologic diagnosis agreed with the tissue biopsy diagnosis. We then applied the same procedures to the last two groups of NAF samples that had no diagnostic duct epithelial cells.

The Papanicolaou-stained NAF’ specimens were scanned using the ACUity quantitative microscope system a t Lawrence Livermore National Laboratory (25). The system consists of a Leitz (Rockleigh, NJ) microscope configured for absorption microscopy, a Sierra Sci- entific (Mountain View, CA) television camera, a Quantex DS/20 Image Digitizer (Sunnyvale, CA) and an LSI 11/02 System (Digital, Maynard, MA) that coordi- nates the components with a menu-driven, interactive program written in Pascal. Images with a diffraction- limited spatial resolution approaching 0.30 pm are achieved with a 4 0 ~ oil-immersion objective (N.A. = 1.0) and a 1 6 ~ ocular (Fig. 2). The image of the cell is scanned by the television camera and is digitized to 256 grey levels corresponding with a sampling density of eight points per micron in the object.

Analysis of the cell image begins by determining its nuclear and cytoplasmic boundaries. The operator views the stored image and selects interactively those threshold values that best define the nuclear and cytoplasmic borders. Further processing of the nuclear image parti- tions it into dark, average, and light regions of chromatin. This is done automatically by partitioning of the nuclear histogram so that points whose brightness is less than 0 . 8 ~ the mean nuclear brightness are assigned dark, points greater than 1.2 x the mean nuclear brightness are assigned light, and intermediate points are assigned average.

In the second step of the image analysis, 12 nuclear parameters, nine cytoplasmic parameters, and one nu-

Table 1

Classification of Cases Selected for Image Cytometry and Numbers of Foam Cells Analyzed

Group cytodiagnosis diagnosis Cases (cells per case) Total

1 Benign Benign 3 27 27 27 81

2 Hyperplasia Hyperplasia 3 23 21 28 72

3 Atypical Atypical 3 28 28 29 85

NAF Breast tissue NAF foam cells

nonproliferative hyperplasia hyperplasia 4 Malignant Carcinoma 3 40 30 23 93 Subtotals 12 331 5 Benign Atypical 3 27 17 26 70 6 Benign Carcinoma 3 26 25 19 70 Subtotals 6 140 hyperplasia

(4)

127 ANALYSIS OF BREAST FOAM CELLS

FIG. 2. Grey level display of the digitized images of foam cells. Cells are from ( 1 ) benign nipple aspirate fluid (NAF) and benign breast tissue, (2) atypical hyperplasia NAF and atypical hyperplastic breast tissue, (3) malignant NAF and carcinoma in tissue, (4) benign NAF and carcinoma in tissue.

clear-to-cytoplasmic ratio are measured for each cell. These parameters are given in Table 2 and are described by Young et a1 (25,261.

Some parameters were developed specifically to relate to conventional cytology, ie, the algorithms were written to describe nuclear and cytoplasmic features (such as size and darkness) that a cytologist uses to make a diagnosis. Other parameters reflect a heuristic approach to probing cell properties, (eg, shape) with no a priori knowledge of what would prove useful for diagnosis. The parameters used for nuclear chromatic analysis are fine

ness, condensity, heterogeneity, and margination. Chro- matin fineness and condensity are slightly different measures derived from the size distribution of the black and white regions. Chromatin heterogeneity is estimated from the sum of the areas of dark and light relative to the whole nucleus. Chromatin margination is estimated from the relative radial distribution of gray level values about the center of mass. Traditionally the cellular diagnosis of cancer and precancerous lesions depends upon nuclear aberrations and especially the chromatin morphology-the size, shape, and distribution of granules. The size of chromatin particles or clumps is measured by condensity and fineness, the irregularity of chromatin particles or clumps is measured by heterogeneity, and chromatin distribution at the nuclear boundary is measured by margination.

Slides were screened systematically and the first 30 to

50 intact foam cells with complete nuclear and cytoplasmic borders were measured from each case. Foam cells numbering 575 were selected from the first four groups. Satisfactory nuclear and cytoplasmic boundaries were determined on 331 cells by the operator adjusting the threshold. The other 244 cell images were rejected because a n acceptable boundary could not be thresh- olded. For the cytoplasmic boundary, this occurred with cells that were slightly out of focus or came too close to another object; for the nuclear boundary, this occurred when nuclear borders were distorted by a vacuole or had a break in the chromatinic membrane or when there were dense particles in the cytoplasm. There were 21 to Table 2

Parameters of Foam Cell Analysis

Nuclear/cytoplasmic Nuclear parameters Cytoplasmic parameters parameters Stain Total stain

Size Area Average stain Perimeter Perimeter2/area Bending energy Shape Sphericity Total Stain Average stain

Area Ratio of nuclear

Perimeter area to total

Perimeter2/area Sphericity Bending energy Eccentricity

Mean absolute curvature

cell area

Texture Chromatin heterogeneity Chromatin clumping Chromatin condensity Chromatin fineness Chromatin margination

(5)

128

40 cells per case and 72 to 93 cells per category for analysis (Table 1).

Parameters showing significant differences among the groups were selected by multivariate analysis. Linear stepwise discriminant analysis was applied to the mean of each parameter for all cells in each disease category, and to the means of each parameter on a per case basis using the unbiased jackknife classification procedure (8,161.

The same analysis was applied to 140 foam cells from groups 5 and 6, which showed benign NAF cytology in association with atypical proliferative disease or carcinoma. The distribution of cells analyzed from these two groups is also shown in Table 1.

RESULTS

The first experiments analyzed foam cell data from NAF in which the NAF and tissue diagnosis agreed (Groups 1, 2, 3, and 4). The individual measurements from these groups showed a high degree of overlap; this suggested that it would be impractical to classify individual cells according to the disease status of the breast. However, significant effects were seen when all the data for each group were pooled and the group means for each parameter were analyzed. Although there are only a few cases in each group, pooled data per individual case also appear significant.

Analysis of the data from all of the parameters re- vealed eleven parameters with discriminant power ( P

<

0.005) for distinguishing hyperplasia, atypical hyperplasia, and malignant groups from the benign group. These parameters included eight nuclear parameters (area, pe- rimeter, stain, average stain, heterogeneity, condensity,

margination, and fineness) and three cytoplasmic pa-

rameters (perimeter, bending energy, and mean absolute curvature).

Linear stepwise discriminant analysis was applied to the group mean of each parameter for each of these four disease groups. Three parameters (heterogeneity, mar-

gination, and nuclear average stain) were selected as the

most useful for distinguishing among the breast disease groups.

We next evaluated the means of each of the cell parameters on a per-case basis. Linear stepwise discriminant analysis was applied to these case means, and the jackknife classification procedure was used. In this procedure, each case is removed and then classified by a discriminant function based on the remaining cases.

Chromatin condensity, fineness, and margination were

three consistently useful parameters in the stepwise discriminant analysis of case means. Using these three parameters, classification accuracy was 100% (Table 3-11 when the breast disease categories were grouped as proliferative compared with nonproliferative. When the breast disease categories were grouped as one without significant disease (benign and hyperplasia) and the other with significant disease (atypical hyperplasia and carcinoma) (six patients in each group), the parameters accurately identified five (83%) of those cases without significant disease and five (83%) of those cases with significant disease (Table 3-11).

The benign nonproliferative group was removed from the analysis and the remaining proliferative disease was subclassified according to presence and absence of significant disease. Two of three cases (67%) could be classified as having no significant disease and four of six (67%) could be classified as belonging to the atypical hyperplasia and carcinoma group (Table 3-111).

The final experiment analyzed data from the six cases in which conventional NAF cytology had failed to diag- nose significant disease (groups 5 and 6). The results of the jackknifed classification comparing the six “NAF negative” atypical hyperplasia and carcinoma cases with the three benign cases are shown in Table 3-IV. A classification accuracy of 100% was achieved. This last and most important comparison grouping provides addi- tional confirmation that breast disease discrimination may be based on foam cells alone.

DISCUSSION

Our most important discovery was that ordinary foam cells in NAF can be used for diagnosis of breast disease when the cells are analyzed by image cytometry. This analytical approach has the added advantage of objectiv- Table 3

Discrimination With Jackknife Classification (Fraction of Cases Classified Correctly (Unbiased Estimates])

Group 1 2 3 Diagnosis Cases Benign 3 Hyperplasia 3 Atypical 3 Carcinoma 3 Atypical 3 Carcinoma 3 hyperplasia hyperplasia Experiments I I1 I11 IV 313 313 (1 vs 5,6) ( 1 vs 2-4) (1,2 vs 3,4) (2 vs 3,4) 516 919 516 213 416 4 5 6 616

(6)

ANALYSIS OF BREAST FOAM CELLS ₁₂₉ ity and reproducibility when compared with conven-

tional diagnostic cytology. In the formerly “unsatisfac- tory” NAF lacking in duct cells, cytometry of foam cells alone can be used successfully in the assessment of breast disease status.

The parameters that we found to be important in discrimination of benign, premalignant, and malignant breast disease are related to chromatin granule size and distribution (fineness, condensity, heterogeneity, and

margination) and density (nuclear average stain). These parameters correlate particularly well with morphologic features that are of diagnostic importance in conventional cytology of duct epithelial cells. Chromatin changes are considered among the most reliable features for recognition of malignancy and premalignant disease in clinical cytology. The size and distribution of chromatin granules are assessed and compared subjec- tively with chromatin in nuclei of normal cells. Hyper- chromatic, coarsely granular, irregularly distributed chromatin reflects a n abnormal epithelial nucleus. Therefore, it is reasonable that the parameters that measure these same features should be the best dis- criminators between disease categories in our study of foam cells.

The importance of shape and texture measures in image cytometry has also been noted by other investigators. Pressman (21) discriminated between benign and malignant cervical epithelial cells using only texture measures derived by Markovian analysis. Bins et a1 (2) were able to discriminate the five classes of normal human leukocytes using only texture measures based on the profile of particle count per nucleus as a function of threshold. Gunzer et a1 (7) differentiated malignant blast cells from four types of leukemias by utilizing a texture measure that estimated chromatin fineness by analyzing local gradients within the nuclear image.

Only a few investigators have explored the possibility that apparently benign epithelial cells might contain morphologic markers of malignant or premalignant disease. Niebergs (17) used conventional microscopy and found “malignancy associated change” in benign cells from many anatomic sites in patients with malignant disease. His criteria were not well understood and few could duplicate his results. Bibbo et a1 (1) used image cytometry to analyze well-differentiated, apparently normal intermediate squamous epithelial cells from women with malignant or dysplastic lesions of the uterine cervix. They extracted many different features from three color images. Significant differences were found when the values of these features were compared with the values for cells from healthy women. Pahlplatz et a1 (18) using textural parameters recently confirmed that image cytometry could discriminate intermediate squamous cells of the normal or dysplastic and malignant uterine cervix.

Although our data is limited, the trends suggest that image analysis restricted to the foam cells in NAF can become a n accurate and reliable method for the diagnosis of benign, premalignant, and malignant breast dis-

ease. This will increase the value of NAF cytology because the proportion of women who can be diagnosed in this way will be greater than for conventional cytology; NAI? usually contains adequate numbers of foam cells but less frequently contains adequate numbers of duct epithelial cells. Whether the foam cell is a mononuclear-phagocyte or a n epithelial cell, it apparently contains markers for benign, premalignant, and malignant breast disease. The techniques of image cytometry we have used have defined these markers as chromatin changes.

We anticipate that image cytometry of foam cells in NAF will offer an improved cytologic method for identifying women a t high risk for breast carcinoma. An accurate method for diagnosis of premalignant disease would provide a strong indication for close surveillance of this group of women. In the future, preventive inter- ventions may become available, and a definitive early diagnosis of premalignant breast disease could lead ul- timately to a reduction in deaths from breast cancer.

ACKNOWLEDGMENTS

We thank Robert Schweitzer, M.D., for access to women attending the Breast Screening Center of Northern Califor- nia, Merritt Hospital, Oakland, California.

LITERATURE CITED

1. Bibbo M, Bartels PH, Sychra J J , Wied GL: Chromatin appearance in intermediate cells from patients with uterine cancer. Acta Cytol 25:23-28, 1981.

2. Bins M, Landeweerd GH, Gelsema ES, van Montfort LH, Halie M R Texture of white blood cells expressed by the counting densi- togram. Cytometryl:321-324, 1981.

3. Buehring GC: Screening for breast atypias using exfoliative cytology. Cancer 43:1788, 1979.

4. Davies JD: Human colostrum cells: their relation to periductal mononuclear inflammation. J Pathol 112:153-160, 1974. 5. Davies JD: Periductal foam cells in benign mammary dysplasia. J

Pathol 117:39-45, 1975.

6. Gaffney EV, Polanowski FP, Burke RE: The isolation and cultiva- tion of cells from human milk (Abstract). Committee of the Breast Cancer Task Force #3, March 3, 1976, Bethesda, Maryland. 7. Gunzer U, Harms H, Haucke M, Aus HM, ter Meulen V: Com-

puter-aided image analysis for the differentiation of mononuclear cells in peripheral blood smears from leukemic patients. Anal Quant Cytol 3:26-32, 1980.

8. Hill MA: BMDP User’s Digest. A Condensed Guide to the BMDP Computer Programs, 2nd Ed. BMDP Statistical Software, Los Angeles, 1982’.

9. King EB, Barrett D, King M-C, Petrakis NL: Cellular composition of the nipple aspirate specimen of breast fluid. I. The benign cells. Am J Clin Pathol64:728-738, 1975.

10. King EB, Barrett D, Petrakis NL: Cellular composition of the nipple aspirate specimen of breast fluid. 11. Abnormal findings. Am J Clin Pathol64:739-748,1975.

11. King EB, Chew KL, Barrett DL, Ernster VL, Petrakis NL: Abnor- mal nipple aspirate cytology and proliferative changes in benign breast disease. Acta Cytol 25:450, 1981.

12. King EB, Chew KL, Ernster VL, Barrett DL, Petrakis NL: Abnor- mal nipple aspirate cytology and proliferative changes associated with malignant breast disease. Acta Cytol 25:721, 1981.

13. King EB, Chew KL, Petrakis NL, Ernster VL: Nipple aspirate cytology for the study of breast cancer precursors. JNCI (in press). 14. King EB, Zimmerman AL, Barrett DL, Petrakis NL, King M-C: Cytopathology of abnormal mammary duct epithelium. In: Preven- tion and Detection of Cancer, Part 11: Detection, Vol. 2, Cancer

(7)

130

Detection in Specific Sites, Nieburgs HE (ed). Marcel Dekker, 1980, pp 1831-1845.

15. King EB, Kromhout LK, Chew KL, Mayall BH, Petrakis NL: Foam cell measurements may be useful in diagnosis of breast cancer, Abstracts from the Combined International Conference on Analytical Cytology and Cytometry IX and the VI International Symposium on Flow Cytometry. Schloss Elmau, Mittenwald, Ba- varia, West Germany, October 18-23, 1983, p 111.

16. Lachenbruch PA: Discriminant Analysis. Hoffner Press, New York, 1975.

17. Nieburgs HE: Recent progress in the interpretation of malignancy associated changes (MAC). Acta Cytol 12:445-453,1968.

18. Pahlplatz MM, Oud PS, Hermkens HG, Nooijs GB, Herman CJ: Characterization of intermediate cells in normal and abnormal cervical smears. Abstracts from the Combined International Con- ference on Analytical Cytology and Cytometry IX and VI Interna- tional Symposium on Flow Cytometry. Schloss Elmau, Mittenwald, Bavaria, West Germany, October 18-23, 1982, p 154.

19. Petrakis NL, Ernster VL, King EB, Sacks ST: Epithelial dysplasia in nipple aspirates of breast fluid: Association with family history and other breast cancer risk factors. J Natl Cancer Inst 68:9, 1982.

20. Petrakis NL, Mason L, Lee R: Association ofrace, age, menopausal status, and cerumen type with breast fluid secretion in nonlactating women, as determined by nipple aspiration. J Natl Cancer Inst 54:829, 1975.

21. Pressman NJ: Markovian analysis of cervical cell images. J. His- tochem Cytochem 24:138-144,1976.

22. Rogers LW, Page D L Epithelial proliferative disease of the breast- A marker of increased cancer risk in certain age groups. Breast 5:2, 1979.

23. Sartorius OW, Smith HS, Morris P, Benedict D, Friesen L: Cyto- logic evaluatioin of breast fluid in the detection of breast disease. J Natl Cancer Inst 59:1073, 1977.

24. Wellings SR, Jensen HM, Marcum RG: An atlas of subgross pathology of the human breast with special reference to possible precancerous lesions. J Natl Cancer Inst 55:231, 1975.

25. Young IT, Gledhiil BL, Lake S, Wyrobeck A: Quantitative analysis of radiation-induced changes in sperm morphology. Anal Quant Cytol 4:207-216, 1982.

26. Young IT, Vanderlaan M, Kromhout LK, Jensen RH, Grover A, King EB: Morphologic changes in rat urothelial cells during car- cinogenesis 11: Image cytometry. Cytometry (submitted).