ABSTRACT The extracellular matrix (ECM) plays a critical role in directing cell behaviour and morphogenesis by regulating gene expression and nuclear organization. Using non-malignant human mammary epithelial cells (HMECs), it was previously shown that ECM-induced morphogenesis is accompanied by the redistribution of nuclear mitotic apparatus protein (NuMA) from a diffuse pattern, in proliferating cells,  to a multi-focal (punctate) pattern as HMECs growth arrested and completed morphogenesis. When these cells were cultured as monolayers on plastic, NuMA distribution was diffuse in proliferating HMECs, while it appeared slightly aggregated upon induction of growth-arrest. Interestingly there was no visual difference in NuMA distribution between proliferating non-malignant and malignant HMECs. Here we present a novel model-based image analysis algorithm which quantifies the punctateness of NuMA and allows clear distinction not only between growth arrested and proliferating non-malignant cells but also between proliferating non-malignant and malignant cells, cultured as monolayers. Cell cultures were imaged in 3D using confocal microscopy, for fluorescently labeled NuMA, Ki-67 and DNA. Nuclear segmentation, based on the DNA staining, allowed image analysis of NuMA staining within individual nuclei. Ki-67 staining was used to identify cells in the cell cycle. The image analysis algorithm was based on a multi-scale Gaussian blurring method and measured intensity variations within each nucleus. Averaging results over cells in each population resolved significant, yet, sub-visual differences in NuMA punctateness. Non-malignant growth arrested cells were most punctate, non-malignant proliferating cells produced intermediate values and malignant cells were the least punctate. This ability to discern cell phenotype based on quantifying the spatial distribution of a nuclear protein has broad application in furthering fundamental understanding of biological processes. INTRODUCTION The ability of a cell to selectively express genes depends not only on genetic makeup but a complicated network of exquisitely regulated signaling pathways. On the one hand, genetic makeup is a constant across the organism whereas signaling pathways depend locally on cell environment. This has many conclude that spatial organization of cells within a tissue plays a superior role to the genome in determining phenotype [Bissell et al 1999, Wang et at 1998]. Nowhere is the importance of proper spatial organization more clearly demonstrated than the association of the progression of a tissue to malignancy with the loss of tissue organization at the cellular level.  Towards understanding the role of cellular organization and how it becomes aberrant, studies have focused on specific proteins, (extracellular, membrane, cytoskeletal, perinuclear and intranuclear) as a function of phenotype in model cell systems. Specifically, it has been shown, using non-malignant human mammary epithelial S1 cells (HMECs), that extracellular matrix (ECM)-induced tissue-like acinar morphogenesis directs the organization of nuclear mitotic apparatus protein (NuMA). The formation of acini encompas 3 steps. In early stage morphogenesis cells proliferate, they then growth arrest and polarize around a central luman apon completion of morphogenesis. NuMA distribution was diffuse in proliferating cells but became increasingly aggregated into a multi-focal (punctate) pattern as HMECs growth arrested and completed morphogenesis [Lelièvre 1998]. When these cells were cultured as monolayers on plastic they did not undergo morphogenesis, however they exhibited both proliferative and growth-arrest phenotypes, controlled by epithelial growth factor (EGF), which were associated with diffuse and slightly punctate NuMA distributions, respectively. The malignant counterpart (T4) of these S1 cells do not undergo acinar morphogenesis but proliferate into disorganized clusters. Interestingly, there was no visual difference in NuMA distribution between proliferating non-malignant and malignant HMECs. To quantify these findings and to answer the question of whether there is a measurable difference in NuMA distributions between non-malignant and malignant HMECs, we have developed a Gaussian-blurring, multi-scale image analysis technique which measures the characteristic size of punctate foci. HMECs were imaged in 3D, using confocal microscopy, for fluorescently labeled NuMA, DNA, and Ki67, an immunological stain for proteins expressed only in the cell cycle. Nuclear segmentation, based on the DNA staining, allowed NuMA texture analysis to be restricted to the nuclear volumes [Ortiz de Solórzano et al 1999]. Averaging results over HMECs in each population resolved significant, yet, sub-visual differences in NuMA punctateness allowing clear distinction between non-malignant growth arrested, non-malignant proliferating and malignant cells cultured in monoloayers on plastic. Non-malignant growth arrested cells were most punctate, non-malignant proliferating cells produced intermediate values and malignant cells were the least punctate. The ability to quantitatively discern cell phenotype based on the spatial distribution of a nuclear protein has broad application in furthering our understanding of fundamental biological processes.

FIGURE 1 When human mammary epithelial cells (HMECs) are cultured within an extracellular matrix they form polarized and growth-arrested tissue-like acini. These cultured acini mimic epithelial cells of the mammary ducts by forming a central lumen and depositing an endogenous basement membrane. This morphogenesis is accompanied by a redistribution of the nuclear mitotic apparatus protein (NuMA) which is reportedly diffuse in early stage morphogenesis, when the cells are proliferating, but which redistributes into foci of increasing size as the cells growth arrest and differentiate [Lelièvre et at 1998]. FIGURE 2 In model-based image analysis, the aim is to quantify images based on specifically defined features. In the case of non-malignant  human mammary epithelial cells (HMECs), the distribution of nuclear mitotic apparatus protein (NuMA) was reportedly diffuse (uniformly distributed) in nuclei of proliferating cells but reorganized into nuclei foci (spots) as the cells growth arrested. To quantify these findings and to answer the question of whether non-malignant and malignant HMECs exhibit different NuMA distributions, we developed a Gaussian-blurring, multi-scale nuclear analysis technique which measures the characteristic size of punctate foci in the image.  For an image to exhibit punctate features, neighbouring points in the image (voxels) must have related brightness values which correlate over the length scale of such features. Conversely, rapid and unrelated voxel value variation across the image defines diffuse. Consequently, the amount of voxel variation in an image can be used to quantify the size of punctate features. As a measure of voxel value variation, we use the root  sum square gradient (RSSG), defined as the root sum square of the difference in neighouring voxel values, along a given direction, throughout the image (Figure 2A). To normalize this for nuclear size, brightness, contrast and background brightness, the RSSG is calculated on the same image after it has been blurred, mathematically, with an Gaussian (like) filter (Figure 2B). Increasing the Gaussian blur factor in this step increases the range of foci sizes to which the analysis is sensitive. Dividing the blurred RSSG (RSSGblurred) with the unblurred RSSG (RSSGunblurred) produces a value from 0 to 1 which we have termed the contrast variation (CV= RSSGblurred/RSSGunblurred) (Figure 2C) which, as will be shown, relates to the size of punctate features in the image. FIGURE 3 To determine the sensitivity of the image analysis algorithm to the size of foci, we analyzed sets of well characterized test images having increasing punctate foci size. For an image to exhibit punctate foci, neighbouring points in the image (voxels) must have related brightness values which correlate over the length scale of such features. Conversely, a diffuse image has rapid and unrelated voxel value variation. The test images were thus constructed by first creating a diffuse image (Figure 3A1) and then convolving this with a Gaussian of increasing width (Figures 3A2 - 3A12). Each image was then analysed to produce CV values as a function of blur (see Figure 2C). The CV values were plotted against the size of the punctate foci, defined by the width of the convolving Gaussian which produced the images (Figure 3B). For this example, CV at blur factor 4  (CV4) was chosen as it produced the largest increase over the punctate foci size of these images. In this size range, CV2 would have  saturated and CV8 would have not increased significantly for the small sized foci. The results were unchanged if the diffuse image was first multiplied by a random background mask (Figure 3C) before being convolved with the Gaussian of increasing width (Figure 3D1 to 3D12).

FIGURE 4 Cell cultures were imaged in 3D using confocal microscopy, for fluorescently labeled NuMA, Ki-67 (a marker of proliferation) and DNA. Nuclear segmentation, based on the DNA staining, allowed image analysis of NuMA staining within individual nuclei. Figures 4A, 4B and 4C show 2D sections from 3D images of non-malignant proliferating, non-malignant growth arrested and malignant proliferating human mammary epithelial nuclei fluorescently stained for NuMA, respectively. Averaging results over cells in each population resolved significant, yet, sub-visual differences in NuMA punctateness. Non-malignant growth arrested cells showed the largest sized foci, non-malignant proliferating cells produced intermediate values and malignant cells showed the smallest sized foci (Figure 4D). These data not only confirm the difference in NuMA distribution for proliferating and growth arrested non-malignant cells, as seen by Lelièvre et al 1998, but show a significant yet sub-visual difference in the distribution of NuMA between proliferating non-malignant and malignant cells. FIGURE 5 To confirm the difference in NuMA distribution between non-malignant and malignant cells in their proliferative state, we analyzed two different non-malignant cell lines, S1 and Revertant-T4*, and two malignant cell lines, T4 and MDA231. Two different monoclonal antibodies against NuMA were used in the case of S1 and T4 cells. The results clearly demonstrate significant differences in measured NuMA distributions from non-malignant to malignant phenotypes, even though these differences were not apparent visually. *When grown in the presence of Tyrphostin, an inhibitor of the EGFR pathway, tumor cells revert to a non-malignant phenotype and are able to form acini in extracellular matrix culture [Wang et al 1998]. CONCLUSIONS: We have developed a model-based image analysis technique which can quantify sub-visual differences in the spatial distribution of a nuclear associated protein. The spatial distribution of nuclear mitotic apparatus protein (NuMA), which can be used as a marker of cell phenotype, is consistently more diffuse in malignant cells than in non-malignant cells. References: Bissell MJ, Weaver VM, Lelièvre SA, Wang F, Petersen OW, Schmeichel KL 1999 Tissue structure, nuclear organization, and gene expression in normal and malignant breast. Cancer Res 59(7 Suppl):1757-1763s; discussion 1763s-1764s  Lelièvre SA, Weaver VM, Nickerson JA, Larabell CA, Bhaumik A, Petersen OW, Bissell MJ 1998 Tissue phenotype depends on reciprocal interactions between the extracellular matrix and the structural organization of the nucleus. PNAS 95(25):14711-6  Wang F, Weaver VM, Petersen OW, Larabell CA, Dedhar S, Briand P, Lupu R, Bissell MJ 1998 Reciprocal interactions between beta1-integrin and epidermal growth factor receptor in three-dimensional basement membrane breast cultures: a different perspective in epithelial biology. PNAS 95(25):14821-6  Ortiz de Solórzano C, Garcia Rodriguez E, Jones A, Pinkel D, Gray JW, Sudar D, Lockett SJ 1999 Segmentation of confocal microscope images of cell nuclei in thick tissue sections. J Microscopy 193:212-26    Acknowledgements: This work was supported by the Director, Office of Energy Research, Office of Health and Environmental Research of the U.S. Department of Energy under contract NO. DE-AC03-76SF00098 to SJL & MJB; NIH grant Ca-67412 to SJL; NIH grant CA-64786 to MJB; U.S. Department of Defense Breast Cancer Research Program DAMD 17-97-1-7103 to SAL and a contract with Carl Zeiss Inc.