2.1.

SL-QPM

SL-QPM utilizes¹⁴ a low spatial-coherence thermal light source and common-path interferometer configuration to produce a speckle-free optical path length map of nuclear architecture with subnanometer sensitivity from the original unmodified clinical cytology specimens. The high sensitivity achieved by SL-QPM derives from its reflectance-mode configuration, low-coherence broadband illumination, and spectroscopic detection. The reflectance mode highlights the interference signals from elastic light scattering properties, while suppressing the effect of absorption in most cytology specimens. The spectroscopic detection enables us to analyze the interference signals due to refractive index variations within the object at every pixel of a microscopic image. With the glass substrate and biological cell of the cytology slides as a reference and a sample, respectively, that share a common path, SL-QPM effectively minimizes the slightest external disruptions that may compromise the ultrahigh sensitivity. The low-spatial-coherence illumination serves as a virtual aperture to remove the speckle noise, thus producing a speckle-free phase image.

The hardware design of the SL-QPM is similar to the previously reported elastic backscattering spectroscopic microscopy,¹⁵ as shown in Fig. 1. The SL-QPM system was described in detail in our previous publication.¹⁴ Briefly, a broadband white light from a Xe arc lamp was collimated by a 4f imaging system and focused on the sample by a low-numerical-aperture (NA) objective (NA = 0.4). The resulting backscattering light was collected and projected by a tube lens onto the slit of an imaging spectrograph (Acton Research, Massachusetts) coupled with CCD camera (Andor Technology, Connecticut), which is mounted on a scanning stage. The magnification of the system is about 44. By linearly scanning the slit of the spectrograph with a 10-μm step size, the backscattering image is acquired. In each scanning step, the CCD camera records a matrix with the x axis corresponding to the wavelength and the y axis corresponding to the spatial position, resulting in a 3-D intensity cube I(x, y, k), where k represents the wave number. A transmission-mode microscope was used to record the conventional cytology image.

Fig. 1

Schematics of SL-QPM instrument: Xe, xenon arc lamp; L1, and L2, lenses; A, aperture; BS, beamsplitter; OB1, objective lens; M, mirror; ST, sample stage; TL, tube lens; RM, removal mirror; CAM, camera; SP, spectrograph; SS, scanning stage; and CCD, charged-coupled device. The additional transmission-mode microscopic components (OB2-M2-M3-RM1) are used to record the conventional cytology images for comparison.

The detected signal I(x, y, k) comes from interference from the reference (i.e., the glass substrate) and sample signals (i.e., light traveling a round-trip along the axial dimension of the cell) with a shared common path and can be related to the signal and reference field [TeX:] $(E_s, E_r)$ $(E_s,E_r)$ as follows:

Fig. 2

Fourier transformed spectrum |F(OPL)| of the original backscattering spectrum I(x, y, k) from (a) the Diff-Quik staining solution and (b) a cell nucleus stained with Diff-Quik stains. (a) The prominent peak at the low-spatial-frequency component of |F(OPL)| comes from the absorption profile of the Diff-Quik staining solution. (b) Evidently, the |F(OPL)| from the Diff-Quik-stained cell nucleus has a distinctly different peak, corresponding to the optical path length of the cell nucleus (as indicated by the black arrow). These results confirmed that our selected optical path length of interest is not affected by the absorption profile of the Diff-Quik staining solution.

2.2.

Optical Path Length (OPL) Map

The SL-QPM essentially records a microscopic image with a reflectance interferometric intensity cube I(x, y, k). The phase value at every pixel can be obtained by Fourier analysis of I(x, y, k) and is converted to the corresponding OPL [OPL(x, y) = 〈n(x, y)〉L(x, y), where 〈n(x, y)〉 is the average refractive index along the axial direction (i.e., the z direction) at the specific pixel (x, y), and L(x, y) is the physical thickness]. As a result, we obtain a 2-D, nanoscale spatial distribution of the OPL (OPL map). To quantify the image characteristics of the OPL map in the nucleus, we extracted three statistical parameters: average OPL 〈OPL〉 over the 2-D spatial distribution of OPL(x, y); standard deviation [TeX:] $\sigma _{{\rm OPL}} $ ${\sigma }_{\mathrm{OPL}}$ over OPL(x, y), representing a global variation; and entropy [TeX:] $E_{{\rm OPL}} $ $E_{\mathrm{OPL}}$ , which is a measure of randomness, representing a more localized heterogeneity derived from texture analysis. Both [TeX:] $\sigma _{{\rm OPL}} $ ${\sigma }_{\mathrm{OPL}}$ and [TeX:] $E_{{\rm OPL}} $ $E_{\mathrm{OPL}}$ quantify the architectural heterogeneity.

2.3.

Texture Analysis of OPL Map of a Cell Nucleus

Based on the spatial distribution of the OPL from a single nucleus, we performed a texture analysis and quantified three statistical parameters. The average OPL 〈OPL〉 and standard deviation [TeX:] $\sigma _{{\rm OPL}} $ ${\sigma }_{\mathrm{OPL}}$ were calculated by taking the mean and standard deviation of all OPL values from a single nucleus. The entropy [TeX:] $E_{{\rm OPL}} $ $E_{\mathrm{OPL}}$ , a parameter describing the randomness, is defined as [TeX:] $E_{{\rm OPL}} = - \sum_{i = 0}^{N - 1} {p(I_i)\,\log _2 p(I_i)} $ $E_{\mathrm{OPL}}=-{\sum }_{i=0}^{N-1}p\left(I_i\right){\log }_{2}p\left(I_i\right)$ , where [TeX:] $I_i $ $I_i$ is a random variable indicating intensity, [TeX:] $p(I_i)$ $p\left(I_i\right)$ is the histogram of the intensity levels in a region, and N is the number of possible intensity levels.¹⁶ The intranuclear 〈OPL〉 describes the average OPL or average density with a single nucleus (assuming the same nuclear thickness). The intranuclear standard deviation [TeX:] $\sigma _{{\rm OPL}} $ ${\sigma }_{\mathrm{OPL}}$ and intranuclear entropy [TeX:] $E_{{\rm OPL}} $ $E_{\mathrm{OPL}}$ describe the global and local heterogeneity of the OPL distribution, respectively.

2.4.

Animal Model of Colorectal Carcinogenesis

All animal studies were performed in accordance with the institutional Animal Care and Use Committee of University of Pittsburgh. All mice were housed in microisolator cages in a room illuminated from 7:00 AM to 7:00 PM (12:12-h light-dark cycle), with access to water and ad libitum. Three C57BL APC wild-type mice and three age-matched C57BL APC ^Min mice at 4 to 5 months old were sacrificed. The small intestines were removed, longitudinally opened, and washed with phosphate-buffered saline (PBS). Epithelial cells were obtained from visually normal mucosa of the small intestine with a cytology brush. The cytology brush was immersed into the Cytolyt solution (Cytec Corporation, Boxborough, Massachusetts). The slides were prepared by a standard thin prep processor (Cytec Corporation). Cells were then stained with Papanicolaou stains and sealed with a mounting medium and coverslip. Cells were examined by an expert cytopathologist and non-cancerous-looking cells were analyzed by SL-QPM. For each mouse, about 20 to 30 cells are randomly selected for SL-QPM analysis.

2.5.

Human Specimens

All studies were performed and all specimens were collected with the approval of the Institutional Review Boards at University of Pittsburgh and University of Washington. Banked frozen mucosal biopsies were obtained from a group of seven patients with ulcerative colitis who underwent colectomy or colonoscopy, four of whom had no evidence of dysplasia, and three of whom were diagnosed with colorectal cancer. Among the three cancer cases, one patient had frankly malignant cells obtained from the cancer site, while the other two patients had abnormal cells from cancer sites, classified as indeterminate by an expert cytopathologist. The cytology slides were made from frozen mucosal biopsies using a touch preparation (TP) method and were subsequently stained by Diff-Quik stains and coversliped. Such methods typically minimize the single-cell thickness variations among different patients for the same type of cell. These slides were reviewed by an experienced cytopathologist who dotted the colon epithelial cells of interest. Approximately 30 to 40 cell nuclei per patient were evaluated for OPL analysis.

2.6.

Statistical Analysis

We performed the statistical analysis using Microsoft Excel 2007. All the statistical analyses were performed based on the Student's t test. Two-tailed P values were used for all analyses. The alpha level was assumed to be 0.05. Two-tailed P values of less than 0.05 were considered as statistical significance.

3.1.

Experiments with APC ^Min Mouse Model

To explore the ability of nanoscale nuclear architectural characteristics quantified by SL-QPM to identify cancer from cytologically indeterminate cells, we used a well-established animal model of colorectal carcinogenesis—the APC ^Min mouse model. It is a genetically modified animal model that represents the human condition of familial adenomatous polyposis syndrome in which the adenomatous polyposis coli (APC) gene undergoes a germ-line mutation leading to a truncation in the APC protein and spontaneous development of intestinal adenomas. We analyzed the cytopathologically indistinguishable intestinal epithelial cells from three age-matched wild-type mice and three APC ^Min mice at 4 to 5 months with the presence of tumors in the small intestine. Figures 3, 3 show Papanicolaou-stained cytological images of representative intestinal epithelial cells from the APC ^Min mice compared to those of the wild-type mice obtained from a bright-field microscope and their corresponding OPL maps from the cell nuclei. Although the microscopic cytological images look similar (as confirmed by an expert cytopathologist), the OPL images that characterize the intranuclear distribution of OPL exhibit distinct differences. Based on these OPL maps, three statistical parameters from the nuclei were calculated: the average OPL 〈OPL〉 over the 2-D spatial distribution of OPL(x, y); the standard deviation of the OPL [TeX:] $\sigma _{{\rm OPL}} $ ${\sigma }_{\mathrm{OPL}}$ representing a global variation and the entropy [TeX:] $E_{{\rm OPL}} $ $E_{\mathrm{OPL}}$ —a measure of randomness, representing a more localized heterogeneity. As shown in Fig. 3, the nuclear heterogeneity quantified by [TeX:] $\sigma _{{\rm OPL}} $ ${\sigma }_{\mathrm{OPL}}$ and [TeX:] $E_{{\rm OPL}} $ $E_{\mathrm{OPL}}$ are significantly increased in approximately 20 to 30 randomly selected intestinal epithelial cells from the APC ^Min mice compared to those from the wild-type mice (p value < 0.001). However, no statistical difference is found in the average OPL 〈OPL〉 (p value = 0.4). These results suggest that the higher nuclear architectural heterogeneity derived from its in situ nanoscale OPL map is associated with carcinogenesis, underlying the potential of this technique in accurate cancer diagnosis at the level of single cell nucleus.

Fig. 3

(a) Representative cytological images (Papanicolaou stains) and (b) the corresponding OPL maps of cell nuclei (as shown in circles) of intestinal epithelial cells from the wild-type mouse and the APC ^Min mouse at 4 to 5 months with the development of tumors. Scale bars in the image indicate 5 μm. The colorbar shows the magnitude of the OPL in the cell nucleus in micrometers. Although the cytology images appear similar (cytopathologically indistinguishable), the OPL map of the cell nucleus from the APC ^Min mice exhibits an increased heterogeneity compared to that from the wild-type mice. (c) The standard deviation of OPL σ _OPL and the entropy E _OPL of the cell nuclei in cytologically normal-appearing intestinal epithelial cells show a statistically significant increase in the APC ^Min mouse (p value < 0.001), associated with the development of carcinogenesis. The error bar represents the standard error. About 20 to 30 cells were randomly selected from each mouse to conduct SL-QPM measurement and OPL analysis. (Color online only)

3.2.

Experiments with Colorectal Cancer in Patients with Ulcerative Colitis

To further confirm the significance of in situ nanoscale nuclear architecture in human cells, we analyzed human cytology specimens from a group of seven patients with ulcerative colitis (UC), an inflammatory bowel disease, undergoing colonoscopy or colectomy. Three patients had been diagnosed with colorectal cancer, while four had no evidence of dysplasia or cancer. We analyzed the OPL maps from three groups of colon epithelial cells, selected by an expert cytopathologist: (1) normal cells [i.e., cytologically normal cells from uninvolved tissue site (no active colitis) in the four patients without dysplasia or cancer]; (2) cytologically indeterminate cells obtained at the cancer site (i.e., cells classified by an expert cytologist as being indeterminate in patients with colorectal cancer); and (3) malignant cancer cells obtained at the cancer site (i.e., cytologically classified as malignant cells by cytologist using conventional diagnostic criteria).

Figure 4 presents the representative OPL maps of benign/normal colonic epithelial cell nuclei in a patient with ulcerative colitis, cell nuclei from a UC patient with concurrent cancer (colonic carcinoma) that was called indeterminate by cytological diagnosis, and malignant cell nuclei that cytologically called malignant from a UC patient with colonic carcinoma. The OPL maps reveal the significantly increased intranuclear heterogeneity and average OPL in the malignant cancer cell.

Fig. 4

(a) Representative cytology images and corresponding OPL maps of a cytologically normal colonic cell from a patient with ulcerative colitis; an indeterminate cell from cancer site, classified by cytology, obtained from a patient with surgically confirmed diagnosis of colonic cancer patient; and a malignant cell from a cancer site, classified by cytology, from a patient with confirmed colonic cancer. The OPL maps show distinct patterns in these three cell groups, with a significantly increased OPL and heterogeneity in cells from cancer patients compared with the normal cells. Scale bars in the image indicate 10 μm. Colorbar shows the magnitude of optical path length in a cell in microns. (b) Scatter plot of the average optical path length 〈OPL〉 the intranuclear entropy E _OPL and the standard deviation of the OPL σ _OPL of all the nuclei from a patient with ulcerative colitis without dysplasia or cancer (blue diamond); a colon cancer patient with those cells diagnosed as indeterminate by cytology (purple square); and a colon cancer patient with malignant cells, confirmed by cytology, from a confirmed colon cancer patient with malignant tumor (red triangle). Each point corresponds to a single nucleus. (c) The statistical analysis of [TeX:] $E_{{\rm OPL}} $ $E_{\mathrm{OPL}}$ , σ _OPL, and 〈OPL〉 was performed using approximately 30 to 40 cells for each patient. The nuclear heterogeneity derived from [TeX:] $E_{{\rm OPL}} $ $E_{\mathrm{OPL}}$ and σ _OPL is progressively increased from normal cells to cytologically indeterminate cells to frankly malignant cells (Student's t test, P < 0.0001). The 〈OPL〉 in malignant cells from cancer patients is significantly increased compared with that in normal cells (Student's t test, P < 0.01). However, there was no statistically significant difference in 〈OPL〉 between normal and cytologically indeterminate cells.

To illustrate the internuclear variation, Fig. 4 shows a scatter plot of OPL parameters [TeX:] $\langle{\rm OPL}\rangle, \sigma _{{\rm OPL}} $ $⟨\mathrm{OPL}⟩,{\sigma }_{\mathrm{OPL}}$ and [TeX:] $E_{{\rm OPL}} $ $E_{\mathrm{OPL}}$ for all the cells from one representative patient in each group: (1) normal epithelial cells from an ulcerative colitis (no cancer) patient; (2) cytologically indeterminate cells from a cancer patient; and (3) malignant tumor cells from a cancer patient, with each cell nucleus represented by a point (〈OPL〉, [TeX:] $\sigma _{{\rm OPL}},E_{{\rm OPL}}$ ${\sigma }_{\mathrm{OPL}},E_{\mathrm{OPL}}$ ). Despite the intrinsic variations among different nuclei, the malignant cells and normal cells are well separated. Importantly, the majority of cytologically indeterminate cells in cancer patients are clearly distinguishable from normal cells from patients without dysplasia. These cells classified as indeterminate by an expert cytolopathologist from a cancer site, resemble the nuclear architectural heterogeneity of the malignant cells.

Figure 4 shows the analysis of three statistical parameters from all cells in seven patients (approximately 30 to 40 cells from each patient). The heterogeneity parameters— [TeX:] $\sigma _{{\rm OPL}} $ ${\sigma }_{\mathrm{OPL}}$ and [TeX:] $E_{{\rm OPL}} $ $E_{\mathrm{OPL}}$ —were significantly increased in cytologically malignant cancer cells (p value < 0.0001) compared with those in noncancerous cells. The same heterogeneity parameters can also distinguish cytologically indeterminate cells in cancer patients from those in patients without dysplasia with a high level of statistical significance (p value < 0.0001). Although 〈OPL〉 is significantly increased in malignant cancer cells (p value < 0.01), 〈OPL〉 does not present any statistical difference in cytologically indeterminate cells obtained at the cancer site when compared with normal-appearing epithelial colonic cells obtained from patients without cancer or dysplasia. These results suggest the importance of the in situ nanoscale nuclear heterogeneity in detecting subtle changes in tumor malignancy.