Circulation times of prostate cancer and hepatocellular carcinoma cells by in vivo flow cytometry

Authors

  • Yan Li,

    1. Department of Chemistry, Fudan University, 220 Han Dan Road, Shanghai, 200433, China
    2. Institutes of Biomedical Sciences, Fudan University, 138 Yi Xue Yuan Road, Shanghai, 200032, China
    Search for more papers by this author
  • Jin Guo,

    1. Department of Chemistry, Fudan University, 220 Han Dan Road, Shanghai, 200433, China
    2. Institutes of Biomedical Sciences, Fudan University, 138 Yi Xue Yuan Road, Shanghai, 200032, China
    Search for more papers by this author
  • Chaofeng Wang,

    1. CAS-MPG Partner Institute for Computational Biology, Key Laboratory for Computational Biology, Shanghai Institutes for Biological Sciences, Shanghai, China
    Search for more papers by this author
  • Zhichao Fan,

    1. Department of Chemistry, Fudan University, 220 Han Dan Road, Shanghai, 200433, China
    2. Institutes of Biomedical Sciences, Fudan University, 138 Yi Xue Yuan Road, Shanghai, 200032, China
    Search for more papers by this author
  • Guangda Liu,

    1. Institutes of Biomedical Sciences, Fudan University, 138 Yi Xue Yuan Road, Shanghai, 200032, China
    2. Shanghai Medical College, Fudan University, 138 Yi Xue Yuan Road, Shanghai, 200032, China
    Search for more papers by this author
  • Cheng Wang,

    1. School of Medical Instrument and Food Engineering, University of Shanghai for Science and Technology, 516 Jungong Road, 200093, Shanghai, China
    Search for more papers by this author
  • Zhengqin Gu,

    1. Department of Urology, Xinhua Hospital, Jiao Tong University, 1665 Kongjiang Road, Shanghai 200092, China
    Search for more papers by this author
  • David Damm,

    1. Department of Computer Science III, University of Bonn, Bonn, Germany
    Search for more papers by this author
  • Axel Mosig,

    Corresponding author
    1. CAS-MPG Partner Institute for Computational Biology, Key Laboratory for Computational Biology, Shanghai Institutes for Biological Sciences, Shanghai, China
    2. Department of Biophysics, Ruhr University Bochum, Germany
    • CAS-MPG Partner Institute and Key Laboratory for Computational Biology, Shanghai Institutes for Biological Sciences, 320 Yue Yang Road, Shanghai, China 200031
    Search for more papers by this author
    • Fax: (86) 21-54920451

  • Xunbin Wei

    Corresponding author
    1. Department of Chemistry, Fudan University, 220 Han Dan Road, Shanghai, 200433, China
    2. Institutes of Biomedical Sciences, Fudan University, 138 Yi Xue Yuan Road, Shanghai, 200032, China
    3. Med-X Research Institute, Shanghai Jiao Tong University, 1954 Huashan Road, Shanghai, 200240, China
    4. School of Biomedical Engineering, Shanghai Jiao Tong University, 1954 Huashan Road, Shanghai, 200240, China
    • Research Institute and School of Biomedical Engineering, Shanghai Jiaotong University, 1954 Huashan Road, Shanghai, China 200240
    Search for more papers by this author
    • Fax: (86) 21-54237100


Abstract

In metastasis, the cancer cells that travel through the body are capable of establishing new tumors in locations remote from the site of the original disease. To metastasize, a cancer cell must break away from its tumor and invade either the circulatory or lymphatic system, which will carry it to a new location, and establish itself in the new site. Once in the blood stream, the cancer cells now have access to every portion of the body. Here, we have used the “in vivo flow cytometer” to study if there is any relationship between metastatic potential and depletion kinetics of circulating tumor cells. The in vivo flow cytometer has the capability to detect and quantify continuously the number and flow characteristics of fluorescently labelled cells in vivo. We have improved the counting algorithm and measured the depletion kinetics of cancer cells with different metastatic potential. Interestingly, more invasive PC-3 prostate cancer cells are depleted faster from the circulation than LNCaP cells. In addition, we have measured the depletion kinetics of two related human hepatocellular carcinoma (liver cancer) cell lines, high-metastatic HCCLM3 cells, and low-metastatic HepG2 cells. More than 60% HCCLM3 cells are depleted within the first hour. Interestingly, the low-metastatic HepG2 cells possess noticeably slower depletion kinetics. In comparison, <40% HepG2 cells are depleted within the first hour. The differences in depletion kinetics might provide insights into early metastasis processes. © 2011 International Society for Advancement of Cytometry

Metastasis is a complicated process that has yet to be completely understood. In metastasis, the cancer cells that travel through the body are capable of establishing new tumors in locations remote from the site of the original disease. To metastasize, a cancer cell must break away from its tumor and invade either the circulatory or lymph system (1–3). Once in the blood stream, the cancer cells now have access to every portion of the body. The cancer cells in the bloodstream must fight the body's defense system and try to reattach itself in a new location. Fewer than 1 in 10,000 cancer cells survive circulation to create a new tumor. The circulation of the blood plays an important role in determining where cancer cells travel. The cancer cells usually are trapped in the first set of capillaries that they encounter downstream from their point of entry. These capillaries are often in the lung, since returning deoxygenated venous blood leaving many organs is returned to the lung for reoxygenation. From the intestines, the blood go to the liver first, thus cancer cells leaving the intestines will go there. Therefore, the lung and the liver are the two most common sites for metastasis in the human body. Many circulating cancer cells cannot finish the entire process of metastasis (4).

The body has many safeguards to prevent cancer cells from metastasizing, yet many cancer cells have the ability to overcome these safeguards. Prostate cancer can metastasize throughout other areas of the body, such as the bone, lung, and liver. Surgical resection, hormonal therapy, chemotherapy, and radiation therapy are the foundation of current prostate cancer therapies. Treatments for prostate cancer cause both short- and long-term side effects that may be difficult to accept. Molecular mechanisms of prostate cancer metastasis need to be understood better and new therapies must be developed to selectively target to unique characteristics of cancer cell growth and metastasis. Research is now focused on understanding in which ways cancer cells have mutated to circumvent the body's defenses and travel to other locations. Optical techniques play an essential role to study molecular mechanisms of the cancer and its metastasis. Previously, researchers have used a number of optical methods to study the cancer and its metastasis (5–7).

The in vivo flow cytometer has the capability to detect and quantify continuously the number and flow characteristics of fluorescently labelled cells in vivo (8–15). The in vivo flow cytometer allows researchers to acquire cytometric information from the circulation in live animals without extracting blood samples (Fig. 1). Previous research demonstrates that using this technique the number of fluorescently labelled circulating cells can be quantified in a real-time and reproducible manner in live animal. Here, we use the “in vivo flow cytometry” combined with optical imaging to study the mechanisms that govern the early steps in tumor cell spread through the microenvironment.

Figure 1.

Schematic of the two-color in vivo flow cytometer experimental setup. Laser light (488 nm or 635 nm) is focused into a slit (5 μm × 72 μm) by a cylindrical lens (CL, focal length 150 mm) and imaged across the selected blood vessel with a microscope objective lens (40X, NA = 0.6). The fluorescence from excited labelled cells in blood stream is collected by the same microscope objective, directed through the dichroic beam splitter DM3, which reflects green LED light to image sample onto CCD camera, (reflection 25%, transmission 75%, Edmund Optics), reflected by a mirror, a second splitter DM2 (transmission ≥ 90% for 488 nm and 635 nm; reflection≥ 90% for 499–556 nm, 580–622 nm, and 652–755 nm, Semrock) and a third dichroic beam splitter DM4 (edge wavelength 516 nm, reflection band>90% for 490–510 nm, transmission band>90% for 520–700 nm, Semrock), and imaged onto a 200 μm × 3000 μm mechanical slit (MS), which is confocal with the excitation slit. F1-3: bandpass filter (F1: 509–552 nm; F2: 500–520 nm; F3: 640–690 nm; Semrock). DM1: diochromatic mirror; edge wavelength 505 nm, reflection band 513–725 nm, transmission band 446–500 nm, Semrock). AL1, 3, 4: achromatic lenses, focal length 150 mm; AL2: achromatic lens, focal length 30 mm. M1-3: mirrors.

Signals obtained from the in vivo flow cytometer are inherently noisy, and thus require computational methods for quantifying the number of labelled cells passing the cytometer. Previously, a semiautomated method has been proposed for this purpose (8), involving a human operator specifying a threshold-line in a two-dimensional feature space that distinguishes labelled cells from noise. In this context, we present a robust and fully automated computational method for cell counting that allows reliable quantification of circulating cells.

MATERIALS AND METHODS

Cell Culture and Animal

A human androgen-dependent prostate cancer cell line LNCaP, a human androgen-independent high metastatic potential prostate cancer cell line PC-3 and a human hepatocellular carcinoma cell line HepG2 were bought from The Cell Bank of Type Culture Collection of Chinese Academy of Science (Shanghai Institute of Cell Biology). A human high metastatic potential hepatocellular carcinoma cell line HCCLM3 (16) was bought from the Liver Cancer Institute, Zhongshan Hospital, Fudan University. The LNCaP, PC-3 and HepG2 cells are cultured at 37°C and 5% CO2 in RPMI-1640 medium (GIBCO) containing 10% fetal bovine serum (HyClone). The HCCLM3 cells were cultured at 37°C and 5% CO2 in high-glucose DMEM medium (GIBCO) containing 10% fetal bovine serum (HyClone). For cell labeling, suspensions of all the cells were ex vivo incubated at 37°C with 0.1 mM DiD (Invitrogen) for 30 min, and then washed by phosphate-buffered saline for three times. Cell number was measured by hemocytometer before tail vein injection. 5 × 105 DiD-labeled cells were intravenously injected into the mice.

Experiments were carried out on 6–8-week-old male BALB/cASlac-nu mice (Shanghai SLAC Laboratory Animal Co.; 20 ± 2 g). For each cell line, more than six mice were used for tests. This study was approved by the Ethical Committee of Animal Experiments of Institutes of Biomedical Sciences, Fudan University.

In Vivo Flow Cytometry

We set up our in vivo flow cytometer to detect fluorescence signal from a given circulating tumor cell (CTC) population in a confocal geometry (Fig. 1) based on previous groups' experience (8, 9). Briefly, fluorescence signal from a given circulating cell population is recorded as the cells pass through the slit of light. Confocal detection of the excited fluorescence enables continuous monitoring of labeled cells in the upper layers of scattering tissue, such as the skin of a mouse ear. The size of the slit at the focal plane of the sample is ∼ 5 × 72 μm. The depth of focus (i.e., the full width at half maximum of the light slit onto the sample in the axial direction) is ∼ 50 μm, a value chosen to match the vessels of interest. The sample is positioned so that the long dimension of the slit traverses the width of the blood vessel and covers the whole interface. All cells passing through the focus slit will be excited and detected at the same time, if they are flowing in cluster (i.e., above or beside each other). Fluorescence is detected with a photomultiplier tube placed directly behind the mechanical slit, sampled at a rate of 5 kHz with a data acquisition card, and displayed and stored on a computer.

In experiments, the BALB/c nude mice were anesthetized with sodium pentobarbital and injected intravenously with DiD-fluorescently labelled tumor cells. After the injection, mice were then placed on a heated stage and an artery in the ear was chosen for obtaining measurements. Fluorescence signal was excited as the labelled circulating cells passed through a laser slit focused across the blood vessel (Fig. 2A). Detecting the excited fluorescence confocally enables researchers to monitor labeled cells in the animal circulation system continuously. Signal was recorded at a rapid rate (5 kHz) to ensure the measurement of fast-flowing cells (Fig. 2B). The blood flow velocity is 2.3–7.0 mm/s (8). The slit width is ∼5 microns. The Nyquist sampling theory indicates that 2,800 data points per second is required to be saved. Therefore, the signal recording rate of 5 kHz is sufficient for the system. The number of fluorescent peaks, along with the height and full width at half maximum of each peak, was determined using algorithms developed in-house.

Figure 2.

A: The mouse is anesthetized and placed on a heated stage. An artery in the ear is chosen for the measurement. The fluorescence signal from DiD-labelled circulating cells is recorded when the cells pass through the slit of laser light focused across the artery. DiD is a fluorescent dye to label the membrane lipid (excitation peak: 640 nm; emission peak: 660–670 nm; Molecular Probes). B: A trace of labelled circulating HCCLM3 hepatocellular carcinoma cell measured by the in vivo flow cytometer, after intravenously introduced into a mouse. The peak within the trace indicates that a DiD-labelled HCCLM3 cell passes through the slit of light and thus gives a burst of fluorescence.

Signal Preprocessing and Wavelet-Based Denoising

To address the problem of quantifying the number of labeled cells passing the in vivo flow cytometer, we developed a computational cell counting method. The algorithm essentially works in two steps: in a first preprocessing step, the signal is denoised and normalized; in a second step, dynamic peak picking is performed to distinguish noise-generated local maxima from those that correspond to labeled cells in the denoised signal.

The preprocessing step first subtracts the median intensity Ms from the absolute value signal s. Furthermore, signal intensities are normalized by scaling the maximum intensity to a reference intensity to obtain the normalized signal s′. After that, Denoho's approach to wavelet based soft-thresholding (17) is applied to the normalized signal. We compute the wavelet coefficients Ws′ of s′ using the Symlet7 (18) mother wavelet. In the wavelet coefficients Ws′, the noise level is estimated in the largest detail coefficients based on the median absolute deviation, which is used as cut-off to obtain the thresholded wavelet coefficients Vs′. Transforming these coefficients back into the time domain yields the normalized and denoised signal s′′.

The denoised signal s′′ still contains local maxima (peaks) that are due to noise (noise peaks), which need to be distinguished from those peaks that result from labeled cells passing the cytometer (cell peaks). For distinguishing cell peaks from noise peaks, we employ a dynamic peak picking approach based on an idea developed in the context of audio signal processing (19). Our algorithm can be described using a finite state automaton as illustrated and described in Figure 3.

Figure 3.

Illustration of the cell counting algorithm. (Top): The original signal s is normalized and transformed using soft-thresholding yielding the denoised signal s′′ (dashed green signal). The thresholding baseline (upper horizontal dashed line) is computed using the local median (horizontal continuous line) plus one standard deviation. The crosses indicate the sign of the discrete derivative as + (blue crosses) or − (green crosses), which are used as input features for the dynamic peak picking. (Bottom): Dynamic peak picking follows a finite state automaton that sequentially reads the features obtained from the discrete derivative and the thresholding baseline. The next state is determined on the current sign of the discrete derivative (transitions labelled + or −) or whether the current sample in s″ exceeds the baseline threshold (transitions labelled < or >). In some transitions, both criteria need to be matched, as indicated by transition labels (+,<) or (+,>). A peak is reported whenever encountering state P indicated by a red circle. Other states correspond to an ascending phase (blue circles) or a descending phase (green circles) of a peak.

The cell counting algorithm has been implemented in Matlab using the Wavelet Toolbox to implement Donoho's soft thresholding.

In Vivo Confocal Imaging

We use the bone marrow confocal imaging to track the depleted circulating prostate cancer cells, since the bone marrow is one of the common metastatic sites for prostate cancer and readily accessible for in vivo confocal imaging. However, the bone marrow accumulation value is not intended to be used as a precise quantitative measure of cells depleted from circulation, because circulating cancer cells may also attach to vessel walls. Mice were anesthetized and a small incision was made in the scalp so as to expose the underlying dorsal skull surface. For cell tracking, 5 × 105 PC-3 or LNCaP prostate cancer cells labeled with DiD were injected into the tail vein. Prostate cancer cells spreding to bone marrow vasculature of the skull was then imaged using a Leica fluorescence confocal microscope while the mice were under anaesthesia on a warmed microscope stage. Images with cellular details were obtained through the intact mouse skull at depths of up to 250 μm from the surface of the skull using a water immersion objective lens (10X; NA = 0.3). Quantitative evaluation was made by counting the number of fluorescent cells per field.

RESULTS

Validation of Cell Counting Procedure

To validate our cell counting algorithm, we compared its performance on datasets, where peaks were identified by Line-Separating Method (LSM, n = 9). This leads to a statistically significant linear correlation with a correlation factor of 0.992, nearly identical cell counts on average (Fig. 4).

Figure 4.

A: Comparison of Line-Separating Method with our fully automated Wavelet-based cell counting method. B: 9 data are processed using both LSM and Wavelet-based method. The cell countings are represented in Passing-Bablock Plot in dots. The regression gives a slope of 0.9997 and an intercept of 4.3662 on Y-axis.

The Depletion Kinetics of Circulating Prostate Cancer Cells

We have used the “in vivo flow cytometer” to study if there is any relationship between metastatic potential and depletion kinetics of circulating tumor cells. Two commonly used prostate cancer cell lines, PC-3 and LNCaP cells are studied with PC-3 cells being generally thought more invasive than LNCaP cells (20, 21).

The depletion kinetics of circulating PC-3 and LNCaP cells in BALB/c nude mice during the first 8 h following injection of the fluorescently labeled cells illustrates the depletion process (Fig. 5). More than 70% PC-3 cells are depleted within the first hour. After the initial depletion, there is a reappearance in the number of circulating cells, quickly followed by a second depletion. By 8 h, 76 ± 3.6% PC-3 cells are depleted from the circulation. Interestingly, the low-metastatic LNCaP cells possess noticeably slower depletion kinetics. In comparison, <30% LNCaP cells are depleted within the first hour. By 8 h, 65 ± 3.4% LNCaP cells are depleted from the circulation. In vivo confocal microscopy imaging shows the cell numbers of DiD-labeled PC-3 cells spreading in the bone marrow areas within mouse skull are considerably higher than those of LNCaP cells at 1 h and 5 h after injection (Fig. 5B and 5C).

Figure 5.

A: Depletion kinetics of circulating high-metastatic PC-3 cells and low-metastatic LNCaP prostate cancer cells in BALB/c nude mice. The normalized numbers of circulating cells per min are shown for 24 h following injection of the DiD-fluorescently-labelled cells to illustrate the depletion process. >70% P3 cells are depleted within the first hour. In comparison, the low-metastatic LNCaP cells possess noticeably slower depletion kinetics. <30% LNCaP cells are depleted within the first hour. B: Cell numbers are counted manually in the bone marrow areas within mouse skull from in vivo confocal microscopy imaging (10) 1 h and 5 h after intravenous injection of PC-3 and LNCaP prostate cancer cells. (C) Representative in vivo confocal microscopy imaging of the bone marrow areas beneath mouse skull at 1 h and 5 h after injection of prostate cancer cells by tail vein. a, b: PC-3 cells at 1 h and 5 h after tail vein injection, respectively. c, d: LNCaP cells in the corresponding area at 1 h and 5 h after tail vein injection, respectively.

The Depletion Kinetics of Circulating Liver Cancer Cells

To further explore if there is any relationship between metastatic potential and depletion kinetics of circulating tumor cells, we have measured the depletion kinetics of two related human hepatocellular carcinoma (HCC, liver cancer) cell lines, high-metastatic HCCLM3 cells (16) and low-metastatic HepG2 cells. The depletion kinetics of circulating HCCLM3 cells in BALB/c nude mice during the first 12 h following injection of the fluorescently labelled cells illustrates the depletion process (Fig. 6). More than 60% cells are depleted within the first hour. After the initial depletion, there is a reappearance in the number of circulating cells, quickly followed by a second depletion. This phenomenon was also observed in previous study on prostate cancer cells (9). By 12 h, 95 ± 1.0% HCCLM3 cells are depleted from the circulation. Interestingly, the low-metastatic HepG2 cells possess noticeably slower depletion kinetics. In comparison, <40% HepG2 cells are depleted within the first hour. By 12 h, 57 ± 6.5% HepG2 cells are depleted from the circulation. When present in circulation for extensively long time, cancer cell might undergo cell death because of lack of survival signal from cell adhesion and the harsh environment imparted by the sheer stress (22, 23). Therefore, the differences in depletion kinetics might provide insights into early metastasis processes.

Figure 6.

Depletion kinetics of circulating high-metastatic HCCLM3 cells and low-metastatic HepG2 cells in BALB/c nude mice. The normalized numbers of circulating cells per min are shown for 12 h following injection of the DiD-fluorescently-labelled cells to illustrate the depletion process. >60% HCCLM3 cells are depleted within the first hour. By 12 h, ∼ 94% HCCLM3 cells are depleted from the circulation. In comparison, the low-metastatic HepG2 cells possess noticeably slower depletion kinetics. <40% HepG2 cells are depleted within the first hour. By 12 h, ∼ 55% HepG2 cells are depleted from the circulation.

DISCUSSION

The in vivo flow cytometer allows researchers to quantify continuously the number and flow characteristics of fluorescently labelled cells in vivo. Here, we have improved the counting method to study the relationship between metastatic potential and depletion kinetics of circulating tumor cells. Interestingly, the high-metastatic cancer cells possess noticeably faster depletion kinetics. The in vivo-flow-cytometry is particularly useful for studying small animal models of tumors. For example, it is often difficult to obtain sufficient blood sample from a single mouse, whose total blood volume is only ∼ 2 ml. This is particularly true when multiple blood samples are needed. With this in vivo technique, multiple measurements can be performed on a single animal over time, minimizing the number of animals and also eliminating variation between individuals.

The metastatic properties of the cancer cells, such as molecule expressed by vasculature endothelium and cancer cells, contribute significantly to the depletion kinetics from circulation (10, 15). However, there are some other factors could affect the accuarcy of in vivo-flow-cytometry. The physiology conditions of mice such as body temperature, heart rate may also affect the blood flow velocity, which might cause measurement errors. Therefore, an animal temperature control apparatus is used in the experiment. In addition, some injected cancer cells might undergo cell death or being cleared by the monocytes.

For computational analysis, it turns out that our fully automated procedure using wavelet-based denoising and dynamic peak picking can reliably detect peaks in in vivo flow cytometry signals. This procedure eliminates the neccessity of setting a separating line by a human operator in the method used previously (8). This has two important implications for in vivo flow cytometry as a whole. First, cell counts are more objective as no human operator is involved in the analysis. Second, it eliminates the need for control experiments, which are necessary to reliably set a threshold line in the line-separating method.

Detection of green fluorescence signal on circulating tumor cells could be technically more difficult, depending on the green fluorescence intensity. It is particularly true for detection through blood because of the strong absorption below ∼ 590 nm (24, 25). Both excitation and emission will be attenuated, respectively on the way in and on the way out. One method to enhance the sensitivity is to utilize the multi-photon detection, although with significantly higher cost. Low et al. developed a multiphoton intravital flow cytometer to quantify rare circulating tumor cells in vivo and achieved moderately higher sensitivity than confocal detection used in our current in vivo-flow-cytometer (13, 24–28). Another method is to use the photothermal/ photoacoustic detection, which also improves the detection depth. Tuchin et al. developed a photothermal image flow cytometer to detect target cells in blood and lymph flow in vivo (29–31). They used nanoparticles to enrich the rare cells and achieved significantly higher sensitivity (32–34).

The study here focuses on measuring circulating tumor cells. However, the methods developed here shall be applicable to various potential biomarkers in circulation for detecting cancer, characterizing pathologic malignant tumors, assessing disease prognosis, and for predicting and measuring response to treatments, thus providing invaluable information about detecting and treating cancer. Furthermore, the developed methods here will be useful to monitor the circulating cells in general.

Acknowledgements

The authors thank Charles P. Lin for collaboration. Yan Li, Jin Guo, and Chaofeng Wang contributed equally to this paper.

Ancillary