PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
J Biophotonics. Author manuscript; available in PMC 2017 August 11.
Published in final edited form as:
PMCID: PMC5547001
NIHMSID: NIHMS889733

Autofluorescence flow sorting of breast cancer cell metabolism

Abstract

Clinical cancer treatment aims to target all cell subpopulations within a tumor. Autofluorescence microscopy of the metabolic cofactors NAD(P)H and FAD has shown sensitivity to anti-cancer treatment response. Alternatively, flow cytometry is attractive for high throughput analysis and flow sorting. This study measures cellular autofluorescence in three flow cytometry channels and applies cellular autofluorescence to sort a heterogeneous mixture of breast cancer cells into subpopulations enriched for each phenotype. Sorted cells were grown in culture and sorting was validated by morphology, autofluorescence microscopy, and receptor expression. Ultimately, this method could be applied to improve drug development and personalized treatment planning.

Keywords: autofluorescence, flow cytometry, breast cancer, tumor heterogeneity, cellular metabolism

Graphical Abstract

An external file that holds a picture, illustration, etc.
Object name is nihms889733u1.jpg

1. Introduction

Tumor heterogeneity can impact treatment response for cancer patients. Tumors can contain multiple subpopulations of cells with distinct phenotypes and sensitivities to drugs, and cells that are resistant to treatment can cause patient relapse [1]. The goal in clinical cancer treatment is to administer drugs that target all cell subpopulations within a tumor, leading to progression-free survival. Therefore, single-cell analysis techniques have become powerful tools for characterizing tumor heterogeneity and developing strategies to eliminate all treatment-resistant cells in a tumor.

Subpopulations of tumor cells can be defined by responsiveness or resistance to anti-cancer treatment. Responsive cells undergo cell death or senescence after treatment, whereas resistant cells continue to proliferate after treatment. Since cancer cells often exhibit altered cellular metabolism, particularly increased aerobic glycolysis (Warburg effect), and many drugs target metabolic pathways, cellular metabolism can be a marker for drug sensitivity [2]. Metabolic signaling pathways involve the cofactors NAD(P)H and FAD, and these molecules naturally exhibit autofluorescence. Previous studies have shown that breast cancer cells that are responsive to treatment exhibit a decreased ratio of NAD(P)H fluorescence to FAD fluorescence, termed the optical redox ratio, compared with cells that are resistant to treatment [3]. The redox ratio has also been used to distinguish subtypes of breast cancer cells with different receptor statuses, including human epidermal growth factor receptor 2 (HER2)-positive, estrogen receptor (ER)-positive, and triple negative cells. Specifically, the triple negative breast cancer cell line MDA-MB-231 has been shown to exhibit a lower optical redox ratio than the HER2-positive breast cancer cell line SKBr3 [3]. To mimic cellular heterogeneity, SKBr3 cells and MDA-MB-231 cells have been mixed in culture, and distinct subpopulations of each cell line have been identified using autofluorescence microscopy of NAD(P)H and FAD [4]. Therefore, single cell measurement techniques of cellular autofluorescence are beneficial to characterize heterogeneity across breast cancer subtypes.

Single cell techniques to measure fluorescence include microscopy and flow cytometry, and each of these has unique advantages. Microscopy provides higher resolution (sub-cellular) and requires a longer dwell time (ms), whereas flow cytometry provides lower resolution (cellular) and requires a shorter dwell time (μs). Microscopy measures smaller sample sizes (hundreds of cells) compared with flow cytometry (thousands of cells). Additionally, microscopy measures adherent cells and can provide information about spatial relationships between cells, whereas flow cytometry measures cells in suspension and can sort cells to enrich cell subpopulations based on a target fluorophore. Overall, microscopy provides high signal to noise whereas flow cytometry provides high throughput and cell sorting. Flow cytometry is well-suited for characterizing cellular heterogeneity because sorting cancer cells by metabolic fluorophores could isolate subpopulations that could then be grown in vitro and used for further analysis. In particular, cell subpopulations could be tested for sensitivity to anti-cancer therapies to determine personalized treatment strategies as well as to develop new therapies for isolated resistant subpopulations.

Current methods for flow sorting tumor cell sub-populations are based on fluorescent staining for specific molecular markers, and some studies have identified markers for treatment resistance. In particular, CD44+ cells have been shown to be tumorigenic and resistant to chemotherapy in breast cancer, head and neck cancer, and pancreatic cancer [58]. CD24 has also been shown to be a marker for tumorigenic potential in breast cancer and pancreatic cancer. Additionally, CD133+ cells have been shown to be tumorigenic and resistant to chemotherapy in pancreatic cancer [9]. However, there are drawbacks to fluorophore staining for flow cytometry. In particular, labeling efficiency can affect the signal intensity from fluorophore staining, thus confounding the interpretation of “positive” and “negative” stained cells. Therefore, autofluorescence measurements can be beneficial by eliminating the need for dyes or stains. Additionally, staining for specific markers could miss cells that maintain treatment resistance yet circumvent the labeled pathway. Therefore, cellular autofluorescence might be an advantageous marker to sort cells based on overall cell metabolism, compared to traditional markers that are highly specific.

Previous studies have also applied flow cytometry based on intrinsic contrast. In addition to measuring cell fluorescence, flow cytometry measures scattering properties of the cells, including forward scattering measurements (FSC), which reflect cell size. These scattering properties have been used to distinguish cells of different sizes and types, including isolating neutrophils from leukocytes [10]. Additionally, NAD (P)H and FAD autofluorescence flow cytometry has been shown to measure response to increasing concentrations of glucose in rat b-cells, INS-1 cells, and rat islet cells [11, 12] Since flow cytometry measures fluorescence intensities per cell, cell size could affect autofluorescence measurements. For example, cells from the bottom ten percent of the autofluorescence intensity distribution have been shown to have decreased size compared with cells from the top ten percent [13]. Therefore, it is important to compare autofluorescence intensities from cells with similar sizes and FSC values.

This study applies flow cytometry for autofluorescence measurements of cell metabolism in breast cancer. Three flow cytometry channels were measured for cellular autofluorescence between two breast cancer cell lines that exhibit either overexpression of HER2 or triple negative status. Additionally, a heterogeneous sample of these two cell lines was sorted based on cell autofluorescence, and the sorted subpopulations, which were enriched for each cell line, were grown in culture. Flow sorting was validated by cell morphology, autofluorescence microscopy, and staining for HER2 receptor expression. These results indicate that flow sorting by cell autofluorescence can separate phenotypic subpopulations of cells. Ultimately, this achievement could be applied to cells from patient tissue to enable more specific testing of tumor heterogeneity in cell subpopulations sorted by treatment response, ultimately driving improved treatment regimens for cancer patients.

2. Methods

2.1 Cell culture

MDA-MB-231 and SKBr3 cells were grown in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin:streptomycin. For flow cytometry experiments, cells were trypsinized and prepared as 106 cells in 1ml phosphate buffered saline (PBS) with 5% FBS. For flow sorting experiments, cells were prepared as 7 · 106 cells in 1ml PBS with 5% FBS. Cells recovered from flow sorting were plated and grown on 35 mm glass-bottomed dishes (MatTek Corp.) for 1 week.

2.2 Flow cytometry and flow sorting

The BD LSRII instrument was used for flow cytometry analysis experiments, and 3 fluorescence channels were analyzed. The DAPI channel used an excitation wavelength of 350 nm and an emission filter of 450/50 nm. The Alexa Fluor 405 channel used an excitation wavelength of 405 nm and an emission filter of 450/50 nm. The Alexa Fluor 488 channel used an excitation wavelength of 488 nm and an emission filter of 505 nm longpass. 10,000 cells were analyzed for each group. Cells were distinguished from debris in the sample by distinct FSC values, since debris can be identified as particles with low FSC values. The cells exhibit increased fluorescence signal compared to debris in the DAPI, Alexa Fluor 405, and Alexa Fluor 488 channels (data not shown), confirming that cellular autofluorescence signals are higher than the detection limit of the flow cytometer. The BD FACSAria III instrument was used for flow sorting experiments, and 1 fluorescence channel was analyzed. The Alexa Fluor 405 channel used an excitation wavelength of 405 nm and an emission filter of 450/50 nm. Initial analysis was done on separate samples of MDA-MB-231 and SKBr3. Then the two cell lines were mixed at a ratio of 50% each, and the heterogeneous solution was analyzed and sorted. Sorting was performed using gates defined by pure cell subpopulations to enrich for each cell subpopulation and minimize contamination from other cells. Gating was determined based on autofluorescence intensity values and was done by an expert with experience in flow cytometry gating methods (JNH). For each cell line 3–5 × 105 cells were recovered after sorting. Cells were also analyzed post-sort, and sorting error was calculated by quantifying the proportion of cells in each sorted subpopulation that fall outside the gates for that subpopulation. Graphs were made in Cytobank (www.cytobank.org). Experiments were repeated in triplicate on separate days, and results were consistent across experiments. Since flow cytometry data uses large sample sizes, biological significance is interpreted from consistent trends across replicate experiments. This is in contrast to microscopy, where small sample sizes justify statistical tests within an experiment.

2.3 Microscopy validation

Sorted cells were grown in culture for 1 week and imaged with brightfield and fluorescence microscopy. For brightfield microscopy, images were acquired using an inverted microscope (EVOS, Fisher Scientific) and 4× objective. For fluorescence microscopy, images were acquired using an inverted two-photon microscope (Bruker, and TiE, Nikon) and 40× oil-immersion objective (1.3 NA). Incident light was provided with a titanium:sapphire laser (Coherent, Inc.). A GaAsP photomultiplier tube (H7422P-40, Hamamatsu) was used to collect fluorescent photons. NAD(P)H was measured using an excitation wavelength of 750 nm and an emission filter of 440/80 nm. FAD was measured using an excitation wavelength of 890 nm and an emission filter of 550/100 nm. NAD(P)H and FAD were measured from the same fields of view. Each image averaged 4 frames, and 9 fields of view were imaged per group. Microscopy images were analyzed on a per-cell basis using a CellProfiler routine described previously [14]. Briefly, the optical redox ratio was calculated by dividing the image of NAD(P)H by the image of FAD for the same field of view. NAD(P)H images were thresholded to identify cell cytoplasms. NAD (P)H, FAD, and redox ratio images were quantified per cell.

2.4 HER2 expression validation

A fluorescently labeled anti-HER2 antibody, HER2-Sense (5 μM, PerkinElmer), was used to validate flow sorting. HER2Sense exhibits optimal excitation at 643 nm and emission at 661 nm. This fluorescence was measured with the Alexa 647 channel, which has an excitation wavelength at 633 nm and a collection filter of 660/20 nm. Cells were stained with HER2Sense for 30 minutes and washed with PBS. MDA-MB-231 and SKBr3 cells were characterized separately, mixed to form a heterogeneous solution, and sorted based on autofluorescence in the Alexa Fluor 405 channel. Next, sorted cells were characterized for autofluorescence in the Alexa Fluor 405 channel and for HER2 expression in the Alexa 647 channel. Sorted cells were grown for 1 week in culture, stained with HER2Sense for 30 minutes, washed with PBS, and imaged with an inverted confocal microscope (Meta, Zeiss) and 40× oil-immersion objective using the cy5 channel (maximal excitation and emission wavelengths of 647 nm and 665 nm). Pure, unstained MDA-MB-231 and SKBr3 cells served as negative controls for flow cytometry and confocal microscopy experiments.

2.5 Statistical analysis

Bar graphs are shown as mean ± standard error. Statistical testing was performed using two-tailed t-tests with an α of 0.05 indicating statistical significance.

3. Results

The triple negative breast cancer cell line MDA-MB-231 and the HER2-positive breast cancer cell line SKBr3 were characterized in the DAPI, Alexa Fluor 405, and Alexa Fluor 488 channels (Figure 1A). NAD (P)H has an excitation maximum at 351 nm and an emission maximum at 440 nm [15]. These spectral properties align with the DAPI flow cytometry channel, which excites at 350 nm and collects emission between 450/50 nm. The Alexa Fluor 405 flow cytometry channel excites at 405 nm and collects emission between 450/50 nm. Fluorescence from these wavelengths has been shown to include a combination of NAD(P)H and FAD signals [16]. FAD has an excitation maximum at 450 nm and an emission maximum at 535 nm [15]. These spectral properties align with the Alexa Fluor 488 flow cytometry channel, which excites at 488 nm and collects emission longer than 505 nm. The SKBr3 cells exhibit an increased fluorescence signal in the DAPI and Alexa Fluor 405 channels compared with MDA-MB-231 cells. SKBr3 cells show a slight increase in the Alexa Fluor 488 channel compared with MDA-MB-231 cells. The FSC measurements show similar cell sizes between the cell lines (Figure 1B).

Figure 1
Cell line characterization. (A) MDA-MB-231 cells exhibit lower fluorescence intensity in the DAPI, Alexa Fluor 405, and Alexa Fluor 488 channels compared with SKBr3 cells. (B) MDA-MB-231 and SKBr3 cells exhibit similar FSC measurements, which reflect ...

The distinct fluorescence intensities of MDA-MB-231 and SKBr3 cells in the Alexa Fluor 405 channel justifies applying this channel for sorting a heterogeneous sample of these two cell lines. Pure samples of MDA-MB-231 and SKBr3 cells show separate peaks for each cell line (Figure 2A, blue, orange). The cell lines were mixed to create a heterogeneous sample, and this sample exhibits two distinct peaks that align with the peaks of each pure cell line (Figure 2A, green). The sample was sorted based on fluorescence intensity in the Alexa Fluor 405 channel. Similarly, the sorted cell subpopulations exhibit peaks that align with the peaks of each pure cell line (Figure 2A, red, purple). Error in the sorting was less than five percent based on post-sort analysis. FSC measurements indicate similar cell sizes between the MDA-MB-231 cells, SKBr3 cells, mixture of the two cell lines, and sorted SKBr3 cells (Figure 2B). The sorted MDA-MB-231 cells exhibit slightly lower FSC values.

Figure 2
Flow sorting. (A) MDA-MB-231 cells and SKBr3 cells measured separately exhibit distinct fluorescence intensities in the Alexa Fluor 405 channel. A mixture of MDA-MB-231 and SKBr3 cells exhibit two peaks, representing each cell type. Flow sorting the mixture ...

The sorted cell subpopulations were grown in culture for one week and validated with brightfield microscopy to visualize cell morphology (Figure 3). Images show pure MDA-MB-231 cells exhibit an elongated morphology, whereas pure SKBr3 cells show a round and grouped morphology. Furthermore, the sorted cells from each subpopulation exhibit morphologies that align with the pure cell subpopulations.

Figure 3
Brightfield microscopy validates cell sorting. Images illustrate an agreement in morphology between pure and sorted MDA-MB-231 and SKBr3 cell lines.

Additionally, fluorescence microscopy was applied to validate cell sorting (Figure 4). Sorted cells were grown in culture for a week and autofluorescence images of NAD(P)H and FAD were acquired. Representative images show the expected cell morphology for each cell line (Figure 4A). The microscopy images were quantified on a cellular level to compare fluorescence intensities between the sorted cell subpopulations. The sorted SKBr3 cells exhibit higher NAD(P)H intensity (Figure 4B), FAD intensity (Figure 4C), and redox ratio (Figure 4D) compared with MDA-MB-231 cells (p < 0.05), and these results agree with values in pure populations of MDA-MB-231 and SKBr3 cells.

Figure 4
Fluorescence microscopy validation of flow sorting a mixture of MDA-MB-231 and SKBr3 cells. (A) Representative images of NAD(P)H and FAD autofluorescence show the expected morphology from MDA-MB-231 and SKBr3 cells grown in culture for 1 week after flow ...

Fluorescence staining of HER2 expression was applied for a final validation of flow sorting (Figure 5). HER2Sense labels HER2-positive cells, and excites and emits at wavelengths longer than NAD(P)H and FAD so it can be spectrally separated in the Alexa 647 channel. Flow sorting based on autofluorescence in the Alexa Fluor 405 channel was repeated with cells stained for HER2Sense (Figure 5A). Unstained controls exhibit minimal fluorescence in the Alexa 647 channel (Figure 5B, brown, pink). Stained SKBr3 cells exhibit increased fluorescence signal in the Alexa 647 channel compared with stained MDA-MB-231 cells (Figure 5B, blue, orange). The cell lines were mixed, and the fluorescence profile of the mixture (Figure 5B, green) matches the sum of the initial populations. The mixture was sorted based on autofluorescence in the Alexa Fluor 405 channel (Figure 5A). After sorting, the Alexa 647 fluorescence for each subpopulation (Figure 5B red, purple) matches the peaks for each pure cell line. FSC measurements (Figure 5C) are similar across the samples, indicating similar cell sizes across the groups. Additionally, sorted cells were grown in culture for a week and fluorescence microscopy of HER2Sense was performed (Figure 5D). Unstained controls show low background signal. As a positive control, stained SKBr3 cells show the localization of HER2 in the cell membrane, whereas stained MDA-MB-231 cells exhibit low signal. Similarly, sorted subpopulations of SKBr3 cells show the localization of HER2 in the cell membrane, whereas MDA-MB-231 cells exhibit low signal.

Figure 5
HER2 staining validates flow sorting. (A) Autofluorescence in the Alexa Fluor 405 channel sorts a mixture of MDA-MB-231 and SKBr3 cells. (B) Staining with a fluorescent anti-HER2 antibody (HER2Sense) validates flow sorting. (C) FSC measurements indicate ...

4. Discussion

Tumor heterogeneity describes multiple subpopulations of cells within a tumor. Subpopulations of cells that are resistant to treatment can cause patient relapse. Therefore, sorting these subpopulations could enable improved treatment regimens that target all cell phenotypes. The goal of this study is to apply flow cytometry and flow sorting to breast cancer cell metabolism based on autofluorescence. The autofluorescence profiles of MDA-MB-231 triple negative breast cancer cells and SKBr3 HER2-positive breast cancer cells were measured using three autofluorescence flow cytometry channels. A heterogeneous mixture of these two cell lines was sorted into subpopulations enriched for each cell line based on autofluorescence flow cytometry. Sorting was validated with flow cytometry, brightfield microscopy, autofluorescence microscopy, and HER2 staining. Overall, this technique could be applied to tumor cells from patient tissue by separating subpopulations of cells, testing sensitivities to anti-cancer treatments, and planning optimal treatment schemes for individual patients.

Flow cytometry measures distinct autofluorescence intensities between the triple negative MDA-MB-231 cell line and the HER2-positive SKBr3 cell line (Figure 1). The DAPI and Alexa Fluor 405 channels exhibit increased fluorescence intensity in SKBr3 cells compared to MDA-MB-231 cells. The increase in signal for the SKBr3 cells compared with MDA-MB-231 cells in the Alexa Fluor 405 channel could result from the greater magnitude of increase in NAD(P)H fluorescence compared with FAD fluorescence. These results match published redox ratio results from these cell lines using confocal microscopy [3]. Although the DAPI channel best aligns with the spectra properties of NAD(P)H, the 350 nm laser is rare in flow cytometers and flow sorters, whereas the 405 nm laser used in the Alexa Fluor 405 channel is common. The FSC measurements show no difference in cell size between the cell lines, which confirms that changes in autofluorescence intensity are not an artifact of cell size. Overall, the separation between MDA-MB-231 and SKBr3 cells in the Alexa Fluor 405 channel indicate that this channel could be used to sort these cell lines based on their autofluorescence.

Flow sorting based on the Alexa Fluor 405 channel separates a heterogeneous mixture of breast cancer cells into subpopulations of HER2-positive SKBr3 cells and triple negative MDA-MB-231 cells (Figure 2). The FSC measurements indicate a slightly smaller cell size for the sorted MDA-MB-231 cells. This can be explained by slightly different gates defined to identify these cells for this replicate. Additional replicates show similar cell sizes for the sorted MDA-MB-231 cells compared with the pure MDA-MB-231 cells, as shown in Figure 5C. It is important to note that shifts in FSC values and fluorescence values represent different magnitudes, since FSC values are plotted on a linear scale while fluorescence values are plotted on a log scale. These results lay the foundation for future applications to dissociate cells from tumor tissue and sort subpopulations with different metabolic phenotypes, receptor statuses, and treatment sensitivities. Furthermore, the isolation of treatment-resistant cells could enable additional characterization of these cell subpopulations to identify targets for drug development as well as tests for sensitivity to drugs and drug combinations that aim to eliminate these resistant cells.

Brightfield microscopy, fluorescence microscopy, and HER2 staining provide additional validations for cell sorting. Cell morphologies of sorted cells grown for one week in culture align with morphologies of each pure cell line (Figure 3) and are consistent with previous morphological studies of these cell lines grown on glass [17]. Autofluorescence characterization, particularly NAD(P)H fluorescence intensity, FAD fluorescence intensity, and redox ratio, of sorted subpopulations enriched for each cell type agrees with published confocal microscopy results from SKBr3 and MDA-MB-231 cell lines (Figure 4) [3]. HER2 staining further validates the sorting of a heterogeneous mixture into subpopulations of each cell line (Figure 5). Previous studies have shown that SKBr3 cells stained for HER2 expression exhibit positive membrane signal, whereas MDA-MB-231 cells exhibit minimal signal [18]. Flow cytometry and confocal microscopy of HER2Sense confirm increased HER2 expression in pure and sorted SKBr3 cells compared with pure and sorted MDA-MB-231 cells. Overall, these microscopy and staining techniques confirm the separation of a heterogeneous mixture of breast cancer cell lines with distinct phenotypes into subpopulations enriched for each cell type based on autofluorescence flow sorting.

Practical applications could incorporate flow sorting using cell autofluorescence with standard techniques using fluorescent labels. Traditional surface labels used in flow cytometry can provide a binary answer that could be applied for characterizing heterogeneity. Autofluorescence measurements provide a continuous variable, which could be advantageous to monitor more subtle differences in tumor metabolic heterogeneity. This method could extend to patient-derived tumor cell populations by isolating tumor cells from other cell types in the tumor. In particular, scattering properties (e.g. FSC) could be used to separate cells with a smaller size, or protocols for labeling surface marker expression could isolate tumor cells. Gradients in autofluorescence intensity values could determine gates for flow sorting. Thereby, autofluorescence flow cytometry could be used in combination with traditional molecular markers to characterize cell subpopulations in primary human tumors. Additionally, this initial demonstration sorts cells with different receptor expression, but this approach could be applied to cells with different metabolic phenotypes regardless of receptor status. Cells with similar receptor status could exhibit different activation levels of downstream molecules, like phosphoinositide 3-kinase (PI3K), which can impact cell metabolism [19]. Thus, autofluorescence profiling could provide information on internal cell signaling and metabolism that is not provided by traditional receptor profiling. Sorting cells with these different molecular mutations based on autofluorescence would be clinically beneficial for identifying treatment-resistant cells within tumors. These autofluorescence-sorted cell subpopulations could be further characterized (e.g. proteomics, genomics) to identify molecular targets for novel treatments.

Tumors can contain cells with different receptor statuses, metabolic profiles, and responses to treatments. Cancer patients often exhibit relapse after treatment, which could be attributed to subpopulations of cells that are resistant to treatment. Single-cell measurements that reflect treatment response would allow improved treatment regimens for cancer patients. This study applies flow cytometry and flow sorting to triple negative breast cancer cells and HER2-positive breast cancer cells. These results measure three flow cytometry channels for cell autofluorescence and indicate that flow cytometry based on cellular autofluorescence distinguishes two breast cancer cell lines with different receptor expressions. Furthermore, flow sorting based on cell autofluorescence separates a heterogeneous mixture into subpopulations enriched for each cell line. Ultimately, this technique could analyze cell heterogeneity from tumor tissue by sorting cell subpopulations based on metabolic profile. This method could be applied to preclinical studies for drug discovery, testing experimental drugs, and optimizing drug combinations. Furthermore, this method could be applied clinically to tissue from cancer patients to develop individualized treatment strategies and improve patient outcomes.

Acknowledgments

Flow cytometry experiments were performed in the Vanderbilt Medical Center (VMC) Flow Cytometry Shared Resource. The VMC Flow Cytometry Shared Resource is supported by the Vanderbilt Ingram Cancer Center (P30 CA68485) and Vanderbilt Digestive Disease Research Center (DK058404). Confocal microscopy experiments were performed in the Cell Imaging Shared Research at Vanderbilt. Funding sources include the NSF Graduate Research Fellowship (DGE-0909667), Mary Kay Foundation (067-14), NIH/NCI (R01 CA185747), DoD (W81XWH-13-1-0194), and Vanderbilt University, as well as NCI GI Special Programs of Research Excellence P50 95103 and R01 CA163563 to RJC.

References

1. Khong HT, Restifo NP. Nat Immunol. 2002;3(11):999–1005. [PMC free article] [PubMed]
2. Zhao Y, Butler EB, Tan M. Cell Death Dis. 2013;4(3):e532. [PMC free article] [PubMed]
3. Walsh AJ, Cook RS, Manning HC, Hicks DJ, Lafontant A, Arteaga CL, Skala MC. Cancer Res. 2013;73(20):6164–6174. [PMC free article] [PubMed]
4. Walsh AJ, Skala MC. Biomed Opt Express. 2015;6(2):559. [PMC free article] [PubMed]
5. Li X, Lewis MT, Huang J, Gutierrez C, Os-borne CK, Wu MF, Hilsenbeck SG, Pavlick A, Zhang X, Chamness GC, Wong H, Rosen J, Chang JC. J Natl Cancer Inst. 2008;100(9):672–679. [PubMed]
6. Al-Hajj M, Wicha M, Benito-Hernandez A, Morrison S, Clarke M. Proc Natl Acad Sci USA. 2003;100(7):3983–3988. [PubMed]
7. Prince ME, Kaczorowski RSA, Wolf GT, Kaplan MJ, Dalerba P, Weissman IL, Clarke MF, Ailles LE. Proc Natl Acad Sci USA. 2007;104(3):973–978. [PubMed]
8. Creighton CJ, Li X, Landis M, Dixon JM, Neumeister VM, Sjolund A, Rimm DL, Wong H, Rodriguez A, Herschkowitz JI, Fan C, Zhang X, He X, Pavlick A, Gutierrez MC, Renshaw L, Larionov Aa, Faratian D, Hilsenbeck SG, Perou CM, Lewis MT, Rosen JM, Chang JC. Proc Natl Acad Sci USA. 2009;106(33):13820–13825. [PubMed]
9. Hermann PC, Huber SL, Herrler T, Aicher A, Ellwart JW, Guba M, Bruns CJ, Heeschen C. Cell Stem Cell. 2007;1(3):313–323. [PubMed]
10. Dorward Da, Lucas CD, Alessandri AL, Marwick Ja, Rossi F, Dransfield I, Haslett C, Dhaliwal K, Rossi AG. J Leukoc Biol. 2013;94:1–10. [PMC free article] [PubMed]
11. Van de Winkel M, Pipeleers D. Biochem Bio-phys Res Commun. 1983;114(2):835–842. [PubMed]
12. Hanson MS, Steffen A, Danobeitia JS, Ludwig B, Fernandez LA. Cell Transplant. 2008;17(12):1337–1347. [PubMed]
13. Tzur A, Moore JK, Jorgensen P, Shapiro HM, Kirschner MW. PLoS One. 2011;6(1):1–9. [PMC free article] [PubMed]
14. Walsh AJ, Skala MC. SPIE Proc 89481M. 2014 Feb;
15. Ramanujam N. Neoplasia. 2000;2(1–2):89–117. [PMC free article] [PubMed]
16. Wu Y, Qu JY. J Biomed Opt. 2006;11(5):054023. [PubMed]
17. Sommers CL, Byers SW, Thompson EW, Torri JA, Gelmann EP. Breast Cancer Res Treat. 1994;31(2–3):325–335. [PubMed]
18. Liu J, Li J, Rosol TJ, Pan X, Voorhees JL. Phys Med Biol. 2007;52(16):4739–4747. [PubMed]
19. Dang CV. Genes Dev. 2012;26(9):877–890. [PubMed]