Portal de Periódicos da CAPES

Quantitative phase imaging for label‐free cytometry

2017; Wiley; Volume: 91; Issue: 5 Linguagem: Inglês

10.1002/cyto.a.23130

1552-4930

Elena Holden, Attila Tárnok, Gabriel Popescu,

Advanced Fluorescence Microscopy Techniques

In order to understand complex biological processes, scientists must gain insights into the functioning of individual live cells. Unlike fixed cell imaging, where a single snapshot of the cell's life is retrieved, live-cell imaging allows investigation of the dynamic processes underlying the cell's function. Label-free imaging avoids the limitations inherent to fluorescent probes (phototoxicity, photobleaching) and maintains an appropriate environment for normal cellular behavior. Typical mammalian cells are transparent phase objects which do not absorb or scatter much light. Enhanced visualization of unlabeled cells was enabled in 1932 by the development of phase contrast (PC) technique by Frits Zernike 1, 2 who, during the decade that followed, collaborated with Carl Zeiss AG resulting in a release of the first phase contrast microscope, revolutionizing the field of live cell dynamic imaging. PC is based on an optical set-up that translates small variations in phase into corresponding changes in amplitude, which consequently can be visualized as differences in image contrast. PC and various derivatives, such as differential interference contrast, became widely adopted techniques for imaging thin live cells in culture. However, these methods only provide a means for visualizing the cells and not making measurements, that is, the phase information they output is qualitative. Traditional PC techniques are not quantitative as they do not provide a direct measurement of a phase delay with enumeration as pixel intensity. This special issue is focused on introducing to our readership the subject of Quantitative Phase Imaging (QPI) 3 and its benefits to cytometry. QPI is a valuable method for studying live cell dynamics, as it provides a noninvasive analysis over a wide range of time scales. This type of analysis is gaining traction very rapidly because it is performed with little to no phototoxicity and requires minimal sample preparation. There are no effects of biological and chemical labels or genetic modification, which would alter cellular behavior. QPI offers the benefit of repeated observations and quantitative analysis of cell cultures over time providing minute-by-minute insight into cell proliferation, cell death, and transient events. Quantitative measurements are based on direct phase image analysis of cell structure. QPI yields optical path difference maps associated with the specimen of interest and, as such, it is sensitive to both local thickness and the refractive index of the sample. Several QPI related publications have previously appeared in Cytometry Part A, paving the way for this new field of applications 4-6. A collection of manuscripts in this special issue attests to the fact that QPI is becoming a prominent technique complementary to traditional cytometry technologies and indispensable in dynamic label-free live-cell analysis applications. It has been recognized recently that, because in QPI one has knowledge of the complex field information, the data can be numerically processed to obtain angular scattering information. This approach, referred to as Fourier transform light scattering (FTLS) is the spatial analog of Fourier transform (infrared) spectroscopy 7. Because the data is collected in the image plane, where at each point there are contributions from various angles simultaneously, the FTLS benefits from orders of magnitude higher sensitivity when compared with traditional, goniometer-based angular scattering measurements. It is well-known that in many scattering-based cytometers, the forward scattering provides global information about the cell (i.e., its size), while large angle scattering reveals the internal structure of the cell. This is a standard approach for obtaining a complete blood count in the clinic. Some cytometry instruments contain detectors at several angles, to yield more structural details and higher accuracy in cell classification. Remarkably, using QPI data, FTLS yields all the scattering angles within the numerical aperture simultaneously. Thus, it is not surprising that QPI is a particularly efficient method for cytometry. Furthermore, dynamic QPI imaging is entirely equivalent to dynamic light scattering and its potential to studying transport phenomena in live cells has been demonstrated 8. There are multiple variants of QPI systems that can be categorized with respect to parameters that describe their performance 9. The main figures of merit are: acquisition rate, transverse resolution, temporal phase sensitivity, and spatial phase sensitivity. Acquisition rate establishes the fastest phenomena that can be studied by a QPI method. According to the Nyquist sampling theorem (or Nyquist-Shannon theorem), the sampling frequency has to be at least twice the frequency of the signal of interest 10, 11. In QPI, the required acquisition rates vary broadly with the application, from 100's of Hz in the case of membrane fluctuations to 1/1,000 Hz when studying the cell cycle. The acquisition rate of QPI systems depends on the modality used for phase retrieval. Off-axis interferometry gives the phase map from a single camera exposure and is thus the fastest. On the other hand, phase-shifting techniques (see scheme in Fig. 1) are slower as they require several intensity images for each phase map. Principles of quantitative phase imaging (QPI). (a) Phase shifting QPI. (b) Off-axis QPI. Both geometries can be implemented in common path and with broad band illumination. (c) Quantitative phase image of a neuron. The insets zoom into the respective selected areas. Color bar indicates optical pathlength in nanometers. (d) Pathlength profile along the line shown in 1c top right inset. In QPI it is desirable to preserve the diffraction limited resolution provided by the microscope 12. Defining a proper measure of transverse resolution in QPI is nontrivial and perhaps worth pursuing by theoretical researchers. Such a definition must take into account that the coherent imaging system is not linear in phase (or in intensity), but in the complex field. Phase-shifting methods are more likely than off-axis methods to preserve the diffraction limited resolution of the instrument. In off-axis geometries, the issue is complicated by the additional length scale introduced by the spatial modulation frequency (i.e., the fringe period). Following the Nyquist sampling theorem, this frequency must be high enough to recover the maximum frequency allowed by the numerical aperture of the objective. By contrast, in phase shifting, the phase image recovery involves only simple operations of summation and subtraction, which preserves the original resolution of the intensity images. Temporal stability is perhaps the most challenging feature to achieve in QPI. In studying dynamic phenomena, the question that often arises is: what is the smallest phase change that can be detected at a given point in the field of view? Sources of phase noise include air fluctuations, mechanical vibrations of optical components, vibrations in the optical table, etc. In order to improve the stability of QPI systems, there are several approaches typically pursued: (i) Passive stabilization, (ii) Active stabilization, (iii) Differential measurements, and (iv) Common path interferometry. Common path systems consist of two fields that travel along physically overlapping paths. In this case, the noise in both fields is very similar and hence automatically cancels in the interference (cross) term. Thus, common path interferometers are inherently stable. Analog to the "frame-to-frame" phase noise discussed in the previous section, there is a "point-to-point" (spatial) phase noise that affects the QPI measurement. This spatial phase sensitivity limits the smallest topographic change that the QPI system can detect within a field of view. Unlike with temporal noise, there are no clear cut solutions to improve spatial sensitivity besides keeping the optics pristine and decreasing the coherence of the illumination light. The spatial non-uniformities in the phase background are mainly due to the random interference pattern (i.e., speckle) produced by fields scattered from impurities on optics, specular reflections from the various surfaces in the system, etc. This spatial noise is worst in highly coherent sources, that is, lasers. Using white light as illumination drastically reduces the effects of speckle while preserving the requirement of a coherence area that is at least as large as the field of view. It is apparent from the discussion so far that there is no perfect QPI method, that is, there is no technique that performs optimally with respect to all figures of merit identified in the last section. We summarize the QPI approaches and their performances in Table 1. In sum, the off-axis methods are fast as they are single shot, phase-shifting preserves the diffraction-limited transverse resolution without special measures, common-path methods are stable and white light illumination suffers less from speckle and, thus, more spatially uniform. As the diagonal table above suggests, we can think of these 4 figures of merit as the "normal modes" of categorizing QPI techniques. However, there are methods that combine these four approaches, seeking to add the respective individual benefits. There are possible combinations of 2 benefits, combinations of 3 benefits and 1 that combines all of them. Unlike traditional PC technologies, the images produced by QPI technologies are quantitative and lend themselves to reliable segmentation routines 13. Luther et al. (this issue, page 412), Guo [Moses] et al. (this issue, page 424), and Allier et al. (this issue, page 433) exemplify this point by showing image acquisition progression from a raw image to the reconstructed QPI image with segmentation applied contours. As opposed to poor separation between signal and background noise in traditional PC images, QPI techniques produce very high contrast images allowing accurate segmentation. As a result, the raw QPI data is translated into groups of representative cytometric parameters that can be measured on a per cell basis. For example, dry mass and optical thickness of a cell represents basic quantitative biophysical parameters. Segmentation routines allow calculation of morphological parameters such as cell area, perimeter, optical volume, circularity, texture, etc., as well as distance, traveled speed and directness of migration. Alanazi et al. (this issue, page 443) present newly developed segmentation algorithms of individual Saccharomyces cerevisiae cells by employing the optical-phase metric of the transmitted wave front rather than its intensity claiming success rates greater than 99%. Using the Phi Optics CellVista SLIM instrument, their findings indicate that the optical-phase thresholding paradigm is computationally less intensive and may assist in the scaling-up of high-throughput imaging with single-cell resolution. A group of researchers led by Cédric Allier reveals the results of their efforts to validate multi-wavelength lens-free video-microscope (this issue, page 433). Their incubator friendly set-up utilizes a multi-wavelength LED illumination with well separated wavelengths acquiring widefield images (29.4 mm2). The proof of performance and accuracy of their holographic reconstruction algorithm was tested over a range of 31 cell lines, including those at highly confluent conditions (up to 700 cells/mm2). Researchers from MIT (Jin et al. this issue, page 450) describe an automated imaging system based on line-field interferometric quantitative phase measurements of adherent cellular samples advanced by a motorized XY stage. The authors suggest an improved image acquisition rate which is demonstrated by characterizing approximately 104 HeLa cells, obtaining dry mass distribution on a per cell basis. The approach is important for the increased statistical power of the experiments. It is also expected to be potentially useful for analysis of cells in microfluidic devices. Janicke et al. (this issue, page 460) from Phase Holographic Imaging emphasize the need for practical methods that allow scientists to handle a large number of samples in standard multi-well plates. They present a validation study of an automated, label-free, kinetic proliferation assay. The study was performed on L929, Jimt-1, and SK-MEL-5 cell lines treated with increasing concentrations of the topoisomerase II inhibitor Etoposide. Image acquisition for 36 hours at 30 min intervals and data analyses were performed using their commercially available digital holographic cytometer. Authors conclude that the growth rate data based on cell number, confluence and average optical cell volume, provides an advantage over current single-parameter and end-point methods. Kastl et al. (this issue, page 470) constructed an elegant and systematic study convincing us that QPI cytometry based on a digital holographic microscopy allows the non-invasive quantification of the response of single cells to changing cell culture conditions, that is, osmolarity variations and different cell confluency. In their experimental model two pancreatic tumor cell lines were analyzed on detached cells for refractive index, volume, and dry mass. The authors demonstrated these parameters to be reliable and absolute biophysical measures of cell culture quality. The study presented by Guo [Moses] et al. (this issue, page 424) attests to the value of the QPI technology for demonstrating the differences between angiogenic and non-angiogenic phenotypes of human osteosarcoma cells. A panel of cell biophysical, morphological and behavior parameters was obtained continuously and simultaneously by a holographic imaging cytometry system over 48 hours at 5 min intervals. The authors compare this new approach with common methodologies, that is, standard phase microscopy in well-established cell proliferation, cell viability, and cell migration assays. The presented results show that that angiogenic cells have a smaller cell area, increased cell thickness, and significantly higher cell motility speeds when compared with dormant cells. They suggest that this new method is simple, cost-efficient and has the potential to offer new insights into molecular mechanisms underlying the escape from tumor dormancy in human cancers which may eventually be translated into important clinical applications. The new study by Roitshtain et al. (this issue, page 482) evaluates the value of 15 biophysical and morphological parameters in live tumor cells and normal control cells using low-coherence off-axis interferometric phase microscopy. Interestingly, this set up allows a single-exposure acquisition mode suitable for the quantitative imaging of cells in suspension. The authors compare normal skin fibroblasts to a primary melanoma cell line, melanoma cell line derived from primary tumor cells to cells from metastatic site on the skin, and finally a cell line established from a primary colon adenocarcinoma to cells cultured from metastatic lymph node of the same patient, providing tumor and control counterparts for comparative studies. Upon implementing a comprehensive machine learning algorithm, this study claims the specificity of cell type classification (healthy/cancer/metastatic) of 81%–99% and 81%–93% sensitivity. A study from Tokyo University (Guo [Goda] et al. this issue, page 494) continues the theme of developing quantitative phase imaging of cells in flow applied to microalgal lipids. In this case, an opto-fluidic time-stretch QPI technique with hydrodynamic-focusing microfluidic chip was optimized and compared with previously reported studies. The system demonstrates high throughput capability of imaging 103 cell/s and is complemented by a machine learning algorithm which utilizes both intensity- and phase-derived parameters on a per cell basis. Authors demonstrate promising results increasing accuracy of the cell classification from 88% with either intensity or phase images alone to 93.3% based on the newly developed method. We learn that the accurate cell classification is advantageous for selective breeding of Euglena gracilis for efficient biofuel production and may be extended to other types of microalgae. Luther et al. (this issue, page 412) present a comprehensive review of validating a recently developed QPI-based digital holographic cytometer HoloMonitor to assist a nanotechnology-based anti-tumor pharmaceutical development programs at Northeastern University. It enabled quantification of the effects of new formulations by tracking rare cells of interest and devising intuitive and informative methodology for displaying phase holographic time-lapse image data in 4-dimenional displays (biophysical feature, x-y-position, and time). These studies are particularly informative because of the time dimension. Kinetic imaging for 2–3 days at 5 min intervals allows precise monitoring of cell cycle related toxicology effects and determining the commencement of compound-specific effects. Cintora et al. (this issue, page 503) present a study of dry mass content and intracellular transport rates in neural networks. The authors developed and employed highly sensitive Spatial Light Interference Microscopy technology (SLIM) and an additional proprietary Dispersion-Relation Phase Spectroscopy (DPS) analysis method to spatially (down to the fraction of a nanometer) and temporally (29.7 hours at 81 min intervals) resolve SLIM data. They examine the average dry mass, and the spatial and temporal distribution of active and passive transport processes under different neuron plating densities. This study provides evidence that cell confluence significantly affects measured parameters with contact inhibition postulated to play a role in reducing growth rates for high-confluence regions. The authors reiterate that their method is label-free and does not require neuronal tracing, particle tracking, or neuron reconstruction, and offers a full field-of-view quantitation at high resolution, provides long-term imaging capabilities with an expectation that the technology will be widely utilized in neurobiology applications. The work by Yang et al. (this issue, page 510) describes analysis of a drug-induced cellular model of Parkinson's disease by optical diffraction tomography, a 3D QPI technique developed and commercialized by the authors of the study. A study by Kandel et al. (this issue, page 519) uses SLIM/DPS configuration but extends their analysis method to investigation of cellular transport in neuron culture in three dimensions resulting in simultaneous analysis of the horizontal and vertical traffic of mass through a cell. This is believed to be the first time-resolved tomograms obtained by full-field white-light QPI. Authors found statistically significant differences between diffusion coefficients and spread of advection velocities in the cell bodies versus neurites and longitudinal versus transverse traffic. Merola et al. (this issue, page 527) examine the optical behavior of Red Blood Cells (RBCs) under optically induced mechanical stress by a combination of digital holography and Holographic Optical Tweezers (HOT). The authors view RBCs as "lenses" filled with uniform liquid showing optical focusing properties. They exploited this RBC lens effect to analyze the wavefront transmitted by healthy RBCs and cells stretched with different HOTs configurations allowing them to precisely quantify various aberrations to characterize the "lens quality." Having the complete control on the lens deformation by HOTs, authors foresee important diagnostic applications by studying the deformed wavefront giving information on the cell "healthiness" and evaluating different cell elasticity and shape changes. More specifically, the authors suggest that the technique may be useful in assessing storage lesions associated with blood storage, retrieving the age of a cell and assessing the stored blood quality. We hope you will find the articles published in this special volume thought provoking. QPI field spans a broad area of interests and biological applications. It is a relatively new member of the cytometry instrumentation techniques and developing very rapidly. Multiple, commercially available QPI systems are now available (Table 2). We anticipate that the subjects of performance metrics (sensitivity, specificity, reproducibility, and reliability of QPI measurements) will be addressed in systematic ways compliant with the principles of cytometry, especially in applications proposed for clinical use. A pressing need has already emerged in developing optimal analysis strategies and intelligent machine learning algorithms of large multi-dimensional data sets. Finally, we wish to thank the dedicated scientists and technology developers for their contributions to the QPI field and making this special issue possible.