Introduction

Viable cells are difficult to visualise microscopically because there is usually little difference in contrast between the cytoplasm and the extracellular milieu. Cellular structures can be imaged and identified after staining or labelling, but these procedures affect the viability of the specimen [13]. In addition to the visualisation challenges associated with cellular transparency, further optical problems are encountered where it is necessary or desirable to image cell preparations through plastic culture-ware. It is extremely useful, however, to be able to image living cells in culture or suspension for a variety of reasons including cell lineage maintenance and to undertake experiments investigating the effects of growth interventions in vitro.

The implementation of optical phase microscopy, which enhances the contrast between cells and media, has been of considerable assistance in the visualisation of viable, unstained cellular specimens. Invented in the 1930’s by Fritz Zernike [21], optical phase microscopy uses a phase plate to alter the passage of light passing directly through a specimen so that it is 1/2-wavelength shifted from light deviated by the specimen. This results in destructive interference and allows the details of the specimen to appear relatively dark against a light background (i.e. a difference in phase retardation is converted into an intensity contrast, or amplitude difference). The extent of the phase-shift induced is determined by a combination of the refractive index and the thickness of a specimen at any point [5]. Thus cellular structures may be rendered more visible using optical phase microscopy than can be achieved using bright-field microscopy. Various implementations of phase microscopy have been utilised to visualise unstained specimens, including differential interference contrast [15] and Hoffman modulation contrast [11]. Although each of these methods allows enhanced visualisation of transparent specimens, they all have inherent problems, including the generation of distorted “halos” at the edges of the cells (where boundary conditions may reflect rather than refract light), making visual analysis difficult. More importantly, the information provided by these techniques is useful primarily for viewing purposes and for qualitative analysis only.

Quantitative phase microscopy (QPM) is a new computational method for visualising translucent specimens: i.e. materials poor in amplitude characteristics but rich in phase information content. QPM has been employed recently in the analysis of inert, geometrical specimens including optical fibres [18] and also in the investigation of inorganic materials [4, 6]. QPM uses an algorithm that is applied to a standard “in-focus” bright-field image, and one positive and one negative de-focus image set. Calculations based on these three images enable the generation of a phase map that contains information about the specimen thickness and refractive index and can allow quantitation of specimen structure [5]. When applied to viable cell preparations, QPM not only enhances cell visualisation [3, 7], but also offers the potential to provide quantitative data about cellular characteristics [6]. One of the benefits of QPM is the absence of cell edge distortion and haloing that is inherent in other forms of optical phase microscopy [7]. As the phase map is calculated from conventional bright-field images, it is devoid of these artefacts and therefore results ultimately in a more accurate representation of cell morphology that is also more amenable to analysis. Compared with the optical phase image, cell contrast is discriminated better in the phase map, allowing for more accurate cell delineation and characterisation.

The goal of this study was to develop and evaluate the usage of QPM for the systematic analysis of cell growth in culture in situ. Human airway smooth muscle (HASM) cells were utilised as a test culture system for the development of QPM methodology. HASM cells can be cultured in a thin monolayer and display a relatively simple internal structure that is difficult to visualize under bright-field conditions. We and others have extensively documented the growth characteristics of this cell type in culture using a variety of disruptive techniques including fluorescence-activated cell sorting (FACS), haemocytometry and DNA and protein synthesis [8, 12, 17, 20]. Here we report the use of QPM techniques to track cell culture growth by repeated time-lapse confluency measurement. This investigation involved optimising, validating and benchmarking a QPM-derived phase map segmentation and thresholding procedure for a non-subjective, non-destructive measurement of cell culture confluency.

Materials and methods

Cell culture and preparation

HASM cells were obtained by collagenase and elastase digestion from bronchi of lung transplant resection patients. Cultures were maintained in phenol red-free DMEM with 10% FCS, supplemented with 2 mM l-glutamine, 100 U/ml penicillin-G, 100 µg/ml streptomycin and 2 µg/ml amphotericin B. Cells were passaged weekly at a 1:4 split ratio by exposure to 0.5% trypsin containing 1 mM EDTA. For experiments measuring confluency, cells were seeded onto plastic culture dishes at 2.5×104–4×104 cells/well in media as above. A period of 24 h was allowed for adherence of cells to the culture dish and measurements were then obtained daily with a media change after 3 days.

Bright-field imaging and phase map calculation

Bright-field images were captured using a black and white 1300×1030 pixel Coolsnap FX CCD camera (Roper Scientific) mounted on a Zeiss Axiovert 100 M inverted microscope utilising a Zeiss Plan-Neofluar (×10, 0.30 NA) objective. To ensure optimal specimen illumination, Köhler illumination conditions were established for each optical arrangement (condenser and objective alignment and condenser stop at 70% field width). To calculate the phase map, one in-focus, and equidistant positive and negative de-focus images were acquired, using a de-focus distance of 5 µm in this instance. This was achieved using a piezoelectric positioning device (PiFoc, Physik Instrumente, Karlsruhe, Germany) for objective translation. Bright-field images were subsequently processed to generate phase maps using QPm software (v. 2.0, IATIA, Box Hill Nth, Victoria, Australia). The phase map generation, based on the set of three bright-field images captured, involved application of a software-automated algorithm to solve the “transport of intensity” equation using previously published techniques. The transport of intensity equation describes how light wave intensity changes with spatial propagation through the optical axis of the microscope. If the in-focus image intensity is measured and the intensity derivative is determined by analysis of the two de-focus images, it is possible to calculate phase using this equation. This recently developed algorithm and methodology has been detailed mathematically and validated in relation to use with cellular and other transparent specimens [2, 6].

In addition to the set of images obtained for phase map calculation, for each specimen an image using conventional optical phase techniques was also acquired (Plan-Neofluar, ×10, NA 0.30) so that calculated and optical phase imaging techniques could be compared. An example and comparative view of the three different image types (bright-field, phase map and optical phase) are shown in Fig. 1. The three bright-field images captured, one in-focus, and the pair of positive and negative de-focus images are shown in Fig. 1A. The lack of structural detail in the bright-field image is notable when compared with the two phase images. The differences in these images, which are barely discernible by visual inspection, provide the substance for the mathematical manipulation necessary to recover the phase information (phase excursion in radians) from the specimen. The distinct cell boundary definition achieved using the QPm software-calculated phase map (Fig. 1B) when compared with an optically derived phase image (Fig. 1C) is also apparent.

Fig. 1A–C
figure 1

Examples of human airway smooth muscle (HASM) culture images obtained using bright-field and phase optics (Plan Neofluar, ×10, NA 0.30). A Bright field: in-focus, positive and negative de-focus images. B Quantitative phase microscopy (QPM)-derived calculated phase map (5 µm de-focus, 550 nm illumination specified, maximum phase excursion 2.171 radians). C Optical phase image (phase ring inserted)

Full size image

Tracking growth of cultured cells over 92 h

Phase map images were analysed to evaluate confluency and to measure the growth of the cultured muscle cells over the period of 92 h. Reproducible location of a reference point within the culture dish was achieved using a mark on the base of the culture plate and by reference to the gradation scale on the microscope stage. This enabled measurements of the same field of cells (those in the field surrounding the centred reference point) over the extended time period at 24-h intervals.

Culture plates were set up so that parallel measurements of confluency and determination of cell number could be performed at each time interval. Following phase image capture, cells were lifted from the culture substrate by exposure to trypsin (0.5% v/v containing 1 mM EDTA) and counted using standard haemocytometry. To ensure uniform growth rates across the six-well plates, all wells were seeded at the same density, from the same cell passage type, and were exposed to identical incubation conditions. One well of the six-well plate was imaged repeatedly for daily confluency measurement, with the remaining five wells harvested one per day for cell number determination. The relationship between cell growth measurements obtained by confluency measurement of phase maps and by haemocytometric cell counting methods was estimated.

Results

Phase map segmentation and thresholding

Inspection of the images presented in Fig. 1 illustrates the difficulties encountered in visualising cultured cell monolayers under bright-field conditions. The cellular outlines and processes are barely discernible in Fig. 1A, despite the optimised Köhler illumination conditions. In Fig. 1B, as is typically observed, the calculated phase map exhibits a much enhanced dynamic contrast range. In this example, the maximum phase-shift detected was 2.171 radians, and the phase map is normalised to “stretch” this shift over the entire grey scale range available. The extent of the phase-shift observed in subconfluent HASM cultures is usually 2.500–7.000 radians. The optical phase image of the same field presented in Fig. 1C offers somewhat improved contrast relative to the bright-field image. This is particularly accentuated (and somewhat distorted) at the cell boundaries, but the optical phase view provides less useful contrast between the internal cellular and non-cellular image features.

Phase maps (i.e. Fig. 1B) were analysed (using the QPm software image analysis tools) to construct pixel intensity histograms (Fig. 2A) to identify phase-shift characteristics associated with cellular structures. Scrutiny of numerous phase map histograms indicated that the initial portion of the steepest gradient of the histogram could be used for reproducible demarcation of cellular from extracellular material. A linear function was fit to the ascending portion of the derivative of the intensity histogram (Fig. 2B) and extrapolated to the x-axis to obtain the threshold grey level at which segmentation of cellular from non-cellular material could be achieved using the phase map (Fig. 2C). This novel calculation provides an entirely non-subjective technique of image segmentation for cell delineation. The extrapolated threshold value was then utilised to construct a binary image (software: Image-Pro Plus v. 3.0, Media Cybernetics, Silver Spring, Md., USA) representing demarcation of cellular material from non-cellular material in the phase image (see Fig. 2D). The binary map generated by these segmentation manipulations is simply used to sum the quantity of “black-delimited” cellular material on the culture plate as a measure of the confluency of the culture, expressed as a percentage of the total field area examined (% section area). For the culture used as a “case” image analysis presented in Figs. 1 and 2, this value was 5.68%, a value typical for cultures at about 20 h post-seeding under these conditions.

Fig. 2A–D
figure 2

Case example of the segmentation process undertaken in order to perform HASM culture confluency measurement using non-subjective QPM-based technique. A A grey level histogram plotted from the phase map presented in Fig. 1B, exhibiting a unimodal distribution. B The derivative of the plot shown in A, exhibiting smooth asymptotes and rapid inflexion through zero gradient. C Graphical illustration of the threshold determination process, by axis intercept using the differentiated pixel intensity histogram. D Binary image generated by segmentation of the phase map using the threshold value determined by axis intercept

Full size image

An 8-bit image (grey scale representing values ranging from 0–255) was found to be optimal for the segmentation procedure summarised above. The analysis was also undertaken using a 12-bit image to increase the contrast range available, and potentially to improve the precision of determination of the threshold point. However, the increase in noise in the 12-bit image histograms generally offset any improvement in the determination of the threshold grey-level. The thresholding technique involving intercept extrapolation of the differentiated pixel intensity histogram was essentially developed empirically. The segmentation achieved by this method consistently matched well with thresholds determined manually by image inspection. Other techniques have been employed with less success, including Gaussian fit and 3 SD subtraction methods. The pixel intensity histogram cannot be assumed to be normal (and is usually not normal), and this lack of normality may limit the applicability of these methods.

Optical vs. phase map image segmentation

This analysis procedure allows for an accurate and non-biased calculation of the threshold point with which to distinguish cells from background. Of crucial importance in achieving a successful thresholding outcome in this process is the quality of data available in the phase map where haloing and cell edge distortion is suppressed allowing for accurate cell delineation. When the same analysis procedure was attempted with an image captured using conventional optical phase techniques, a reliable outcome could not be achieved (Fig. 3). The reduced contrast between cellular and non-cellular material in the optical phase image renders the curve-fitting procedure unstable, while the effect of uneven illumination intensity in the optical image produces regional variation in the thresholding process. Inspection of the derivatives generated from the optical phase image intensity histograms revealed that these plots are somewhat more complex than those extracted from the phase maps and exhibit multiple peaks (Fig. 3A). Thus the process of intercept extrapolation cannot be applied easily to these plots and it is not possible to employ a non-subjective thresholding method. The difficulties associated with segmentation of optical phase images are exemplified in Fig. 3B, where it is apparent that the binary image produced by thresholding incorporates extracellular regions at the top left of the image (Fig. 3B) and fails to delineate boundaries at other locations in the lower portion of the image. The combined effect of this lack of field uniformity and the inaccuracy of cell delineation, over-estimates the confluency status of this culture specimen markedly (more than twofold) compared with the phase map determination using similar thresholding methods.

Fig. 3A, B
figure 3

Segmentation of the optical phase image shown in Figure 1C. A The derivative of the pixel intensity histogram obtained from Figure 1C, exhibiting multiple peaks and gradient instability. B Segmented binary image derived from Fig. 1C, using threshold value estimated manually, corresponding to the axis intercept associated with extrapolation of pre-inflexion differential peak (grey level 70)

Full size image

Tracking growth of cultured cells

Phase-map thresholding and segmentation techniques were applied to measure the progressive increase in confluency of HASM cell cultures from several different patient cell lines. Following re-passaging and seeding at standardized density, culture growth was tracked by repeated imaging over a 92-h period. As shown in Fig. 4A, an approximately linear growth response was observed over this period, with the degree of confluency increasing from about 8% at 24 h to around 17% after 92 h.

Fig. 4
figure 4

HASM growth plots. A Relation between cell culture confluency and culture passage period (92 h). Three different patient-derived cell culture lines. B Correlation of cell culture confluency calculated using QPM-derived phase map segmentation procedures with cell number determined by haemocytometry (r 2 Pearson’s correlation coefficient)

Full size image

Figure 4B illustrates the correlation between the QPM-calculated culture confluency and cell number determined by haemocytometry for the same culture wells throughout this growth period for the three lines tracked in Fig 4A. A high degree of correlation (r 2=0.95) is observed between these two growth measures. These findings indicate that in circumstances where proliferative growth is conventionally (and destructively) assessed by cell harvest and counting, the use of in situ QPM imaging methods provides a reliable and non-destructive surrogate.

Discussion

Quantitative phase—a new imaging modality

A novel, non-destructive technique for measuring confluency status of cell cultures using newly developed quantitative phase microscopy methods has been established. Using QPM, a non-subjective procedure for image thresholding and segmentation has been devised and applied to the measurement of smooth muscle cell growth in culture. A key factor in the successful determination of cell confluency using QPM is the production of a computational phase map that offers significant advantages over optical phase contrast views produced under similar conditions. In the phase map, the reduction in cell edge distortion and haloing, evident in optical phase view, allows for more precise discrimination of cell boundaries, improved stability of image analysis and more accurate cell delineation.

In the study of cellular morphology, image segmentation methods have been employed widely in the analysis of a variety of image types to extract structural information [1, 10, 19]. Typically these methods are applied to stained, fixed specimens or fluorescently tagged, viable specimens. In such samples relatively sharp gradients between background and foreground intensity signals are apparent, often observed as separate peaks in the pixel intensity histogram. In these situations, subjective investigator judgement is usually called for to set a threshold level of appropriate discrimination, adjusted on an image-by-image basis depending on signal contrast and intensity. For unstained, viable cellular specimens, where use of phase microscopy is necessary to achieve visualisation, thresholding and segmentation manoeuvres have been relatively unsuccessful. The extraction of the pixel intensity histogram double differential for this purpose has not been described previously. Thus the major methodological advance reported here combines the application of a recently devised computational phase microscopy technique with the usage of a new non-subjective image thresholding procedure to enable rapid and reproducible morphologic analysis of cultured cells in situ. QPM can be implemented using an optically simple microscope, which requires provision for bright-field viewing only, and offers ease of access to the stage for other cellular manipulations. Where experimental protocols require greater temporal resolution, it is also possible to expedite the image acquisition process by simultaneous capture of the in-focus and de-focus image ensemble. Furthermore, the non-subjective attribute of the phase map segmentation technique allows for implementation of macro-controlled batch processing of culture assessment.

HASM cultures provide a multicellular system particularly amenable to interrogation using QPM techniques. Recently, we have evaluated the theoretical and empirical limitations which apply to the phase imaging of cellular and non-cellular specimens [6, 16]. For cellular specimens, such as the HASM cultures used here, that are thinner than the depth of field of the objective used, it would be expected that good phase recovery can be achieved and that the phase measurement corresponds to the full optical thickness of the object. In circumstances where illumination level is not critically low, it is predicted that phase recovery should be independent of illumination intensity variation and optimal with condenser aperture settings in the range 0.4–0.8. Our previous findings with inert specimens confirm this prediction [6]. The observations presented in this study are also consistent with this expectation, demonstrating that the optical phase images are considerably more susceptible to intensity variation contamination than the corresponding phase maps even though Köhler illumination conditions were maintained stringently. The source of the intensity variation evident in the optical phase images (i.e. Fig. 3B) is not clear, but shadowing from the steep-sided culture wells, media surface and meniscus irregularities or non-uniformity of the plastic base of the culture plates may be involved and are virtually unavoidable in these environments. It is the ability of QPM to separate the intensity and phase image components that confers the primary utility in evaluating translucent specimens in the cell culture systems investigated here.

Non-destructive cell growth measurement

The capacity to apply QPM to the analysis of unstained, cultured cell preparations in situ is of considerable value. Other methods used to monitor cell growth including FACS and 3H-leucine uptake have destructive end-points, rendering them unusable in longitudinal studies of changes in cell growth. These other methods have additional limitations which restrict their application. FACS is useful in determining cell size, but provides minimal three-dimensional information and lacks sensitivity when dealing with a heterogeneous cell population. The 3H-leucine method relies on measurement of leucine uptake into the cell as an indicator of protein synthesis and therefore cellular growth. This method also has limitations as many factors other than cell growth may influence the uptake of 3H-leucine into a cell and altered uptake could indicate altered turnover rather than growth per se. One of the benefits of using QPM to measure confluency of these cells is the ability to be able to track the growth and/or regression of growth of an identifiable cell population longitudinally throughout treatment intervention—an option not available when growth measurement requires cell harvest.

The values obtained for cell culture confluency obtained using QPM-derived phase maps correlated closely with cell counts determined by haemocytometry. The growth of HASM cells in culture probably involves a combination of both hyperplastic and hypertrophic growth. Whilst it is not possible without additional structural information to distinguish between these two growth phenotypes, the present findings certainly indicate that the proliferative process is associated with a proportional increase in total cell area which can be efficiently and non-destructively tracked by QPM-based analysis. The slope of the relationship between confluency and cell number can vary between culture lines, indicating a different balance between hypertrophic and hyperplastic growth processes (data not shown). Whilst these findings are of general relevance to all cell growth systems, the results presented here in relation to HASM cells are of particular interest to smooth muscle growth phenotype research [9, 14]. Increased efficiency of assessment of cell growth and an improved understanding of the regulation of HASM growth patterns will obviously be of considerable importance in our evolving understanding of airway smooth muscle structural remodelling and inflammatory responsiveness in the aetiology of asthma.

Further development of the QPM-based analysis methodology has the potential to provide even more refined measures of cellular growth. Nuclear identification and localisation in conjunction with confluency measurement would permit cell growth to be characterised in terms of proliferative behaviour. In addition, it is also theoretically possible to recover data relating to cell thickness (and therefore volume) from the calculated phase values at every location in the phase map. Both these developments await further advances in our ability to measure and compare the refractive indices of cellular structures and organelles using QPM and other technologies.