Sage Journals: Discover world-class research

Abstract

Image-processing programs are used to identify and classify eukaryotic cell colonies as spots following seeding at low density on dishes or in multiwell plates. The output from such approaches, however, is generally limited to 1–2 parameters, and there is no ability to extract phenotypic information at the single colony level. Furthermore, there is a lack of user-friendly pipelines for analysis of clonogenicity in the context of high-content analysis. This article describes an experimental and multiparametric image analysis workflow for clonogenic assays in multiwell format, named the Colony Assay Toolbox (CAT). CAT incorporates a cellular-level resolution of individual colonies and facilitates the extraction of phenotypic information, including the number and size of colonies and nuclei, as well as morphological parameters associated with each structure. Furthermore, the pipeline is capable of discriminating between colonies composed of senescent and nonsenescent cells. We demonstrate the accuracy and flexibility of CAT by interrogating the effects of 2 preclinical compounds, Nutlin-3a and ABT-737, on the growth of human osteosarcoma cells. CAT is accessible to virtually all laboratories because it uses common wide-field fluorescent microscopes, the open-source CellProfiler program for colony image analysis, and a single fluorescent dye for all the segmentation steps.

Keywords

colonies high content wide-field microscope CellProfiler p53

Introduction

The mammalian colony outgrowth assay is a widely used in vitro approach within the oncology research field.^1,2 Indeed, the clonogenic assay is considered a “gold standard” in terms of evaluating the efficacy of treatments, including radiation and novel potential antitumor drugs. This methodology, however, relies predominantly on manual counting of colonies, which is a time-consuming and subjective process. When evaluating the effects of drug combination treatments over several concentrations and conditions using colony formation assays, the increased number of samples rapidly becomes a burden in terms of time and reagent cost. Thus, there is a clear need for the development of platforms that enable automated image capture and analysis. In this regard, several groups have developed automated threshold methods to segment images of colonies. Computer-aided colony-counting methods have addressed some difficulties associated with the segmentation of overlapping colonies or of colonies at the periphery of a culture dish. Two independent groups improved the analysis either by means of an edge segmentation coupled to a compact Hough transform³ or by using a model-based segmentation process based on prior knowledge of colony shape.⁴ Although these automated counters decrease the subjectivity of colony counting, they can process only 1 dish at a time. This has been somewhat improved by Dahle et al.,⁵ who used a flatbed scanner to count colonies in 12 culture dishes simultaneously. Laser scanning microplate cytometry also improves throughput; indeed, this technology has been used to determine the colony number in a 24-well format through the application of a volume algorithm.⁶ All of these approaches, however, suffer from poor resolution in terms of high-content information. Although they do provide information on area and density distribution in addition to the number of colonies, the methods lack depth of information regarding single-colony composition as well as cell number, cell shape, cell size, and DNA content. Recently, Katz et al.⁷ were able to extrapolate more information from a single clonogenic assay by using high-content microscopy and an integrated software for automated colony counting. The cost of this type of integrated technology, however, together with the use of expensive dyes for colony staining, is prohibitive for many laboratories that wish to perform automated clonogenic assays. Thus, several groups have implemented automated colony analysis of crystal violet–stained colonies in culture dishes using freely distributed programs either as standalone solutions (CellProfiler^8,9 and OpenCFU¹⁰) or as macros in ImageJ (National Institutes of Health, Bethesda, MD¹¹). Recently, the ImageJ plugin entitled ColonyArea was developed for quantitative analysis of the area covered by foci of cell colonies in scanned 6- to 24-well plates stained by crystal violet.¹² This plugin demonstrates the simplicity and efficiency of analysis routines that can provide information regarding colony area and an indirect estimation of cellular density within colonies by means of crystal violet staining intensity. Still, in view of the wealth of high-content information potentially available in clonogenic assays, it would be useful to improve image resolution while preserving throughput; such optimization should also be accessible to biologists with no specialized knowledge of programming languages.

To meet these needs, we have developed the Colony Assay Toolbox (CAT), which is an improved end-to-end approach for clonogenic assays that enhances the cost-effectiveness in terms of materials, instruments, and software, and also decreases the total time required for experimental procedures and analytical follow-up. CAT also allows the user to discern phenotypic information for individual colonies that would otherwise require specialist stains or direct observation by a trained operator. These improvements were achieved by using (1) a common wide-field fluorescent microscope with a 4× objective, (2) a 96- or 384-multiwell plate format to improve throughput, (3) an algorithm for analysis segmentation developed in the open-source CellProfiler image analysis software,^8,9 (4) and the ability to use a single dye to concurrently label for nuclei and segment for colonies.

The performance of the CAT system and the robustness of the pipeline were evaluated by comparing 2 colony segmentation methods, regular plastic versus optical multiwell plate formats, and different microscope and camera systems. We also compared the output and performance of CAT to a commonly used adenosine triphosphate (ATP)-based cell viability assay. CAT was able to accurately determine the in vitro potency of the p53-activating agent, Nutlin-3a, yielding values almost identical to those in previously published studies.^13,14 Furthermore, CAT revealed that although treatment with Nutlin-3a alone leads to senescence, a combined treatment with the BCL2 antagonist ABT-737 abrogates this response and instead triggers tumor cell death.

Materials and Methods

96-Well Colony Formation Assay, Drug Treatments, and Viability Assay

U2OS osteosarcoma cell lines (see Suppl. Material for cell culture conditions) were trypsinized, counted, and plated at different densities ranging from 25 to 400 cells/well on plastic (tissue culture–treated Falcon #3072; Corning, Corning, NY) or clear optical (tissue culture–treated Costar #3904; Corning) 96-well plates. Most of the experiments were performed in plastic plates unless otherwise specified. Colony formation was monitored for 6–7 days, after which colony analysis and viability assays were carried out. Cell densities between 25 and 150 cells/well were found to be optimal for our requirements. Single-drug treatments with Nutlin-3a and combinatorial treatments with ABT-737 were performed following seeding at 150 cells/well.

Cells were allowed to attach overnight and then treated with 0, 0.5, 1.5, 2.5, and 5 μM of Nutlin-3a (SML0580-5MG; Sigma-Aldrich, St. Louis, MO) alone or in combination with 0.5, 1.5, 2.5, and 5 μM of ABT-737 (sc-207242; Santa Cruz, Dallas, TX). All of the solutions were prepared in such a manner that the final DMSO concentration was the same (0.01%) in all treatments. After 6–7 days, the colonies were fixed and stained.

Interwell homogeneity and even cell distribution within wells were achieved using an automated liquid dispenser (Multidrop Combi Reagent Dispenser; Thermo Scientific, Waltham, MA).

The CellTiterGlo viability assay used in this article is described in the Supplementary Material.

Immunofluorescence Protocol

To perform image acquisition and automated image analysis of colonies, cells were labeled with DAPI dye and (when reported) with HCS CellMask Red Stain by using the immunofluorescence protocol detailed in the Supplementary Material.

Fluorescent Microscopy

The 96-well assay plates were scanned by 1 of the following wide-field inverted fluorescent microscopes: a Nikon-Ti Eclipse (Tokyo, Japan), an Olympus Scan^R (Tokyo, Japan), or an AMG EVOS digital microscope (Life Technologies, Carlsbad, CA). A 4× objective was used in all of the systems to acquire a large area of the well. Depending on the experiment, 1 or 2 fluorescent channels were acquired per well: DAPI for nuclear staining and CellMask for whole cell staining. Further details are described in the Supplementary Material.

Image Processing

The algorithms for image segmentation described in this work were implemented as modules in CellProfiler image analysis software.^8,15

All of the images acquired were initially processed to correct for the uneven illumination. To remove differences in light intensities throughout the well, we used a regular correction function for both the DAPI and CellMask channels (CorrectIlluminationCalculate Module). The algorithm processed the raw images by rescaling the intensity at each pixel of the image. Next, we applied an average smoothing function that resembled the uneven illumination pattern of the raw images. Finally, the resulting correction pattern was applied to the raw images (CorrectIlluminationApply Module).

When the images were scanned with the EVOS digital inverted microscope, 2 steps were added to the pipeline before the correction illumination module: 1 to convert the color images using the ColorToGray Module (because the digital output of the images is JPEG format), and 1 to correct the alignment of the blue and red channels using the Align Module because of a small shift in the x–y coordinates. The alignment correction was not required for images generated by the Nikon or Olympus microscopes.

The image-processing algorithm called dMask (for DAPI mask) was designed to use the DAPI channel alone to segment for both nuclei and colonies simultaneously. The method was compared and parameters optimized with the same pipeline workflow but using the dual-staining CellMask–DAPI to segment for colonies and nuclei, respectively. This latter algorithm was termed CdMask (for CellMask and DAPI mask) and was required for the calibration procedure solely at the start of a new screening experiment. Specifically, CdMask was used to fix parameters for a good estimation of colonies’ shapes and sizes. These parameters are then used to adjust the smoothing function (Smooth Module) and intensity thresholds (IdentifyPrimaryObjects Module) of the dMask method to obtain the same number and size of colonies reported by the calibration pipeline.

dMask pipeline ( Fig. 1A , left panel): DAPI channels corrected for uneven illumination were initially processed by applying a threshold to create a binary image in which the background is set to 0 and colonies are set to 1 (ApplyThreshold Module).

Figure 1.

(A) Workflow of DAPI mask (dMask) and CellMask–dMask (CdMask) segmentation methods. The left panel shows the dMask method: The DAPI emission channel only (load raw images) was corrected for illumination variations, then was converted to a binary layer and processed using a smooth function (pre-processing steps). The DAPI processed images were used to detect colonies as primary objects. Next, the primary objects were used as a mask for the corrected DAPI channel to segment nuclei belonging to the identified colonies. Finally, both colony and nuclei segmentations were correlated (segmentation steps). The right panel shows the CdMask method. The principal difference with the dMask method was the use of CellMask and DAPI emission channels to differentiate between colonies and nuclei, respectively. Scale bar, 100 µm. (B) Flowchart of the calibration procedure. Three cells densities were seeded in the calibration wells (red-blue colored). At the endpoint of the assay, the resulting colonies were stained by both CellMask and DAPI dyes; meanwhile, the remaining wells were labeled by the single DAPI dye. Only the calibration wells were used to run CdMask and dMask simultaneously. Once the optimized parameters in the dMask algorithm were fixed, the latter was used on the entire plate to run the segmentation on the DAPI images.

Subsequently, the thresholded images were submitted to a Gaussian smooth filter with a kernel size at least twice the mean diameter of U2OS nuclei (Smooth Module). This size-related smoothing filter was required so that colony segmentation obtained by the machine was consistent with the size and number of colonies identified by trained researchers, and it also compared well with CellMask staining. Next, the smoothed images were used to identify colonies as primary objects (IdentifyPrimaryObjects Module) by using the Otsu intensity threshold method. Clumped colonies were separated based on shape features, and typical diameters ranging between 125 and 2500 pixels were optimal to distinguish all colony sizes present in our images.

Next, colony segmentation was used to mask the DAPI-corrected channels (Mask Module) and to identify discrete nuclei within colonies (IdentifyPrimaryObjects Module). The Otsu intensity threshold method was able to robustly distinguish foreground from background regions; to remove background artifacts due to intraexperimental variations in the staining procedure or in object numbers in the field of view (FOV; i.e., no or few colonies in the FOV), some minor adjustments in the minimum and maximum allowable thresholds may be required on a day-to-day basis.

Clumped nuclei were separated based on intensity values, and we found that typical diameters ranging between 6 and 35 pixels were optimal for distinguishing all nuclei present in our images.

Finally, we assigned relationships between children objects (nuclei) and the associated parent objects (colonies) based on spatial reference (RelateObjects Module).

CdMask calibration pipeline ( Fig. 1A , right panel): In this method, CellMask channels were first corrected for uneven illumination and then processed by applying a threshold to create a binary image (ApplyThreshold Module). Subsequently, the threshold images were submitted to a Gaussian smooth filter (Smooth Module) and used to identify colonies as primary objects (IdentifyPrimaryObjects Module). Again, clumped colonies were separated based on shape features, and typical diameters ranging between 125 and 2500 pixels were optimal for distinguishing all colony sizes present in our images. Afterward, colony segmentation based on the CellMask channel was used to mask the initial DAPI-corrected channels (Mask Module) and to identify the nuclei within colonies (IdentifyPrimaryObjects Module).

Finally, relationships were assigned between children objects (nuclei) within the proper parent objects (colonies) based on spatial reference (RelateObjects Module).

Data and Statistical Analysis

Data generated during the CellProfiler pipelines were exported to .csv files and analyzed using OriginPro data analysis and graphing software by OriginLab (Northampton, MA).

Further details are described in the Supplementary Material.

Results

Calibration Procedures to Set and Optimize the dMask Pipeline

Our method for segmentation of colony images was developed within the CellProfiler environment^8,9 (www.cellprofiler.org). The pipeline was named dMask because it was able to perform segmentation of both colonies and nuclei by recording the single DAPI intensity channel. Once acquired, the DAPI images were loaded and analyzed by the pipeline.

Following pre-processing steps, the resulting DAPI smoothed images were segmented for colonies ( Fig. 1A , left panel). Afterward, the colonies were used as a mask on the DAPI channel. Finally, the masked DAPI was segmented to identify single nuclei within every colony (for detailed steps, see Materials and Methods).

The challenge during the design of the dMask pipeline was to define the smoothing function in such a way that it expanded the nuclear staining to the entire area of the cell within the membrane. Because nuclear staining does not permit the detection of cellular boundaries, a whole-cell staining using CellMask was necessary to acquire the “cell shape” parameter. Thus, at the beginning of a new colony assay screening experiment, we labeled and recorded both DAPI and CellMask channels for only a selected range of colony densities, whereas the rest of the wells in the plate were stained with only DAPI ( Fig. 1A , right panel, and Fig. 1B ). In our setting, we found that only 3 different colony densities, in the so-called calibration wells, were sufficient to simultaneously compare CdMask and dMask ( Fig. 1B ). Using the same image set, we tuned the parameters of the dMask algorithm to obtain results that were virtually identical to those obtained using CdMask. Briefly, once the image was pre-processed, the object size was fixed at the same value for both algorithms. Then, the smooth function and threshold parameters were adjusted in dMask to obtain a number and area of all colonies in the FOV that were identical to those reported using the CdMask calibration pipeline. For this first setting, the pipeline should be used in “test mode” to allow pausing of the analysis at each step to fine-tune these parameters in real time. This initial setup was fast (on a timescale of minutes), and measurements are easily accessible because of the IdentifyPrimaryObjects module that is available during the analysis and that summarizes the number and area of the objects (Suppl. Fig. S1).

With this approach, we obtained a suitable overlap among the number of colonies detected by the dMask process and the CdMask segmentation, with a maximum differences of 1–2 colonies between the 2 methods. Moreover, there was excellent overlap between the 2 pipelines with regard to estimation of the total area covered by colonies within the FOV (Suppl. Fig. S2). Once the parameters were validated, we used only the dMask algorithm to analyze the entire multiwell plate ( Fig. 1B ).

Accordingly, data obtained using the single DAPI stain and the dMask segmentation algorithm were as robust as those obtained using the standard dual-staining CdMask method.¹⁶

Moreover, dMask significantly decreases not only the material costs (by using exclusively DAPI staining after the initial calibration plate is run) but also the time for the experimental procedure (≅50 min), image acquisition (1 single fluorescent channel), and data analysis (the latter used between 20 and 60 min for a 96-well plate) (Suppl. Table 1A and 1B).

Quantitation of Colony Growth in a 96-Well Format

Colony outgrowth is one of the gold standard in vitro tests for the efficacy of potential therapeutic agents. We thus reasoned that miniaturizing this assay to a multiwell format would increase the throughput of clonogenic assays during compound testing. Initially, we tested the ability of the algorithm to detect colony formation in a commonly used plastic 96-well plate format (Falcon #3072). Specifically, U2OS cells were seeded at 25, 50, 100, 200, and 400 cells/well on day 0. After 6–7 days, we stained the cells and scanned the wells. The acquired images were then loaded and run using the automated dMask method, and the images were segmented for nuclei and colonies, as described above ( Fig. 2A ).

Figure 2.

Estimation of colony formation for different cell densities. Several colony densities were obtained by seeding 25, 50, 100, 200, and 400 cells/well. DAPI mask (dMask) was run to segment for colonies and nuclei. (A) Representative segmentation images of nuclei (Panel a) and colonies (Panel b) at 4 different cell densities. (B) Plots of the mean nuclei number detected for each cell density. (C) Plots of the mean percentage of the area covered by colonies for each cell density. (D) CellTiterGlo viability measurements for each cell density. The images were acquired by the EVOS microscope with a 4× objective. The data (mean and SE) are derived from triplicate wells.

For the data analysis, we decided to use a simplified alternative to colony number by determining the percentage of area covered by the colonies on a per-well basis and already used by other groups.^12,17–19 Indeed, we wanted to design the present analysis in such a way that a single algorithm could provide an estimation of colony formation regardless of the precise density of colonies per well. Moreover, the relationship between the number of cells seeded and the number of colonies formed is not one-to-one due to differences in plating efficiency; thus, increasing the initial cell density could result in fewer colonies than expected.²⁰ In addition, wells in which cells were seeded at low cell density (i.e., 150 cells/well in a 96-well format) and left untreated were almost confluent by 6 days, precluding an accurate quantification of colony numbers. Thus, the area occupied by each colony most accurately reflected “colony formation units”; also, the output data were independent from numbering clustered colonies in high-density samples.

Hence, for each cell density, we measured 2 indices that characterized the colony growth. First, we calculated the proliferative potential by extracting the number of nuclei per colony. Second, colony-forming potential was evaluated by analyzing the total area covered by colonies in each well. The initial threshold number of nuclei that an object had to contain to be scored as a colony was ≥40, because this is the minimum number that is consistent with growth throughout a 6–7-day period. Subconfluent growth conditions (ca. 10⁴ cells/well) were found by seeding from 25 to 150 cells per well at day 0 ( Fig. 2B ). Within this range, the area occupied by colonies increased linearly with cell density and covered from 20% to 80% of the total FOV area ( Fig. 2C ). Similar results were obtained for U2OS colony-forming units’ growth on 96-well optical plates (Suppl. Fig. S3). To validate our method, we compared the same data set with CellTiterGlo, a reagent commonly used to measure cellular growth based on ATP production. As shown in Figure 2D , we observed a linear correlation between cell number and viability estimation within the range of 25–150 cells/well. The excellent agreement among the methods validated the dMask pipeline and the accuracy of colony area as an indicator of colony-forming potential (Suppl. Fig. S4). Furthermore, we were also able to estimate the doubling time (t_d) of U2OS cells in 96-well plates by counting the nuclei number at each seeding density, based on the following equation:

t_{d} = \frac{t * \ln 2}{ln (N_{t} / N_{0})}

where t is the growth period; N₀ is the initial number of cells; and N_t is the final number of cells. The results indicated a t_d of approximately 24 h and 26 h for U2OS growing on plastic and optical 96-well formats, respectively (data not shown).

Drug Dose–Response Curve Validation

The utility of dMask for evaluating compound activity was therefore investigated using a preclinical compound to treat U2OS osteosarcoma cells, which express the wild-type allele of the p53 tumor suppressor (p53^WT).²¹ We tested Nutlin-3a (Nut-3a), a nongenotoxic MDM2 antagonist that induces p53-dependent growth arrest or apoptosis, depending on the cell type.

Initially, we verified colony growth for each subconfluent cell density over a range of Nut-3a concentrations (data not shown). The goal here was to identify doses that partially suppressed colony outgrowth; this would facilitate the detection of combination treatments that further enhanced the cytotoxicity of Nut-3a. For this purpose, we chose to seed 150 cells/well at day 0, because seeding at a lower cell density led to virtually complete suppression of outgrowth with Nut-3a as a single agent (data not shown).

To validate the robustness of our dMask staining and pipeline in determining the IC₅₀ for Nut-3a, we compared the data with those obtained by using the CdMask dual-staining and analysis pipeline.

The percentage of area occupied by colonies in each well was plotted against the Nut-3a dose. The dose–response and associated inhibitory curves for both the dMask and CdMask segmentation methods are shown in Figure 3A . The 2 curves, which were highly correlated with one another (Suppl. Fig. S4), overlapped and yielded Nut-3a IC₅₀ values of 1.5±0.02 µM and 1.6±0.06 µM for dMask and CdMask, respectively. These results are virtually identical to those previously obtained by using other viability assays for Nut-3a or its analogs in cell lines with p53^WT.^13,14 Although treatment with 1.5 µM Nutlin resulted in a 50% reduction of colony outgrowth, increasing the concentration to 2.5 µM or higher led to a 90–100% suppression of outgrowth.

Figure 3.

Nut-3a dose–response curves. Quantitation of the percentage of the area covered by the colonies following Nut-3a treatment. (A) The colony assay was performed in 96-well, nonoptical, clear-bottom plastic plates. The percentage of the area occupied by colonies was extracted by simultaneously running the DAPI mask (dMask; black circles) and CellMask–dMask (CdMask; gray circles) methods. The threshold diameter to be scored as a colony was established for both methods as >400 µm. The images were acquired by the Ti-Eclipse microscope with a 4× objective. The data (mean and SE) are derived from triplicate experiments and were compared by performing a t-test. (B) Nut-3a dose–response curves were obtained by using 2 different microscope systems and a 96-well optical-quality plate format. The percentage of the area covered by the colonies following Nut-3a treatment was obtained by running dMask methods on images derived from different cameras. The same samples were acquired by using both a large FOV (Zyla-CMOS camera: 4.2×3.5 mm; black circles; Ti-Eclipse microscope) and a smaller one (ORCA-ER camera: 2.2×1.7 mm; gray circles; Scan R microscope). The percentage of the area occupied by colonies was then compared and plotted for the 2 systems. The diameter of a colony was established for both methods as >400 µm. The data (mean and SE) are derived from triplicate wells. (C) Three sizes of colony diameters were set: >200 µm (white squares), >400 µm (gray squares), and >500 µm (black squares) to segment for colonies by using the dMask method. The percentages of the area covered by colonies following titration of Nut-3a are compared. The data (mean and SE) are derived from triplicate experiments.

The consistency, accuracy, and robustness of the dMask method to account for drug IC₅₀ were also underscored by performing Nut-3a titrations in optical-quality 96-well plates on 2 different image acquisition systems. Specifically, the same data set was acquired using both the Ti-Eclipse and Scan^R microscopes equipped with 2 different camera sizes: the Zyla CMOS camera and ORCA-ER camera, respectively. The resulting dimensions for a single FOV acquired with a 4× objective were 4.2×3.5 mm and 2.2×1.7 mm for Zyla and ORCA, respectively. Thus, 4 FOVs per well were acquired with the Scan^R to cover the same area acquired by a single FOV in the Zyla CMOS camera (Suppl. Fig. S5). The dMask algorithm was then used on both Scan^R and Ti-Eclipse images, and the resulting dose–response inhibitory curves are plotted for comparison in Figure 3B . Despite the 2 different acquisition systems, the curves overlap perfectly with a high level of correlation (Suppl. Fig. S4) and show an IC₅₀ of 1.3±0.13 µM, which is almost identical to the one obtained in the plastic well format. The small deviation in the IC₅₀ values between the plastic ( Fig. 3A ) and optical plate ( Fig. 3B ) formats was due to differences in cell adhesion for the 2 tissue culture–treated substrates. Indeed, for the same cell density (i.e., 50 cells/well), we observed a reduction from 50% to 30% of the area occupied by colonies in the plastic and the optical plate formats, respectively ( Fig. 2C and Suppl. Fig. S3).

We also tested whether the IC₅₀ calculations were affected by altering the size of the cutoff used to define colonies. To that end, we ran dMask segmentations in parallel when the size cutoffs for mean colony diameter were set to 200 µm, 400 µm, or 500 µm (Suppl. Fig. S6A). These limits corresponded to approximately 20, 40, and 50 nuclei per colony, respectively, assuming a mean nuclear diameter of 10 µm.^22,23 As shown in Figure 3C , the area covered by different colony sizes followed the same trend and a similar Nut-3a IC₅₀. Indeed, we obtained IC₅₀ values of 1.6±0.14 µM, 1.5±0.05 µM, and 1.4±0.10 µM for colonies with diameters of >200 µm, >400 µm, and >500 µm, respectively.

Because the IC₅₀ was not significantly affected with any cutoff selected, and CellTiterGlo measurements performed on the same data set yielded similar values (Suppl. Fig. S6B), we performed the next experiments by defining a “colony” as an object with ≥40 cells. Indeed, this 40-cell threshold is consistent with both the doubling time of the investigated cell line and the traditional criteria adopted by several other laboratories.^1,4,6,7,12

Extraction of Multiparametric Data from Colony Outgrowth Assays

We further compared the data derived from our analysis on Nut-3a treatments to those obtained using CellTiterGlo.

In Figure 4 , the CellTiterGlo signal is plotted together with the nuclei number as obtained using the dMask method. We observed Nut-3a dose-dependent inhibition for both of these measurements. From each single response, however, we obtained singular parametric information. Indeed, proliferation (nuclei count) and metabolic activity (CellTiterGlo) were affected at low Nut-3a doses (≅0.5 µM), with a relative IC₅₀ of 0.7±0.05 µM and 0.9±0.004 µM, respectively ( Fig. 4 ), compared to colony formation (IC₅₀ = 1.5±0.05 µM; Fig. 3 and Suppl. Fig. S6B). Proliferation, however, was clearly more sensitive to low-dose (0.5 µM) Nut-3a when compared to cell metabolism, in agreement with previous studies.^6,24 Indeed, cessation of proliferation is one of the most immediate responses to drugs and is therefore the most sensitive indicator of drug activity, even at low doses.⁶ In the absence of Nut-3a, the cells underwent approximately 7 cell divisions in 7 days. In contrast, the colonies that formed in the presence of 1.5 µM Nut-3a contained ≅2×10³ cells, consistent with 4 divisions by day 7 (Suppl. Fig. S7). This observation confirms that the steady accumulation of p53 activity progressively triggers cell cycle arrest, which extends to 100% of the population by day 4 ( Fig. 4 ). Indeed, in the presence of 2.5 or 5 µM Nut-3a, <1×10³ cells were counted, meaning that only 2 or 3 cell divisions could have occurred before proliferation stopped (Suppl. Fig. S7). Similar proliferation kinetics have been shown in other cell lines in the presence of Nut-3a.^25,26

Figure 4.

Relationship between metabolism and proliferation in Nut-3a treatment. The figure shows 2 parameters, proliferation (nuclei counts), and metabolic state [adenosine triphosphate (ATP) content], in the presence of increasing drug concentrations; the nuclei number obtained by dMask and the viability measurement obtained with CellTiterGlo are represented by dark gray and light gray bars, respectively. The ratio between ATP content and nuclei number gives a 3-fold increase in ATP demand for 1.5–2.5 µM of Nut-3a-treated cells compared to untreated cells or those treated with lower doses of drug. The data (mean and SE) for Nut-3a titrations are derived from triplicate experiments and are normalized to the control (0 µM Nut-3a).

An added advantage of the segmentation process that we developed is the ability to extract a series of morphological information, such as circularity, major and minor axis length (Suppl. Fig. S8), and the area of nuclei within discrete colonies. Figure 5A shows the dot plots of the mean nuclei area for each colony. This reveals that even at low (0.5 µM) doses of Nut-3a, the mean area of the nuclei is clearly larger. By setting a threshold in the untreated population, we observed a significant increase in nuclear area in 40–64% of the population between 0.5 and 2.5 µM of Nut-3a. Enlargement of both the cell and the nucleus (hypertrophy) are indicators of senescence,^26,27 and we observed this morphology specifically in Nut-3a-treated colonies ( Fig. 5B ). Thus, we further investigated whether dMask could be used to determine whether individual colonies might contain senescent cells.

Figure 5.

Phenotypic colony composition. (A) Box plot of mean nuclei area distribution. The average area of nuclei per colony was obtained by DAPI mask (dMask) segmentation in Nut-3a titration experiments. The data are derived from triplicate experiments and were compared by performing a t-test. (B) Senescence cell morphology within colonies. Representative images of the senescent phenotype. Raw images and colony segmentations are shown on the top and bottom of Panel a and Panel b, respectively. Magnification of the raw images shows a senescent phenotype for cells treated with 1.5 µM of Nut-3a (Panel a). The senescence morphology is not observable in untreated colonies or in those treated with ABT-737 (Panel b). Scale bar, 100 μm. (C) Box plot of the Senescence Index (SI) per colony following Nut-3a treatment.

To estimate this, we defined a Senescence Index (SI) [Equation (2)] for each colony as follows:

S I = 1 - (\frac{N u c l e i_{c o u n t s}}{C o l o n i e s_{A r e a}}) \times 100

We assumed that a senescent colony in a drug-treated sample would occupy a similar overall area as, but have far fewer nuclei than, its nontreated counterpart.

The results of the SI analysis are shown in Figure 5C . Nut-3a treatment significantly increased the SI, which rose from 30% of colonies at 0.5 µM to 80% at 2.5 µM Nut-3a compared to untreated samples. Moreover, by fitting the mean SI per Nut-3a concentration, we obtained an IC₅₀ value for senescence induction of 0.64±0.05 µM (Suppl. Fig. S9A). This is similar to the IC₅₀ for proliferative arrest but lower than that for inhibition of ATP production (see above). This result suggests that cessation of proliferation and induction of senescence are the primary responses to this particular drug. To determine whether the enlargement of cells that we detected was drug dependent, we treated U2OS with the BCL2 antagonist, ABT-737, another pro-apoptotic drug.^28,29 The SI in response to ABT-737 was essentially unchanged compared with vehicle-treated controls (Suppl. Fig. S9B). To validate these results, we performed the β-galactosidase (β-gal) assay in addition to qPCR analysis as a common reporter of senescence measurements.^27,30,31 β-gal signals and overexpression of genes involved in senescence were present only in Nut-3a-treated cells but not in ABT-737-treated cells (Suppl. Fig. S10).

Finally, we assessed whether the toolbox could quantify the effects of combination treatments. For this purpose, we evaluated the phenotypic response of U2OS to a mixture of Nut-3a and ABT-737. Figure 6A shows the effect on the area occupied by colonies in single treatments. Increasing concentrations of Nut-3a affected the colony-forming potential, as shown above; in contrast, ABT-737 alone had no effect on the growth of U2OS at doses up to 5 µM. We verified colony growth over low doses of Nut-3a (0.5–1.5 µM) with a range of ABT-737 concentrations (data not shown). A combination of ABT-737 (up to 5 µM) with 0.5 µM Nut-3a did not show any effects on colony formation (data not shown). By contrast, ABT-737 significantly enhanced the toxicity of a dose of Nut-3a that, when used alone, caused senescence. Specifically, the combination of 2.5 µM ABT-737 and 1.5 µM Nut-3a elicited a significant 30% reduction in growth compared with Nut-3a alone at this dose ( Fig. 6B ; and Suppl. Table 2 for statistical support).

Figure 6.

Evaluation of drug combinations. (A) The effect of increasing the concentration of either Nut-3a (black circles) or ABT-737 (white circles) alone is shown for comparison. (B) The graph reports the quantitation of colony growth potential induced by the combination of Nut-3a (1.5 µM) with increasing concentrations of ABT-737. The data (mean and SE) are derived from triplicate experiments.

Discussion

The CAT system we describe here is characterized by a cellular-level resolution of colonies in multiwell format, which facilitates the extraction of high-content information from colony formation assays.

Following the initial calibration of the system with a pan-cellular stain such as CellMask, this workflow can be performed using only a nuclear stain. This method potentially could be applied to samples stained with a single dye that is able to label both the nucleus and cytoplasm, thereby obviating the need for a calibration procedure (data not shown). The cell type–dependent variability in the relative nuclear versus cytosolic staining efficiency of a specific single dye, however, will require careful consideration on a case-by-case basis. Thus, to provide a robust and universal method to automate colony analysis, we opted to introduce a nuclear stain combined with a short step for cell calibration (which is neither cost-prohibitive nor challenging in terms of analysis). Combined with the excellent performance of the dMask algorithm on regular (i.e., non-optical-quality), clear-bottom plastic plates, this makes CAT extremely low-cost.

In addition, the complete CAT workflow can be performed in a short period of time (comparable with other automated approaches^10,12); indeed, only half a day was required from the staining procedure to the data analysis, which is a dramatic reduction compared to the time required for manual approaches (Suppl. Table 1A and 1B).

In a single experiment, the CAT toolbox enables the user to optimize the measurement of the colony growth potential of any cell line in different multiwell formats (Suppl. Fig. S11) and in the absence or presence of drugs. This is possible because the analysis is focused on the percentage of the area covered by colonies in the FOV ( Figs. 2C and 3 ), rather than on the absolute colony count. The rationale for this is that colony counting is problematic when the colonies are at high density (as we observed, e.g., in untreated cells by day 7 of the assay). Our comparisons of different microscopes, standard plastic versus optical-quality plates, and staining procedures demonstrate the robust and highly correlated nature of the data, making CAT a readily accessible cross-platform tool ( Fig. 3 ). This is underscored by the ability of CAT to report IC₅₀ values for Nut-3a that closely match those obtained with this drug, or its analogs, in previous reports.^13,14

The image segmentation and analysis we used enabled us to define and measure, in a colony assay format, an SI without the requirement for complex additional staining protocols. The hallmarks of senescence, such as hypertrophy of cells and nuclei,^27,30,31 are induced by activated oncogenes and chemotherapeutic drugs, and thus have direct clinical relevance. We observed this phenotype in Nut-3a-treated (but not in ABT-737-treated) cells ( Fig. 5B and Suppl. Figs. S9 and S10). Indeed, by combining two senescence-related parameters (proliferative potential as measured by nuclear count, and the morphological features defined by colony area), we could detect a significant increase in the proportion of senescent colonies in Nut-3a-treated cells ( Fig. 5C ). This confirms previous reports of senescence induced by this drug in tumors with p53^WT.^27,30,31 Moreover, by measuring the metabolic activity of the colonies, we detected >3-fold higher ATP production per cell in senescent colonies ( Fig. 4 ). This higher energy demand was not observed in untreated cells or in cells treated with low doses of the drug ( Fig. 4 ). Similar observations were recently published by Dorr and colleagues,³² who observed that ATP production was higher in tumors that underwent senescence following chemotherapeutic treatments. Together, these results confirm that by using only a single nuclear stain, the CAT system is able to identify treatments that induce proliferative arrest and also indicate whether such treatments induce cell senescence in colony assay formats. Furthermore, although ABT-737 is reported to induce senescence,^28,29 our data also show that this is likely to be cell line dependent.

Induction of tumor cell senescence has been suggested as a potential chemotherapeutic strategy. The persistence of senescent cells that harbor oncogenic lesions, however, could increase the risk of tumor recurrence. Thus, shifting the phenotypic response from senescence toward apoptosis using combination therapies may be beneficial. We show here that the CAT method can be used to predict such drug combinations. For example, when compared with either single agent alone, the area covered by U2OS colonies was lower when Nut-3a (1.5 µM) was combined with ABT-737 ( Fig. 6B ). The remaining colonies, however, still exhibited a senescent phenotype (data not shown). Notably, senescence was not observed with ABT-737 alone. Together, these data confirm that in these cells, ABT-737 induced neither apoptosis ( Fig. 6A ) nor senescence (Suppl. Figs. S9 and S10), but can sensitize cells to Nut-3a.^25,32 This indicates that the major response of U2OS following Nut-3a treatment is senescence, but this can be converted to apoptosis on combination with a BH3-like BCL2 antagonist.²⁹

In some contexts, subtle phenotypic changes, such as changes in the nuclear length–width ratio, may precede more “obvious” effects, such as cell death or inhibition of proliferation. Although this was not the case with Nut-3a in our experiments (Suppl. Fig. S8), our ability to detect these types of parameters will be useful to categorize and then cluster compounds that have significant yet more “nuanced” cellular effects.

We note that CAT was developed using U2OS, which is representative of a number of “workhorse” tumor cell lines used for image-based screening. Such lines perform well with wide-field imaging systems due to their planar morphology and well-separated nuclei. The extension of CAT to cell lines in which nuclei within colonies are close to one another or to 3D culture models will require careful consideration on a case-by-case basis.

We tested CAT in other human tumor cell lines with more tightly clustered nuclei (HeLa adenocarcinoma and A375 melanoma), however; our system was able to segment colonies and extract proliferative information in these cell types and in higher density multiwell formats (Suppl. Fig. S11).

In conclusion, the Colony Assay Toolbox (CAT) enables the multiparametric analysis of mammalian cell colony-forming potential. The multiwell format of this assay, combined with a simple and cost-effective experimental procedure, renders CAT accessible to virtually all laboratories. In contrast to previous methods, a single dye for nuclear staining and a simple pipeline (dMask) were sufficient for the extraction of high-content information, even in the context of a colony assay, due to the level of single-cell resolution obtained; the procedure also leaves other fluorescence channels available for the additional labeling of organelles or proteins. Furthermore, the simple workflow allows the user to set up the image analysis without prior knowledge of programming languages. Moreover, the use of wide-field microscopes and our integration with CellProfiler, an open-source image analysis environment, make this protocol widely accessible. The CAT system can thus be used as a high-content alternative to quantify colony growth in the context of combinatorial drug treatments, and it can be exploited in both compound and functional genomic screens to identify regulators of senescence. The pipelines used in this work are also available freely for download both at the CellProfiler Web site (www.cellprofiler.org) and at our Web site (http://genomics.iit.it/core-platforms/high-throughput-screening-unit.html).

Footnotes

Supplementary material for this article is available on the Journal of Biomolecular Screening Web site at .

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

References

Meyskens

F. L.

Jr. Thomson

S. P.

Moon

T. E.

Quantitation of the Number of Cells within Tumor Colonies in Semisolid Medium and Their Growth as Oblate Spheroids. Cancer Res. 1984, 44, 271–277.

Shoemaker

R. H.

Wolpert-DeFilippes

M. K.

Kern

D. H.

. Application of a Human Tumor Colony-Forming Assay to New Drug Screening. Cancer Res. 1985, 45, 2145–2153.

Barber

P. R.

Vojnovic

Kelly

. Automated Counting of Mammalian Cell Colonies. Phys. Med. Biol. 2001, 46, 63–76.

Bernard

Kanduser

Pernus

Model-Based Automated Detection of Mammalian Cell Colonies. Phys. Med. Biol. 2001, 46, 3061–3072.

Dahle

Kakar

Steen

H. B.

. Automated Counting of Mammalian Cell Colonies by Means of a Flat Bed Scanner and Image Processing. Cytometry A. 2004, 60, 182–188.

Wylie

G. P.

Bowen

W. P.

Determination of Cell Colony Formation in a High-Content Screening Assay. J. Lab. Auto. 2005, 10, 203–206.

Katz

Ito

Lau

K. S.

. Increased Efficiency for Performing Colony Formation Assays in 96-Well Plates: Novel Applications to Combination Therapies and High-Throughput Screening. Biotechniques. 2008, 44, ix–xiv.

Carpenter

A. E.

Jones

T. R.

Lamprecht

M. R.

. CellProfiler: Image Analysis Software for Identifying and Quantifying Cell Phenotypes. Genome Biol. 2006, 7, R100.

Vokes

M. S.

Carpenter

A. E.

Using CellProfiler for Automatic Identification and Measurement of Biological Objects in Images. Curr. Protoc. Mol. Biol. 2008, Ch. 14, Unit 14 17.

10.

Geissmann

OpenCFU, a New Free and Open-Source Software to Count Cell Colonies and Other Circular Objects. PLoS One. 2013, 8, e54072.

11.

Cai

Chattopadhyay

Liu

W. J.

. Optimized Digital Counting Colonies of Clonogenic Assays Using ImageJ Software and Customized Macros: Comparison with Manual Counting. Int. J. Radiat. Biol. 2011, 87, 1135–1146.

12.

Guzman

Bagga

Kaur

. ColonyArea: An ImageJ Plugin to Automatically Quantify Colony Formation in Clonogenic Assays. PLoS One. 2014, 9, e92444.

13.

Vassilev

L. T.

B. T.

Graves

. In Vivo Activation of the p53 Pathway by Small-Molecule Antagonists of MDM2. Science. 2004, 303, 844–848.

14.

Ding

Zhang

Liu

J. J.

. Discovery of RG7388, a Potent and Selective p53-MDM2 Inhibitor in Clinical Development. J. Med. Chem. 2013, 56, 5979–5983.

15.

Kamentsky

Jones

T. R.

Fraser

. Improved Structure, Function and Compatibility for CellProfiler: Modular High-Throughput Image Analysis Software. Bioinformatics. 2011, 27, 1179–1180.

16.

Abraham

V.C.

Ghosh

R.N.

Sammak

. High Content Screening to the Analysis of Stem Cell Growth & Differentiation. Thermo Scientific Cellomics. 2008. Release note. www.thermosci.jp/cellomics/docs/008_StemCell_March08.pdf

17.

Ichinose

Liu

X. H.

Hagihara

. Extracellular Bad Fused to Toxin Transport Domains Induces Apoptosis. Cancer Res. 2002, 62, 1433–1438.

18.

Bettger

W. J.

Boyce

S. T.

Walthall

B. J.

. Rapid Clonal Growth and Serial Passage of Human Diploid Fibroblasts in a Lipid-Enriched Synthetic Medium Supplemented with Epidermal Growth Factor, Insulin, and Dexamethasone. Proc. Natl. Acad. Sci. USA. 1981, 78, 5588–5592.

19.

Liao

Creek

K. E.

. Human Keratinocytes and Tumor-Derived Cell Lines Express Alternatively Spliced Forms of Transforming Growth Factor-Alpha mRNA, Encoding Precursors Lacking Carboxyl-Terminal Valine Residues. Oncogene. 1999, 18, 5554–5562.

20.

Pomp

W. J.

Ouwerkerk

I. J.

Hoogstraten

. Cell Density Dependent Plating Efficiency Affects Outcome and Interpretation of Colony Forming Assays. Radiother. Oncol. 1996, 40, 121–125.

21.

Soussi

Focus on the p53 Gene and Cancer: Advances in TP53 Mutation Research. Hum. Mutat. 2003, 21, 173–175.

22.

Maeshima

Iino

Hihara

. Nuclear Size, Nuclear Pore Number and Cell Cycle. Nucleus. 2011, 2, 113–118.

23.

Edens

L. J.

White

K. H.

Jevtic

. Nuclear Size Regulation: From Single Cells to Development and Disease. Trends Cell Biol. 2013, 23, 151–159.

24.

Hashimura

T. Y.

Yoshida

Soft Agar Colony Formation of Mouse Epidermal Cells during the Early Phase of Two-Stage Chemically Induced Carcinogenesis. Japan J. Cancer Res. 1985, 76, 321–323.

25.

Wade

Rodewald

L. W.

Espinosa

J. M.

. BH3 Activation Blocks Hdmx Suppression of Apoptosis and Cooperates with Nutlin to Induce Cell Death. Cell Cycle. 2008, 7, 1973–1982.

26.

Korotchkina

L. G.

Demidenko

Z. N.

Gudkov

A. V.

. Cellular Quiescence Caused by the Mdm2 Inhibitor Nutlin-3A. Cell Cycle. 2009, 8, 3777–3781.

27.

Efeyan

Ortega-Molina

Velasco-Miguel

. Induction of p53-Dependent Senescence by the MDM2 Antagonist Nutlin-3a in Mouse Cells of Fibroblast Origin. Cancer Res. 2007, 67, 7350–7357.

28.

van Delft

M. F.

Wei

A. H.

Mason

K. D.

. The BH3 Mimetic ABT-737 Targets Selective Bcl-2 Proteins and Efficiently Induces Apoptosis via Bak/Bax if Mcl-1 Is Neutralized. Cancer Cell. 2006, 10, 389–399.

29.

Adams

J. M.

Cory

The Bcl-2 Apoptotic Switch in Cancer Development and Therapy. Oncogene. 2007, 26, 1324–1337.

30.

Kranz

Dobbelstein

Nongenotoxic p53 Activation Protects Cells against S-Phase-Specific Chemotherapy. Cancer Res. 2006, 66, 10274–10280.

31.

Kumamoto

Spillare

E. A.

Fujita

. Nutlin-3a Activates p53 to Both Down-Regulate Inhibitor of Growth 2 and Up-Regulate mir-34a, mir-34b, and mir-34c Expression, and Induce Senescence. Cancer Res. 2008, 68, 3193–3203.

32.

Dorr

J. R.

Milanovic

. Synthetic Lethal Metabolic Targeting of Cellular Senescence in Cancer Therapy. Nature. 2013, 501, 421–425.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

6.62 MB

0.00 MB

Open Access to High-Content Clonogenic Analysis

Abstract

Keywords

Introduction

Materials and Methods

96-Well Colony Formation Assay, Drug Treatments, and Viability Assay

Immunofluorescence Protocol

Fluorescent Microscopy

Image Processing

Data and Statistical Analysis

Results

Calibration Procedures to Set and Optimize the dMask Pipeline

Quantitation of Colony Growth in a 96-Well Format

Drug Dose–Response Curve Validation

Extraction of Multiparametric Data from Colony Outgrowth Assays

Discussion

Footnotes

Declaration of Conflicting Interests

Funding

References

Supplementary Material