Table 1.
Problems encountered during pipeline development.
Fig 1.
Mosaic image generated by EVOS.
On left, is a 4x tiled scan (a composite of 4x3 images) of a single well of a 96-well plate containing H2B-GFP labeled MCF-7 cancer cells. We have overlaid a gray dashed line to indicate the edge of each image. A 40x phase contrast image of these cells is displayed in the upper right panel, whereas the fluorescent GFP channel is provided at bottom right. The perimeter of the well appears as a bright green fluorescent, labeled ‘well halo’ (indicated by the yellow dashed line).
Fig 2.
Fluorescence produced by cell media.
a) Representative tiled-images of wells containing different media (phenol red-free ‘ClearDMEM;’ standard ‘DMEM’ containing phenol red; ‘DMEM/F12;’ phosphate buffered saline ‘PBS;’ ‘‘M87’ breast epithelial medium; and FlouroBrite™ ‘Imaging DMEM’ b) Fluorescent intensity within the colored circles (shown in 2a) was measured by ImageJ (n = 6 for each condition). * statistically significant (p<0.05, t-student test).
Fig 3.
Debris in unfiltered medium interferes with cell counts.
Unfiltered DMEM (supplemented with 10% FBS) contains particulates that the imaging software will incorrectly identify as cells (purple outlines).
Fig 4.
Luminosity variance of EVOS autoexposure.
On left, two replicate wells captured using EVOS’ autoexposure setting. On right, Matlab histogram analysis of pixel count and fluorescence intensity of these two replicates. Δ symbol emphasizes difference in mean fluorescence intensities between replicates (introduced by the automatic exposure setting). * p<0.0001 (t-test of random background fields (free of cells), n = 18 fields for each replicate).
Fig 5.
The impact of autofocus on scan time.
The time to scan an entire plate varies among the five EVOS focus settings (Autofine each image, Autofine center only, Find sample each image, Find sample center only, and Manual).
Fig 6.
Compression error caused by software settings.
On left, an image of a 96-well produced by EVOS. On right, a saved image of a well (from the same plate/scan) that has compressed the well into an ovoid shape. This problem was corrected by increasing the computer’s virtual memory allocation.
Fig 7.
Comparing well A1 from eight 96-well plates shows the variability of the well location in the respective images. The growth area of each well is indicated by the color outlines (left). When these are overlaid, the variability is evident (right).
Fig 8.
MCF-7 cells were processed via the optimized imaging pipeline. The final step outlines each cell nuclei and generates a merged image of the outlines(purple outlines) and H2B-GFP fluorescent nuclei (green). Inset is a magnification of the area outlined within the white box.
Fig 9.
Assessment of cancer cell growth via image analysis pipeline.
MCF-7 (left) and MBA-MD-231 (right) cells were seeded at low density (100 cells per well, 96-well plate). They were scanned every 2 days to quantify their growth. Images were processed via the developed pipeline, as described in methods. Data is expressed as population doublings. Mean ± SD (n = 12).