Invasion-Block and S-MARVEL: A high-content screening and image analysis platform identifies ATM kinase as a modulator of melanoma invasion and metastasis

Abstract

Robust high-throughput assays are crucial for the effective functioning of a drug discovery pipeline. Herein, we report the development of Invasion-Block, an automated high-content screening platform for measuring invadopodia-mediated matrix degradation as a readout for the invasive capacity of cancer cells. Combined with Smoothen-Mask and Reveal, a custom-designed, automated image analysis pipeline, this platform allowed us to evaluate melanoma cell invasion capacity posttreatment with two libraries of compounds comprising 3840 U.S. Food and Drug Administration (FDA)-approved drugs with well-characterized safety and bioavailability profiles in humans as well as a kinase inhibitor library comprising 210 biologically active compounds. We found that Abl/Src, PKC, PI3K, and Ataxia-telangiectasia mutated (ATM) kinase inhibitors significantly reduced melanoma cell invadopodia formation and cell invasion. Abrogation of ATM expression in melanoma cells via CRISPR-mediated gene knockout reduced 3D invasion in vitro as well as spontaneous lymph node metastasis in vivo. Together, this study established a rapid screening assay coupled with a customized image-analysis pipeline for the identification of antimetastatic drugs. Our study implicates that ATM may serve as a potent therapeutic target for the treatment of melanoma cell spread in patients.

Keywords: high-throughput screening; imaging; melanoma metastasis.

Fig. 1.

Optimization and characterization of fluorescent gelatin-based invadopodia assay. (A) Representative confocal immunofluorescence images of WM983C cells plated on fluorescent gelatin matrix (gray). The images highlight the presence of F-Actin (blue) within regions of gelatin degradation (black foci, arrows); (B) Representative confocal immunofluorescence image (extended focus) of gelatin (gray), F-actin, MMP14, and cortactin (colors as indicated) in WM983C cells; (C) Bar graph depicting the gelatin degradation capacity of commonly used melanoma cell lines. Error bars represent the mean ± SEM of at least n = 3 independent experiments; (D) Immunofluorescence confocal imaging of B16-F10 melanoma cells highlighting the colocalization of F-Actin, cortactin, MMP14, and regions associated with gelatin degradation. (Scale bar, 10 μm).

Fig. 2.

Development, integration, and validation of invasion-block and the S-MARVEL pipeline. (A) Schematic highlighting the workflow for high-throughput, high-content drug screening and imaging (Invasion-Block); (B) Representation of customized image analysis pipeline S-MARVEL for the automated quantification of gelatin degradation; (C and D) Two representative image panels of invadopodia imaging assay, pre- and post-S-MARVEL processing. Artifacts caused by shadows from uneven imaging (arrows pointed in Left panels) can be removed and the invadopodia structures that are affected (arrows pointed in Right panels) can be nicely retained and revealed. Inverted masks are represented here for better visualization; (E) Invadopodia index of MMPi treated wells normalized to dimethyl sulfoxide (DMSO) control in 14 independent screening plates from screening of 3840 compounds as indicated. The index was acquired prior to S-MARVEL processing; (F) Invadopodia index of MMPi treated wells normalized to DMSO control in 14 independent screening plates from screening of 3840 compounds as indicated. The index was acquired after S-MARVEL processing. All calculations were carried out using masked images; (G) Z-factor of Invasion-Block screenings of 3840 compounds pre- and post-S-MARVEL smoothing. Student’s t test was employed to determine the statistical significance. A difference was considered significant if P < 0.05 and ****P < 0.0001.

Fig. 3.

S-MARVEL offers superior performance at removing background/artifacts of Invasion-Block images. (A) Representative images of background/artifact removal using different image analysis software as indicated. S-MARVEL was able to remove uneven fluorescence signals (yellow squares) and artifacts (red squares) and reserve invadopodia degradation signals. (B) Z-factor calculated from 144 images processed by S-MARVEL or Rolling Ball (ImageJ) of 18 wells of invasion-block assay (9 wells of positive control and 9 wells of negative control). (C) Extended focus images of light-sheet data from orthotopic amelanotic melanoma tumors from Lyz2^gfp/+ mT/mG mice. The images highlight the ability of the S-MARVEL pipeline in removing various artifacts observed in original datasets (white arrows). (D) Extended-focus and single-plane images of the intravital bone marrow imaging dataset highlight the ability of the S-MARVEL pipeline in removing artifacts observed in original datasets (white arrows). (E) Extended focus and single plane images of explant lymph node imaging dataset. The panel demonstrates the ability of the S-MARVEL pipeline in removing artifacts observed in original datasets (white arrows). Student’s t test was employed to determine the statistical significance. A difference was considered significant if P < 0.05 and ****P < 0.0001.

Fig. 4.

Invasion-Block and S-MARVEL identify multiple kinase inhibitors capable of suppressing invadopodia formation in melanoma cells. (A) Scatter plot of the Invasion-Block screening of 3840 drug compounds. The analysis identified a significant reduction in matrix degradation activity by Bosutinib, AZD 6482, PI 828, and KU55933, highlighted in red. Other MMP inhibitors like CP471474 and SD 2590 were also identified. The y axis represents the Log₁₀ of invadopodia index normalized to DMSO negative control wells, and the x axis represents all 3840 compounds, in a sequential order of screening plating; (B) Normalized invadopodia index of screening hits from Invasion-Block of 3840 compounds, all conditions are normalized to DMSO control (indicated as a dotted line); Error bars represent the mean ± SD of n = 3 independent wells; (C) Representative images of compound hits in the 3840 Invasion-Block screening, compounds are as indicated; blue is nuclei (DAPI) and white dots represent degradative foci; (D) Scatter plot of the Invasion-Block screening of 210 kinase inhibitors. The analysis identified a significant reduction in matrix degradation activity by Enzastaurin, GSK2126458, PD173955, PD173955 analogue 1, Bisindoylmaleimide X, AZD 6482, TGX 221, KU55933, and KU60019, highlighted in red. The y axis represents the Log₁₀ of invadopodia index normalized to DMSO negative control wells, and the x axis represents all 276 compounds (some inhibitors were screened in replicates), in the order of screening plating; (E) Normalized invadopodia index of hits from Invasion-Block of 210 kinase inhibitors, including PKC (Enzastaurin and Bisindolylmaleimide-X), Abl and c-Src (PD173955), c-Src (PD173955 analogue 1), PI3K (GSK2126458 and TGX221), and ATM (KU55933 and KU60019) inhibitors. Error bars represent the mean ± SD of n = 3 independent wells, all conditions are normalized to DMSO control (indicated as dot line); (F) Representative images of compound hits in the 210-compound Invasion-Block screening, compounds are as indicated. Blue represents nuclei (DAPI), and white dots represent degradative foci. Please note that for the purpose of improved visualization, the masked degradation foci images have been inverted (white dots on black background).

Fig. 5.

Enzastaurin, TGX221, KU60019, and ATM KO inhibit melanoma cell invasion in vitro. (A and B) Percentages of cells with positive gelatin degradation posttreatment with DMSO, pan-MMP inhibitor, Enzastaurin, TGX221, and KU60019 at concentrations of 0.3, 1, 3, and 10 µM (as indicated) in B16-F10 and WM983C cells. DMSO treatment was used as control; Right panels, Representative confocal immunofluorescence image (extended focus) of gelatin (gray) and ECM degradation (black) in B16-F10 or WM983C cells under different treatments at 10 µM (as indicated). Blue represents nuclei (DAPI). (Scale bar, 10 μm); (C) Phase contrast images of WM983C spheroid invasion 72 h after embedding in collagen. Cells with treatments of 10 µM DMSO, Enzastaurin, TGX221, and KU60019 as indicated. Arrows indicate that the distance cells have invaded away from the edge of the spheroid. (Scale bar, 100 μm); (D) Quantification of the distance cells have invaded away from the spheroid edge (normalized to the DMSO control). Error bars represent the mean ± SEM of at least n = 3 independent experiments. (E) Western blots showing the level of ATM in WM983C cells expressing Non-target control (Non-target), ATM CRISPR-knockout #1 (KO#1) and #2 (KO#2). Representative of at least 3 independent experiments; (F) Phase contrast images of WM983C spheroid invasion 72 h after embedding in collagen. Cells express Non-targeting control, ATM CRISPR-knockout #1 and #2 are indicated as Non-target, ATM KO#1 and ATM KO#2. Arrows indicate the distance cells have invaded away from the edge of the spheroid. (Scale bar, 100 μm); (G) Quantification of the distance cells have invaded away from the spheroid edge (normalized to the Non-targeting control). Error bars represent the mean ± SEM of at least n = 3 independent experiments. ANOVA and Dunnett’s test were employed to determine the statistical significance. A difference was considered significant if P < 0.05 (*P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001).

Fig. 6.

ATM KO blocks melanoma cell invasion and metastasis in vivo. (A) Western blots showing the level of ATM in B16-F10 cells expressing vector control (vector), ATM CRISPR-knockout #1 (KO#1) and #2 (KO#2). Representative of at least 3 independent experiments; (B) Average tumor volumes of each group of mice postinjection of B16-F10 empty vector control (vector control), ATM CRISPR guide RNA KO#1 (KO#1) and ATM CRISPR guide RNA KO#2 (KO#2). Error bars represent the mean ± SEM of 18 mice per group; (C) Representative bright-field color images of positive and negative lymph node metastasis, (scale bar, 500 µm); (D) Percentage of positive and negative lymph node metastasis within each indicated group. N = 18 for each group; Data pooled from 3 independent experiments. (E) ATM mRNA expression levels in primary tumors and metastases of melanoma patients from TCGA datasets; (F) ATM gene expression levels in melanoma microarray dataset (GDS3966) via Robust Multichip Average (RMA) normalization. Student’s t test was employed to determine the statistical significance. A difference was considered significant if P < 0.05.