Exploring the potential Nanolive technology holds for phenotypic drug discovery

Nanolive’s label-free, multiplexed data offers users a non-invasive means of gathering high content data (11 metrics) without time-consuming experiments or costly fluorescent reagents, which perturb cell responses. In particular, it allows researchers to phenotypically profile small molecules:

• At multiple time-points without biasing results by choosing a single timepoint in end-point techniques
• Without confounding effects of photobleaching or phototoxicity
• In a robust and standardized manner using meaningful metrics of the highest biological relevance, i.e., changes in cell morphology and cell composition

The goal of the study was to evaluate whether Nanolive’s label-free live cell image analysis could be combined with machine learning to identify the mechanism of action of small molecule drugs.

Dr. Padrón and his team tested this by exposing the non-small cell lung cancer cell line SW1573 to three antimitotic drugs: paclitaxel (PTX), colchicine (CLC), and vinblastine (VBL), whose mechanisms of action differ. PTX is a microtubule stabilizer agent; it binds to tubulin, which induces mitotic arrest by preventing the formation of a centered mitotic spindle. In contrast, CLC and VBL are microtubule-disrupting agents; they bind to specific β-tubulin binding sites, which disrupts the formation of microtubules and blocks the cell cycle.

Images were captured every 3 mins for 20 h and then analyzed by Nanolive’s software, EA, which outputs 11 metrics covering all aspects of cell morphology (area, perimeter, form factor, extent, compactness, and eccentricity) and composition (mean RI, average dry mass density, dry mass, and granularity).

The metrics easily captured the phenotypic differences between the two groups of drugs; exposure to PTX caused a reduction in cell area, with a rounded and diminished nucleus area surrounded by a contracted cytoplasm, whereas CLC and VBL induced a nuclear reduction, cytoplasm contraction and an overall reduction in cell area.

When the experiments were completed, the authors used an approach called Fast Fourier Transform (FFT) Analysis to analyze the data. The 11 phenotypic metrics output by EA were fed into an algorithm which broke each parameter down into two measures (maximal amplitude and phase). These values were then used to cluster compounds with similar modes of action with great success.

This study is an important first step towards developing computer-assisted phenotype annotation of the mode of action of compounds. Congratulations to all the authors involved.