Currently, little systems-level understanding exists of the organization and composition of signaling networks that regulate cell morphology. Christopher Bakal, a research fellow in the Department of Genetics at Harvard Medical School, and his colleagues developed quantitative morphological profiling methods to systematically investigate the role of individual genes in the regulation of cell morphology in a fast, robust, and cost-efficient manner.
The researchers genetically screened 249 gene over-expression or double-stranded RNA treatment conditions using the Drosophila BG-2 cell line. First, they stochastically labeled samples with green fluorescent protein to enable individual cells to stand out in the crowded and overgrown samples, acquired images of these cells using conventional fluorescence microscopy, and used software to identify the boundaries of individual cells.
For each individual cell, they computed 145 different quantitative features that reflected basic aspects of cell geometry, detailed aspects of cellular protrusions, or the distribution of GFP intensity within the cellular boundaries.
To transform these 145 features into biologically meaningful morphological indicators, they trained a set of neural networks to use informative subsets of the features to discriminate cells from particular reference TCs from sets of other reference TCs
Finally, for each NN and TC, they calculated a NN Z-score, which is the variance-adjusted difference between the mean NN score of all cells in the TC and the mean NN score of all cells in the data set.
Bakal this week answered a few questions for CBA News about his team’s work, which was published in the June 22 issue of Science.
Can you give me a little background on this work?
It came from my own interest in signaling networks and networks of genes. I think that there is an ever-growing appreciation that we cannot understand any type of biology within the context of just one gene or just a few genes. It is really networks of proteins and signaling networks involving hundreds, if not thousands, of different components, that control any aspect of cellular behavior.
One question that we are interested in is: How do you study at the same time these cellular machines composed of so many components? We really need new techniques in order to understand the relationships that exist among all of these different molecular components.
I am particularly interested in cell shape. Why, for example, is a neuron such a different shape than a muscle cell or a skin cell? I wanted to understand the actions of the many genes that give the cells their characteristic shapes. That is the academic/biological problem. We just need new methods to study these networks.
The other parallel thing that has happened is that a lot of cell biology today is done in a qualitative manner. I really wanted to make cell biology more quantitative in nature. We were looking for ways to address both of those problems simultaneously, and that is how we got into the business of measuring cell shape.
How is this relevant to drug discovery?
It is actually very relevant to drug discovery, which is something that we are already interested in. Many screens out there look for small molecules as inhibitors of certain processes. Many of these assays yield a simple readout of a drug’s effect. For example, does the cell live or die? These assays are based on simple reporters.
This is really a complex readout of what is going on inside a cell in response to a small molecule. What we hope to do is combine, for example, a drug-based screen with, for example, an RNAi-based screen.
Say you have a drug that gives you a certain shape, which we measure quantitatively. What if in parallel, we did a RNAi screen knocking out individual genes and identified one gene that gives you the same shape quantitatively as the drug when you knock it out?
You could probably say that drug is very likely a target of that gene based on that quantitative readout.
And it a very cheap assay. That’s one thing we want to stress. We are learning a lot about cellular behavior from an assay that is very easy and inexpensive to do.
Why did you use the techniques that you did?
We are hardly the first people to decide that we wanted to quantitate cell morphology. However, the issue remains of how you get a computer to see in an image what a human would see. We as humans can see very detailed things in cells that computers cannot. In addition, the samples that we use are very complex in nature.
So we decided to make things easier for the computer by labeling a particular population of cells to greatly simplify the image. We call that stochastic labeling, because instead of looking at all the cells in a particular treatment, we only look at a random sampling of them.
By looking at this random sampling, and getting high-quality, easy-to-interpret information about it, we can make judgments about the whole population.
It is still a major challenge in terms of the imaging of cells to properly and accurately segment and get information from cells.
When would it be appropriate to use this technique?
Any time you are looking for detailed and accurate information about cell shape or specific cellular structures, and you really want to accurately measure the morphology and shape of cells.
When would it be inappropriate?
There are a lot of assays out there now that look, for example, at nuclear translocation or the phosphorylation of one particular marker.
What was gratifying was discovering that you can really learn a lot about what is going on inside the cell from cell morphology alone. For example, you could name a protein, and you could predict where it would act inside the cell, where it would exist inside the cell, what its binding partners would be, and what its function is for hundreds of different genes.
You could learn a lot about your favorite gene or assay from running this assay.
What would be the next step in this work?
The next step would be to increase the scale of the assay. Right now this has only been done on a relatively limited subset of genes. Doing it on small molecules would certainly be something to investigate. Another thing would be to use it on clinical samples.
Are you planning to publish further work?
Yes, but we are not quite there yet. We are collecting the data right now.