Automatically detecting subcellular morphologies with u-shape3D

To draw robust biological conclusions from 3D optical microscopy data, we need automated computational tools such as u-shape3D.
Automatically detecting subcellular morphologies with u-shape3D

When we first started this project, there was one light-sheet microscope in the lab. Its name was Louis, short for Louis XIV, the opulent French “Sun” king, who built the extravagant Palace of Versailles. Although we later more formally named the microscope meSPIM (microenvironmental Selective Plane Illumination Microscopy), everyone in the lab did and still does refer to it as Louis.  

Louis the microscope with co-author Erik Welf in the background.

As expected of a Sun King, Louis produces exquisitely detailed fluorescence images. The images are so detailed that most people’s first instinct is just to sit and marvel at them. As our awe slowly turned into a desire to actually learn new biology, however, we ran into a problem. Simply looking at a 3D movie takes a lot of effort and time. A 2D microscopy movie can, for the most part, be understood by directly watching it, but there is no one way to visualize a 3D movie on a 2D screen. The only way of appreciating the multitude of details in a 3D microscopy movie is to interact with it using software that allows you to visualize it in many different ways. This process takes so long that it’s easy to get lost in the data. While interacting with a movie, many times someone would say something to the effect of ‘this reminds me of a detail I saw three weeks ago in another cell.’ But finding that previously noticed detail is such a massive operation that we eventually stopped trying. (I named some of the cells so that we could more easily refer to them, which did help, but only somewhat.) 

A blebbing cell, which we named Ornament after a holiday decoration, colored by local actin localization.

We solved our problem of too much intricate detail by building computational tools that allow us to extract information from many cells. If you'd like to learn more about these tools, please go read our paper! Rather than fruitlessly trying to compare multiple cells by looking at them, we can now test our hypotheses by asking the software a series of targeted questions that have quantitative answers. 

That computational analysis is required to effectively even look at our data is relatively new to the field of cell biology, but it’s certainly not new to other fields. Genomics, for example, has many established computational tools for exploratory data analysis. However, whereas hopefully no one thought they could make sense of the human genome by sitting down and reading it, convincing people that they need to move on from simply admiring images like Louis’ is tough. There are more and more microscopes that can produce images as marvelous as Louis’, however, and it would be a loss for biology if we didn’t develop the software necessary to fully use them. The tools described in this paper, and other tools that are now being developed, are just the beginning of this process.