It’s soon going to be one year since we started screening for drug candidates that slow or reverse the progress of Niemann-Pick C disease in simple animal models.
The last six months here at the Perlstein Lab have been especially busy. We are now knee deep into the screening process, only surfacing momentarily for a breath of air.
So, like our fly scientist Tom blogged a couple weeks ago, I wanted to write about the process of worm screens, what factors we considered, what parts were difficult and so on..
Recap
As I mentioned in my last worm post, the phenotype in the nematode models of NPC had been well studied. They include slowed growth, smaller brood sizes, proclivity to entering a dormant or ‘dauer’ state under favorable conditions and last but not the least, a sensitivity to cholesterol deprivation. Thus, when we generated our proprietary worm mutants in collaboration with Knudra Transgenics, we first validated that our worms replicated the phenotype of previously published npc1 (aka ncr-1) mutants.
Identifying a quantifiable phenotype
While we validated that our new strain phenocopied the mutants generated by Jim Thomas et al, we also worked on how to make our phenotype rapidly quantified. In other words, how do we develop a high throughput assay around it? Normally when we study worms, we do so in a petri dish that is filled agar and has a lawn of bacteria on the surface. The bacteria is there because that is what the worms eat. Here is a picture of worms on a plate. Those wavy lines that you see are made by worms crawling through a lawn of bacteria:
In these images, you can also see that when raised on agar, animals grow well in the presence of cholesterol. However, when deprived of cholesterol, they are smaller and reproduce with smaller brood sizes. Thus, in theory, we could select endpoints — number of eggs, number of worms, size of worms — to measure.
While the phenotype looks robust, how does one measure it? We dabbled with taking images of the field of view and using automated image-based algorithms to detect worms and measure them. Here is an representation of how a processed image would look:
There are several open source image processing algorithms that can extract quantitative information regarding the number of worms, their lengths, etc and spit out metrics that allow us to make comparisons across two treatments. So, we were on to something, right?
Kinda-sorta…but not completely. While we can certainly use this method to get to our end goal, that is to screen worms with compounds that rescue the growth defect, there are some logistical and analytical limitations to getting data from such an assay format rapidly.
Identifying logistical limitations
First, to do an assay in such an agar plate format, we would need mg quantities of drugs in our library. We would need to solubilize and mix every drug into molten agar and then expose our worms to them. For a 50,000-100,000 compound library this becomes a very labor-intensive step. Next, in order for us to conduct this kind of analysis, we would have to take multiple images of each plate. After each plate was done, we would need to manually move the next plate into focus, conduct the same analysis, batch process all the images and then conduct our data analysis. So, you can see how this would very quickly become tedious…
Now, there are companies that make automated imaging and phenotypic analysis software, particularly for worms. In fact those tools are particularly sophisticated if you want to determine fine details of worm locomotion, angle of bends, pumping speeds, swimming speeds etc. However, while they would give us a lot of detail, they would not help us achieve scale. Our goal is to rapidly screen hundreds of thousands of compounds and find ones that rescue the deficit most robustly.
So, we decided to grow our worms in liquid culture as opposed to on solid agar plates. This is by no means new. Worms have been cultured in liquid media for a while now. Liquid cultures offer a number of advantages, particularly for scaling: you can use a multi-well plate, like 384-well or 96 well. You can reduce the volume of worms in liquid to as little as 50μL and still be able to study a population of worms. Further, we can use nanoliter quantities of drug to achieve a reasonable final screening concentration. This allows us to conserve our drug library consumption as well.
The solution (quite literally)
So, taking advantage of the sensitivity of NPC worms to cholesterol deprivation, we came up with a cholesterol-deficient liquid culture system to expose worms to drug-like libraries. By using a 384-well plate format, we can screen as many as 6,400 drugs a week. Each plate has appropriate internal positive and negative controls. Here is how we work through our assay.
Work flow:
Step 1: Each plate first goes through the Echo550 acoustic dispenser by Labcyte, where it gets the predetermined quantity of drug from source plates, along with positive and negative controls.
Step 2: Plates go through the MultiFlo by Bio-Tek, where they get a fixed volume of media for the worms to grow in.
Step 3: Drumroll 1 for the Biosorter by Union Biometrica, a laser-based fluidic system capable of dispensing single organisms with a high degree of accuracy. Using the Biosorter, we are able to obtain equal numbers of worms per well of a 384-well plate. I dispense age-matched larval stage 1 worms into the 384-well plates. After dispensing, each plate is sealed and placed on a plate shaker for 6 days. Plates are vortexed intermittently as well to provide proper mixing. Five days after this, I must look for changes in growth of the worms.
Step 4: Imaging
Because the endpoint of this assay is image based, we looked high and low for a high throughput imager that would suit our purpose. The process of arriving at an imaging system is a story in itself, but we’ll save that for another blog post.
After much searching, we zeroed in on (…Drumroll 2 for…) the IXM by Molecular Devices. This imager is capable of supporting multiple objectives, multiple plate types and transmitted light as well as multiple wavelengths. Thus, we will be able to use this imager for our current and future needs. Another important feature that Molecular Devices provided was the promise of customized software to analyze our worm images.
Here is an example of an array of images taken with the IXM:
It takes roughly 5 minutes to image an entire 384 well plate at the lowest objective at a single wavelength. Luckily, I needed no more than that. The next phase was to work out an image analysis algorithm that would work for our purposes.
Step 5: Image analysis
We worked with a fantastic application scientist who developed many versions of an image processing algorithm for us. Here is the final version of an analyzed image. On the left is an image of a positive control, i.e. NPC mutant worms whose growth has been rescued and on the left is an image of a negative control, i.e. NPC mutant worms, in the absence of cholesterol, showing limited growth.
Assuming a fixed area for each well image, one can calculate the area occupied by the population of worms. By using variable strains and variable numbers of worms, we determined that the area occupied by worms correlated linearly with the growth of worms. Further, I mined the images visually to confirm that the numbers correlated with visual image, that the algorithm was eliminating non-worm debris and other artifacts that sometimes come along for the ride in our assays.
Side note: Dust is my mortal enemy. Because my endpoint is dependent on image analysis, I cannot have anything that is non-worm in my 384-well plates. This includes any kind of dust, any string, any particulate matter basically. So, I resorted to the following precautions:
- Filtering all my solutions prior to autoclaving them,
- Using tubing and consumables that are bagged, boxed and guarded like gold… I keep them locked in a drawer.
- Not opening any bottle more than twice to prevent incoming dust.
- Using a static plate seal at the bottom of the plates to prevent any scratches, any particulates and
- Blasting the bottom of the plates with compressed air prior to imaging.
In addition to our automated image analysis and assay precautions, we have built a number of checks in place to confirm a) that each image metric is true to the image itself and not a result of artifacts, b) that the growth profile of worms in any given assay consistent both within and across assays.
Et Voila!
We were now ready to perform our screen. Starting this January, we have screened about 20,000 compounds in worms. We held our breaths while the results of the first 10K compounds rolled in. We were elated to find that we had a hit rate of 1.5% and a ‘strong positive’ hit rate of 0.1%. Further, this rate remained consistent for the next 10K compounds that we screened.
We are now in the process of re-testing and validating these hits while we continue to power through the rest of our library. Stay tuned for our progress!
In the meantime, if you have questions or imaging or analysis tips for us, please drop us a line!
