High content microRNA inhibitor screening with cell arrays miRCURY LNA™ microRNA Inhibitor Libraries
Dr. Xavier Gidrol
Dr. Xavier Gidrol
joined the Institute of Life Science Research and Technologies in Grenoble, France in 2009 where he directs the Biomics Laboratory, which uses microfluidics, micromanufacturing and MEMS (MicroElectroMechanical Systems) to study the impact of genetic and micro-environmental determinants on carcinogenesis, at the scale of a few or even single cells. Prior to this Xavier managed a large Functional Genomics laboratory during 8 years at Genopole d’Evry in Paris.
Xavier – as manager of a large functional genomics laboratory you have implemented a variety of DNA microarray platforms over the years and made them available to your colleagues long before they became readily commercially available. In recent years you have devoted a lot of your efforts to the development of cell microarrays as a tool for performing high throughput reverse transfection of live cells growing on glass slides coupled with automated microscopic phenotypical analysis.
1. Can you tell us what prompted this change in focus and why you find this technology so important today?
Two factors motivated this evolution:
1 – In the mid 2000s DNA microarrays became commercially available and affordable. Therefore it was no longer so relevant for us to develop and produce these microarrays in an academic setting.
2 – With the release of the human genome sequence in 2003 and the development of all these multiple microarray technologies and proteomics the stage was set for a new field of functional genomics.
But the real bottleneck was and still is functional analysis of individual genes. To properly address this issue, there was an urgent need to develop experimental platforms that would enable massive parallel gain and loss-of-function studies in human cells. We were greatly inspired by a David Sabatini Nature paper
(2001) in which large scale parallel DNA transfections were performed on a lawn of cells growing on a glass slide the size of a microarray. It addressed the problems we were facing and I realized that we had all the necessary expertise and equipment in the laboratory to improve that format being specialists of DNA microarrays manufacturing and analysis.
Since then the need to develop tools for large-scale phenotypic screens to identify key factors that dictate cell fate is becoming more obvious to us every day.
2. Despite all the hype a few years ago about cell arrays and reverse transfection high throughput RNAi screens are still typically performed in 96 or 384 well plates using liquid-handling robots. I guess it is fair to say that cell arrays have yet to live up to their promise of accelerating high throughput functional genomics. In your view what are the practical obstacles that hamper the widespread use of this technology?
I guess that the availability and ease of use of robots dedicated to the handling and analysis of 96 or 384 well plates simply makes it more convenient to work with microtiter plates. In contrast, cell microarrays still require dedicated skills and a lot of manual setting and tuning. Nevertheless, to me it still worth it as I see important advantages using cell microarrays. For a start, instead of having 96 different cell cultures and as a consequence well to well variability, we have about 3000 features (islands of transfected cells) in one single culture medium, with an extremely large medium/cell volume ratio, which somehow “buffers” some of the variability.
3. Can you reveal some of the features of the particular cell array system you have been developing?
Our cell microarray platform has evolved considerably since the above mentioned first publication on reverse transfection of cells on glass slides, which concerned plasmid transfection, using gelatin as a polymer and growth of a continuous monolayer of cell over the entire array. We now spot with a dedicated arrayer, mixtures of various proteins from the extracellular matrix (ECM), transfection agents and nucleic acids (siRNA, LNA™ or plasmid) on a glass slide coated with a thin layer highly refractory to cell growth. So the cells will grow as isolated islands onto the spots or micropattern of ECM protein (Figure 1).
We use 75 different combinations of ECM and transfection agents to optimize cell seeding, spreading and transfection. Next, we optimize phenotypic readouts, with proper positive and negative controls. Only then we are ready for high-throughput screening.
4. Give us an idea of the dimensions here – what is the diameter of each cell island, what is the distance between each island and approximately how many cells grow in each island?
The diameter of each micropattern (island or feature) can vary from 80 µm to 700 µm. Of course the smaller the micropattern, the more islands/array (up to 31000 with 80 µm features) but fewer cells per island, which in turn might be problematic for statistical relevance.
We currently on a routine basis use a 400 µm micropattern which is a good compromise between the number of feature per array (about 3000) and the number of cells per feature, about 100 to 200 depending on cell size and cell confluence.
5. Microscopic inspection of ~3000 cell islands on a minute microarray – that sounds like a non-trivial challenge. Tell us a little about your automated microscopic setup?
We have collaborated with a French small medium enterprise (IMSTAR) for the automatic microscope capture and automatic cell island recognition. We use a motorized confocal microscope, which automatically captures a 3000 features array in about 3 hours. Of course for this type of application the microscope had to be hooked up with a computer with considerable RAM memory and storage capacity.
6. Why is it important that the cells grow as islands rather than a monolayer?
This is a huge advantage in connection with the automated microscopic capture –especially with transfections leading to loss-of-function. Microscopes have trouble identifying black holes in a lawn of fluorescent cells. In addition some of the concerns about cell microarrays are the risk of confounding effects of cross contamination during printing of the array, transfected cells unsettling and moving to another part of the array and cross talk between adjacent transfected and untransfected cells. Our “cell island” arrays go a long way to reduce these problems to a tolerable minimal level.
7. How do you culture cells on a microarray – did you develop a specialized culture chamber where humidity, CO 2 concentration and temperature can be regulated?
No, microarrays are dropped in large Petri dishes, covered with a suspension of cells in culture medium and placed in a regular CO 2
8. Can you give us an idea of the cost of producing cell arrays and the minute amounts of individual oligonucleotides consumed in the printing of each array?
It is difficult to say as we never really calculated it. In addition, this format is really flexible - it depends on the number of features/array, whether you want to perform gain or loss-of-function screens and the type and number of phenotypes we are looking at etc. However for instance, siRNA microarrays would require 10 fold less siRNA and 1600 fold less transfection agent than 96 well plates.
But cost aside, one other key advantages of this microtechnology format is the dramatically reduced requirement of cells (200 cells in an island compared to ~10 4
cells in a 384 well). This enables screening with primary cells directly from patients with no need for prior lengthy cell expansions reducing the invariable phenotypical changes that takes place in vitro to a minimum. Actually in our last study, we performed screens with primary cells from both healthy subject and patients.
9. How about transfection efficiency? Normally primary cells are difficult to transfect. Does the fact that you can produce your arrays with 75 different combinations of ECM and type of transfection reagent and varying amount of siRNA/microRNA inhibitor actually make the fine tuning of transfection conditions more efficient?
It certainly does, while remaining affordable in that format. Furthermore, I guess that the reduced entropy and the continuous and slow diffusion of both nucleic acid and transfection agent increase transfection efficiency. I have no hard evidence, but this is my “gut-feeling”.
10. For how long can these arrays be stored and still give reproducible results?
We have stored “printed array” as long as one year in vacuum desiccators, then seeded cells onto them and observed efficient transfection of both plasmid and siRNA.
11. With this platform you have recently successfully conducted large screenings on various prostate cancer cell lines with a siRNA library and the Exiqon miRCURY LNA™ miRNA Inhibitor Library. Why prostate cancer? And which phenotypic markers did you screen for?
We are specializing in epithelial cancer. We wanted to use cell lines corresponding to different stage of the pathology as well as primary cell from patients. For us prostate cancer was a convenient model, since we had the opportunity to obtain primary cells from patients, thanks to an efficient collaboration with some prominent urologists in Paris.
We performed two high-throughput (1400 siRNA targeting 700 human kinases and 870 LNA™ miRNA inhibitors) and high-content screens, as we analyzed 4 phenotypes: Cells were fixed and stained with DAPI (nuclei) and for EdU (marker of proliferation), Keratin 18 and Keratin 19 as markers of differentiation (Figure 2).
12. The miRNA inhibitor screen involved 1 cell line, 2 different primary cells, 5 replicate slides with ~3000 features per slide (including various types of controls and each LNA was spotted in 3 replicas per array yielding a total of 15 replicates per LNA™ microRNA inhibitor). Each feature on the array contained about 100-200 cells which were automatically captured, each individual cell in the feature was counted and scored for four different phenotypes and other morphology parameters generating an astounding 3x10 8 data points (individual cells and their associated phenotype). Tell us about the automated image analysis behind the generation of this enormous data set and the subsequent statistical analysis required to make sense of it all.
Yes, all together the kinase and microRNA inhibitor screens has generated close to 1 Terabyte (10 12
byte) of data. In connection with the microRNA screen alone , each cell line generated tables containing around 2 million rows (one for each individual cell on the 5 replicate microarrays) and 15 columns (one for each phenotypic readout of that given cell). It was just impossible to analyze this amount of data manually so we got help from two French academic groups at the “Ecole des Mines” in Paris: the Mathematical Morphology Centre for the automated image analysis (individual cell segmentation and quantification in each channel) and the Bioinformatics Center for the data analysis (extracting statistically significant hits out of noise).
13. How long time did it take to perform the actual screenings of the miRNA inhibitor library and how many people were involved?
Setting up and optimizing cell microarrays for a given cell type and a given phenotype takes time, but once everything is tuned it is relatively fast. Actually, as we had set up everything for the siRNA screen already, a single post-doc managed to screen the entire Exiqon LNA™ miRNA inhibitor library within 2 months. I’m pretty sure that this could be reduced to a couple of weeks with 2 persons working full time.
14. How about the image analysis and the biostatistics – how much time and resources went into that?
This is definitely the rate limiting step. For the image analysis special cell recognition algorithms had to be developed for each of the 7 cell types included in the screen. Then each of the four phenotypic readouts was analyzed at single cell level (Figure 4). Finally, the resulting massive amounts of data had to be analyzed. All this required one year, with 4 persons working full time on the image analysis and 2 more engineers on the data analysis. However if we had to repeat the exercise with the same cell type and phenotypes it would take only a couple of weeks for one person to analyze the data.
15. Without revealing too many secrets can you give us an idea of the type of results you obtained as reward for this massive undertaking?
Depending on cell lines, phenotypes and statistical threshold we obtained from 0 to 50 hits. As we have used up to 15 replicates for each siRNA or LNA™ miRNA inhibitor, hits usually came out with p-values around 10 -7
- 10 -8
and sometime as low as 1X10 -22
. Interestingly with the kinome-targeted siRNA collection, we observed an expected negative correlation between proliferation and differentiation phenotypes. Some siRNA increased proliferation while reducing the level of differentiation and vice-versa.
We did not observe this with the library of miRNA inhibitors, but this is perhaps not really surprising given that each miRNA probably regulates many genes and that inhibition of miRNAs therefore in general is likely to have pleiotropic effects. However thanks to the LNA™ miRNA inhibitor screen, we have discovered several new miRNAs playing a major role in the regulation of the proliferation/differentiation balance in prostate cells.
16. Cancer cell lines are in general considered to be rather poor models of true cancers. Did you observe any interesting correlations and differences between the cell lines and primary cells?
Yes the results of these screens emphasize the fact that primary cells are very different from cell lines. Since we were able to work with primary cells extracted from healthy individuals and patients our results should be of direct relevance to translational medical research.
17. Obviously these screens have laid down the foundation of years of downstream work. What is your strategy going forward?
Not an easy question and the strategy may vary for siRNA and miRNA inhibitors. The short term strategy is to validate by qRT-PCR and Western blot analysis that the observed phenotype is indeed due to the knock-down (KD) of the corresponding mRNA and protein or to the inhibition of the corresponding miRNA and subsequent up-regulation of putative protein target(s). At the same time and whenever possible we would like to reproduce the KD-triggered phenotype with a different type of specific inhibitor (small molecule) in order to sort out potentially disturbing “off-target” effects. This is feasible with kinase or enzyme-targeting siRNAs, but not possible with miRNA inhibitors.
Longer term, we are currently trying to set-up a systematic secondary in vivo screen to validate derived gene functions using a Drosophila transgenic RNAi collection.
18. How do you envision the future of cell array technology – is the time ripe for more widespread applications and what will the next generation cell arrays look like?
I envision a great future for cell array technology and I recently wrote a review about it (2D and 3D cell microarrays, Curr Opin Pharmacol. 2009 5:664-8
). To date we and others have done large scale functional screens using either 96 well plates or cell microarrays with 2 dimensional cultures. Limited to 2D, there are questions we cannot even dream to ask, as for instance what are the genes or the miRNA playing a role in epithelial stratification? In lumen formation? In acini devlopment? Etc.
Using microtechnology we are currently developing 3D cell microarrays (Figure 5, Biomaterials. 2010 12:3156-3165) to perform high-throughput functional analysis screens in 3D-cultures. We are also developing single cell screening formats in order to increase reproducibility and to dramatically reduce the difficulties associated with image analysis. It would also be nice to perform RNAi or miRNA inhibitor high-throughput screening with suspension cells. All of this can hardly be achieved in 96 or 384 well-plates, right?