Here, I have outlined an approach to perform high-throughput combinatorial RNAi screens using an optimized expression vector harboring pairs of shRNAs. In this vector, two shRNAs, separated by a spacer are expressed from the same promoter. To increase mature small RNA levels of both shRNAs, I tested a range of spacer length ranging from 0 to 800nt. Spacers of 200 to 500nt allowed good processing of both hairpins and I selected the 400nt spacer for to use in the expression vector. In addition, I designed a cloning strategy that allows large libraries comprised of all pair-wise combinations of sets of genes to be easily assembled.
Our initial experiments with miR30 based shRNAs highlight the importance of selecting potent shRNAs and optimizing synthetic miRNA backbones for efficient small RNA produc- tion. Building on large shRNA efficacy datasets, shERWOOD, a machine learning algorithm predicting hairpin potency was developed. Coupled with a canonical miR30 backbone, named ultramiR, it permits any gene to be knocked-down with high efficiency. These tools have wide applications beyond dual shRNA screens, both for genome-wide screens and one-by-one ex- periments. We have thus built a fifth-generation genome-wide library of sanger-sequenced verified shERWOOD shRNAs in the ultramiR scaffold. These libraries harbor on average 5 shRNA for each gene of the human and mouse genome for a total of ∼75 000 human and ∼60 000 mouse shRNAs. The availability of individual sequence-verified shRNAs also allows screens to be performed in an arrayed format, or the generation of custom pools of shRNAs targeting a specific gene-set. This was particularly useful to obtain some of the shRNAs used in this project.
With optimized dual shRNA expression vector in hand, the first gene set we interrogated was comprised of all pair-wise combination of druggable genes over-expressed in four melanoma cell line compared to melanocytes. For each pair of genes, two different pairs of shRNA were included in two libraries to minimize false-positive and false-negative effects. Each library was screened in parallel in the A-375 cell line. From the screen, ∼300 deleterious gene-interactions consistently observed in both libraries were identified. To validate these results, I performed an orthogonal CRISPR/Cas9 screen using pCRoatan, a lentiviral vector expressing pairs of sgRNAs that I and others were developing for another project (chapter 3). Some inconsis- tencies were observed between the CRISPR screen and the RNAi screens, especially for some constructs that consistently depleted with shRNAs but not with sgRNA. This can perhaps be explained by the different molecular mechanisms used to manipulate gene expression. Knock- ing down a gene with shRNAs leads to a widespread reduction of the target mRNA to similar levels across all cells. However, the targeted mRNA is generally not removed entirely and some level of expression can remain. In contrast, CRISPR experiments using fluorescent reporters to track knockout shows that knockout effiency varies cell to cell, perhaps depending on Cas9 level of expression within each cell, although this can be somewhat mitigated by generating clonal cell lines expressing Cas9. Double-strand breaks generated by Cas9 are repaired by the
error-prone non-homologous end-joining pathway which leads to the introduction of indels at target site. These indels can induce both frame-shift as well as in-frame repairs with vary- ing efficiency depending on the sgRNA used as a guide. A pool of cells with a Cas9 induced knock-out is thus a mosaic of cells each harboring a different genomic scar at the targeted locus, which can lead to a variety of phenotypic outcome. This is particularly true when Cas9 is di- rected to two genes simultaneously, which can lead to four different repair events in each cell. Improvements in sgRNA design is likely to further improve such combinatorial CRISPR screen by increasing editing efficacy. Additionally, the potency of shRNAs used in the RNAi screen had been experimentally tested using a sensor assay. Perhaps using similar high-throughput assays to select sgRNAs will improve the consistency of CRISPR screens.
One of the advantages of using CRISPR technology rather than RNAi in combinatorial ex- periments is that guides have a single variable region of 20bp which allows multiple sgRNA to be printed on the same DNA chip. This allows great flexibility in library design and con- structs can be barcoded with known sequences which eliminates the intermediary cloning step required to barcode pairs of shRNA. Although this is technically feasible for two hairpins, se- quencing of many shRNA constructs shows that when shRNAs are cloned from chips, a large number of constructs have errors when Sanger-sequenced. This can be due to the complex structure hairpins which makes DNA synthesis less reliable or the Gibson assembly cloning. The intermediary cloning step is thus necessary to generated dual shRNA libraries as it selects for error-free construct thus limiting potential false-negative effects of having mutations in the hairpin sequences.
Combining the results of both RNAi and CRISPR screens, we focused of pairs of genes that showed synergistic deleterious effect on cell proliferation. Although all selected pairs pass stringent criteria in terms of number of significantly depleted construct per pair, depletion rates, and increased lethality of combination when compared to individual targeting of both genes, further experiments need to be performed to validate and characterize these hits. As a first step, I am performing one-by-one dual knockdown experiment to eliminate hits that were an artifact of the large scale screen that can be noisy given the total number of constructs considered. In addition, some pairs can already be targeted by small molecule inhibitors, which provides a third orthogonal validation strategy. Although four cell lines had initially been
selected, the screens have only been performed on A-375, and lethal pairs should be tested on a wider range of cell lines.
The tools developed in the project will allow for the high-throughput discovery of tar- gets to be used in combinatorial drug therapies. This approach focuses on interfering with entire molecular networks and will provide insights on critical pathway nodes that could be targeted in combination to overcome resistance mechanisms that are observed in single target treatments.