MicroRNAs (miRNAs) were discovered as a class of small regulatory molecules ten years ago [1–3]. These ~21 nucleotide (nt), small RNAs recognize partially complementary sequences on target mRNAs [4, 5]. Following the initial discovery of miRNAs, substantial effort has gone into characterization of the canonical miRNA pathway [6, 7] and into miRNA discovery; by identifying miRNAs in more species and by adding to the list of known miRNAs . Although cDNA cloning and northern blotting techniques can be used to detect the most abundant miRNAs, the advent of massively parallel sequencing technologies has propelled the miRNA field, allowing for both discovery and quantification of all miRNAs in a given sample [9, 10].
With the bulk of the miRNAs revealed in commonly studied species, the next challenge lies in elucidating the biological processes in which miRNAs play a role. Current bioinformatic approaches rely on the identification of partially complementary sequences in mRNAs to predict miRNA targeting. Yet these approaches still come with one limitation; the exact parameters governing targeting remain unknown. Several prediction algorithms have been developed to overcome this difficulty by ascribing different weights to key parameters, such as binding energy between target and miRNA, conservation of the target site, quality of the "seed pairing", et cetera . Still no single algorithm emerges as the best performer , and most algorithms predict thousands of targets for each miRNA . Combining different target prediction algorithms generates shorter list of targets by creating more stringent cut-offs. This can provide some enrichment in true positives, but at the cost of more false negatives . In addition to bioinformatics prediction, several approaches to genome-wide experimental miRNA target identification have been developed. These experiments utilize Argonaute pull-down assays (HITS-CLIP and PAR-CLIP) [15, 16], changes in mRNA levels , and protein expression after introduction or ablation of a specific miRNA [18–20]. These studies support that miRNAs indeed function by targeting hundreds of genes. Still, it is a daunting task to derive a function for a miRNA from these long lists of potential target genes. Despite progress in systematic approaches to find sets of related gene that are enriched within these long target lists , we are still far from satisfactory in silico prediction of miRNA function.
Alternatively, differential expression of a miRNA is commonly used to infer its function [22, 23]. Identification of conditions where a specific miRNA is expressed versus an opposing condition where it is not, offers some clues as to the potential action of the miRNA. While this approach has been very successful in leading investigators to uncover miRNA functions, it still requires direct experimentation to prove effects due to miRNA activity beyond providing an only coincidental biomarker.
Another approach to determine the function of a miRNA is by knocking it down [24–26], or knocking it out, of the genome of a model organism [27–29]. Experimental knockdown of miRNAs may confirm or invalidate predicted functions, but it requires prior knowledge where a miRNA is expressed. Even with this knowledge, sufficient knockdown to demonstrate an observable effect is not guaranteed. Complete knockout delivers a clean result, but may not result in an obvious phenotype. Adding to this challenge is the possibility that many miRNAs may elicit only subtle changes or are redundant with other family members entirely. Indeed, only a fraction of all C. elegans miRNA families display pronounced abnormal phenotypes when deleted . Given these challenges, knocking out a miRNA in mice or in a human cell line may often prove a fruitless endeavor.
In order to unravel the functions of specific miRNAs, while overcoming much of the challenges discussed above, we proposed to introduce or overexpress miRNAs in a system of interest. Moreover, we argue that it is more efficient to examine the effect of any miRNA for a predetermined phenotype, rather than blindly investigating one miRNA at a time. Such screens have been performed on different scales. Most are based on transfection of miRNA mimics ; synthetic RNAs that usually have a modified backbone. Although this approach ensures the presence of the mature miRNA in the target cells, a miRNA mimic is not processed via the canonical miRNA biogenesis pathway. This multi-cleavage process starts with the recognition of a hairpin in the primary transcript and ends with a mature miRNA produced from one or both of the arms of the hairpin . Bypassing this physiological processing step has several implications. For instance, several variants of a miRNA from the same primary transcript can arise due to variations in the processing pathway, such as arm switching, non-templated additions of adenines or uracils, and variations in the 3' and 5' cleavage sites [33, 34]. Besides the loss of miRNA variants, the most-widely used miRNA mimics have modified backbones to increase stability within the cell. Consequently, they are not cleared naturally from the cells as endogenous miRNAs. Furthermore, it is possible that miRNA mimic transfection achieves cellular concentrations beyond physiological relevance. On the other hand, loading into the RISC complex, which is essential for biological activity, is not necessarily efficient for mimic transfection, as this process is known to be coupled to hairpin processing [35, 36]. To express transcripts for endogenous processing into mature miRNAs, plasmid  or viral vectors [38, 39] can be employed. While the retroviral library described by Agami and colleagues is a valuable tool for miRNA functional screens in pooled format, the viral supernatants are not available as individual isolates amenable to arrayed high-throughput screens .
Here, we describe the construction and application of a lentiviral human miRNA expression library. This library contains 660 annotated human miRNAs and 422 candidate miRNAs [10, 40]. All are expressed from their genomic backbone, ensuring physiological processing of the miRNAs. The library is organized for high-throughput screening to provide a resource for the systematic elucidation of miRNA function. In addition, all lentiviral miRNA expression constructs can be applied individually to evaluate primary results. Lastly, we demonstrate the utility of this library in various types of screens to present the miRNA expression library as a versatile tool to study miRNA function.