CRISPR-based editing reveals edge-specific effects in biological networks

Unraveling the properties of biological networks is central to understanding normal and disease cellular phenotypes. Networks consist of functional elements (nodes) that form a variety of diverse connections (edges) with each node being a hub for multiple edges. Herein, in contrast to node-centric network perturbation and analysis approaches, we present a high-throughput CRISPR-based methodology for delineating the role of network edges. Ablation of network edges using a library targeting 93 miRNA target sites in 71 genes reveals numerous edges that control, with variable importance, cellular survival under stress. To compare the impact of removing nodes versus edges in a biological network, we dissect a specific p53-microRNA pathway. In summary, we demonstrate that network edges are critical to the function and stability of biological networks. Our results introduce a novel genetic screening opportunity via edge ablation and highlight a new dimension in biological network analysis.


Introduction
Using the established cell lines (LIB-WT and LIB-p53 -/-), we focused on the role of miR-34a in the overall p53-miRNA network (Figure 1a). miR-34a is transcriptionally activated by p53 and induces an antiproliferative phenotype including senescence, cell cycle arrest at the G1 phase, and apoptosis (41,42). In turn, overexpression of miR-34a increases p53 protein level and stability (33). Importantly, our established cell lines (LIB-WT and LIB-p53 -/-) have low-level baseline miR-34a expression (Supplementary Figure 5) while the ectopic miR-34a delivery results, on average, in a 71-fold increase. We adopted a survival assay mediated by ectopic exposure to miR-34a mimics. Specifically, we treated both cell lines (LIB-WT and LIB-p53 -/-) to 25 nM of miR-34a mimic for 6 days. Cells were harvested at day 0 (before miRNA treatment) and at day 6. For each sample, sgRNA constructs were amplified from genomic DNA and subjected to the Illumina NGS Amplicon Sequencing to assess the relative abundance for each of the 93 sgRNA target sequences (Supplementary Table 2). The most enriched or depleted sgRNA targets, defined by fold-changes between Day 6 and Day 0 larger than 10, were identified for both LIB-WT and LIB-p53 -/cells Supplementary Table 3).
Intriguingly, RBX1 (RING-box protein 1), a RING subunit of SCF (Skp1, Cullins, F-box) E3 ubiquitin ligases, was highly enriched in both cell lines. Although not a direct target of miR-34a, overexpression of RBX1 has been demonstrated to increase cancer cell survival (43), and thus could serve as a general response mechanism to cellular stress induced by ectopic miR-34a. Additionally, for a subset of gene targets, we observed differential response to miR-34a treatment between the LIB-WT and LIB-p53 -/cells (Supplementary Table 3). For example, the sgRNA targeting the anti-apoptotic gene Bcl-2 was enriched in the LIB-p53 -/cells after miR-34a treatment while no enrichment was observed in the LIB-WT cells (Supplementary Figure 4).

Node perturbations versus edge edits in biological networks
Our edge editing approach revealed (Figure 2b) several clones that are enriched or depleted after prolonged exposure to ectopic miR-34a. To assess the impact of edge removal (through ablation of miRNA:target interactions) we focused on Bcl-2, a gene that in response to miR-34a has different response between the two cell lines (Figure 2b) and is known to be involved in cell survival (44).
Returning to the HCT116 wild-type (WT) and HCT116 p53 -/cells, we removed the miR-34a target site from the Bcl-2 locus. We prepared a single sgRNA construct designed against the Bcl2 3'UTR and established stable cell lines (BCL2tgt-WT and BCL2tgt-p53 -/-) using the same viral delivery system. Sanger sequencing of PCR products spanning the sgRNA target site showed that edits (indels) occurred immediately upstream of the PAM (Supplementary Figure 6, AGG, highlighted) in both cell lines.
Treating delivery of ectopic miR-34a mimic as perturbation of a network node and removal of the miR- with the miR-34a node, with a p-value of 0.002). In this case, perturbing the miR-34a node results in the same behavior for both p53 wild-type and p53 -/cells.
In the context of edge perturbation, we observe that the response of the two cell lines to ectopic miR-34a is sensitive to the presence of the miR-34a/Bcl-2 edge. Specifically, removing the ability of miR-34a to regulate .0% without the miR-34a/Bcl-2 edge, and 71.6% with the miR-34a/Bcl-2 edge, with a p-value of 0.90). We note that the same conclusions can be drawn when quantifying the early-or lateapoptotic cells (Supplementary Figure 9).
To further explore the response to miR-34a in BCL2tgt-WT cells, we quantified expression of the Bcl-2 mRNA in response to miR-34a treatment using qRT-PCR. As expected, miR-34a suppresses the expression of Bcl-2 mRNA in the WT and p53 -/cells by 55% and 40%, respectively (Supplementary Figure 10). In the BCL2tgt-p53 -/cells, ectopic miR-34a has a minimal effect on Bcl-2 mRNA level (95% compared to the control-treated sample, with a p-value of 0.71). In the BCL2tgt-WT cells, ectopic miR-34a results in a significant down-regulation of Bcl-2 expression (62% compared to the control-treated sample, with a p-value of 0.028), possibly due to additional p53-miR-34a regulatory mechanisms.

Discussion
Biological networks consist of nodes and the interactions between them (i.e. edges). Conventional screening methods remove one node at a time, disrupting all edges connected to that node, and therefore producing a relatively blunt effect. An inhibitor that perturbs or removes a single node yields diverse and systemic changes in the whole network through both direct and indirect connections (45), which may explain the heterogeneity observed with single-molecule associated therapeutics.
Our approach reveals edge-specific effects related to the pro-apoptotic p53 and anti-apoptotic Bcl-2 proteins, focal nodes of apoptotic signaling. Normally, p53-dependent inhibition of Bcl-2 and induction of BAX, PUMA and NOXA overcome the anti-apoptotic threshold set by Bcl-2 family members. Conceivably, the difference in apoptosis observed in the BCL2tgt-WT and BCL2tgt-p53 -/cells under miR-34a treatment (Figure   3b) may be explained by the presence of wild type p53-dependent upregulation of PUMA or NOXA in p53 WT cells and not in p53 -/cells. Additionally, p53 could disrupt the binding of POU4F1 to the promoter of Bcl-2 and thus indirectly down-regulate Bcl-2 expression (Supplementary Figure 10).
Taken together, our results show that the disruption of the edge between miR-34a and Bcl-2 can revert the miR-34a triggered apoptotic effects in a p53-deficient cell model. In WT cells, introducing miR-34a presumably indirectly triggers the suppression of the expression of Bcl-2 via an alternative p53-related pathway, which subsequently leads to increased apoptosis.
In conclusion, our CRISPR-mediated edge screening can be used to dissect critical biological interactions essential to cell survival. More generally, we demonstrate that the subtle effect of our edge removal methodology offers superior resolution and granularity in the analysis of biological networks and can lead to the identification of previously hidden interactions and opportunities for intervention.

Preparation of the CRISPR plasmid library
The CRISPR plasmid library was prepared by following the lentiCRISPRv2 cloning protocol provided by Dr. to prepare the CRISPR plasmid library. To confirm library complexity, the plasmid library was subjected to Sanger sequencing (Genewiz) using primer P1 and analyzed using FinchTV (Geospiza).

Generation of the CRISPR lentiviral screen library
To generate the lentiviral vectors, HEK293T cells were grown to 50-70% confluence and then transfected with 3.3 µg of the CRISPR plasmid library, 3.3 µg of the pMD2-VSVG plasmid, and 3.3 µg of the psPAX2 plasmid using 20 mL of JetPRIME (Polyplus, catalog number: 114-01). 24 h later, the medium was removed and replenished with 5 mL of complete growth medium. In the next 3 days, the growth medium containing lentiviral vectors was harvested, and 5 mL of fresh complete growth medium was replenished. The final pooled 15 mL growth medium was centrifuged at 3,000 rpm for 15 min at 4°C to remove cell debris. The supernatant was filtered through a 0.45 µm filter, dispensed into 1-2 mL aliquots and stored at -80°C. Viral titers were

Sanger amplicon sequencing
To confirm the complexity of the LIB-WT and LIB-p53 -/cell line libraries, total genomic DNA was isolated from LIB-WT and LIB-p53 -/cells using the DNeasy Blood & Tissue Kit (Qiagen, catalog number: 69504). The cDNA fragments harboring the sgRNA target sequences were PCR amplified by using ~100 ng of the genomic DNA and primers P2 and P3. PCR conditions were one cycle of 30 seconds at 98°C, 40 cycles of 10 seconds at 98°C, 30 seconds at 60°C, and 30 seconds at 72°C. The 181 bp product was then subjected to direct Sanger sequencing using primer P2 and analyzed using FinchTV (Geospiza). To determine editing efficiency, total genomic DNA was isolated from BCL2tgt-WT and BCL2tgt-p53 -/cells using the DNeasy Blood & Tissue Kit (Qiagen, catalog number: 69504). cDNA fragments harboring the miR-34a target site within the 3'UTR of BCL2 were PCR amplified by using ~100 ng of genomic DNA and primers P8 and P9. The 191 bp product was then subjected to direct Sanger sequencing using primer P9 and analyzed using FinchTV (Geospiza). Figure 1. p53-miRNA network and CRISPR-based edge screens. (a) Complexity of the p53-miRNA network with nodes comprising the indicated miRNAs and their 71 target genes and edges based on experimentally verified and high-confidence predicted direct interactions, derived using Qiagen Ingenuity Pathway Analysis. (b) CRISPR-based lentiviral libraries were prepared using the lentiCRISPRv2 system. The stably integrated CRISPR sgRNA constructs were recovered by PCR and the sgRNA targets were identified using NGS.  The node-based approach shows that addition of the miR-34a node induces apoptosis in both p53 WT and deficient cells. In contrast, the edge-based approach reveals that introduction of the miR-34a/Bcl-2 edge induces apoptosis only in the p53-deficient cells and not in p53-WT cells.