Identification of a novel NAMPT inhibitor by CRISPR/Cas9 chemogenomic profiling in mammalian cells

Chemogenomic profiling is a powerful and unbiased approach to elucidate pharmacological targets and the mechanism of bioactive compounds. Until recently, genome-wide, high-resolution experiments of this nature have been limited to fungal systems due to lack of mammalian genome-wide deletion collections. With the example of a novel nicotinamide phosphoribosyltransferase (NAMPT) inhibitor, we demonstrate that the CRISPR/Cas9 system enables the generation of transient homo- and heterozygous deletion libraries and allows for the identification of efficacy targets and pathways mediating hypersensitivity and resistance relevant to the compound mechanism of action.

37-YO30 and other derivatives with bulkier acyl groups (e.g. R = O-tButyl) demonstrated a general tolerance for large substituents at this position and therefore validated compound design for immobilization onto affinity beads at this position.

Determination of LB-60-OF61 potency
To correctly dose compounds for the chemogenomic profiling experiment, cytotoxic effects of LB-60-OF61, ZA-87-IW08 and ND-37-YO30 on HCT116 cells (#CCL-247, ATCC) were assessed in 384 well plates by seeding 1500 cells/well and adding serially diluted compound in a dose range from 0.37 nM to 20 µM. The compound solvent dimethyl sulfoxide (DMSO) was normalized to 0.2 %. 72 hours after compound addition cell viability was measured using the CellTiter-Glo assay according to the manufacturer's instructions (#G7573, Promega). Data was analyzed and IC 50 s (compound concentration at which 50% inhibition was observed) calculated using the logistic regression curve fit analysis of the Tibco Spotfire package (version 6.5.3, TIBCO Software Inc.). As the LB-60-OF61 inhibition curve was very steep it was difficult to pinpoint the IC 30 and IC 50 concentrations. We thus repeated testing in 6 well plates, seeding 100'000 HCT116 cells/well testing a narrow dose range (0, 10, 20, 30, 40 and 100 nM) as directed by the 384 well experiment. 72 hours after compound addition viability was assessed using a Vi-cell XR Cell Viability Analyzer (Beckman Coulter) and curves plotted as described above. The IC 30 and IC 50 concentrations obtained from the 6 well experiments were then used for the chemogenomic profiling experiment.

Generation and characterization of the HCT116-Cas9 clone
The Cas9 gene encoding the S. pyogenes CRISPR associated protein 9 RNA-guided DNA endonuclease

sgRNA library design and construction
The genome-wide sgRNA library targeting 18,360 protein-coding genes was constructed using chipbased oligonucleotide synthesis to generate spacer-tracrRNA-encoding fragments that were PCRamplified and cloned as a pool into the BpiI site of the pRSI16 lentiviral plasmid (Cellecta). The modified tracrRNA scaffold published by Chen et al. 5 was used. Olfactory receptors were omitted from the library. The sgRNA designs were based on published sequences 6 and five sgRNAs were selected per gene targeting the most proximal 5' exons. 277 genes did not have published sgRNA sequence information and new sgRNAs were designed for these targets that contained an NGG PAM motif, filtering for GC content greater than 40% and less than 80%, eliminating homopolymer stretches greater then 4, and removing any guides with off-target locations having fewer than 4 mismatches across the genome. Sequencing of the plasmid pool showed robust normalization with >90% clones present at a representation of +/-5-fold from the median counts in the pool.

Viral Packaging
sgRNA libraries were packaged into lentiviral particles using HEK293T cells as described previously 7 .

Chemogenomic profiling
For the profiling experiment one positive HCT116-Cas9 clone was expanded to 2.0 e10 8 cells and transduced with the lenti-viral sgRNA library (described in detail above) with a coverage of 5 sgRNAs/gene and a multiplicity of infection of 0.5. As described above, the library was designed as two sublibraries each covering 52'000 sgRNAs. Complexity of each subpool was kept always above at least 1000 cells/sgRNA. Each subpool was cultured in 1 CellSTACK flask (#3313, Corning) seeded with 6.7 e10 7 cells on day -1. On day 0 each CellSTACK was infected with one of the subpools. 4 Days after sgRNA infection and continuous selection on puromycin (2 µg/ml), transduction efficiency was assessed by flow cytometry and the experiment only continued if >95% cells were RFP positive.
Cultures of each subpools were then split into 6 CellSTACK flasks at a cell density of 3.5 e10 7 cells.
One day later LB-60-OF61 was added at an IC 30 or IC 50 to both subpool cultures, the control cultures
The R software package DESeq2 9 was used to evaluate differential sgRNA representation between the compound treated and the untreated samples. A robust z-score for each sgRNA was calculated using the median and mean-absolute deviation across the log2 fold changes of the library combined results.
To summarize the results at the gene level we applied a methodology derived from siRNA screening analysis named redundant siRNA activity (RSA) analysis 10 . It models the probability of a gene 'hit' based on the collective activities of multiple siRNAs/sgRNAs per gene. All sgRNAs in our pool were initially ranked according to their individual signals. Then, the rank distribution of all 5 sgRNAs targeting the same gene was examined and a P-value was assigned. Thus, P-value indicates the statistical significance of all 5 sgRNAs targeting a single gene being unusually distributed toward the top (RSA up) or bottom (RSA down) ranking slots. To visualize the gene significance and result strength, we plot the RSA up value against the Q3 z-score for each gene for the investigation of gene deletion that promotes resistance and the RSA down value against the Q1 z-score for each gene for investigating genes that upon deletion increase sensitivity to the compound treatment. The Supplemental Table S1 contains the annotated data for the all conducted experiments.

Single-cell analysis of the NAMPT locus
Time-dependent editing of the NAMPT locus by the CRISPR/Cas9 system was analyzed using a single sgRNA with the following sequence: 5'-GGCCAGGAGGATGTTGAACT-TRACR-3'. Of the five sgRNAs targeting NAMPT in the genome-wide pool this sgRNA resulted in the biggest Log2 fold change in the genome-wide profiling experiments described above. This sgRNA was transduced The resulting FASTQ files were demultiplexed using the barcoded primers and the demultiplexed FASTQ files were aligned against the human reference genome (GRCh38) using BWA version 0.7.4 11 with the parameter setting: -t 1 -A 4 -B 16 -O 24 -E 1 -T 120 -L 20 to allow for alignments with long indels. The alignments were then processed by a custom script to extract all putative variants.
Variants that were inconsistent between the two reads of a read pair were excluded. The mutations were then filtered based on the following criteria: The position of the variation had to be within ± 50bp of the sgRNA alignment, the coverage had to be at least 100, the number of reads containing the variation at least 10, and the percentage of reads containing the variation at least 5% of the reads covering the position. The resulting variants were then aggregated into haplotypes based on the reads in which they occurred together. Reads without a variation were assigned to the wild-type (WT) haplotype. Based on the inspection of the data for the empty wells of the C1 IFC plate, we excluded haplotypes with coverage of less than 2048 for all IFC plates except for the first replicate of day 18 for which we reduced the coverage threshold to 512 due to the low sequencing depth of this experiment. Furthermore, we also excluded haplotypes which contained less than 15% of the reads covering the NAMPT locus. This resulted in one to four haplotypes per cell. Each haplotype was then manually classified by its functional consequence on the gene product as WT, functional, or nonfunctional. Finally, the functional classifications of the haplotypes of a cell were manually combined into a functional classification for the cell. http://tools.proteomecenter.org/software.php) using a false positive threshold of <1% for protein identifications and removing obvious contaminants. For each peptide sequence and modification state, reporter ion signal intensities from all spectral matches were summed for each reporter ion type and corrected according to the isotope correction factors given by the manufacturer. Only peptides unique to a given protein within the total dataset of identified proteins were used for relative protein quantification. Peptide fold changes were calculated (treatment over DMSO control) and subsequently renormalized within each replicate analysis using the median fold change of all quantified peptides to compensate for differences in total protein yield for each affinity purification. Protein fold changes were calculated as median peptide fold change and p-values were calculated using a one-way T-test (arbitrarily set to 1 for non-significant single peptide quantitations) and adjusted using the Benjamini-Hochberg False Discovery Rate (FDR). Data were visualized for further analysis using TIBCO Spotfire v3.2.1. Figure 2d depicts the two experiments plotted as Log10 fold change over DMSO control for 2327 human proteins (distinct protein accession numbers) quantified based on 2 or more peptides with reporter ion intensities in all iTRAQ channels in both replicates. Solid lines denote no competition (Log10 fold change 0), dashed lines denote 50% competition (Log10 fold change -0.3).
Full dose response quantitative proteomics data for all proteins quantified in the two experiments are given in Supplemental Table S3.
Supplemental Figure S1. Gel-analysis of lysates from chemoproteomics experiments. A) ZA-87-IW08 and ND-37-YO30 used in the chemoproteomics experiments; for ZA-87-IW08 the actual affinity resin after immobilization on NHS-activated sepharose is depicted. B) Gel separation of 10% of each individual pulldown material identifies a strongly enriched protein in the 55kDa range (arrow) that shows visible competition with free ND-37-YO30. The molecular weight of the band corresponds to that of NAMPT (55521 Da). Taken together with the fact that NAMPT is among the most abundant proteins in the pulldown material based on semi-quantitative spectrum count (see Table S3), these data indicate that he ZA-87-IW08-based resin is able to strongly enrich functional NAMPT protein. C) Densiometric analysis of the bands from panel B. Decreasing intentsties of the NAMPT bands in presence of increasing concentrations of ND-37-YO30 indicates dose responsive competition by ND-37-YO30 across the two replicates.

NAD rescue experiment
Serially diluted compound (1:2 dilution factor) was dispensed into 384 well plates containing 1'500 cells/well prefilled with 40 µl growth medium containing supplemented with or without 10 µM nicotinic acid (#N4126, Sigma-Aldrich). The experiment was set up to obtain triplicate data points.
Plates were incubated for 48 hours followed by cell viability determination using CellTiter-Glo (#G7570, Promega) according to manufacturer instructions.

Identification of resistance conferring mutations in NAMPT
The cDNA of wildtype or H191R mutant NAMPT was cloned into a pLENTI6-derived expression well plates at a density of 8000 cells/well prefilled with 90 µl/well growth medium (RPMI+10%FBS+Pen/Strep). Serially diluted compound (1:2.5 dilution factor) was dispensed and to a subset of plates 5ug/ml doxycycline added. The experiment was set up to obtain triplicate data points. Plates were incubated for 72 hours followed by cell viability determination using CellTiter-Glo (#G7570, Promega) according to manufacturer instructions.

In silico docking approaches
LB-60-OF61 was analyzed for its potential to engage the catalytic pocket of the published NAMPT crystal structure (PDB code : 2GVJ 12 ) by in silico docking approaches the using the Glide program (Schrödinger, LLC, New York, NY 10036, USA). The proposed docking solution and the key interaction residues are presented in Supplemental Figure S2 below. Histidine 191 that was identified to lead to significant resistance when mutated (Figure 2f) lies directly adjacent to the benzyl moiety of the LB-60-OF61 backbone and mutation to arginine would lead to a steric clash just like shown for the published NAMPT inhibitor GNE-618 13 . Figure S2. Illustration of the docking model of compound1 bound in the NAMPT protein. The binding pocket is indicated by mesh surface. The ligand and its key interaction residues are represented in stick model and colored in pink and grey, respectively Supplemental references