Direct Inhibition of Tumor Hypoxia Response with Synthetic Transcriptional Repressors

Many oncogenic transcription factors (TFs) are considered to be undruggable due to their reliance on large protein-protein and protein-DNA interfaces. TFs like hypoxia-inducible factors (HIFs) and X-box binding protein 1 (XBP1) are induced by hypoxia and other stressors in solid tumors and bind to UPRE/HRE motifs to control oncogenic gene programs. Here, we report a strategy to create synthetic transcriptional repressors (STRs) that mimic the bZIP domain of XBP1 and recognize the UPRE/HRE motif. A lead molecule, STR22, binds UPRE/HRE DNA sequences with high fidelity and competes with both TFs in cells. Under hypoxia, STR22 globally suppresses HIF1a binding to HRE-containing promoters/enhancers, inhibits hypoxia-induced gene expression and blocks pro-tumorigenic phenotypes in TNBC cells. In vivo, intratumoral and systemic STR22 treatment inhibited hypoxia-dependent gene expression, primary tumor growth and metastasis of TNBC tumors. These data validate a novel strategy to target the tumor hypoxia response through coordinated inhibition of TF-DNA binding. Overall design: RNA-seq: For the in vitro HeLa cell experiment, total RNA was extracted from cell culture samples treated as described in the main text using the RNeasy Plus Mini Kit (Qiagen). Three independent biological replicates were performed per experimental condition for a total of 9 RNA samples. For the in vivo 4T1 tumor experiment, five PBS and five STR22-treated tumors were collected from ten BALB/c mice. Fresh tumors were placed inside M Tubes (Miltenyi Biotec, 130-093-236), then homogenized with TRI Reagent (Zymo Research, R2050-1) by a gentleMACS Dissociator (Miltenyi Biotec 130-093-235). Total RNA extraction with genomic DNA removal was performed using the Direct-zol RNA MiniPrep Plus kit (Zymo Research, R2072). RNA sample quality check, library construction, and sequencing were performed by the University of Chicago Genomics Facility following standard protocols. Samples from each experiment were sequenced in two runs on a NovaSeq 6000 sequencer to generate paired-end 100bp reads. For each sample, raw FASTQ files from two flow cells were combined before downstream processing. RNA-seq data were analyzed as previously reported and briefly described below. A local Galaxy 20.05 instance was used for the following steps. Quality and adapter trimming were performed on the raw sequencing reads using Trim Galore! 0.6.3. The reads were mapped to the human genome (UCSC hg19 with GENCODE annotation) or the mouse genome (UCSC mm10 with GENCODE annotation) using RNA STAR 2.7.5b. The resulting mapped reads from each sample were counted by featureCounts 1.6.4 for per-gene read counts. The raw counts were analyzed for differential expression between experimental conditions using DESeq2 1.22.1, which also generated a normalized gene expression matrix. The Morpheus software (Broad Institute) was used to draw gene expression heatmaps using the DESeq2-normalized gene expression data. For each gene, the normalized expression values of all samples were transformed by subtracting the mean and dividing by the standard deviation. The transformed gene expression values were used to generate heatmaps. Significant genes and their expression fold change results were analyzed by Ingenuity Pathway Analysis (IPA) software (Qiagen) to predict changes in Canonical Pathways and Diseases and Biological Functions. ChIP-seq: DNA sample quality check, library construction, and sequencing were performed by the University of Chicago Genomics Facility following standard protocols. Samples were sequenced on a NovaSeq 6000 sequencer to generate paired-end 100bp reads. RNA-seq data were analyzed using a local Galaxy 20.05 instance using the following steps: quality and adapter trimming were performed on the raw sequencing reads using Trim Galore! 0.6.3. IP and input reads for each sample were mapped to the human genome (UCSC hg19 with GENCODE annotation) using BWA-MEM 0.7.17.1. To visualize ChIP-seq results, the mapped reads files were counted and the resulting TDF files graphed using Integrative Genomics Viewer 2.9.4. The mapped reads were converted to a SAM file format using samtools 1.2. Peak calling, motif analysis, and annotations were performed by Homer 4.11.1 using the IP and input SAM files for each sample. Settings for peak calling were included in the submitted peak files.