Candidate cis enhancer regions for DPYD (NM_000110.4) were identified using GeneHancer [20] and Ensembl Regulatory Build [21] data mapped to GRCh37/hg19. We defined the upstream boundary of the target region using enhancer data supported by multiple methods and gene-enhancer links supported by multiple methods as implemented within GeneHancer (i.e., putative enhancer regions with “double-elite” status; Figure 1A). We then mapped annotated transcription factor binding sites in the target region using ChIP-seq data from the Ensembl Regulatory Build as an additional layer of evidence (Figure 1A). In all, three candidate proximal upstream enhancer regions were identified, which will be referred to herein as E9, E16, and E20 based on the relevant location upstream of the DPYD transcription start site (Figure 1A).

CRISPR-interference (CRISPRi) and -activation (CRISPRa) were used to determine which of these regions were capable of regulating DPYD expression. For CRISPRi, we generated cell lines that stably expressed the Krüppel-associated box (KRAB) domain of Kox1 fused to nuclease-deficient Cas9 protein (dCas9-KRAB; [22]). For CRISPRa, we created cell lines that stably expressed dCas9 fused to the histone acetyltransferase E1A binding protein P300 (dCas9-P300; [23]). Six guide RNAs (gRNAs) were used to target dCas9 fusion proteins to each region of interest, and quantitative reverse transcription PCR (qRT-PCR) was used to monitor changes in DPYD expression following transfection.

Targeting of dCas9-KRAB (CRISPRi) to the E9 region significantly decreased DPYD expression relative to control for four of six gRNAs (Figure 1B). Targeting of dCas9-KRAB to E16 resulted in a significant decrease in DPYD expression for one of six gRNAs (Figure 1C), and no significant reductions in DPYD expression were noted following transfection of gRNAs for E20 in dCas9-expressing cells; however, two gRNAs increased DPYD expression (Figure 1D). For reciprocal CRISPRa experiments, targeting dCas9-P300 to E9 significantly increased DPYD expression for four of six gRNAs tested (Figure 1E), whereas gRNAs specific to E16 and E20 did not elicit any significant changes in DPYD expression (Figures 1F and 1G). Consistent with the results in HepG2 cells, CRISPRi and CRISPRa targeted to the E9 region significantly altered DPYD expression in HCT116 cells (Figures S1A and S1D), whereas no significant changes were noted when targeted to E16 or E20 (Figures S1B, S1C, S1E, and S1F).

To confirm that the expected epigenetic changes were induced at the E9 region, we performed chromatin immunoprecipitation (ChIP) coupled with quantitative PCR (ChIP-qPCR) in dCas9-KRAB and -P300 expressing cells transfected with E9 gRNAs. As expected, targeting dCas9-KRAB to E9 resulted in a localized increase in H3K9me3 compared to controls, indicating a shift from active to inactive chromatin at E9 (Figure 2A; the primer nearest E9 is denoted by the gray box). Targeting of dCas-P300 to E9 caused changes consistent with epigenetic activation of the region, including a localized gain of H3K27ac-marked chromatin (Figure 2B). Similar changes in localized histone modifications were noted when dCas9-KRAB and dCas9-P300 HCT116 cells were directed to E9 (Figure 2C–D). Collectively, these results indicate that the E9 region might act as a cis regulatory element modulating DPYD expression.

In an expression quantitative trait loci (eQTL) genome-wide association study (GWAS) conducted in human liver specimens, Etheridge et al. previously identified three independent haplotype blocks that significantly associated with altered DPYD expression [9]. One of the blocks identified in that study spanned the E9 region, prompting us to investigate if genetic variants in E9 could potentially perturb DPYD regulation. Based on the biological role of enhancers, we postulated that a causal functional single nucleotide polymorphism (SNP) could interfere with transcription factor binding. Within the Ensembl Regulatory Build [21], we identified a region of approximately 150 bp showing evidence for binding by multiple transcription factors in HepG2 cells, but not in other cell types tested (Figure S2). Within this region, we identified the variant rs4294451, which is a tag SNP for the previously identified eQTL block [9], prompting us to hypothesize that this variant might be the causal eQTL SNP.

To assess the contribution of rs4294451 alleles to DPYD regulation, we first performed reporter assays by cloning the DPYD promoter and E9 enhancer region containing either the reference or variant allele for rs4294451 (T or A, respectively) into a luciferase reporter vector (Figure 3A). Reporter vectors containing the T-allele yielded significantly higher luciferase activity compared to vectors containing only the DPYD promoter (p=0.00054), whereas those containing the A allele within E9 showed a more modest increase (p=0.070, Figure 3B). Directly comparing E9-containing reporters, vectors containing the T allele showed significantly higher activity than those with the A allele (p=0.046, Figure 3B), suggesting that the variant could impact DPYD expression.

To directly characterize the impact of rs4294451 alleles on DPYD expression at the endogenous locus, we used CRISPR-mediated genome editing to create matched isogenic HCT116 knock-in cell models for each genotype. Consistent with the results from the reporter assay (Figure 3B), cells homozygous for the A allele had significantly lower DPYD expression compared to cells that were homozygous for the T allele (p=0.0011, Figure 3C). Heterozygous rs4294451 A/T cells displayed intermediate expression compared to cells with homozygous A/A and T/T genotypes (p=0.00013 and p=0.028, respectively; Figure 3C). At the chromatin level, cells homozygous for the A allele showed reductions in H3K27ac and accumulation of H3K9me at E9 compared to cells that were homozygous or heterozygous for the T allele (Figure 3D–3E). This finding is consistent with the A allele being associated with a less active epigenetic state at E9. Cells carrying a heterozygous genotype for rs4294451 displayed an intermediate epigenetic state, which is consistent with one epigenetically active and one inactive copy of E9 (Figure 3D–3E). Similar epigenetic differences between rs4294451 genotypes were noted at E9 in liver specimens obtained from human donors (Figure S3A–S3B).

To precisely determine the impact of rs4294451 alleles on DPYD expression in vivo, we measured allele-specific expression using reverse transcriptase digital droplet PCR (RT-ddPCR) using donor liver tissues. Because rs4294451 falls outside of the coding region, we used coding region variants as proxies to measure strand-specific expression. These coding variants consisted of DPYD-c.85T>C (rs1801265) and c.496A>G (rs2297595), both of which are in linkage disequilibrium (LD) with the rs4294451 T allele (c.85C: D′=0.92, R²=0.80; c.496G: D′= 0.67, R²= 0.21). Imperfect LD between rs4294451 and proxies allowed us to use samples that are heterozygous for the coding region SNP, but homozygous for rs429441 genotype, as an additional level of control. In samples measured using the c.85 genotype, the fractional abundance of the C-allele ranged from 0.542 to 0.602 in samples heterozygous for rs4294451 (Figure 3F). Expressed in terms of relative expression, the C-allele was expressed at levels 18– 51% higher than the T-allele (i.e., 0.542/0.458=1.18 and 0.602/0.398=1.51). In contrast, the sample that was homozygous for the rs4294451 T allele showed allelic expression at the c.85 locus that was indistinguishable from the DNA control (Figure 3F), indicating equal expression of both transcripts. In measurements using c.496, a significant allelic-imbalance of 0.565 (CI: 0.546–0.584) in favor of the variant c.496 G-allele compared to the DNA control (CI: 0.488–0.511) was observed (Figure 3G). This again indicated higher expression of the linked rs4294451 T allele. Notably, liver specimens that were homozygous at the rs4294451 locus again showed c.496 allelic expression that was indistinguishable from the DNA control (Figure 3G). These data indicate that rs4294451 significantly affects the expression of DPYD in the liver.

We next investigated the impact of rs4294451 genotype on intra-chromatin interactions between the E9 region and the DPYD promoter using quantitative analysis of chromatin conformation capture (3C-qPCR) in knock-in cells. The schematic in Figure 4A shows the location of primers used for 3C-qPCR relative to the DPYD transcription start site (TSS) and E9 region, as well as the location of HindIII digestion sites. When using primers anchored at the DPYD promoter, the E9 region showed a significantly stronger interaction with the promoter in cells carrying the T allele compared to those homozygous for the A allele (p=8.9 x 10⁻⁴ comparing TT to AA; ANOVA p=3.0 x 10⁻⁴ across all three genotypes; Figure 4C). With primers anchored to the E9 region, cells homozygous for the rs4294451 T allele showed stronger promoter interaction than those carrying the A allele (p=4.9 x 10⁻⁴ comparing TT to AA; ANOVA p=1.2 x 10⁻⁴ across all three genotypes; Figure 4D).

Having demonstrated that the rs4294451 genotype impacts DPYD expression, we next evaluated the impact on 5-FU sensitivity using real-time cellular analysis (RTCA). RTCA provides a measure of the number of viable cells present in a culture. We have previously demonstrated the utility of this technology for measuring differences in drug-sensitivity to 5-FU and that those differences directly correlate with DPD enzyme function within cells [14, 17]. Consistent with lower expression of DPYD, knock-in cells homozygous for the rs4294451 A allele were significantly more sensitive to 5-FU than cells homozygous for the T allele (IC₅₀ concentrations were 9.1 and 95.0 μM 5-FU, respectively; p<0.0001; Figure 4E). Heterozygous T/A cells showed an intermediate IC₅₀ value of 29.9 µM 5-FU (Figure 4E).

Based on ENCODE data, rs4294451 is localized to a region that previously has been shown to be bound by multiple transcription factors (Figure S2). To determine if known transcription factor binding sites could be affected by allele status at rs4294451, we computed binding scores for JUND, MYBL2, HNF4A, CEBPB, FOXA1, and FOXA2 for sequences comprising the E9 region containing the T and A alleles at rs4294451. CEBPB, FOXA1, and FOXA2 showed differential predicted binding scores between the rs4294451 A- and T-containing query sequences (data not shown). For CEBPB, the differential binding site had a higher score than other sites in the queried region that did not overlap with rs4294451, suggesting that the SNP could affect CEBPB binding in the region. While predicted binding in the region varied for FOXA1 and FOXA2 depending on rs4294451 genotype, other binding sites that did not overlap with the variant site showed stronger predicted binding, indicating that the SNP was likely not affecting the critical binding site in the region. Binding scores did not differ by genotype for JUND or HNF4A, and no binding site above the threshold score was detected in the region for MYBL2.

Based on the above, we sought to determine if CEBPB could regulate DPYD expression through the E9 region and if regulation was affected by rs4294451 genotype. We first used reporter assays in conjunction with CEBPB. CEBPB expression significantly increased reporter activity relative to GFP control for luciferase vectors containing the DPYD promoter, indicating that the promoter contains CEBPB recognition sites (p=9.1×10⁻⁴; Figure 5A). In cells overexpressing CEBPB, plasmids containing the promoter and E9 region resulted in higher luciferase activity compared to those containing only the DPYD promoter, regardless of the presence of the A or T allele (p=0.0036 and p=0.0017, respectively), suggesting that rs4294451 genotype does not completely disrupt CEBPB regulation through E9 (Figure 5A). Comparing results for the A and T alleles at rs4294451 in the presence of overexpressed CEBPB indicates that the T allele is likely more responsive to CEBPB (Figure 5A; p=0.036).

To directly examine CEBPB binding to the E9 region and to determine if binding affinity differs between the rs4294451 A and T alleles, we performed ChIP with CEBPB antibodies in isogenic knock-in HCT116 cells for the rs4294451 A/A, A/T, and T/T genotypes. Cells homozygous for the reference T allele showed significantly higher CEBPB occupancy at both E9 and the DPYD promoter than A/T or A/A cells (Figure 5B–5C). These results suggest that rs4294451 genotype determined the binding potential for CEBPB at the E9 enhancer region, which, in turn, affects CEBPB-driven expression of DPYD from the promoter.

To further characterize the role of CEBPB in regulating DPYD, we disrupted CEBPB expression using two independent shRNAs (denoted as “sh1” and “sh2”) in HCT116 knock-in cells carrying rs4294451 A/A, A/T, and T/T genotypes. Knockdown of CEBPB was confirmed at the protein (Figure 6A) and mRNA (Figures 6B–6D) levels. Knockdown of CEBPB significantly reduced DPYD expression in rs4294451 T/A and T/T cells, but not in rs4294451 A/A cells (Figures 6E–6F). CEBPB knockdown also reduced occupancy of the transcription factor at both E9 and the DPYD TSS in rs4294451 A/T and T/T cells, but not A/A cells, and the level of CEBPB occupancy at both regions under CEBPB knockdown conditions is similar to that in control shRNA-treated A/A cells (Figures S4A–S4B).

We next determined the extent to which CEBPB contributed to 5-FU chemoresistance in cells with the T allele of rs4294451 (Figure 4E). In rs4294451 A/A cells, knockdown of CEBPB did not affect 5-FU sensitivity (Figure 6H). In contrast, CEBPB knockdown significantly reduced the IC₅₀ for 5-FU in rs4294451 T/T cells (p<0.0001; Figure 6I).

Data from the Genome Aggregation Database (gnomAD [24]) were used to estimate the frequency of the rs4294451-T allele in global populations (Figure S5A). The highest minor allele frequency (MAF) was noted for African/African American individuals (40% MAF), where the MAF was lowest in East Asian individuals (7%). For comparison, individuals of European (Non-Finnish) ancestry, the population with the highest number of individuals reported in gnomAD, had a MAF of 23%. Within Latino-Admixed American gnomAD subjects, similar differences in MAFs were noted in local ancestry-informed frequency data (Figure S5B).

HEK293T, HCT116, and HepG2 cell lines were obtained from the American Tissue Collection Center (ATCC, Manassas, VA). All cell lines were cultured in Dulbecco’s modified Eagle’s medium (DMEM; Corning, Corning, NY) supplemented with 10% fetal bovine serum (FBS; MilliporeSigma, Billerica, MA), 2 mM L-glutamate (Corning), and 1x penicillin/streptomycin solution (MilliporeSigma). Cells were maintained at 37°C with 5% CO₂. To support cell attachment and expansion, HepG2 cells were grown on plates coated with 5% Matrigel (Corning).

Human liver tissues used for ChIP analyses were processed through Dr. Mary Relling’s laboratory at St. Jude Children’s Research Hospital, part of the Pharmacogenetics of Anticancer Agents Research (PAAR) Group and were provided by the Liver Tissue Cell Distribution System funded by NIH Contract #N01-DK-7-0004/HHSN267200700004C and by the Cooperative Human Tissue Network. The acquisition and use of specimens for this manuscript was conducted with the approval of the University of North Carolina at Chapel Hill IRB (study number 10-2253), which has designated the use of these livers for the current analyses as nonhuman subject research and the need for direct consent for use in this study was waived. Human liver tissues used for allele-specific expression were obtained from the University Clinic of Visceral Surgery and Medicine, Inselspital, Bern, Switzerland. Specimens were from donated material from patients who had undergone liver surgery at Inselspital and signed a written consent form for remnant tissues to be used in research (KEK-BE:2016-02202). Patients with impaired liver function due to cirrhosis or other conditions were excluded from analyses.

Total RNA was extracted from cells using the Direct-zol RNA Kit (Zymo Research, Tustin, CA) following the manufacturer’s protocol. Reverse transcription into cDNA was performed using the Transcriptor Reverse Transcriptase kit (Roche, Indianapolis, IN) and random hexamer primers (Roche) according to the manufacturer’s instructions. Quantitative PCR (qPCR) was performed on a LightCycler 480 System (Roche) using LightCycler 480 SYBR Green I Master Mix reagents (Roche). Primers used for RT-qPCR are listed in Supplementary Table S1. RNA expression was normalized to the reference gene L32, and relative gene expression was calculated using the 2^−ΔΔCT method. For all analyses, three independent experiments were performed; for each experiment, gene expression was assessed in triplicate.

Whole-cell lysates were extracted using the RIPA lysis buffer system (Santa Cruz Biotechnology, Dallas, TX), separated by SDS-PAGE, and transferred to PVDF membrane (MilliporeSigma). Blots were blocked using Odyssey Blocking Buffer (LI-COR Biosciences, Lincoln, NE) and incubated with primary antibody at 4°C overnight. Primary antibodies consisted of anti-CEBPB (PA5-27244; 1:1,000 dilution; Thermo Fisher Scientific, Waltham, MA), anti-alpha Tubulin (ab4074; 1:7,500 dilution; Abcam, Waltham, MA), and anti-cas9 (sc-517386; 1:1,000 dilution; Santa Cruz Biotechnology, Dallas, TX). Membranes were washed and incubated with anti-mouse and anti-rabbit secondary antibodies (#926-32212 and #926-68073; both 1:5,000 dilution; LI-COR Biosciences) for 1 hour at room temperature. Blots were imaged using the Odyssey infrared imaging system (LI-COR Biosciences).

The gRNA expression empty vector lentiGuide-Puro was a gift from Feng Zhang (Addgene_52963). pCMV-FLAG LAP2 was a gift from Joan Massague (Addgene_15738). The oligonucleotides targeting enhancer regions (E9, E16, and E20) were designed using GuideScan [47], hybridized, phosphorylated, and cloned into lentiGuide-Puro via BsmBI sites. The luciferase expression vector pGL4.10 was purchased from Promega (Madison, WI). Lentivirus vectors expressing shRNAs targeting CEBPB were obtained from the University of Minnesota Genomics Center (Sh1: TRCN0000007440 and Sh2: TRCN0000007442).

HepG2 cell lines that overexpress dCas9-KRAB and dCas9-P300 were generated by lentiviral transduction using lenti-EF1a-dCas9-KRAB-Puro and pLV-dCas9-p300-P2A-PuroR, respectively. To generate lentiviral particles, HEK293T cells were co-transfected with lenti-EF1a-dCas9-KRAB-Puro plasmid (a gift from Kristen Brennand; Addgene_99372) or pLV-dCas9-p300-P2A-PuroR plasmid (a gift from Charles Gersbach; Addgene_83889), psPAX2 (a gift from Didier Trono; Addgene_12260), and pMD2.G (a gift from Didier Trono; Addgene_12259) using TransIT-Lenti Transfection Reagent (Mirus Bio, Madison, WI). A 3:1 ratio of transfection reagent to total plasmid was used for all transfections. For all transfections, medium was changed 14 hours after transfection, and viral supernatants were collected 34 hours later. Supernatants were filtered using 0.45 µm PVDF filters (MilliporeSigma) to remove debris/cells and used directly for transductions. For transductions, HCT116 cells or HepG2 cells were seeded at a density of 4×10⁵ cells per well in 6-well plates and incubated with 500 μL virus-containing supernatant, 12.5 μg/mL polybrene (MilliporeSigma), and 1.5 mL fresh DMEM culture medium. Medium was changed after 24 hours. Cells were treated with 1 μg/mL puromycin to initiate selection for transduced cells 48 hours after transduction. Expression of dCas9-KRAB and dCas9-P300 was confirmed by western blotting. Guide RNAs (gRNAs) for each target region were identified and designed using GuideScan [47]. Oligonucleotides corresponding to each gRNA were obtained from IDT (Coralville, IA), hybridized, phosphorylated using T4 polynucleotide kinase (New England Biolabs, Ipswich, MA), and ligated into digested BsmBI-digested (New England Biolabs) lentiGuide-Puro vector (a gift from Feng Zhang; Addgene_52963). Cell lines stably expressing dCas9-KRAB or dCas9-P300 were transfected with plasmids encoding gRNAs or lentiGuide empty vector using TransIT-X2 (Mirus Bio). RNA was extracted 2 days after transfection, and DPYD expression was measured by RT-qPCR.

ChIP assays were performed using the ChIP-IT Express Enzymatic Kit (Active Motif, Carlsbad, CA) following manufacturer’s directions. One million cells were harvested, washed, and crossed linked using 1% formaldehyde (Thermo Fisher Scientific) in serum-free medium for 10 minutes followed by quenching with 125 mM glycine for 5 minutes at room temperature. The chromatin was digested using enzymatic shearing cocktail provided by the kit to an average size of 200–1000 bp. Two percent of the sheared chromatin was retained as input control. Approximately 25 μg sheared chromatin was incubated with 2 μg H3K27ac antibody (ab4729; Abcam), 2 μg H3K9me3 antibody (ab8898; Abcam), 2 μg CEBPB antibody (PA5-27244; Thermo Fisher Scientific), or 2 μg control normal Rabbit IgG antibody (antibody 2729; Cell Signaling Technology, Danvers, MA), in the presence of protein G magnetic beads, ChIP buffer, and protease inhibitor cocktail (all Active Motif) for 4 hours at 4°C. Magnetic beads were washed, chromatin was eluted, cross-linking was reversed, and proteinase K treatment performed using reagents provided in the kit following manufacturer’s directions. DNA was purified using the QIAquick PCR Purification Kit (Qiagen, Germantown, MD). Purified DNA was used for subsequent qPCR reactions using SYBR Green I Master Mix on a LightCycler 480 System (Roche). Enrichment was calculated using the following formula: (a) % ChIP = 2^{(Input Ct − ChIP Ct)} * (dilution factor) (100); (b) % IgG = 2^{(Input Ct − IgG Ct)} * (dilution factor) (100); (c) Fold Enrichment = % ChIP ÷ % IgG. Primers used for ChIP-qPCR are listed in Supplementary Table S1.

The DPYD promoter region, consisting of the 1154 bp of genomic DNA directly upstream of the DPYD TSS, was amplified by PCR, digested with EcoRV and HindIII (New England Biolabs), and cloned into compatible sites on the pGL4.10 vector (Promega). The 1392-bp region comprising the E9 region was PCR amplified from genomic DNA and cloned upstream of the DPYD promoter using KpnI and SacI sites (New England Biolabs). A vector containing the A allele of rs4294451 within E9 was confirmed by Sanger sequencing. The vector containing the rs4294451 T allele was generated by site-directed mutagenesis and confirmed by sequencing. All primers used in vector construction are listed in Supplementary Table S1. For reporter assays, 10⁵ HEK293T cells were seeded into 24-well plates and co-transfected with pGL4.10-based plasmids and pRL-SV40 Renilla luciferase plasmid (Promega). After 48 hours, luciferase activity was measured using the Dual-Glo Luciferase Assay (Promega) following manufacturer’s recommendations on a Synergy HTX Multimode Plate Reader (Agilent Technologies, Santa Clara, CA).

Knock-in cell lines were generated using CRISPR/Cas9 gene editing. Homology-directed repair (HDR) donor templates and target-specific Alt-R crRNA were designed using the Alt-R HDR Design Tool (IDT). Equimolar amounts of crRNA (IDT) and common Alt-R tracrRNA (IDT) were annealed to form the gRNA duplex. RNP complexes were formed by combining gRNA with Alt-R S.p. Cas9 Nuclease V3 (IDT) to a final Cas9:gRNA ratio of 4:4.8. RNA complex and Alt-R HDR Donor Oligos were transfected into HCT116 cells by electroporation (Lonza Nucleofector 96-well Shuttle System; Lonza, Bend, OR) using parameters provided by the manufacturer. Seventy-two hours after transfection, serial dilutions were performed to obtain single-cell clones. Clones were expanded, genomic DNA was isolated, and rs42944551 genotype was determined by rhAmp Genotyping (assay ID: Hs.GT.rs4294451.A.1; IDT) using rhAmp Genotyping Master Mix and universal probe Reporter Mix (both IDT).

Allele-specific expression of DPYD was measured using reverse transcriptase droplet digital PCR (RT-ddPCR) by targeting variants in the coding region of DPYD (c.85T>C and c.496A>G). Linkage disequilibrium (LD) between variants was calculated using LDpair implemented within LDlink [48]. DNA was extracted from donor liver tissues using the QIAamp DNA Mini Kit (Qiagen). For RNA extraction, tissues were lysed in QIAzol (Qiagen), and RNA was extracted using the miRneasy Kit (Qiagen) with on-column DNA digestion using RNase-free DNase (Qiagen). RNA quality was assessed using an Agilent 2100 Bioanalyzer running 2100 Expert Software v.B.02.10 using Agilent RNA 6000 Nano kits.

DNA samples were genotyped for c.85, c.496, and rs4294451 loci using TaqMan SNP Genotyping assays (Applied Biosystems, Waltham, MA). Tissues that were heterozygous for at least one of the coding region SNPs (i.e., c.85T>C and/or c.496A>G) were suitable for allele-specific expression analysis because expression from both alleles could be discriminated using the coding-region markers in mRNA. Allele-specific expression was measured using RT-ddPCR with the One-Step RT-ddPCR Advanced Kit for Probes (Qiagen) on a QX200 ddPCR Droplet Reader (BioRad, Hercules, CA). Fractional abundance was calculated using QuantaSoft software (BioRad). Poisson distributions were determined using Quantasoft and were used to define 95% confidence intervals. To address possible biases associated with differing probe efficacies caused by differential probe binding affinities or specificities, fractional abundances are also reported relative to those measured in DNA.

Quantitative analysis of chromatin conformation capture (3C-qPCR) was performed as described by Hagère et al. [49] with minor modifications. Cells were trypsinized using 0.25% w/v trypsin-EDTA (Thermo Fisher Scientific) and resuspended in 1% FBS (MilliporeSigma) in DMEM (Corning) for counting by flow cytometry using a NovoCyte 3000 RYB system (Agilent Technologies). Cells (10⁷) were pelleted by centrifugation at 300 ξ g at 22°C for 5 minutes. Supernatant was discarded, and cell pellets were resuspended in 500 µL of 10% FBS in DMEM. Single-cell suspensions were obtained by filtration through a 40-µm cell strainer (Corning). Crosslinking was performed by adding 9.5 mL of 1% formaldehyde in 10% FBS in PBS per 10⁷ cells. Reactions were incubated at 22°C while tumbling for 10 minutes. Reactions were transferred to ice and crosslinking was quenched by the addition of ice-cold glycine (MilliporeSigma) to achieve a final concentration of 0.125M. Samples were centrifuged at 300 ξ g at 4°C for 5 minutes. Supernatant was removed and discarded. Crosslinked cell pellets were lysed in 5 mL cold lysis buffer (10 mM Tris-HCl, pH7.5; 10 mM NaCl; 5 mM MgCl; 0.1 mM EGTA; 1x Roche Complete protease inhibitor cocktail). Lysis was allowed to proceed for minutes on ice with intermittent gentle pipetting to obtain homogeneous suspensions of nuclei. Nuclei were pelleted at 500 ξ g at 4°C for 5 minutes and resuspended in 0.5 mL of 1.2x NEBuffer r2.1 (New England Biolabs) containing 3% sodium dodecyl sulfate (SDS, MilliporeSigma). Samples were incubated at 37°C for 1 hour with shaking. Following incubation, Triton X-100 (MilliporeSigma) was added to a final concentration of 2%. Reactions were incubated at 37°C for 1 hour with shaking. To digest DNA, 600 U of HindIII (New England Biolabs) was added, and reactions were incubated for 16 hours at 37°C with shaking. To inactivate digestion, SDS was added to a final concentration of 1.6%, and reactions were incubated at 65°C for 25 minutes with shaking. Excess SDS was sequestered by the addition of Triton X-100 to a final concentration of 1%. Samples were divided into two aliquots, one for ligation and the other for non-ligation control. Ligation was performed on a 7-fold dilution of HindIII-digested chromatin using 100 units of Quick T4 DNA ligase (New England Biolabs) at 16°C for 16 hours, followed by 1 hour at 22°C. Proteinase K (300 µg; New England Biolabs) was added to ligation mixtures and non-ligated controls, and samples were incubated at 65°C for 16 hours to reverse crosslinking. DNA was subsequently purified by adding 300 µg RNase A (Thermo-Fisher) followed by a 30-minute incubation at 37°C and subsequent phenol-chloroform extraction as described [49]. DNA pellets were washed with 70% ethanol and dissolved in 10 mM Tris pH 7.5.

For qPCR, directional primers were designed within each fragment as depicted in Figure 4A. Two anchors, one localized to the fragment containing the DPYD promoter and the other to the E9 region, were selected and paired with primers designed to “walk” across the length of the surrounding region. Reactions were carried out in 10 µL reaction volumes, consisting of 5 pmol each primer and 1 µL of a 1:50 dilution of each 3C sample. Amplification was performed using LightCycler 480 SYBR Green I Master Mix (Roche) on a LightCycler 480 thermocycler (Roche). PCR conditions were 95°C for 10 minutes, and 40 cycles of 95°C for 15 seconds, 65°C for 1 minute, and 72°C for 15 seconds. Enrichment was calculated as 2^{−(cp (ligated DNA) − cp (non-ligated DNA))}. Enrichment with the fragment containing the nearest primer was used as the control for interaction frequency for further normalization between replicate experiments. Specifically, when primer 5 was used as the anchor, data were normalized to the average of values obtained from using primers 5 and 4. When primer 7 was used as the anchor, data were normalized to the average of values from primers 7 and 8. All experiments were performed in triplicate. Primer sequences and positions relative to the DPYD transcription start site (TSS) for qPCR are listed in Supplementary Table S2.

Cell viability was monitored using the xCELLigence MP Real-Time Cell Analysis (RTCA) system (Agilent) as previously described [14]. Background impedance values for RTCA E-View plates (Agilent) were obtained using complete DMEM prior to plating cells. Cells were seeded at a density of 5,000 cells per well and incubated for 20 hours, at which time medium was removed and replaced with medium containing serial dilutions of 5-FU ranging from 1.25 μM to 100 μM. Impedance values (expressed as cell index, CI, units) were recorded every 15 minutes over the course of the experiment to monitor proliferation. Results represent the average of three independent cultures. To account for minute differences in plating and potential cell loss during drug addition, relative CI was calculated as the CI measured 48 hours after 5-FU divided by the CI recorded immediately after treatment. Four parameter logistic non-linear regression analysis was used to determine IC₅₀ concentrations (GraphPad Prism version 9, GraphPad Software, San Diego, CA).

Lentiviral particles for CEBPB knockdown were generated by transfecting shRNA plasmids (TRCN0000007440 and TRCN0000007442) or scramble shRNA control (a gift from David Sabatini; Addgene_1864 [50]), psPAX2 (a gift from Didier Trono; Addgene_12260), and pMD2.G (a gift from Didier Trono; Addgene_12259) into HEK293T cells using TransIT-Lenti (Mirus Bio, Madison, WI). A 3:1 ratio of transfection reagent to total plasmid was used for all transfections. For all transfections, medium was changed 14 hours after transfection, and viral supernatants were collected 34 hours later. Supernatants were filtered using 0.45 µm PVDF filters (Millipore) to remove debris/cells and used directly for transductions. For transductions, target cells were seeded at a density of 4×10⁵ cells per well in 6-well plates and incubated with 500 μL virus-containing supernatant, 12.5 μg/mL polybrene (MilliporeSigma), and 1.5 mL fresh DMEM culture medium.

Potential transcription factor binding sites within the E9 region were determined using TFBSTools [51]. DNA sequence corresponding to the 101-nucleotide region centered on rs4294451 was retrieved from genome build GRCh38.p13 (NC_000001.11). A second DNA string was created to mimic the sequence corresponding to the rs4294451 A allele. Position frequency matrices for each transcription factor were retrieved from JASPAR CORE 2022 [52] and converted to log-scale position weight matrices using the toPWM method implemented in TFBSTools. JASPAR includes binding site information from multiple sources. ENCODE data [53] were available for CEBPB (matrix ID MA04661) and JUND (MA0491.1). In absence of ENCODE data, alternatives including PAZAR [54] (HNF4A, MA0114.2; FOXA1, MA0148.3), REMAP [55] (FOXA2,MA0047.3), and data from an individual publication [56] (MYBL2, MA0777.1) were used. EP300 is a cofactor that does not recognize a specific DNA motif on its own; instead, it interacts with various DNA-binding factors to modify chromatin and facilitate activation of target genes. As such, P300 does not have a DNA-binding motif and was not included in analysis. Nucleotide sequences were scanned using the patterns presented in the position weight matrices to identify putative transcription factor binding sites. Forward and reverse strands were searched, and the 70^th percentile between the minimum and maximum possible value for a matrix was used as the minimum threshold score. Empirical p-values for each score were calculated by an exact method using the TFMPvalue R package. R version 4.2.2 was used for analyses

Significance was defined as p<0.05 unless otherwise noted in the text. Pairwise comparisons were performed using unpaired two-tailed Student’s t tests calculated using GraphPad Prism version 9. One-way ANOVA statistics were calculated using GraphPad Prism. Summary statistics pertaining to allele-specific expression were calculated using Quantasoft software as described in the Allele-specific gene expression section. Transcription factor binding predictions and associated analyses were performed in R version 4.2.2 as described above.