Journal: bioRxiv

Article Title: Cooperative FOXA1–HNF4A binding emerges from motif spacing and nucleosome architecture

doi: 10.64898/2026.05.27.728252

Figure Lengend Snippet: (a) Top: schematic of the K562 doxycycline-inducible system from Hansen et al. (2022a), in which FOXA1 and HNF4A were induced individually or together in cells lacking endogenous expression of either factor. Bottom: representative CUT&Tag tracks at one peak from each category, showing FOXA1 antibody signal (blue) and HNF4A antibody signal (orange) across the three induction conditions. Co-bound sites (peaks present in both dual-induction antibody tracks; 50% reciprocal overlap on narrowPeak intervals) were classified by their dependence on single-TF expression. FOXA1-enabled (FE, n = 1,510): bound by FOXA1 in the FOXA1-only condition. HNF4A-enabled (HE, n = 2,727): bound by HNF4A in the HNF4A-only condition. Cooperative (CB, n = 1,824): bound by neither factor in either single-TF condition. Redundant (n = 1,875): bound by both factors in their respective single-TF conditions. (b) Per-peak baseline (uninduced) ATAC-seq signal by category. ATAC-seq from GSE182188, same K562 doxycycline-inducible system. (c) Change in per-peak ATAC-seq signal upon dual induction (ΔATAC = induced − uninduced). Dashed line: no change. In b and c, box plots show median (centre line), interquartile range (box), and 1.5×IQR whiskers; violins show the underlying data distribution. Brackets show two-sided Mann–Whitney U tests comparing Cooperative against each other category (****p < 0.0001). (d) Log₂ fold-enrichment of each site category over genome-wide background across seven summary chromatin states consolidated from the Broad 15-state ChromHMM K562 segmentation (wgEncodeBroadHmm). Fold enrichment = (fraction of category overlapping state) / (genomic fraction of state). Cell values are fold enrichments; colour, log₂(fold enrichment). (e) Mean MNase-seq nucleosome occupancy in a ±1 kb window centred on each peak summit, by category. MNase-seq from Mieczkowski et al. 2016 (GEO GSM2083140) . Lines show category means; shaded bands show ±SEM. Sites with usable bigWig coverage (≥50% non-NaN bins): FE n = 1,420; HE n = 2,560; CB n = 1,781; RD n = 1,822. Signal binned at 10 bp and Gaussian-smoothed (σ = 20 bp).

Article Snippet: Recombinant human full-length FOXA1 (Origene TP306045) and HNF4A (Origene TP317863) were used for binding reactions.

Techniques: Expressing, MANN-WHITNEY, Genome Wide

Journal: bioRxiv

Article Title: Cooperative FOXA1–HNF4A binding emerges from motif spacing and nucleosome architecture

doi: 10.64898/2026.05.27.728252

Figure Lengend Snippet: (a) Dual-head binding CNN architecture. One-hot encoded 1,001 bp sequences (summit ± 500 bp) pass through three convolutional blocks (64/128/128 filters; kernel sizes 19/11/7; each: Conv → BatchNorm → ReLU → MaxPool(4) → Dropout 0.25), global average pooling, and two independent task-specific MLP heads with sigmoid output. Training: 138,489 sequences (peaks from all four categories vs. cis-regulatory negatives from uninduced K562 ATAC-seq); chromosome-based splits (test: chr1, chr8, chr9; validation: chr2, chr3). (b) ROC (left) and precision-recall (right) on the held-out test set. FOXA1 head: AUROC = 0.868, AUPRC = 0.792; HNF4A head: AUROC = 0.878, AUPRC = 0.731. (c) DeepLIFT attribution heatmaps by category, after SVA filtering of the HNF4A-Enabled set (see Supplementary Fig. 4): FOXA1-Enabled (n = 1,507), HNF4A-Enabled (n = 2,105; 622 SVA-overlapping sites removed), Co-Bound (n = 1,775), Redundant (n = 1,865). Left: FOXA1 head importance (blue); right: HNF4A head importance (orange). Each row is one site; rows are sorted by position of peak attribution. Each head’s attribution is strongest at its single-TF-enabled category; both heads contribute at Co-Bound sites. (d) Total CNN head attribution within ±250 bp of the peak summit by category. FOXA1 head (left) is most active at FOXA1-Enabled sites (median 3.83 vs. 2.44 at HNF4A-Enabled); HNF4A head (right) is most active at HNF4A-Enabled sites (median 4.49 vs. 2.50 at FOXA1-Enabled). The per-site cognate-head attribution fraction (cognate-head attribution / total attribution) is higher at HNF4A-Enabled than FOXA1-Enabled sites (63.7% vs. 58.4%; two-sided Mann–Whitney p = 2.3 × 10⁻²⁴). (e) FIMO-based motif counts within ±250 bp of the peak summit (FIMO p < 10⁻³; JASPAR MA0148.1, MA0114.2). FOXA1-Enabled sites carry more FOXA1 motifs (median 3) than HNF4A motifs (median 2); HNF4A-Enabled sites show the reverse (median 4 vs. 2). The cognate-motif fraction is correspondingly higher at HNF4A-Enabled sites (71.4% vs. 57.1%; p = 7.8 × 10⁻⁸⁷). In d and e, box plots show median (centre line), interquartile range (box), and 1.5×IQR whiskers; violins show the underlying data distribution.

Article Snippet: Recombinant human full-length FOXA1 (Origene TP306045) and HNF4A (Origene TP317863) were used for binding reactions.

Techniques: Binding Assay, Biomarker Discovery, MANN-WHITNEY

Journal: bioRxiv

Article Title: Cooperative FOXA1–HNF4A binding emerges from motif spacing and nucleosome architecture

doi: 10.64898/2026.05.27.728252

Figure Lengend Snippet: Spacing: centre-to-centre between lowest-p FOXA1 (MA0148.1) and HNF4A (MA0114.2) motifs per peak, summit ± 500 bp, max 500 bp inter-motif (best-score pairing; FIMO p < 10⁻³). (a) Kernel density of per-peak motif spacing by category; lines mark medians. Cooperative is shortest (146 bp) vs FOXA1-enabled (180), HNF4A-enabled (182; SVA-filtered), Redundant (197); n = peaks with both motifs per category. (b) Per-bin log₂(observed/expected) in 5 bp bins against a 1,000-permutation per-peak null (motif positions shuffled within the 1,001 bp window). Dark red, FDR-enriched (BH q < 0.05); dark blue, depleted; pale, n.s. Cooperative shows 12 enriched bins at 15–60 bp; FOXA1-enabled and Redundant show 0; HNF4A-enabled shows 1 (Suppl. Fig. 6 for unfiltered). (c) Same pipeline at endogenously co-bound sites. Top: K562 Cooperative, replotted from (b). Second: HepG2 (FOXA1–HNF4A ChIP-seq, GSE104247; 9,373 motif pairs), 14 enriched bins. Third: HDMA fetal hepatocyte caCREs (Liu et al. 2026, Nature; 44,165 peaks in clusters LI_1/3/4/6 from 29,926 cells, PCW15–22; 24,327 motif pairs), 10 enriched bins concentrated at 15–60 bp. Bottom: HDMA fetal brain caCREs (BR_0–BR_17; 74,035 peaks, 14,015 motif pairs), 0 enriched bins.

Article Snippet: Recombinant human full-length FOXA1 (Origene TP306045) and HNF4A (Origene TP317863) were used for binding reactions.

Techniques: ChIP-sequencing

Journal: bioRxiv

Article Title: Cooperative FOXA1–HNF4A binding emerges from motif spacing and nucleosome architecture

doi: 10.64898/2026.05.27.728252

Figure Lengend Snippet: (a) Single-site Pioneer-seq library design. A single FOXA1 binding site (blue; TGTTTACTTTG, JASPAR MA0148.1) or a single HNF4A binding site (orange; GAGTCCAAAGTCCAG, JASPAR MA0114.2) was placed at each of 182 centre positions (−85 to +96 bp relative to the dyad) on three reconstituted nucleosomal templates: Widom-601, 5S rDNA, and mouse mammary tumor virus (MMTV)-A. A paired nonspecific control sequence (a partial ETS motif; ACCGGAAGTG, JASPAR MA0098.3) was placed at matched positions on the same templates. Each row of the schematic represents one library member. (b) Relative shift (RS) as a function of binding-site centre position relative to the nucleosome dyad. Top row: FOXA1 (blue) and the paired nonspecific control (grey). Bottom row: HNF4A (orange) and the paired nonspecific control. Points show the mean of n = 3 biological replicates; vertical error bars, SEM. Vertical dashed lines mark the dyad (position 0) and the canonical nucleosome boundaries (±73 bp). RS is defined as −log₂((T / T_NS) / (N / N_NS)), where T and T_NS are read counts of the test and paired nonspecific-control nucleosomes in the unshifted band of the TF-treated lane, and N and N_NS are the corresponding counts in the no-TF (null) lane (Methods). (c) Binding ability per template, defined as the mean excess RS over the nonspecific control, ⟨RS_TF − RS_NS⟩, averaged across all 182 positions. Bars show the mean; error bars, SEM propagated from per-position SEMs. p-values, one-sided paired Wilcoxon signed-rank test (alternative: FOXA1 > HNF4A; 182 paired positions per template); the directional hypothesis was prespecified from cellular observations (Hansen et al., 2022a) of FOXA1’s lower per-motif binding requirement.

Article Snippet: Recombinant human full-length FOXA1 (Origene TP306045) and HNF4A (Origene TP317863) were used for binding reactions.

Techniques: Binding Assay, Virus, Control, Sequencing

Journal: bioRxiv

Article Title: Cooperative FOXA1–HNF4A binding emerges from motif spacing and nucleosome architecture

doi: 10.64898/2026.05.27.728252

Figure Lengend Snippet: (a) Cobinding Pioneer-seq library design. On each of three nucleosomal templates (Widom-601, 5S rDNA, mouse mammary tumor virus (MMTV)-A; light-to-dark grey shading), a FOXA1 site (blue) and an HNF4A site (orange) were placed adjacently with a fixed 5 bp gap between the two sites, at each of 77 outermost-site positions (bp 21–97 from the dyad). Each row in the schematic represents one library member. (b) Pioneer-seq relative shift (RS) as a function of the outermost site’s distance from the dyad. Green: FOXA1–HNF4A composite (TGTTTACTTTG–N₅–GAGTCCAAAGTCCAG; JASPAR MA0148.1 + MA0114.2). Blue: FOXA1 alone (MA0148.1). Orange: HNF4A alone (MA0114.2). Grey: paired nonspecific control (ACCGGAAGTG; JASPAR MA0098.3). Per-position values are the mean of n = 3 biological replicates with SEM error bars. Vertical dashed line marks the canonical nucleosome edge (bp 73). Large green dots mark positions where the cobinding signal exceeds the sum of single-TF signals on the linear scale (2^FH > 2^F + 2^H; paired z-test with delta-method error propagation; Bonferroni-corrected across the 77 positions per template, α = 0.05). Cartoons at right depict the four binding conditions, colour-matched to the trace lines. (c) Genomic-nucleosome library. Each row represents one of n = 179 nucleosomes selected from K562 Cooperative-category peaks containing exactly one FOXA1 motif (blue) and exactly one HNF4A motif (orange) (FIMO p < 10⁻³, JASPAR MA0148.4 for FOXA1 and MA0114.4 for HNF4A; nucleosome dyads inferred by DANPOS from K562 MNase-seq (Mieczkowski et al. 2016) at occupancy score ≥ 0.7; Methods). (d) Cobinding RS on the genomic-nucleosome library versus the FOXA1 motif’s distance from the inferred dyad (left) and the HNF4A motif’s distance from the inferred dyad (right). Points, individual nucleosomes; line, ordinary least squares fit; grey band, 95% CI. In-panel: Pearson r, two-sided p, and n. Banner: Δr = r_FOXA1 − r_HNF4A and two-sided Fisher z-test comparing the two Pearson correlations.

Article Snippet: Recombinant human full-length FOXA1 (Origene TP306045) and HNF4A (Origene TP317863) were used for binding reactions.

Techniques: Virus, Control, Binding Assay