Single-molecule states link transcription factor binding to gene expression

A) Gene expression (MFI) for Citrine for constructs with variable TetO number at 24 and 48 h with dox. B) ChIP-seq signal (reads overlapping the reporter per million mapped reads across the genome) for H3K27ac (left) and Pol II-S5P (right) at the reporter with and without the addition of doxycycline. C) Average methylation on a synthetic amplicon containing a single CTCF site (purple). Error bars represent the 95% CI from three biological replicates. D) Genome-wide relationship between the average ATAC-seq signal (RPM) and average methylation probability from SMF at ATAC-seq peaks. Peaks were grouped into 100 quantile bins (green dots). E) Methylation probabilities at CpGs from the fully methylated (pUC19) and fully unmethylated (Lambda phage) EM-seq controls. F) Percent of pairs of GpCs flanking the same TetO site that have identical methylation status on the same molecule. Error bars represent the 95% CI obtained by averaging all pairs of GpCs flanking all TetOs on all molecules. G) Average rTetR-VP48 occupancy across amplicons for two different lysis and methylation conditions.

A) Cartoon schematic of probabilistic binding model. For each read (single-molecule methylation signal), we enumerate all possible underlying molecular configurations of TFs and nucleosomes and assign each a likelihood of being observed. We then select the state with the maximum likelihood. B) The fraction of molecules with a nucleosome overlapping each position in the enhancer with no TetO sites. C) The fraction of molecules which have enhancers either a) completely covered by nucleosomes, b) only bound by TFs, or c) covered with a mix of the two as a function of the number of TetOs. D) Average TF occupancy across amplicons for both rTetR-VP48 and rTetR as a function of the duration of the methylation reaction. Dotted line represents the conditions used in our assay (7.5 min). E) Average methylation across the reporter molecule with six TetO binding sites under three different fixation and methylation conditions: native SMF with 7.5 min methylation time (green), 15 min 1% formaldehyde fixation and 1 hr methylation (purple), and 15 min 1% formaldehyde fixation and 6 hr methylation (orange). F) Average methylation in the promoter after subclustering the active (nucleosome-free) promoters using k-means. G) Fraction of active promoters that are found in each k-means cluster (above), faceted by the number of TetOs in the enhancer. Error bars represent the 95% CI from four biological replicates.

A) Relationship between average TF occupancy and ATAC-seq signal (RPM) for a selection of single enhancer amplicons from multiple backgrounds. Each dot represents a single enhancer. Error bars represent the 95% CI from four (SMF) or two (ATAC-seq) biological replicates. B) Schematic describing how rTetR-VP48 is installed with PiggyBac transposase and expressed in K562 cells, selected with hygromycin, and assayed for activity with mCherry under a dox-responsive promoter. C) Flow cytometry distributions for mCherry fluorescence with and without dox for two technical replicates. D) Histogram of the predictions of number of TFs bound per molecule for the enhancer with 7 TetO sites from the Simple Competition model (gray) fit to the binding data of rTetR-VP48 and these predictions with Gaussian noise (SNR = 1) added to TF energy (black), effectively varying the concentration of the TF per cell/molecule (see Methods). E) Average rTetR-VP48 occupancy with variable TetO number at 24 and 48 h with dox. F) Probability of each TetO site being occupied on nucleosome-free molecules for the enhancer with six TetO sites (two biological replicates). Dashed line indicates the average occupancy. G) Fraction of molecules with >0 TFs bound for enhancers with increasing TetO number (four biological replicates) fit to either the Simple Competition (r² = 0.44) or Nucleosome Destabilization (r² = 0.99) models. H) Occupancy distributions across all enhancers (left) with matching simulations from the Nucleosome Destabilization model (right). Each row represents a single enhancer and each column represents the number of TFs bound. The pixel intensity at (row i, column j) indicates the fraction of molecules with i TetO sites that have j TFs bound. Rows sum to 1. I) Full molecular state representations for 10,000 measured (left) or simulated (according to the Nucleosome Destabilization, right) molecules with six TetO sites, where each column represents a TetO site and each row represents a molecule. Sites are colored by their occupancy status. J) Best-fit parameters for binding of rTetR-VP48 using the Nucleosome Destabilization model (four biological replicates). K) Three alternate TF cooperativity models (left) and their fits to average occupancy and occupancy distributions (right) for rTetR-VP48.

A) Fraction of molecules with >0 TFs bound for rTetR only, rTetR-VP48 treated with BAF inhibitor, and rTetR-VP48 treated with p300 inhibitor. Fits are either to the Simple Competition (rTetR only) or Nucleosome Destabilization (rTetR-VP48 + inhibitors) models (r² = 0.86, 0.96, and 0.96, respectively). Data are from two biological replicates. B) Occupancy distributions across all enhancers (top) with matching simulations from the Nucleosome Destabilization model (bottom) for rTetR only, rTetR-VP48 treated with BAF inhibitor, and rTetR-VP48 treated with p300 inhibitor. C) Full molecular state representations for 10,000 measured (top) or simulated (bottom) molecules for rTetR only, rTetR-VP48 treated with BAF inhibitor, and rTetR-VP48 treated with p300 inhibitor. Each column represents a TetO site and each row represents a molecule. Sites are colored by their occupancy status. D) Flow cytometry for FLAG-tagged TF expression levels for cells expressing rTetR or rTetR-VP48. Dotted line represents the level of background staining. Data are from two biological replicates. E) Average occupancy (top) and occupancy distribution (bottom) of rTetR (two biological replicates) with a fit from the Nucleosome Destabilization model (r² = 0.83 and 0.96, respectively). F) Average TF occupancy across variable TetO number for rTetR-VP48 (left) and rTetR only (right) with and without treatment with BAF inhibitor. Data are from two biological replicates. G) Best-fit parameters from the Nucleosome Destabilization model after treatment with the BAF and p300 inhibitors (two biological replicates). Gray dashed line represents rTetR-VP48 parameter fits for comparison. H) Average occupancy (left) and occupancy distribution (right of rTetR-VP48 in the presence of the BAF inhibitor BRM014 (two biological replicates) with a fit from the Simple Competition model (r² = 0.78 and 0.51, respectively). I) Fraction of molecules with nucleosomes overlapping the enhancer as a function of the number of TetO sites for rTetR-VP48, rTetR-VP48 treated with BAF inhibitor, and rTetR-VP48 treated with p300 inhibitor. J) Average H3K27ac signal across the synthetic reporter locus with and without p300 inhibition. Data are from two biological replicates, and the signal is normalized to spike-in mouse chromatin.

A) Relationship between rTetR-VP48 (black) and rTetR (green) occupancy and active promoter fraction. B) Relationship between rTetR-VP48 occupancy and active promoter fraction across transcription inhibition conditions with: no inhibitor (black), 1.5 h flavopiridol (cyan) and 6 h (brown) and 25 h (tan) alpha-amanitin. C) TF occupancy (left) and active promoter fraction (right) across transcription inhibition conditions relative to no inhibitor for constructs with 5–8 TetOs. Alpha-amanitin for 25 h (tan) likely reduces rTetR-VP48 concentration due to global reduction of transcription. D) Relationship and linear fit (black line) between active promoter fraction and RNA expression (RT-qPCR) on constructs with 0–8 TetOs (two technical replicates). E) Schematics of alternative models relating TF binding to promoter activation. Cooperative Activation (left) assumes activation is cooperative with the number of TFs bound on average (k_on*avgTFⁿ). Thresholded Activation (right) assumes activation of the promoter occurs when there is at least 1 TF present (k_on*frac_{>=1 TF}). F) Model fits for additional models in E on the relationship between average rTetR-VP48 occupancy and active promoter fraction. Fit parameters for cooperative model: k_on/k_off = 0.13 ± 0.006 TF⁻¹, n = 1.1 ± 0.04. Fit parameter for thresholded model: k_on/k_off = 0.5 ± 0.1. G) Chi-squared per degree of freedom comparing fits in F. H-I) Instantaneous relationship between the number of rTetR-VP48s bound and fraction promoter active on the same molecule across backgrounds (H) or across TetO number (I). J) Example plots to compare to data in Fig. 3h represent the expected relationship if promoter rates are much faster than TF binding rates (left) or vice versa (right). K-L) EMSA of rTetR (concentration noted above each lane) binding to 60 bp target DNA (1 nM) without a TetO site (left) and with a TetO site (right) in the presence and absence of doxycycline (1:50 rTetR to dox concentration). M) EMSA of rTetR (160 nM) binding to 60 bp target DNA (1 nM) with a TetO site across varying dox concentrations (ratio relative to rTetR concentration noted under each lane) (left). K-L) In vitro rTetR:TetO binding from EMSAs varying rTetR concentration in the presence (black) and absence (gray) of dox (K) or varying dox concentration (L) with binding isotherm fits (y = x/(K_D+x) where K_D is the affinity of rTetR to DNA and the affinity of dox to rTetR is assumed to be much smaller. M) Apparent TF concentration from relative in vivo rTetR-VP48 concentration (from binding energy in Nucleosome Destabilization model) across dox concentrations with binding isotherm fit. N) Flow cytometry distributions for a construct with 7xTetO without inhibitor (gray), with p300 inhibition (cyan) or with BAF inhibition (purple).

A) Distributions of the number of ISREs within a 500 base pair window around TSSs of ISGs identified from bulk RNA-seq (pink) and of non-ISGs (gray). B) Relationship between methylation time and measured average wide (left) and narrow (right) footprints across amplicons after 24 h of IFN-β stimulation. Dashed line is the methylation time chosen for all experiments. C) Bulk methylation data for the construct with 4 ISRE sites, with (black) and without (gray) IFN-β. Error bars are standard deviation between two biological replicates. D) Nucleosome occupancy across ISRE number before and after six hours of stimulation (two biological replicates). E) Likelihood of occupancy across six ISRE binding sites in nucleosome-free molecules pre- (black) and post-stimulation (pink) (two biological replicates). The line indicates average occupancy. F) Relationship between ISRE number and active promoter fraction across with six hours of IFN-β (two biological replicates). G) Enrichment of active promoter clusters (classified using rTetR-VP48 data) at six hours relative to zero hours of IFN-β (two biological replicates). H) Relationship between narrow footprints and active promoter fraction with (black) and without (gray) 6 h of IFN-β (two biological replicates). I) Relationship between active promoter fraction and RNA expression (RT-qPCR, two technical replicates). Black line is a linear fit. J) Relationship between ISRE copy number and gene expression (RT-qPCR, two technical replicates). Black line is the coupled Additive Activation model and linear fit from fraction promoters active to gene expression with input of measured average wide footprints. K) Relationship between ISRE number and the fold change in average wide footprints with six hours IFN-β and BAF inhibition (purple) or p300 inhibition (cyan) relative to without inhibitors. L) Relationship between ISRE number and gene expression (as measured by flow cytometry at 24 h of IFN-β) with BAF inhibition (purple), p300 inhibition (cyan), and without drug (black) (two technical replicates).

A) Relationship between ISRE number and average narrow footprints for the average of two biological replicates (left) and the distribution of narrow footprints bound for 4x ISRE sites (right) prior to stimulation fit by the Simple Competition model (black) and Nucleosome Displacement model (green). For both models, the nucleosome energy is fixed to the fit value in rTetR-VP48 experiments. B-C) Goodness of fit of number bound distributions in A by r² across ISRE copy numbers (B) and chi-squared per degrees of freedom (C). D) Fit parameters (averaged across two biological replicates) from Nucleosome Displacement model (gray bars); black dots are rTetR-VP48 parameters. E) Relationship between ISRE number and the average narrow footprints pre-stimulation under conditions of BAF inhibition (purple) or p300 inhibition (cyan). Dotted line is data without inhibitors. F) Relationship in E shown as a fold change compared to no inhibitor across constructs with three to six ISREs.

A) Differential ATAC-seq accessibility of TF motif types after IFN-β stimulation (two biological replicates). B) Differential RNA-seq expression after IFN-β (two biological replicates). C) Average ATAC-seq accessibility relative to ISREs genome-wide, Tn5-bias corrected and RPM normalized for samples with (red) and without (pink) IFN-β (two biological replicates). D) Average ATAC-seq accessibility with (red) and without (pink) IFN-β relative to the TSSs of genes upregulated with stimulation (identified by RNA-seq) that contain at least three ISREs within the 500 bp window. Gray line is average ATAC-seq accessibility for non-ISG promoters that are matched to the pre-stimulation expression level (TPM) of plotted ISG promoters. Shaded error regions are 95% confidence intervals from bootstrapping (two biological replicates). E) Average ATAC-seq accessibility as in D with no inhibitor (gray), BAF inhibition for 12 h (purple), and p300 inhibition for 24 h (cyan). F) Average ATAC-seq accessibility as in D for conditions in E after IFN-β addition. G) Average ATAC-seq accessibility relative to ISREs genome-wide as in C for conditions in F. H) ATAC-seq tracks of normalized chromatin accessibility without (top) and with (bottom) IFN-β with BAF inhibition for 12 h (purple), p300 inhibition for 24 h (cyan), and no inhibitor (black) for IFI6, ISG15, and USP18. I) RNA expression of endogenous ISGs before (pink) and after (red) IFN-β (two biological replicates). J) ATAC-seq tracks of normalized chromatin accessibility over an IFN-β timecourse for IFI6, ISG15, and USP18. Black lines denote the region that was installed at the reporter.

A) Reporter with ISG15 and USP18 promoter containing intact and scrambled ISREs replacing minCMV (inset). Average methylation pre- (light) and post- (dark) IFN-β for intact (red) and scrambled (gray) ISREs (two biological replicates and two scrambles). Dashed line set by construct without ISREs. B) Gene expression (MFI) of reporters as in A. C) Aggregated methylation data obtained for reporter constructs with endogenous ISG promoters and proximal enhancers present (IFI6, ISG15, USP18) with intact (top) and scrambled (bottom) ISREs with (pink) and without (black) IFN-β. D) Gene expression (flow cytometry) of ISG promoter reporters after 0, 8, and 24 h of IFN-β with BAF inhibition (purple), p300 inhibition (cyan), and no inhibitor present (gray). Black dashed line is the MFI of WT K562s.

A) rTetR-VP48 occupancy without (black) and with (gray) dox during sample processing across TetO numbers (two biological replicates). B) Representative temporal delay between rTetR-VP48 occupancy (blue) and promoter activation (black) for constructs with six and eight TetOs (two biological replicates). C) Potency (k_on/k_off) fit from Additive Activation Model across dox timecourse. Error bars are standard deviations. D) Measured data (top) and fit kinetic models (bottom) for TF occupancy (r² = 0.99), promoter activation (r² = 0.90), RNA (r² = 0.95), and protein levels (r² = 0.92) over time and across TetO number. E-F) Data and fits (lines) for average TF occupancy (E) and active promoter fraction (F) over time for the construct with seven TetOs without inhibitor (black), with p300 inhibition (cyan), and with BAF inhibition (purple) (two biological replicates). G) Potency (k_on/k_off) fit from Additive Activation Model for conditions in E-F. Error bars are standard deviations. H) Representative lack of temporal delay between wide footprints (pink) and promoter activation (black) for constructs with four and six ISREs (two biological replicates). I) RNA expression as measured by RT-qPCR over time (two technical replicates) for variable ISRE numbers. J) The relationship between average wide footprints and active promoter fraction across IFN-β activation (two biological replicates).