Fig. 4: Surface condensation enables robust sequence-specific localization of Klf4.

a, Schematic depicting the in situ condensation assay: MBP-Klf4-GFP-coated DNA is transferred to a Klf4-free buffer containing 3C protease that can cleave off the MBP tag. b, Top, confocal images before (time, t = 0 s) and after (t = 350 s) transfer to the 3C solution. Bottom, representative kymograph of the 3C-dependent rearrangement of Klf4-GFP on DNA (Extended Data Fig. 10a–f and Supplementary Video 4). c, Top, colour map representation of the average intensity profile (N = 22 experiments) 350 s after cleavage induction. Bottom, colour map representation of the coarse-grained binding energy landscape (h_i; Fig. 2h). d, Average intensity along the DNA for Klf4-GFP (red) and MBP-Klf4-GFP (black) as a function of concentration. Light dots, individual experiments; filled circles, binned medians (Klf4-GFP: 11–34 experiments per bin; MBP-Klf4-GFP: 3–9 experiments per bin); error bars, 95% confidence interval; red line, fit of Klf4-GFP data using the two-state heterogeneous model (Supplementary Note); black line, fit of MBP-Klf4-GFP data to the Hill–Langmuir model (Supplementary Methods). e, Correlation between the coarse-grained binding energy landscape (h_i; Fig. 2h,c, bottom) and the average intensity profile (quantified by Pearson’s correlation coefficient) for Klf4-GFP (red) and MBP-Klf4-GFP (black) as a function of protein concentration (Extended Data Fig. 10g,h). Circles, correlation resulting from the average intensity profiles selected from a moving window along the concentration axis (N = 15 and 12 experiments per bin for Klf4-GFP and MBP-Klf4-GFP, respectively); shaded area, 95% confidence interval (bootstrapping); red line, prediction from the two-state heterogeneous model for condensation using parameter values extracted from the fit in Fig. 2d; black line, prediction from the two-state heterogeneous model for adsorption (Supplementary Note provides the parameter values and Extended Data Fig. 9d,e).