Fig. 4: DNA-binding and protein interactions patterns.

a Comparison of SEP3-AG and SEP3-AGI^AP1 seq-DAP-seq binding intensity (log₁₀ of reads per kb per million of reads mapped in bound regions) and colour coded by purple-blue (SEP3-AG^IAP1 specific) to orange-red (SEP3-AG specific) according to log₁₀ of SEP3-AG^IAP1/SEP3-AG. b Density plot showing data as per a. c Logos derived from PWM-based models obtained for SEP3-AG and SEP3-AG^IAP1. d Predictive power of TFBS models. Models are built using 600 sequences best bound by each of the two heterocomplexes and are searched against 1073 SEP3-AG (orange) and 1073 SEP3-AG^IAP1 (blue) specific regions, defined as the top 15% of sequences that are most strongly bound by one complex relative to the other. Matrix-based models (PWM and TFFM) are not able to differentiate SEP3-AG and SEP3-AG^IAP1 binding, whereas k-mer-based analysis is able to better predict binding for the respective datasets. e SEP3-AG favours intersite spacings of 47 and 57 bp based on SEP3-AG-specific regions. SEP3-AG^IAP1 favours intersite spacings of 25 and 34 bp based on SEP3-AG^IAP1I specific regions. f Top, published SELEX-seq for SEP3-AP1 and SEP3-AG⁷⁸ comparing the normalised score ratios (SEP3-AG/SEP3-AP1) for SELEX-seq and score ratios (AG/AP1) ChIP-seq at 1500 SEP3 best bound loci in ChIP-seq show a positive correlation, suggesting that SEP3-AP1 and SEP-AG bind different sequences in vivo and that in vitro binding is able to differentiate bound sequences that are more SEP3-AP1-like versus SEP3-AG-like. Bottom, SEP3-AG and SEP3-AG^IAP1 seq-DAP-seq coverage as per SELEX-seq scores. A positive correlation is observed suggesting that, in vitro, the swap of AP1 I domain in AG is able to recover some of the binding specificity of SEP3-AP1. g Yeast two-hybrid assays using AG, AP1 and AG^IAP1 as bait against MIKC^C MADS TFs in Arabidopsis. Data show that AG^IAP1 loses AG interactors and gains AP1 interactors.