SelexGLM differentiates androgen and glucocorticoid receptor DNA-binding preference over an extended binding site

Abstract

The DNA-binding interfaces of the androgen (AR) and glucocorticoid (GR) receptors are virtually identical, yet these transcription factors share only about a third of their genomic binding sites and regulate similarly distinct sets of target genes. To address this paradox, we determined the intrinsic specificities of the AR and GR DNA-binding domains using a refined version of SELEX-seq. We developed an algorithm, SelexGLM, that quantifies binding specificity over a large (31-bp) binding site by iteratively fitting a feature-based generalized linear model to SELEX probe counts. This analysis revealed that the DNA-binding preferences of AR and GR homodimers differ significantly, both within and outside the 15-bp core binding site. The relative preference between the two factors can be tuned over a wide range by changing the DNA sequence, with AR more sensitive to sequence changes than GR. The specificity of AR extends to the regions flanking the core 15-bp site, where isothermal calorimetry measurements reveal that affinity is augmented by enthalpy-driven readout of poly(A) sequences associated with narrowed minor groove width. We conclude that the increased specificity of AR is correlated with more enthalpy-driven binding than GR. The binding models help explain differences in AR and GR genomic binding and provide a biophysical rationale for how promiscuous binding by GR allows functional substitution for AR in some castration-resistant prostate cancers.

Figure 1.

SELEX-seq reveals differences in AR- and GR-DBD (DNA-binding domain) DNA-binding specificity. (A) SELEX-seq. A 70-bp dsDNA library with 23-bp randomized region was incubated with the DBD of AR or GR and separated into monomer and dimer species by EMSA. Dimer-bound DNA was recovered, quantified by qPCR, amplified as the library for the next round, and repeated for eight rounds. Each round of library, including the initial dsDNA library, was sequenced. (B) EMSA gel showing the enrichment of dimer-bound sequences after each round of selection for GR-DBD. The intensity of the shifted band plateaus after round 7. A high-affinity palindromic sequence served as a control to locate the dimer band. (*) An artifact during the synthesis of control sequence but not observed in the SELEX library. (C,D) Information gain, or Kullback-Leibler divergence, from R0 to R8, as a function of oligonucleotide length. (E) Boxplot showing the multiplicity of unique 23-mers in each of the last three rounds of SELEX-seq selection for AR and GR. Even for the most highly selected library (AR R8) fewer than 10% of all reads have 10 copies or more, indicating that the libraries are not overselected. (F) Venn diagram showing the overlap of sequences for AR- and GR-DBD with at least 100 sequencing counts. (G) Scatterplot of sequences that were commonly bound (yellow from F) by AR- and GR-DBD.

Figure 2.

SelexGLM shows differences in DNA recognition between AR and GR throughout their binding sites. (A–D) Energy logos for AR-DBD (top) and GR-DBD (bottom), obtained by fitting biophysical models for protein–DNA interaction to the SELEX read counts using an iterative generalized linear modeling approach based on Poisson regression, implemented as SelexGLM. Highly similar logos were obtained using two separate rounds of data. See Supplemental Figure S5 for logos generated using round 4 to 8 data. (E) Cumulative distribution functions for the contribution of half-site (squares), spacer (triangles), and flanking (circles) sequences on AR-DBD (red) and GR-DBD (blue) binding energy. (F) Validation of the contribution of flanking A tracts and spacer to AR- and GR-DBD binding performed by quantitative electrophoretic mobility shift assay (EMSA). Loss of flanking A tracts is more detrimental to AR- than GR-DBD (one vs. two), whereas changing spacer can have detrimental effects on the binding of both (one vs. three). Error bars, SEM based on at least three repeats of each experiment. (*) P-value ≤0.05, (**) P-value ≤0.01, two-sided t-test.

Figure 3.

Difference in DNA shape readout between AR and GR. (A) Difference in ΔΔG/RT values between AR and GR at each nucleotide position, normalized by their mean across all four bases. (B) Quantitative EMSA was used to measure the affinities of AR- and GR-DBDs for four sequences that maximally favor AR (Shape-1) and GR (Shape-4) and test the importance of half-site minor groove width (Shape-3 and -4). Each measurement was performed at least three times. Error bars, SEM. Average minor groove width profile for the top/middle/bottom 5% of sequences in terms of affinity for AR (C) and for GR (D).

Figure 4.

ITC analysis reveals distinct DNA-binding thermodynamics between AR- and GR-DBD. (A) The raw heat titration signals (top) and normalized heat of injection profiles (bottom) of AR- and GR-DBD bound to a given DNA sequence. Standard errors are estimated by NITPIC (Brautigam et al. 2016). (B) The K_D of AR-DBD and GR-DBD for four sets of sequences fit from the ITC data. AR-DBD affinity is increased with flanking As and an optimal spacer, whereas GR is insensitive. (C) Enthalpy, ΔH, is calculated from the heat of binding for each DNA sequence. Flanking sequences decrease ΔH for AR-DBD, enhancing affinity. Smaller indicates a greater contribution to affinity. (D) Entropy, ΔS, is calculated from the K_D (thus ΔG) and ΔH. GR-DBD affinity is more entropically driven. Larger indicates a greater contribution to affinity. (*) P-value ≤0.05, (**) P-value ≤0.01, (***) P-value ≤0.001, (****) P-value ≤0.0001, two-sided t-test. Error bars, SD represent the standard deviation from at least three experiments.

Figure 5.

Differences of intrinsic specificity between AR- and GR-DBD are reflected in the respective cellular genomic binding profiles. (A) Ability of the AR (red) and GR (blue) PSAMs to identify true ChIP-seq peaks as measured by the receiver operator characteristic (ROC) of binary classifiers identifying peaks versus adjacent regions. (B) Comparison of the median enrichment over background from ChIP-seq (y-axis) to the relative affinity for the strongest sequences within the ChIP-seq peak calculated from the position specific affinity matrix for AR (red) and GR (blue). In vitro affinity is binned by decile (e.g., 10 is the sequence space representing the top 10% highest affinity sites) (C) Multiple linear regression coefficients in models that use AR (red) and GR (blue) PSAM scores to predict AR (left) and GR (right) ChIP-seq peak enrichments. P-Values were calculated using t-tests.