Study sequence rules of estrogen receptor α–DNA interactions using dual polarization interferometry and computational modeling

Analytical Biochemistry 433 (2013) 121–128

Contents lists available at SciVerse ScienceDirect

Analytical Biochemistry journal homepage: www.elsevier.com/locate/yabio

Study sequence rules of estrogen receptor a–DNA interactions using dual polarization interferometry and computational modeling Hong Yan Song a, Wenjie Sun b, Shyam Prabhakar b,⇑, Khin Moh Moh Aung a, Xiaodi Su a,⇑ a b

Institute of Materials Research and Engineering, Agency for Science, Technology, and Research (A⁄STAR), Singapore 117602, Singapore Genome Institute of Singapore, Agency for Science, Technology, and Research (A⁄STAR), Singapore 138672, Singapore

a r t i c l e

i n f o

Article history: Received 5 September 2012 Received in revised form 11 October 2012 Accepted 12 October 2012 Available online 23 October 2012 Keywords: Dual polarization interferometry Thermodynamic Modeling of ChIP-seq Transcription factor–DNA interaction Estrogen receptors Estrogen response elements

a b s t r a c t Estrogen receptor a (ERa) is a ligand-activated transcription factor. In a classical model, ERa regulates gene expression by binding to DNA sequences called estrogen response elements (EREs). A perfect ERE contains a palindromic consensus sequence of 50 -GGTCAnnnTGACC-30 . A slight variation in ERE sequence alters ERa binding affinity and, thus, the gene transcription activity. In this study, all possible singly mutated EREs of 15 sequences (three possible base substitutions at each of one to five positions of one half-site) were created. Dual polarization interferometry (DPI) was used to measure the receptor binding to generate an in vitro binding energy model. A motif discovery algorithm, Thermodynamic Modeling of ChIP-seq (TherMos), was used to compute the binding energy model from in vivo genome-wide ERa binding data. The in vitro affinity model measured by DPI correlates very well with the TherMos prediction (in vivo model), with a rank correlation coefficient of 0.91, which indicates that the DPI-determined model is reliable and powerful in understanding of ERa binding in vivo in the whole genome. This is the first report of DPI study of protein–double-stranded DNA (dsDNA) interactions. The assay protocols developed are efficient for screening a large quantity of DNA sequences with single base variation sensitivity. Ó 2012 Elsevier Inc. All rights reserved.

Estrogen receptor a (ERa)1 is a ligand-activated transcription factor that plays key roles in many physiological processes and is one of the most important nuclear hormone receptors in breast cancer biology [1]. In a classic model, ERa regulates the transcription of target genes by directly binding to DNA sequences called estrogen response elements (EREs) [2]. A perfect ERE contains a palindromic consensus sequence separated by a three-base pair spacer, 50 GGTCAnnnTGACC-30 [3,4]. Despite the consensus ERE delineated from conserved cis-regulatory elements found in chicken and Xenopus vitellogenin A2 genes, the majority of in vivo EREs deviate from the consensus, with one half-site identical to that in the consensus and the second half-site having nucleotide variant(s) [2]. An important line of research in estrogen-related gene regulation is to identify an in vivo genome-wide binding affinity model of ERa–DNA interactions, that is, determine the binding affinity of ERa to the consensus ⇑ Corresponding authors. Fax: +65 68720785 (X. Su). E-mail addresses: [email protected] (S. Prabhakar), [email protected] (X. Su). 1 Abbreviations used: ERa, estrogen receptor a; ERE, estrogen response element; EMSA, electrophoretic mobility shift assay; SPR, surface plasmon resonance; DPI, dual polarization interferometry; TherMos, Thermodynamic Modeling of ChIP-seq; TE, transverse electric; TM, transverse magnetic; RI, refractive index; dsDNA, doublestranded DNA; sulfo-GMBS, N-(4-maleimidocaproyloxy) sulfosuccinimide, sodium salt; ETH, ethanolamine; HPLC, high-performance liquid chromatography; SDS, sodium dodecyl sulfate; PSEM, position-specific energy matrix. 0003-2697/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.ab.2012.10.021

ERE and various ERE variants [5,6] that ultimately determines the transcription rate of estrogen-related genes [7]. Of the many studies of ERa binding with EREs, single base mutation impact is of particular interest. Many naturally occurring EREs differ from the consensus sequence by only one nucleotide. For example, the ERE of the human pS2 gene has an A ? G variation (GGTCAgcgTGGCC) [8,9], and the ERE of the Xenopus vitellogenin B1 gene has a G ? A mutation (AGTCActgTGACC) [10]. The affinity depletion of ERa to these singly mutated EREs has been confirmed using electrophoretic mobility shift assay (EMSA) [8– 10] and surface plasmon resonance (SPR) spectroscopy [11]. Despite many studies on ERa–DNA interactions and particularly single mutation impact, the studies are not systematic. The spacer and flanking sequences in each singly mutated ERE vary from one to another, rendering the affinity model less accurately reflective of nucleotide impact [2]. In this study, to construct a more accurate affinity model and uncover the sequence rule of ERa–DNA interactions in terms of single mutation impact systematically, a total of 15 singly mutated EREs were created from a synthesized consensus ERE. Each of them carries one base variation in the 50 end half-site with all possible base substitutions and at each of the five positions (positions 1 to 5 from left to right). The spacer and flanking sequences in these variants are identical. The binding affinities of all these EREs were measured using an in vitro method, particularly dual polarization interferometry (DPI), which is similar to, yet

122

Measuring ERa–DNA interactions using DPI / H.Y. Song et al. / Anal. Biochem. 433 (2013) 121–128

more powerful than, SPR spectroscopy (SPR) [12–14]. At the same time, the binding energy model (affinity model) was computed from in vivo genome-wide ERa binding data [6] using a new motif discovery algorithm, namely Thermodynamic Modeling of ChIPseq (TherMos) (unpublished results). TherMos is specially designed to derive binding energy models with greater accuracy from in vivo ChIP-seq data, which is by far one of the most precise methods to identify the genome-wide transcription factor binding sites [15]. DPI is a surface-sensitive optical technique similar to SPR in terms of real-time measurement of biomolecular interaction on solid substrate [12–14]. It has been increasingly used for studying biomolecular interactions, including oligonucleotide DNA and its targets [16,17], genome DNA and small molecule DNA binders [18,19], carbohydrate and protein [20,21], protein and ligand [22,23], antigen and antibody [24,25], and peptide and membranes [26,27]. The simultaneous measurement of change in phase for both transverse electric (TE) and transverse magnetic (TM) polarizations of light allows the refractive index (RI) and thickness of the adsorbed molecular layers to be calculated. Furthermore, layer density and mass can be calculated based on the linear relationship between RI and density, from which comprehensive binding characteristics (i.e., kinetics and affinity) and, more important, structural properties such as DNA orientation can be determined. In view of the current status of DPI applications, the secondary contribution of this study is the development of DPI protocols for studying protein–double-stranded DNA (dsDNA) interactions exemplified by ERa–EREs. We have, for the first time, developed protocols for dsDNA immobilization and subsequent protein binding in either a ‘‘direct binding’’ mode (i.e., monitoring ERa binding to immobilized ERE) or a ‘‘competition binding’’ mode (i.e., monitoring ERa binding to immobilized ERE in the presence of free competitive ERE sequences in solution) [7]. These protocols allow for determination of absolute affinity of ERa to the consensus ERE and relative affinity of ERa toward all singly mutated ERE variants. The confirmation information of the immobilized dsDNA film (extracted from mass, density, and thickness values) and its impact on ERa binding is discussed. This is the first systematic study of an ERa–ERE binding affinity model concerning single nucleotide variant impact using a combined in vitro experimental method (DPI) and computational prediction from in vivo genome-wide ERa binding data. The high degree of correlation between in vitro and in vivo binding energy models confirms the position-specific single mutation impact. The high degree of correlation also indicates that the DPI-determined model is reliable and powerful in understanding of ERa binding in vivo in the whole genome.

Materials and methods Materials Purified recombinant human estrogen receptor (ERa) was purchased from PanVera (Madison, WI, USA). It was stocked in Hepes buffer containing 10% glycerol. For long-term storage, the ERa was stored in aliquots of 10 ll at 80 °C. Before use, the aliquots were thawed in an ice bath and were diluted using Hepes-T buffer (40 mM Hepes containing 10 mM MgCl2, 100 mM KCl, and 0.1% [v/v] Triton X-100) to form solutions of the desired concentrations. Sulfo-GMBS [N-(4-maleimidocaproyloxy) sulfosuccinimide, sodium salt, a hetero-bifunctional cross-linking reagent] was purchased from Pierce. Streptavidin and ethanolamine (ETH) were purchased from Sigma–Aldrich. A total of 17 EREs (35 bp: 1 ‘‘wtERE,’’ 1 ‘‘scrERE,’’ and 15 singly mutated EREs) were synthesized by Sigma Life Science. The sense sequences and their descriptors are given in Table 1. wtERE contains a perfect core sequence, 50 -GGTCAnnnTGACC-30 , and scrERE

Table 1 ERE sequences used in this study. ID

Sequence

wtERE mut1 mut2 mut3 mut4 mut5 mut6 mut7 mut8 mut9 mut10 mut11 mut12 mut13 mut14 mut15 scrERE

50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT 50 -AGTAAGCT

ccaGGTCA TTA TGACCtgg AGCTTACT-30 ccaTGTCA TTA TGACCtgg AGCTTACT-30 ccaCGTCA TTA TGACCtgg AGCTTACT-30 ccaAGTCA TTA TGACCtgg AGCTTACT-30 ccaGTTCA TTA TGACCtgg AGCTTACT-30 ccaGCTCA TTA TGACCtgg AGCTTACT-30 ccaGATCA TTA TGACCtgg AGCTTACT-30 ccaGGGCA TTA TGACCtgg AGCTTACT-30 ccaGGCCA TTA TGACCtgg AGCTTACT-30 ccaGGACA TTA TGACCtgg AGCTTACT-30 ccaGGTGA TTA TGACCtgg AGCTTACT-30 ccaGGTTA TTA TGACCtgg AGCTTACT-30 ccaGGTAA TTA TGACCtgg AGCTTACT-30 ccaGGTCG TTA TGACCtgg AGCTTACT-30 ccaGGTCT TTA TGACCtgg AGCTTACT-30 ccaGGTCC TTA TGACCtgg AGCTTACT-30 ccaTAGCG TTA CGCTAtgg AGCTTACT-30

has both the 50 and 30 half-sites completely scrambled. When these two sequences were biotinylated at the 50 end, they are denoted as wtERE-b and scrERE-b, respectively. Besides the wtERE and scrERE, 15 mutant sequences (mut1–mut15), each having a base substitution (italic) in the 50 end half-site, were created. The sense strands and the antisense strands were annealed in phosphate-buffered saline (PBS, pH 7.4) containing 10 mM ethylenediaminetetraacetic acid (EDTA, pH 7.5) and stored at 20 °C. Before use, it was thawed and diluted to the desired concentration with Hepes-T buffer (40 mM Hepes containing 10 mM MgCl2, 100 mM KCl, and 0.1% [v/v] Triton X-100). Dual polarization interferometry All DPI measurements were performed using an AnaLight 4D dual polarization interferometer (Farfield Group, Manchester, UK). The details of the instrument and theory have been reported elsewhere [12–14]. Briefly, DPI is an optical technique based on Young’s interference theory (two-slit experiment in optics) [28]. The key part of the instrument is a sandwich sensing chip with the structure of two horizontally stacked silicon oxynitride waveguides separated by a silica layer. The light source is a helium neon laser (k = 632.8 nm) that is switched rapidly between two polarizations, TM and TE, passing through the two waveguides (reference and sensing). By measuring the phase change of light interference fringes in TM and TE in the far field relative to the reference channel, the RI and thickness (<0.01 nm) of an adsorbed molecular layer on the sensing surface was characterized. Furthermore, mass (<0.1 pg/mm2) of the adsorbent is resolved due to the linearity of the RI and density relationship. There are two experimental channels or two microfluidic sensing cells on each DPI AnaChip, which was addressed by general high-performance liquid chromatography (HPLC) fluidic components and a syringe pump (PHD2000, Harvard Apparatus). Samples are loaded and injected via HPLC injection valves into a sample loop with a maximum volume of 200 ll. All fluids, including sample and running buffer, are continuously flowing over the top sensing waveguide of both channels, and the interference fringes are recorded in real time when material is added or removed from the waveguide. In this study, thiol-modified AnaChips (FB 80) were used. The measurement temperature was set at 20 °C. The sensor chip and the Hepes running buffer were calibrated using 80% (w/w) ethanol/water and water before each experiment. After calibration, the flow rate was changed to 50 ll/min and adjusted depending on experimental requirements.

Measuring ERa–DNA interactions using DPI / H.Y. Song et al. / Anal. Biochem. 433 (2013) 121–128

123

channel. In both the direct binding assay and competition assay, multiple cycles of protein binding were realized through successful regeneration of the immobilized DNA surface, that is, dissociation of bound ER by applying 0.1% sodium dodecyl sulfate (SDS) (2min injection at 50 ll/min).

dsDNA immobilization on DPI chip Biotin–streptavidin affinity coupling was applied for dsDNA immobilization. On thiol-modified DPI AnaChip (Scheme 1A), streptavidin was immobilized through amine coupling. In particular, sulfo-GMBS solution (4 mg/ml) in Hepes was injected over both channels for 12 min at a flow rate of 15 ll/min (step a in Scheme 1A). The surface was then rinsed with the Hepes buffer for 5 min, followed by streptavidin (330 nM) addition for 30 min at 5 ll/min (step b). On rinsing the streptavidin-immobilized surface, the remaining activated sites on both channels were blocked by ETH (1 M, pH 8.5) for 6 min at a flow rate of 25 ll/min. After that, sample loops were thoroughly washed with distilled water. Once the response had stabilized, preannealed dsDNA (i.e., wtERE-b or scrERE-b, 500 nM) was injected over one channel for 18 min at 10 ll/min for protein binding, whereas the other channel was left blank (no DNA immobilization) as a control (step c). On reattaining a stable baseline, the running buffer was changed to Hepes-T buffer for ERa–ERE binding.

In vivo binding energy (affinity) computed using TherMos The ERa binding energy (affinity) model was predicted by applying TherMos on the in vivo ERa ChIP-seq library in MCF7 cells [6]. Using TherMos, a thermodynamic model of transcription factor–DNA binding was fitted to the ERa ChIP-seq data [29]. In particular, TherMos predicts the binding energy model or the position-specific energy matrix (PSEM) by fitting the observed binding profile in the ChIP-seq library through nonlinear regression. Once trained, each entry of the PSEM DDGij represents the free energy contribution of each nucleotide i at position j in the sequence (Table 2). For a specific sequence, the binding free energy change DDG (in units of RT, where R is the gas constant and T is the temperature in Kelvin) relative to the reference sequence can be calculated by adding up the DDGij accordingly. For example, the DDG for the wtERE is zero by adding the corresponding DDGij in the PSEM shown in Table 2. For ERa PSEM, the wtERE is selected as the reference sequence and the DDG for the wtERE is set to zero. A larger DDG represents a weaker binding affinity compared with the wtERE.

In vitro ERa–DNA binding assays using DPI Two assay schemes were used in this study, namely ‘‘direct binding assay’’ (Scheme 1B) and ‘‘competition assay’’ (Scheme 1C). In the direct binding assay, ERa at various concentrations (13–164 nM) was flowed through for 3 min at a flow rate of 50 ll/min, followed by rinsing. The mass curves at different concentrations were used to generate the binding kinetics using DPI’s kinetic association software as reported previously [18]. Typically, the software uses an iterative curve-fitting procedure to derive different observed rate constants kon for each ERa concentration. By plotting the kon against the varying concentrations of ERa, a straight line with a slope of kass and a y axis intercept of kdiss is typically produced according to the equation kon = (kass [C] + kdiss). In the competition assay, ERa at a fixed concentration of 100 nM was preincubated with 3-fold of competitor EREs (i.e., different mutant sequences: mut1, mut2, . . ., or mut15) for 15 min at room temperature prior to injection into the wtERE-immobilized

Results and discussion ERE (dsDNA) immobilization on DPI chip As shown in Scheme 1A, biotinylated ERE (i.e., wtERE-b in this case) was immobilized on streptavidin-modified thiol AnaChip (streptavidin was covalently coupled to the chip via amine coupling using sulfo-GMBS linker, a hetero-bifunctional cross-linker that contains amine and sulfhydryl reactive functionalities). Prior to DNA immobilization, the streptavidin-modified surface is blocked by ETH. Fig. 1A is a DPI sensorgram (i.e., resolved mass

A SH SH SH SH SH SH SH

sensor chip

a

c

b

C

B

0.1% SDS

0.1% SDS

Sulfo-GMBS ERE-b

ETH

-

COO mutERE

SA ER

Scheme 1. (A) Schematic illustration of dsDNA immobilization on thiol DPI AnaChip through streptavidin–biotin interaction, where streptavidin is immobilized through amine coupling. Steps a, b, and c are denoted as sulfo-GMBS, streptavidin, and dsDNA immobilization. (B and C) Direct binding assay (B) and competition assay (C) for ERa. For both panels B and C, DNA surface was regenerated by applying 0.1% SDS. SA, streptavidin.

Measuring ERa–DNA interactions using DPI / H.Y. Song et al. / Anal. Biochem. 433 (2013) 121–128

124 Table 2 PSEM of ERa predicted by TherMos. wtERE

G

G

T

C

A

T

T

A

T

G

A

C

C

Position

1

2

3

4

5

6

7

8

9

10

11

12

13

T C A G

1.74 (mut1) 3.15 (mut2) 1.92 (mut3) 0.00

2.36 (mut4) 2.91 (mut5) 1.78 (mut6) 0.00

0.00 2.21 (mut8) 1.48 (mut9) 0.92 (mut7)

2.14 (mut11) 0.00 4.00 (mut12) 2.44 (mut10)

2.18 (mut14) 2.89 (mut15) 0.00 0.71 (mut13)

0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00

0.00 0.71 2.18 2.89

4.00 2.44 2.14 0.00

1.48 0.92 0.00 2.21

1.78 0.00 2.36 2.91

1.92 0.00 1.74 3.15

Note: Each entry of the matrix shows the binding energy change DDGij for nucleotide i at position j. The wtERE is selected as the reference sequence, and its DDGij is set as zero (highlighted in bold).

over time) showing the above layer-by-layer procedures (curve part I), followed by ERa binding (curve part II). For the steps up to ETH blocking, the treatment is identical for both channels, whereas only one channel is further subjected to wtERE immobilization (denoted as ‘‘wtERE’’ channel) for ERa binding and the other is left blank as a negative control (denoted as ‘‘Ctrl’’ channel). The mass, thickness, and density values measured for sulfo-GMBS, streptavidin, and wtERE-b films are shown in Table 3. The sulfo-GMBS injection led to a steady mass increase of 0.75 ± 0.02 ng/mm2 (measured at the end of a washing step to remove loosely attached molecules). The thickness of this molecular film is 1.24 ± 0.11 nm, and the density is 0.67 ± 0.10 g/cm3. The measured thickness is close to the extended length of sulfo-GMBS [24]. The successive streptavidin injection (330 nM) resulted in a thickness increase of 3.30 ± 0.19 nm, a mass increase of 1.32 ± 0.03 ng/mm2, and the average density of 0.49 ± 0.02 g/cm3. Comparing with the oriented streptavidin immobilized on biotinylated chip with a specific interaction (6 nm in thickness), the measured thickness of streptavidin on sulfo-GMBS is lower [28]. However, the mass loading and the density value are reasonable, indicating that it is not a sparse layer but rather a dense layer where the protein may interpenetrate into the sulfo-GMBS layer or deform when coupling to sulfo-GMBS. After ETH blocking, the injection of wtERE-b (500 nM) (in the experiment channel, denoted as ‘‘wtERE’’) led to a mass increase of 0.57 ± 0.01 ng/mm2 and a thickness increase of 0.93 ± 0.09 nm. Taking the molecular weight of 60 kDa for streptavidin and 22 kDa for wtERE-b into consideration, the molar ratio of bound wtERE on each streptavidin is 1.2, indicating that on average approximately 1 dsDNA is immobilized per streptavidin molecule. The thickness increase of 0.93 ± 0.09 nm

Table 3 Mass, thickness, and density of sequential surface layers measured by DPI.

Sulfo-GMBS Streptavidin wtERE-b

Mass (ng/mm2)

Thickness (nm)

Density (g/cm3)

0.75 ± 0.02 1.32 ± 0.03 0.57 ± 0.01

1.24 ± 0.11 3.30 ± 0.19 0.93 ± 0.09

0.67 ± 0.01 0.49 ± 0.02 0.49 ± 0.02

Note: Results are the means ± standard errors of at least four experiments. The relative standard deviations of mass values for all layers are within 3%.

after wtERE-b immobilization is close to the diameter of dsDNA (1 nm). It is noted that the density of the overall film on dsDNA immobilization changed slightly (±0.02 g/cm3, 4% of overall density of 0.49 g/cm3), as shown in Fig. S-1 of the supplementary material (the average value is shown in Table 3). The increase of film thickness and the slight change in density indicate that the DNA molecules are lying on the streptavidin layer, forming an extended film above streptavidin, rather than inserted in as an intercalated film that would lead to no thickness change and a higher density. This spatial arrangement about dsDNA is very important for protein to find the binding site. Determination of ERa–wtERE binding affinity using direct binding assay After re-calibration of the surface (both wtERE and Ctrl channels) using protein binding buffer Hepes-T, ERas ranging from 13 to 164 nM were injected sequentially (Fig. 1A, part II), followed by rinsing and regeneration at the end of each ERa binding cycle. As expected, the binding signal (i.e., mass, ng/mm2) in the wtERE

Fig.1. (A) Time-resolved mass change of stepwise processes of wtERE-b immobilization (part I, Hepes buffer) and the subsequent ERa binding (part II, Hepes-T buffer). Solid curve, experiment channel with immobilized wtERE-b; dotted curve, control channel without DNA. Part I: a, sulfo-GMBS (4 mg/ml) injection; b, streptavidin (330 nM) addition; c, ETH blocking; d, wtERE-b immobilization (500 nM, inject only at experiment channel); e, buffer changed to Hepes-T. Part II: f, ERa injection (concentrations from left to right: 13, 26, 35, 45, 56, 69, 102, 118, 134, and 164 nM); g, at the end of each ERa binding and after rinsing with Hepes-T buffer, 0.1% SDS was added to regenerate the DNA surface. The start and end points of each injection are indicated by solid and broken arrows, respectively. (B) Overlaid binding curve (resolved mass) of ERa in wtEREimmobilized channel (solid line) and control channel (without wtERE, dotted line) at different concentrations.

Measuring ERa–DNA interactions using DPI / H.Y. Song et al. / Anal. Biochem. 433 (2013) 121–128

channel is much larger than that in the Ctrl channel, evidencing the binding of the receptor to DNA. Furthermore, on regeneration using 0.1% SDS at the end of each binding cycle, the background signal in binding buffer is largely restored, reflecting the removal of bound protein and the exposure of the DNA. In the successive receptor binding cycles, higher binding signals are obtained with the increase of receptor concentration. Fig. 1B is the overlay of ERa binding curves in the wtERE DNA channel and Ctrl channel at different concentrations, from which the dMass curves are deducted (Fig. 2A). From the dMass sensorgram (Fig. 2A) (i.e., the amount of receptor bound specifically to wtERE with subtraction of the nonspecific adsorption and bulk RI effects from the Ctrl channel), kinetic analysis was performed using AnaLight association analysis software. An average association rate constant kass of 8.77  103 M1 s1 and dissociation rate constant kdiss of 6.33  10–4 s1 were obtained. The kinetically determined affinity value Kd is 70 nM. This Kd value is consistent with literature values for ERa with the Xenopus vitellogenin A2 ERE under similar binding buffer conditions (kass of 8.17  103 M1 s1, kdiss of 5.06  10–4 s1, Kd of 61.9 nM) [30,31]. To obtain this valid affinity constant, the flow rate of ERa binding has been optimized (50 ll/min) to minimize mass transport. Using a lower flow rate (25 ll/min), a slower kass (1.81  104 M1 s1) and lower affinity (Kd of 166 nM) were obtained due to diffusion control. To determine the sequence dependency of receptor–DNA interactions, ERa binding with scrERE-immobilized surface was measured with a control surface (no DNA immobilized) as reference

125

(see Fig. S-2 in supplementary material). Fig. 2B shows the overlaid sensorgram of dMass of ERa binding to scrERE at the tested ERa concentrations. In this experiment, the amount of immobilized scrERE (0.55 ng/mm2) is close to that of wtERE (0.57 ± 0.01 ng/ mm2). The consistency of the wtERE and scrERE coverage or mass is essential for assessing sequence dependency of ERa–DNA binding through protein binding signals. The much lower binding signal to scrERE (Fig. 2B) relative to wtERE (Fig. 2A) confirmed that the measured ERa binding is sequence specific. In vitro determination of binding energy (affinity) matrix of ERa to all singly mutated ERE variants using DPI competition assay To assess the relative affinity of ERa to all singly mutated EREs, we adopted the competition assay (Scheme 1C) previously developed for SPR experiments [11]. In particular, ER (100 nM) preincubated with different competitor EREs at an ERa/ERE molar ratio of 1:3 was injected into the wtERE and Ctrl channels. The free ERE in solution completes with the surface-immobilized ERE to the protein binding site. Fig. 3A shows the competitive binding of ERa with mut1, mut3, mut5, mut6, mut8, mut9, mut10, mut14, and mut15 as competitors. Fig. 3B is the binding curve for mut2, mut4, mut7, mut11, mut12, and mut13 as competitors. In both experiments, ERa without competitor and ERa with wtERE and scrERE competitors were measured as reference. Depending on the affinity of the competitor EREs, the reactivity of ERa was inhibited to different extents that lead to different extents of reduction in surface binding signals (dMass). The higher the affinity of the competitive ERE, the smaller the measured dMass. For example, when ERa was preincubated with a strong binder wtERE (the 13th cycle in Fig. 3A), surface-bound ERa is minor (dMass = 0.044 ng/mm2) due to the strong affinity of wtERE in solution; whereas when ERa was preincubated with scrERE (the 14th cycle in Fig. 3A) that has limited affinity, a substantial surface binding (dMass = 0.257 ng/mm2) is observed, similar to that observed without competitor (dMass = 0.267 ng/mm2, 1st cycle in Fig. 3A). Fig. 4A is a bar chart summarizing the dMass of bound ERa without and with competitor DNA (all 15 mutant EREs and wtERE and scrERE) shown in Fig. 3A and B. The varied bar height (varied dMass) shows more clearly the differential affinity of each mutant ERE. For example, when mut12 (i.e., C to A change at position 4) is used as competitor, the surface-bound ERa is high (dMass = 0.2465 ng/mm2), being close to that of scrDNA as competitor (dMass = 0.2284 ng/mm2), indicating that mut12 has low affinity similar to that of scrDNA. This finding agrees well with the gel shift assay result. It is believed that the C to A change diminishes the ERa binding due to the steric hindrance between the R211 residue of ERa and nucleotide A [32]. For mut2 (G to C change at position 1) and mut5 (G to C change at position 2), relatively high dMass values were observed (0.143 and 0.1617 ng/mm2, respectively), indicating that these two DNAs have relatively weak binding affinities to ERa. It is believed that the weak affinity is due to the electrostatic repulsion between the positively charged residue lysine and the positively charged nucleotide C at these two positions [32]. Computing binding energy (affinity) matrix from in vivo ERa–ERE binding data using TherMos

Fig.2. Overlaid binding curves of ERa to immobilized wtERE (A) and scrERE (B) at different concentrations (50-ll/min flow rate) as depicted in mass change (dMass, nonspecific binding was subtracted). At the end of 3 min binding, running buffer (Hepes-T) was allowed to flow over. The binding signals, dMass of ERa to wtERE in panel A, are 0.012, 0.029, 0.049, 0.075, 0.125, 0.178, 0.238, and 0.325 ng/mm2 from the low to high concentrations.

The accuracy of the in vitro binding energy model is important when the model is applied to the study of ERa functionality on a genome-wide basis in vivo. To validate our in vitro DPI-determined model, a binding energy model was computed from in vivo genome-wide ERa binding data using TherMos. Using the Thermodynamic Modeling algorithm (i.e., TherMos), the PSEM of ERa was estimated and is shown in Table 2. The PSEM is obtained by forcing

126

Measuring ERa–DNA interactions using DPI / H.Y. Song et al. / Anal. Biochem. 433 (2013) 121–128

Fig.3. Time-resolved mass change (dMass) of competition binding of ERa (100 nM) in the absence/presence of a 3-fold excess of competitors. (A) From left to right cycles, the injections are ERa (100 nM) without competitor DNA (1st and 14th cycles) and ERa preincubated with mut10 (2nd cycle), mut14 (3rd and 15th cycles), mut15 (4th cycle), mut1 (5th cycle), mut3 (6th and 13th cycles), mut5 (7th cycle), mut6 (8th cycle), mut9 (9th cycle), mut8 (10th cycle), scrERE (11th cycle), and wtERE (12th cycle). (B) From left to right cycles, the injections are ERa without competitor DNA (1st cycle) and ERa preincubated with scrERE (2nd and 5th cycles), mut2 (3rd, 4th, and 6th cycles), mut4 (7th cycle), mut7 (8th cycle), mut12 (9th and 12th cycles), mut11 (10th cycle), mut13 (11th cycle), and wtERE (13th cycle). The blue curve (dMass, ng/mm2) is a differential binding from the experiment channel (with wtERE, solid red) to a reference channel (no wtERE, dotted red). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

a palindrome (i.e., binding to the left and right half-sites is equivalent). No preference of binding to the spacer was assumed as well. Therefore, the contribution to the free energy change at the three spacer positions is zero for all four nucleotides. The free energy changes relative to the wtERE for all 15 variants and the scrERE can be calculated by adding up the corresponding values in the PSEM. Fig. 4B is a bar chart of the calculated free energy for all 15 variants and the wtERE and scrERE. For the wtERE, DDG = 0 is indicative of the highest affinity. For the scrambled sequence, DDG = 10.30 was obtained as an indication of the least ERa binding affinity. With single base substitution of different types and at different positions, the binding affinity is reduced to different extents, as indicated by the varied DDG values. For instance, the C to A change at position 4 (mut12) has the largest reduction of binding affinity compared with the wtERE. However, an A to G change at position 5 (mut13) has a minor effect on the decrease of binding affinity. In this study, we observed that for all 15 ERE variants, the position- and type-specific affinity trend measured by in vitro DPI (Fig. 4A) correlates very well with the energy matrix calculated by TherMos (Fig. 4B). For example, for mut12 as competitor, as mentioned earlier, the surface-bound ERa is very high (dMass = 0.2465 ng/mm2) due to the low affinity of this ERE. Indeed, for this ERE, the binding free energy from TherMos is high (DDG = 4.00). In other words, both the DPI and TherMos results indicate a low affinity for this ERE. On the other hand, for mut7 (T to G mutation at position 3) and mut13 (A to G change at position 5), negligible surface binding was observed and low free

energy (DDG values = 0.92 and 0.71) were obtained, both indicating that these two EREs are stronger binder. In a quantitative analysis of the DPI and TherMos results for all 15 ERE variants, the Spearman rank correlation coefficient between the two sets of data is 0.91 (where 1 represents perfect correlation and 0 represents no correlation at all). The high degree of correlation between in vitro and in vivo binding energy models confirms our discovery of the single mutation-related sequence rule, which agrees very well with the gel shift measurements of ERa–ERE binding affinities in vitro at positions 1 (G > T > A > C), 2 (G > A > T > C), 3 (T > G > A > C), 4 (C > T > A  G), and 5 (A > G > T > C) [32]. The high degree of correlation also indicates that the DPI-determined model is reliable and powerful in understanding ERa binding in vivo in the whole genome. Reliability of DPI mass values It is worth mentioning that during the continuous receptor binding and regeneration cycles using the immobilized DNA on DPI thiol AnaChip, a noticeable baseline drift is observed in both the wtERE and Ctrl channels (Figs. 1 and 3). This could be indicative of protein accumulation through nonspecific binding that cannot be completely removed using the regeneration solution of 0.1% SDS. To understand the influence of the nonspecific binding or baseline drift on receptor binding amount through DNA sequence-specific recognition, identical protein samples were injected at different cycles, and mass loading (Mass, ng/mm2) relative to the respective baseline and, more important, the dMass were compared. As shown

Measuring ERa–DNA interactions using DPI / H.Y. Song et al. / Anal. Biochem. 433 (2013) 121–128

127

ER α

A

B

Fig.4. (A) Bar chart of dMass (ng/mm2) of ERa bound to wtERE-immobilized surface in the presence of a 3-fold excess of competitors (mut1–mut15, wtERE, and scrERE) extracted from Fig. 3. The error of repeated measurement (2–4 times) is approximately 5% for most of the competitor sequences. (B) Bar chart of ER binding free energy change toward all mutant ERE sequences with single base variation in the 50 end half-site, as predicted by TherMos.

in the competition experiments (Fig. 3A), ERa without competitor was injected at the 1st and 14th cycles. Despite the baseline drift, the dMass values of the two identical injections (0.267 and 0.278 ng/mm2) are consistent. A similar observation was obtained for ERa with competitor mut14 at the 3rd and 15th cycles (0.142 and 0.137 ng/mm2) and for ERa with mut3 at the 6th and 13th cycles (0.050 and 0.054 ng/mm2) in Fig. 3A. Good reproducibility is also obtained (Fig. 3B) for ERa–mut2 at the 3rd cycle (0.139 ng/ mm2) and 6th cycle (0.148 ng/mm2) and for ERa–mut12 at the 9th cycle (0.238 ng/mm2) and 12th cycle (0.256 ng/mm2). These results confirm that the dMass values obtained at different regeneration cycles are reliable despite the baseline drift and that 0.1% SDS regeneration is sufficient to dissociate receptors previously bound on DNA and renders the DNA being entirely exposed for the next cycle of receptor binding.

Conclusions This work has a 2-fold impact. First, this is the first systematic study of a binding energy/affinity model or sequence rule of ERa–ERE interactions in terms of single mutation impact using a combined in vitro DPI method and computational prediction from in vivo ERa–DNA binding data. The high degree of correlation between in vitro and in vivo binding energy models confirms the sequence rule of positions 1 (G > T > A > C), 2 (G > A > T > C), 3 (T > G > A > C), 4 (C > T > A  G), and 5 (A > G > T > C) in the 50 end ERE half-site. The high degree of correlation also indicates that the DPI-determined model is reliable and powerful in understanding ERa binding in vivo in the whole genome. Second, this is the first report of a DPI study of protein–dsDNA interactions exemplified by ERa–EREs. Powerful experiment protocols have been

128

Measuring ERa–DNA interactions using DPI / H.Y. Song et al. / Anal. Biochem. 433 (2013) 121–128

developed for dsDNA (ERE) immobilization and competitive ERa binding toward a conclusion of single base mutation impact on ERa–ERE binding affinity for all possible single mutation sequences. Acknowledgments X. Su acknowledges the Agency for Science, Technology, and Research (A⁄STAR), Singapore, for financial support under Grant JCOAG03_FG02_2009. H.Y. Song thanks Marcus Swann from the Farfield Group for providing valuable discussions in DPI data interpretation and for optimizing experimental conditions for kinetic analysis. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.ab.2012.10.021. References [1] K.A. Power, L.U. Thompson, Ligand-induced regulation of ER and ER is indicative of human breast cancer cell proliferation, Breast Cancer Res. Treat. 81 (2003) 209–221. [2] C.M. Klinge, Estrogen receptor interaction with estrogen response elements, Nucleic Acids Res. 29 (2001) 2905–2919. [3] L. Klein-Hitpass, G.U. Ryffel, E. Heitlinger, A.B.C. Cato, A 13-bp palindrome is a functional estrogen responsive element and interacts specifically with estrogen receptor, Nucleic Acids Res. 16 (1988) 647–663. [4] L. Klein-Hitpass, M. Schorpp, U. Wagner, G.U. Ryffel, An estrogen-responsive element derived from the 50 flanking region of the Xenopus vitellogenin A2 gene functions in transfected human cells, Cell 46 (1986) 1053–1061. [5] C.E. Mason, F. Shu, C. Wang, R.M. Session, R.G. Kallen, N. Sidell, T. Yu, M.H. Liu, E. Cheung, C.B. Kallen, Location analysis for the estrogen receptor reveals binding to diverse ERE sequences and widespread binding within repetitive DNA elements, Nucleic Acids Res. 38 (2010) 2355–2368. [6] R. Joseph, Y.L. Orlov, M. Huss, W. Sun, S.L. Kong, L. Ukil, Y.F. Pan, G. Li, M. Lim, J.S. Thomsen, Y. Ruan, N.D. Clarke, S. Prabhakar, E. Cheung, E.T. Liu, Integrative model of genomic factors for determining binding site selection by estrogen receptor-a, Mol. Syst. Biol. 6 (2010) 456. [7] K. Dahlman-Weight, V. Cavailles, S.A. Fuqua, V.C. Jordan, J.A. Katzenellenbogen, K.S. Korach, A. Maggi, M. Muramaatsu, M.G. Parker, J.A. Gustafsson, International Union of Pharmacology: LXIV. Estrogen receptors, Pharmacol. Rev. 58 (2006) 773–779. [8] S.M. Hyder, C. Chiappetta, G.M. Stancel, Interaction of human estrogen receptors a and b with the same naturally occurring estrogen response elements, Biochem. Pharmacol. 57 (1999) 597–601. [9] M.A. Loven, J.R. Wood, A.M. Nardulli, Interaction of estrogen receptors and with estrogen response elements, Mol. Cell. Endocrinol. 181 (2001) 151–163. [10] T.C. Chang, A.M. Nardulli, D. Lew, D.J. Shapiro, The role of estrogen response elements in expression of the Xenopus laevis vitellogen in B1 gene, Mol. Endocrinol. 6 (1992) 346–354. [11] H.F. Teh, W.Y. Peh, X. Su, J.S. Thomsen, Characterization of protein–DNA interactions using SPR with various assay schemes, Biochemistry 46 (2007) 2127–2135. [12] J. Popplewell, N. Freeman, S. Carrington, G. Ronan, C. McDonnell, R.C. Ford, Quantification of the effects of melittin on liposome structure, Biochem. Soc. Trans. 33 (2005) 931–933. [13] C.J. Terry, J.F. Popplewell, M.J. Swann, N.J. Freeman, D.G. Fernig, Characterisation of membrane mimetics on a dual polarisation interferometer, Biosens. Bioelectron. 22 (2006) 627–632.

[14] M.J. Swann, L.L. Peel, S. Carrington, N.J. Freeman, Dual-polarization interferometry: an analytical technique to measure changes in protein structure in real time, to determine the stoichiometry of binding events, and to differentiate between specific and nonspecific interactions, Anal. Biochem. 329 (2004) 190–198. [15] D.S. Johnson, A. Mortazavi, R.M. Myers, B. Wold, Genome-wide mapping of in vivo protein–DNA interactions, Science 316 (2007) 1497–1502. [16] B. Lillis, M. Manning, H. Berney, E. Hurley, A. Mathewson, M.M. Sheehan, Dual polarisation interferometry characterisation of DNA immobilisation and hybridisation detection on a silanised support, Biosens. Bioelectron. 21 (2006) 1459–1467. [17] H. Berney, K. Oliver, Dual polarisation interferometry size and density characterisation of DNA immobilisation and hybridization, Biosens. Bioelectron. 21 (2005) 618–626. [18] J. Wang, X. Xu, Z. Zhang, F. Yang, X. Yang, Real-time study of genomic DNA structural changes upon interaction with small molecules using dualpolarization interferometry, Anal. Chem. 81 (2009) 4914–4921. [19] J. Wang, P.D. Coffey, M.J. Swann, F. Yang, J.R. Lu, X. Yang, Optical extinction combined with phase measurements for probing DNA–small-molecule interactions using an evanescent waveguide biosensor, Anal. Chem. 82 (2010) 5455–5482. [20] J.F. Popplewell, M.J. Swann, Y. Ahmed, J.E. Turnbull, D.G. Fernig, Fabrication of carbohydrate surfaces using non-derivatized oligosaccharides, and their application to measuring the assembly of sugar–protein complexes, ChemBioChem 10 (2009) 1218–1226. [21] E.A. Yates, C.J. Terry, C. Rees, T.R. Rudd, L. Duchesne, M.A. Skidmore, R. Lévy, N.T. Thanh, R.J. Nichols, D.T. Clarke, D.G. Fernig, Protein–GAG interactions: new surface-based techniques, spectroscopies, and nanotechnology probes, Biochem. Soc. Trans. 34 (2006) 427–430. [22] K. Karim, J.D. Taylor, D.C. Cullen, M.J. Swann, N.J. Freeman, Measurement of conformational changes in the structure of transglutaminase on binding calcium ions using optical evanescent dual polarization interferometry, Anal. Chem. 79 (2007) 3023–3031. [23] M.J. Swann, L.L. Peel, S. Carrington, N.J. Freeman, Dual polarization interferometry: an analytical technique to measure changes in protein structure in real time, to determine the stoichiometry of binding events, and to differentiate between specific and non-specific interactions, Anal. Biochem. 329 (2004) 190–198. [24] H.Y. Song, X. Zhou, J. Hobley, X. Su, Comparative study of random and oriented antibody immobilization as measured by dual polarization interferometry and surface plasmon resonance spectroscopy, Langmuir 28 (2012) 997–1004. [25] X.B. Zhao, F. Pan, B. Cowsill, J.R. Lu, L. Garcia-Gancedo, A.J. Flewitt, G.M. Ashley, J. Luo, Interfacial immobilization of monoclonal antibody and detection of human prostate-specific antigen, Langmuir 27 (2011) 7654–7662. [26] S.B. Nielsen, D.E.J. Otzen, Impact of the antimicrobial peptide Novicidin on membrane structure and integrity, Colloid Interface Sci. 345 (2010) 248–256. [27] M.K. Baumann, M.J. Swann, M. Textor, E. Reimhult, Pleckstrin homology– phospholipase C-1 interaction with phosphatidylinositol 4,5-bisphosphate containing supported lipid bilayers monitored in situ with dual polarization interferometry, Anal. Chem. 83 (2011) 6267–6274. [28] G.H. Cross, A.A. Reeves, S. Brand, J.F. Popplewell, L.L. Peel, M.J. Swann, N.J. Freeman, Rational design of an estrogen receptor mutant with altered DNAbinding specificity, Biosens. Bioelectron. 19 (2003) 383–390. [29] B.C. Foat, A.V. Morozov, H.J. Bussemaker, Statistical mechanical modeling of genome-wide transcription factor occupancy data by MatrixREDUCE, Bioinformatics 14 (2006) e141–e149. [30] M. Melamed, S.F. Arnold, A.C. Notides, S.J. Sasson, Kinetic analysis of the interaction of human estrogen receptor with an estrogen response element, Steroid Biochem. Mol. Biol. 57 (1996) 153–159. [31] B.J. Cheskis, S. Karathanasis, C.R.J. Lyttle, Estrogen receptor ligands modulate its interaction with DNA, Biol. Chem. 272 (1997) 11384–11391. [32] D. Nguyen, M. Bail, G. Pesant, V.N. Dupont, É. Rouault, J. Deschênes, W. Rocha, G. Melançon, S.V. Steinberg, S. Mader, Rational design of an estrogen receptor mutant with altered DNA-binding specificity, Nucleic Acids Res. 35 (2007) 3465–3477.