As stated above, fluorescence polarization has been successfully utilized in multiple high-throughput screens, including some whose goal was the identification of inhibitors of protein–DNA interactions (18
). Thus, in our search for inhibitors of the apoptosis inducing factor (AIF)–DNA interaction, we initially attempted to develop an FP-based HTS. Unfortunately, DNA sequences with two different fluorescent tags did not give a noticeable change in fluorescence polarization upon incubation with AIF (see Supplementary Figure 1
). We ascribed the failure of the FP method in this case to the low affinity of AIF for any DNA sequence, as demonstrated by the molar ratios required in EMSA (48
). In addition, the FP assay with AIF is complicated by the fact that AIF itself is fluorescent due to the presence of its flavin cofactor. We thus sought to develop a general and high-throughput technique for the identification of inhibitors of protein–DNA interactions that would be able to avoid these complications.
A new class of disposable microplate-based optical biosensors based on the unique properties of photonic crystals (PC) have been recently developed by Cunningham et al.
). Like other optical biosensors, including those utilized in SPR, PC biosensors detect biomolecular interactions on the surface of a transducer through changes in dielectric permittivity with respect to the liquid media. A PC is composed of a periodic arrangement of dielectric material that effectively prevents propagation of light at specific wavelengths and directions. When illuminated with white light, appropriately configured photonic crystals are able to reflect narrow band light whose wavelength is directly dependent on the local density of adsorbed biomolecules (). Association of macromolecules to the sensor surface modulates the peak wavelength value (PWV) of the reflected light, allowing for detection of binding by a shift in the PWV. Photonic crystal biosensors incorporated onto standard format 96-, 384-, or 1536-well microplates have been used to detect antibody–antigen, small molecule–protein, and whole cell interactions on the biosensor surface without the use of fluorescent labels (50
). In the work described herein, photonic crystal technology is applied to the detection and analysis of protein–DNA interactions. To demonstrate the scope of this method, we chose two very different protein–DNA interactions: the bacterial MazEF complex, which binds to its promoter DNA in a sequence-specific manner, and the human AIF, a protein that binds nonspecifically to chromosomal DNA.
Figure 1 a) Schematic of the PC biosensor. A broadband LED illuminates the biosensor from the bottom, and reflected light is collected and transferred to a spectrometer where the PWV is measured. b) Image of PC biosensor films adhered to the bottom of black 384-well (more ...)
MazEF is a bacterial toxin–antitoxin system thought to be responsible for the maintenance of resistance-encoding plasmids in certain infectious bacteria (51
). Originally identified on the E. coli
chromosome, MazEF is a heterohexameric, ~77
kDa complex consisting of 1 MazE (antitoxin) dimer, and two MazF (toxin) dimers (54
). MazF is an RNase which is released from MazE upon plasmid loss, resulting in inhibition of bacterial growth (55
). In addition to its toxic action, the MazEF complex also regulates its expression by binding to its own promoter sequence (55
AIF is a mammalian mitochondrial NADH-oxidoreductase that also has a key role in caspase-independent cell death (56
). The 67 kDa form of AIF is produced in the cytoplasm, where it then translocates to the mitochondria and carries out its oxidoreductase function as well as possible upkeep of complex I (58
). Upon cellular insults (such as DNA damage), AIF is cleaved off the inner mitochondrial membrane and released into the cytoplasm as a 57 kDa protein. Once in the cytoplasm, AIF translocates to the nucleus and binds to DNA in a sequence-independent fashion, causing stage I chromatin condensation, eventually leading to cell death. AIF is thought to contact DNA through electrostatic interactions, as mutations of surface lysine and arginine residues abrogate DNA binding in vitro
and in cell culture (49
). These surface residues are contained within the FAD-binding domain and the C-terminal domain of AIF, and it is proposed that 12 base pairs of double-stranded DNA (dsDNA) can be bound in this stretch of AIF (49
). Although the crystal structure of AIF has been solved (49
), no cocrystal structure has been obtained for AIF and DNA. Small molecule inhibitors of the AIF–DNA interaction are of great interest due to the involvement of AIF in multiple disease state models including Parkinson's disease (60
), ischemia/reperfusion injury (61
), amyotrophic lateral sclerosis (62
), and Huntington's disease (63
); however, no small molecule inhibitors of the AIF–DNA interaction have been reported.
The PC optical biosensors used in this work have been described previously (50
). Briefly, the sensor contains a one-dimensional surface grating structure with a period of 550 nm (, panel a). It is produced via
a room-temperature replica molding process using a UV-curable polymer on a transparent polyester sheet. The low refractive index polymer grating structure is subsequently coated with a film of high refractive index TiO2
to achieve the final sensor structure. The completed sensor is cut from the polyester sheet and attached to the bottom of a standard 384-well microplate (, panel b). The readout instrument (SRU Biosystems BIND Reader) (50
) illuminates microplate wells from below with a broadband light source coupled to eight optical fibers, each illuminating a ~2
mm diameter region of the PC surface at normal incidence. Reflected light is collected by a second optical fiber, bundled next to the illuminating fiber, and measured by a spectrometer. An automated motion stage enables parallel collection of reflectance data at timed intervals to acquire kinetic information from all 384 wells. In , panel c illustrates the general experimental setup of DNA-binding assays performed using PC biosensors.
Sequence-Dependent DNA Binding: MazEF
MazEF was shown previously to bind to its own promoter sequence using an electrophoretic mobility shift assay (EMSA) (55
). MazE has some intrinsic DNA binding ability, while formation of the MazEF complex dramatically increases DNA binding (55
). For the experiments described herein, the same promoter sequence used by Zhang and co-workers (55
) was purchased with one end functionalized with biotin. The sensor surface was coated with 12 μL of 1 μM biotinylated DNA (12 h at 4 °C) and blocked with Starting Block (Pierce Biotechnology) for 2 h at 4 °C. MazEF was expressed and purified as described (76
) and then added to DNA-containing wells at the specified concentration (, panel a) for 1 h at 25 °C. In , panel a shows the association of the MazEF protein complex with biosensors coated with promoter DNA. This association was inhibited by preincubating MazEF with increasing concentrations of free promoter DNA for 15 min (, panel b). The binding of MazEF to its promoter sequence was also specific; when a control sequence of GC-rich DNA with one end biotinylated was complexed to the PC biosensor, MazEF exhibited only minimal binding (, panel c). The kinetics of MazEF binding to the promoter-bound biosensor were also monitored over the course of 30 min (, panel c).
Figure 2 a) MazEF associates with its promoter sequence bound to the PC biosensor surface in a dose-dependent fashion. b) Preincubation of MazEF (1.8 μM) with its nonbiotinylated promoter sequence reduces the association of MazEF with the promoter-bound (more ...)
Sequence-Independent DNA Binding: AIF
Analyses of the DNA binding properties of the 57 kDa form of AIF (AIFΔ1-121) have been performed previously using EMSA (48
). AIF binds DNA nonspecifically, as different sequences of free oligomer are able to prevent AIF binding to a DNA ladder (48
). Therefore, a biotinylated, randomized 30 bp sequence of dsDNA was chosen as the DNA target of AIF; it has been shown that AIF is capable of binding DNA of this length (48
). Preparation of the biosensor surface was analogous to the MazEF experiments described above, except the specified concentrations of AIF were incubated with biotinylated DNA for 30 min at 25 °C. As monitored by the PC biosensor, the association of AIF with biotinylated DNA was found to be pH dependent (, panel a), and a pH of 6.3 was found to give modest PWV shifts while maintaining protein stability. This pH dependence is not surprising due to the fact that the pI of AIFΔ1-121 is 7.8, and a pH lower than the pI would favor binding to a DNA substrate. In , panel b shows the association of AIF with biotinylated DNA; this interaction is also inhibited by a 15 min preincubation with free DNA (, panel c). AIF is thought to bind DNA in a cooperative fashion, due to the fact that a large molar excess of AIF is required to detect binding (48
). Because MazEF binds to its promoter sequence specifically, and no known cooperative interaction has been postulated, we propose the difference in PWV shift values between MazEF and AIF are because of the difference in the relative affinities of these proteins for their DNA targets.
Figure 3 a) AIF associates with a randomized DNA sequence bound to the biosensor surface in a pH dependent fashion. The pH chosen for further assays was 6.3, the pH at which AIF exhibits moderate PWV shifts and is stable over the course of the assay. b) Increasing (more ...)
Demonstration of HTS Potential: Screening for Inhibitors of the AIF–DNA Interaction
The data in and demonstrate that the PC biosensor can be successfully used to detect protein–DNA interactions. With these experiments in place, we moved to develop a high-throughput screen that could be used to identify compounds that prevent the AIF–DNA interaction. As with previous experiments, a 1 μM solution of biotinylated DNA was immobilized on streptavidin-coated PC biosensors, and Starting Block was then added to reduce nonspecific interactions between AIF and the biosensor surface. AIF (3.51 μM) and putative small molecule inhibitors (25 μM) were incubated together for 15 min at 25 °C in a clear 384-well plate (Falcon); reference wells for each compound were also prepared in the same 384-well plate. These solutions were then transferred to the DNA-containing 384-well biosensor plate. Compounds that inhibit the AIF–DNA interaction would prevent the PWV shift observed in the AIF–DNA binding event. In this fashion, approximately 1000 compounds (obtained from an in-house compound collection (66
)) were screened in duplicate at a concentration of 25 μM. All experimental wells were normalized against the following two reference wells: AIF with no biotinylated DNA (to account for the nonspecific interactions of AIF with the streptavidin coated biosensor), and biotinylated DNA with compounds (to account for nonspecific interactions with the DNA or biosensor surface). The quality of the screen was assessed via the Z′
-factor (see eq 2
in Methods), a unitless coefficient reflective of the assay's signal dynamic range and data variability. This particular assay attained a median score of 0.65, regarding this as an “excellent” assay (67
). Most wells showed very little variation in the PWV shift, implying no prevention of the AIF–DNA interaction (, panels a–c). However, one compound in this collection, aurin tricarboxylic acid (ATA, , panel d), was found to inhibit the AIF–DNA interaction (, panels b and c). In the screen ATA displayed ~80% inhibition of AIF–DNA binding and was the only compound to exhibit significant inhibition out of the ~1000 compounds screened. Representative PWV values are shown for a group of compounds not containing ATA (, panel a) and the group of compounds that contains ATA (, panel b). The PWV shifts were then converted to a percent inhibition of AIF, and these data are graphed for all ~1000 compounds (, panel c). The PC biosensor was then used to assess the effect of a range of concentrations of ATA; this analysis revealed that ATA inhibits AIF–DNA binding with an IC50
of 23 μM (, panel a).
Figure 4 a) A representative group of ~1000 compounds screened for their ability to inhibit the AIF–DNA interaction. The black bar represents a control in which there is no compound present; that is, only AIF (3.51 μM) and 1% DMSO are incubated (more ...)
Figure 5 a) Dose–response curve of the inhibition of AIF–DNA binding by ATA as measured by PC biosensor technology. ATA inhibits AIF in this assay with an IC50 of ~23 μM. b) AIF retards the migration of linearized pUC19 plasmid (more ...)
As there are no known small molecule inhibitors of AIF, we sought to confirm the results of the high-throughput PC biosensor screen. Thus, EMSA was used to probe the ability of ATA to inhibit the AIF–DNA interaction. The migration of linearized pUC19 plasmid DNA was retarded by increasing concentrations of AIF (, panel b). Holding the concentration of AIF constant and increasing the amount of ATA prevented the association of AIF with the plasmid DNA, as measured by this gel-shift assay (, panel c). ATA inhibited AIF–DNA binding with an IC50 of approximately 50 μM in this assay, as determined by densitometry (, panel d). A structurally related analogue of ATA, p-rosolic acid, was unable to inhibit AIF–DNA binding in the gel assay (, panel c). The binding of ATA to AIF was then confirmed by ITC. ATA was shown to bind to AIF with a Kd = 19 ± 5 μM (, panel a), while p-rosolic acid showed little affinity for AIF (, panel b). ATA is the first small molecule known to bind to AIF and to prevent the AIF–DNA interaction.
Figure 6 ITC measurements of AIF binding to either ATA or p-rosolic acid. a) ATA binds to AIF with a Kd = 19 ± 5 μM, while b) p-rosolic acid shows little affinity for AIF. The affinity of AIF for ATA was calculated using a single-site model using (more ...)
Given the difficulty in identifying inhibitors of protein–nucleic acid interactions, a facile, general, method for identifying them would be of great value to the chemical biology and medicinal chemistry community. The data presented herein indicate that the photonic crystal biosensor assay is suitable for the rapid identification of inhibitors of protein–nucleic acid interactions. Photonic crystal technology is analogous to SPR-based methods of detecting binding events, with the key advantage of full compatibility with the standard 384-well format, allowing for high-throughput screening of large compound libraries. In its current incarnation, the biosensor readout instrument allows for the screening of >120 plates per 8 h, translating to a maximum of ~22,000 individual wells per day. Notably, the PC biosensor was able to identify a relatively weak AIF ligand (Kd = 19 μM) that inhibits the AIF–DNA interaction. As with other optical biosensors, detection of binding is ultimately based on differences in molecular weight; thus a decreased signal will be obtained if the ligand is much smaller than its protein or nucleic acid binding partner. While only demonstrated herein for protein–DNA interactions, analogous experiments with protein–RNA and protein–protein interactions can easily be envisioned. In addition to applications in protein–DNA disruption, screens for compounds that enhance protein–DNA interactions would also be feasible with this technology.
We have utilized the PC biosensor technology to discover the first inhibitor of AIF–DNA binding, although the relatively weak potency of ATA and its documented promiscuity (ATA has been found to inhibit targets including von Willebrand factor (68
) gp120 and interferon-α (69
), and other DNA binding proteins (70
)) will likely preclude its use as a cytoprotectant. In addition, the proposed polymeric nature of ATA (73
) correlates well with the mechanism of AIF binding, as AIF also binds to a negatively charged polymer. This is further shown by the fact that AIF is not inhibited by carboxylic acid containing compounds that are not known to polymerize (see Supplementary Figure 2
). However, there are very few disruptors of protein–nucleic acid interactions, and cases that are well described typically involve nucleic acid binding, not protein binding, as the mechanism of inhibition (4
Interestingly, our data indicate that ATA inhibits the AIF–DNA interaction by binding directly to AIF. Although in this particular case ATA inhibits the AIF–DNA interaction by binding to the protein, PC technology would also be able to detect compounds that inhibit protein–DNA interactions through DNA binding, as small molecule binding to macromolecular targets induces a much smaller shift in PWV (~0.1 nm) (50
) than the PWV shift upon protein binding to DNA (~1.0–3.0 nm, see and ).
In summary, a PC biosensor assay was developed for the purpose of detecting protein–nucleic acid interactions, and this assay was utilized in HTS mode to discover a novel inhibitor of the AIF–DNA interaction. Photonic crystal technology is able to detect both low- and high-affinity protein–DNA interactions, as demonstrated for AIF and MazEF, respectively. In the case of AIF, the photonic crystal biosensor technology avoids problems due to the intrinsic fluorescence of the protein itself, which complicate analogous fluorescence polarization experiments. PC biosensors in the microplate format retain all the advantages of SPR biosensors in the flow cell format except for determination of kinetic on/off rate information, with the highly valuable addition of being readily compatible with high-throughput screening platforms. This technology should find general applicability in high-throughput screens for inhibitors of protein–nucleic acid and protein–protein interactions.