Context influences on TALE–DNA binding revealed by quantitative profiling

Julia M. Rogers, Luis Barrera, Deepak Reyon, Jeffry D. Sander, Manolis Kellis, J. Keith Joung, Martha L. BulykNature Communications 2015 · DOI

Abstract

Transcription activator-like effector (TALE) proteins recognize DNA using a seemingly simple DNA-binding code, which makes them attractive for use in genome engineering technologies that require precise targeting. Although this code is used successfully to design TALEs to target specific sequences, off-target binding has been observed and is difficult to predict. Here we explore TALE-DNA interactions comprehensively by quantitatively assaying the DNA-binding specificities of 21 representative TALEs to ∼5,000-20,000 unique DNA sequences per protein using custom-designed protein-binding microarrays (PBMs). We find that protein context features exert significant influences on binding. Thus, the canonical recognition code does not fully capture the complexity of TALE-DNA binding. We used the PBM data to develop a computational model, Specificity Inference For TAL-Effector Design (SIFTED), to predict the DNA-binding specificity of any TALE. We provide SIFTED as a publicly available web tool that predicts potential genomic off-target sites for improved TALE design.

Introduction

Transcription Activator-Like Effectors (TALEs) are modular proteins that have become cornerstones of genome engineering due to their programmable DNA-binding domains. For years, the field operated under the "canonical code" model, which posits a simple one-to-one correspondence between Repeat Variable Diresidues (RVDs) and DNA bases—for example, NI recognizing Adenine and HD recognizing Cytosine. However, while this code facilitates the design of TALEs for specific targets, it has proven insufficient for predicting off-target binding events. Traditional predictive tools often assumed that each repeat within a TALE array acted independently, failing to account for the broader protein context. This study addresses this critical gap by investigating how factors such as repeat position, protein length, and neighboring sequences influence binding specificity.

Methods and the SIFTED Model

To move beyond the limitations of the simple code, the researchers employed a high-throughput, quantitative approach using custom-designed Protein-Binding Microarrays (PBMs). They profiled 21 representative TALE proteins against approximately 200,000 unique DNA sequences. This methodology allowed for the precise measurement of binding affinities across a vast landscape of potential targets and mismatches, providing the data necessary to derive energy-based Position Weight Matrices (PWMs).

Overview of the experimental design, data generation, and predictive modeling workflow.

The resulting data informed the development of SIFTED (Specificity Inference For TAL-Effector Design), a computational suite utilizing Elastic Net regularization. Unlike previous models, SIFTED explicitly incorporates context-dependent features:

  • Protein Length: Accounting for the total number of repeats.
  • Positionality: Recognizing that a repeat's location within the array affects its contribution to binding.
  • Neighbor Effects: Modeling the influence of immediately adjacent N-terminal and C-terminal repeats on a given RVD's specificity.

Key Insights: The Complexity of TALE Binding

The study's findings demonstrate that the "simple code" is an oversimplification. Key insights include:

  • Context-Dependent Specificity: The specificity of an RVD like HD is not fixed; it varies significantly based on its position in the protein and the identity of its neighbors.
  • Length and Specificity Trade-offs: Longer TALE proteins generally exhibit lower overall specificity. As the number of repeats increases, the protein becomes more tolerant of mismatches, likely due to an accumulation of total binding energy.
  • Polarity Effects: Mismatches at the 3' end of the DNA target (corresponding to the C-terminal end of the TALE) are typically better tolerated than those at the 5' end.
  • A-run Targeting: The research found that TALEs can target sequences of multiple Adenines (A-runs) with unexpectedly high specificity, challenging some previous design assumptions.

Results and Validation

SIFTED represents a significant advancement over existing tools such as PROGNOS, TALE-NT 2.0, and TALgetter. The model's PWMs predicted PBM probe intensities with a high degree of accuracy (median $R^2 = 0.877$). Furthermore, in validations against cell-based TALEN nuclease activity data, SIFTED demonstrated superior sensitivity and specificity in identifying genomic off-target sites across various thresholds.

Conclusion

By quantifying the influence of protein context on TALE-DNA binding, this study provides a more accurate framework for genome engineering. While intrinsic binding affinity is a primary driver, the authors note that future models must also consider cellular factors like chromatin accessibility. SIFTED is now available as a public web tool, offering researchers a more robust method for designing TALEs with minimized off-target risks.

Summary updated

Abstract

Transcription activator-like effector (TALE) proteins recognize DNA using a seemingly simple DNA-binding code, which makes them attractive for use in genome engineering technologies that require precise targeting. Although this code is used successfully to design TALEs to target specific sequences, off-target binding has been observed and is difficult to predict. Here we explore TALE-DNA interactions comprehensively by quantitatively assaying the DNA-binding specificities of 21 representative TALEs to ∼5,000-20,000 unique DNA sequences per protein using custom-designed protein-binding microarrays (PBMs). We find that protein context features exert significant influences on binding. Thus, the canonical recognition code does not fully capture the complexity of TALE-DNA binding. We used the PBM data to develop a computational model, Specificity Inference For TAL-Effector Design (SIFTED), to predict the DNA-binding specificity of any TALE. We provide SIFTED as a publicly available web tool that predicts potential genomic off-target sites for improved TALE design.
BibTeX
@inproceedings{rogers2015context,
  author = {Julia M. Rogers and Luis Barrera and Deepak Reyon and Jeffry D. Sander and Manolis Kellis and J. Keith Joung and Martha L. Bulyk},
  title = {{Context influences on TALE–DNA binding revealed by quantitative profiling}},
  booktitle = {Nature Communications},
  year = {2015},
  doi = {10.1038/ncomms8440},
}
Figure