Activity-Stability Tradeoffs via Enzyme Proximity Sequencing

Summary¶

Vanella et al. (2024) developed enzyme proximity sequencing (EP-Seq), a high-throughput deep mutational scanning (DMS) platform using yeast surface display and peroxidase-mediated tyramide radical proximity labeling to decouple folding stability from catalytic activity.[1] Applied to a comprehensive single-substitution library (6,399 missense mutations) of Rhodotorula gracilis D-amino acid oxidase (DAOx, an 80 kDa homodimer), EP-Seq quantified pervasive activity-stability tradeoffs across the active site and dimer interface while identifying distant, non-destabilizing activity hotspots.[1]

EP-Seq Experimental Workflow & Mathematical Formulation¶

EP-Seq processes a pooled enzyme library via two parallel 4-bin FACS sorting branches following galactose-induced yeast surface display (Aga2-DAOx fusion construct):[1]

                ┌─────────────────────────────────────────────────────────────┐
                │        Barcoded Yeast Display Mutant Library (6,399)        │
                └──────────────────────────────┬──────────────────────────────┘
                                               │
                       ┌───────────────────────┴───────────────────────┐
                       ▼                                               ▼
         ┌───────────────────────────┐                   ┌───────────────────────────┐
         │ Expression / Display      │                   │ Catalytic Activity        │
         │ (Anti-6xHis + AF594 Ab)   │                   │ (DAOx + H2O2 + HRP +      │
         │                           │                   │  Tyramide-AF488 Phenoxyl) │
         └─────────────┬─────────────┘                   └─────────────┬─────────────┘
                       │                                               │
                       ▼                                               ▼
         ┌───────────────────────────┐                   ┌───────────────────────────┐
         │ 4-Bin FACS Sorting        │                   │ 4-Bin FACS Sorting        │
         │ (Gate 1: 99% Neg, G2-G4)  │                   │ (Gate 1: 99% Neg, G2-G4)  │
         └─────────────┬─────────────┘                   └─────────────┬─────────────┘
                       │                                               │
                       └───────────────────────┬───────────────────────┘
                                               │
                                               ▼
                ┌─────────────────────────────────────────────────────────────┐
                │ PacBio Long-Read (UMI-Variant Lookup Table, Q20 Filter)     │
                │ + Illumina NovaSeq 6000 UMI Bin Counting                    │
                └──────────────────────────────┬──────────────────────────────┘
                                               │
                                               ▼
                ┌─────────────────────────────────────────────────────────────┐
                │ Weighted Mean Fluorescence (β_v) & Fitness Score (F_v)     │
                │ F_v = β_v / β_wt                                           │
                └─────────────────────────────────────────────────────────────┘

Single-Cell Proximity Labeling Reaction¶

Displayed DAOx variants oxidize D-alanine to produce $\text{H}_2\text{O}_2$, which fuels horseradish peroxidase (HRP)-mediated oxidation of tyramide-Alexa Fluor 488 into short-lived phenoxyl radicals.[1] Diffusion limitations confine covalent labeling to the cell wall of the expressing cell, establishing a single-cell catalytic readout.[1]

Mathematical Fitness Scoring¶

For each sorted bin $i$, cell count $c_{v,i}$ is estimated from raw sequencing reads $r_{v,i}$:[1] $$c_{v,i} = r_{v,i} \times \left( \frac{c_{\text{tot},i}}{r_{\text{tot},i}} \right)$$ Weighted mean fluorescence $\beta_v$ and consensus fitness $F_v$ relative to wild-type $\beta_{\text{wt}}$ are derived across bins using gate median fluorescence values $\omega_i$:[1] $$\beta_v = \frac{\sum_{i=1}^{4} c_{v,i} \omega_i}{\sum_{i=1}^{4} c_{v,i}}, \quad F_v = \frac{\beta_v}{\beta_{\text{wt}}}$$

High Biological Reproducibility & Monogenic Validation¶

Sequencing Reproducibility: High-throughput sequencing of 15-nucleotide UMIs across sorted bins yielded Pearson correlations of $r = 0.94$ (expression) and $r = 0.96$ (activity) between independent biological replicates.[1]
Monogenic Clonal Validation: 12 single missense variants (S48C, L153R, A187K, A187E, G199E, Q200W, S215A, T237M, S268Y, P292E, L310P, G315P) were synthesized, expressed, and evaluated independently.[1] Measured fitness scores correlated tightly with bulk flow cytometry display ($r = 0.89$) and fluorometric Amplex Red D-alanine oxidation activity assays ($r = 0.95$).[1]

Quantitative Activity-Stability Tradeoff Mapping¶

Analysis of 6,399 single missense variants mapped distinct biophysical tradeoffs across DAOx structural domains:[1]

Active Site & FAD Pocket (< 4 Å): Mutations within 4 Å of the FAD cofactor severely impaired catalytic activity (average activity score $-66\%$ relative to baseline, $-0.557$ vs. $-0.336$) while improving surface display/folding stability ($+17\%$, $-0.149$ vs. $-0.180$).[1]
Substrate Binding Pocket (D-Alanine): Mutations contacting D-alanine reduced activity by $36\%$ ($-0.457$) while boosting expression by $+60\%$ ($-0.060$).[1]
Homodimer Interface (< 5 Å): Mutations at the obligate head-to-tail dimer interface reduced activity by $7\%$ ($-0.363$) while increasing expression by $+61\%$ ($-0.070$), proving that DAOx evolution accepts thermodynamic instability to maintain a catalytically active quaternary assembly.[1]
Biophysical Correlates: Solvent-accessible surface area (SASA) correlated strongly with both activity ($r = 0.64, P < 0.0001$) and expression ($r = 0.43, P < 0.0001$). Structural mobility (B-factor) correlated positively with activity ($r = 0.41$). Distance to FAD correlated positively with activity ($r = 0.57$), confirming that activity intolerance is tightly concentrated around the cofactor.[1]
FoldX Energy ΔΔG Discrepancy: FoldX $\Delta\Delta G$ calculations (PDB 1C0P) correlated moderately with measured expression ($r = 0.43$) but failed to capture catalytic activity loss ($r = 0.12$), illustrating that stability algorithms cannot predict functional catalytic impairment.[1]

Normalized Activity Deconvolution & Engineering Hotspots¶

By dividing non-logarithmic activity fitness by expression fitness, EP-Seq deconvoluted catalytic efficiency from expression level, yielding 2,029 variants with normalized activity $> 1.0$:[1]

Substrate Tunnel Gating Loop (Positions 99, 216, 217, 218): Four of the top 6 activity-enhancing positions (S216, D217, P218, and adjacent N99) form a dynamic loop at the entrance of the substrate access tunnel.[1] Mutations here enhance D-alanine access without altering active-site geometry or fold stability.[1]
Allosteric Interface Hotspots (Positions 109, 110): K109 and D110, located opposite the dimer interface (28.2 Å from the active site), induce long-range conformational adjustments that enhance catalytic turnover.[1]

Synthesis: This empirical dataset provides direct experimental validation for computational disentanglement models like DETANGO, which factorizes protein language model embeddings into stability and functional components. It also confirms the central thesis of Predicting Catalytic Competence of Enzyme-Ligand Complexes: stability-only metrics (e.g., Rosetta/FoldX $\Delta\Delta G$) systematically misclassify catalytic-site mutations — which empirically enhance expression while destroying catalysis — as neutral or beneficial.

Code & Data Availability¶

Interactive Heatmaps: Accessible at nash-lab.github.io/DAOx-DMS/heatmaps.[1]
Raw Data & Code: High-throughput PacBio and Illumina raw reads, C scripts (ppba.c, pib.c, rib.c), and lookup tables are archived on Zenodo at doi.org/10.5281/zenodo.8388902.[1]

Citations¶

[1] Vanella, R., Küng, J., Schoepfer, M., Doffini, V., Ren, J., Nash, M.A. (2024). Understanding activity-stability tradeoffs in biocatalysts by enzyme proximity sequencing. Nature Communications, 15, 1807. Supports: EP-Seq proximity labeling method; PacBio/Illumina UMI lookup table pipeline; 6,399 missense variant dataset; domain-specific tradeoff metrics (FAD pocket, D-alanine site, dimer interface); 12 single-mutant validations; FoldX $\Delta\Delta G$ comparisons; substrate gating loop and allosteric hotspots. Location: Abstract; Results ("Enzyme proximity sequencing (EP-Seq) workflow", "DAOx deep mutational scanning library", "Activity-stability tradeoffs", "Deconvolution of catalytic activity and expression", "Engineering hotspots"); Figures 1--6; Supplementary Figures 1--12; Supplementary Tables 1--5.