Programmable protein stabilization with language model-derived peptide guides

Abstract

Dysregulated protein degradation via the ubiquitin-proteasomal pathway can induce numerous disease phenotypes, including cancer, neurodegeneration, and diabetes. While small molecule-based targeted protein degradation (TPD) and targeted protein stabilization (TPS) platforms can address this dysregulation, they rely on structured and stable binding pockets, which do not exist to classically "undruggable" targets. Here, we expand the TPS target space by engineering "deubiquibodies" (duAbs) via fusion of computationally-designed peptide binders to the catalytic domain of the potent OTUB1 deubiquitinase. In human cells, duAbs effectively stabilize exogenous and endogenous proteins in a DUB-dependent manner. Using protein language models to generate target-binding peptides, we engineer duAbs to conformationally diverse target proteins, including key tumor suppressor proteins p53 and WEE1, and heavily-disordered fusion oncoproteins, such as PAX3::FOXO1. We further encapsulate p53-targeting duAbs as mRNA in lipid nanoparticles and demonstrate effective intracellular delivery, p53 stabilization, and apoptosis activation, motivating further in vivo translation.

Fig. 1. Engineering of the duAb architecture.

A Building from prior work, a YFP nanobody (YFP Nb) was linked to potent deubiquitinase catalytic domains using different linker candidates. Created in BioRender. Hong, L. (2025) https://BioRender.com/v65m595. B KCNQ1-YFP stabilization by YFP Nb-based stabilizers in HEK293T cells determined by flow cytometric analysis. Cells were co-transfected with a pcDNA3-Nedd4L vector in the presence or absence of 4 μM PR-619 DUB inhibitor as indicated. Data are the average of independent replicates (n = 3). L1 = GAPGSG, L2 = GSGSG. (C) KCNQ-YFP stabilization by YFP Nb-based stabilizers, specifically comparing the YFP Nb-L2-OTUB1 fusion with the OTUB1 C91S and OTUB1 D88A/C91S/H265A (ASA) mutants. Cells were co-transfected with a pcDNA3-Nedd4L vector. Data are the average of individual replicates (n = 3). For B, C, normalized cell fluorescence was calculated by dividing the percentage of YFP+ cells in a sample by that of (-) DUB with no DUB inhibitor for control cells. Statistical analysis was performed using the two-tailed Student’s t-test using GraphPad Prism 10 software, with calculated p values are represented as follows: *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****, p ≤ 0.0001. Samples with p value representations above their respective bars reflect comparisons between the control and that sample; all other p value notations compare those specific samples. Please refer to source data for numeric p values.

Fig. 2. Engineering of the duAb architecture using target-specific peptides.

A Instead of using a YFP Nb, which is not a therapeutically relevant binder, target-specific peptides can instead be leveraged for a more programmable method of protein stabilization. Created in BioRender. Hong, L. (2025) https://BioRender.com/v65m595. B β-catenin-sfGFP stabilization in HEK293T cells comparing the four different DUB domain candidates linked to βcat_SnP_7 measured by flow cytometric analysis. Cells were transiently transfected in the presence or absence of 4 μM PR-619 DUB inhibitor. Data are the average of independent replicates (n = 3). C β-catenin-sfGFP stabilization by βcat_SnP_7-linked stabilizers, specifically comparing the βcat_SnP_7-L2-OTUB1 fusion with the OTUB1 C91S and OTUB1 ASA mutants. Data are the average of individual replicates (n = 3). D Time-course experiment demonstrates that potent duAb activity can be achieved within three days of treatment. Data was collected by extracting treated HEK293T cells at t = 6, 12, 24, 36, 48, and 72 h post transfection. E TOP-GFP assay for quantifying Wnt signaling in HEK293T cells. Stabilization of endogenous β-catenin results in higher levels of Wnt signaling and increased GFP levels, measured by flow cytometry. Data are the average of independent replicates (n = 3). Created in BioRender. Hong, L. (2025) https://BioRender.com/v65m595. F TOP-GFP signals in HEK293T cells measured by flow cytometric analysis. Cells were transiently transfected in the presence or absence of 4 μM PR-619 DUB inhibitor. Data are the average of independent replicates (n = 3). G TOP-GFP signals in HEK293T cells comparing the βcat_SnP_7-L2-OTUB1 fusion with the OTUB1 C91S and OTUB1 ASA mutants measured by flow cytometric analysis. Cells were transiently transfected, and data are the average of independent replicates (n = 3). For B, G, normalized cell fluorescence was calculated by dividing the percentage of sfGFP+ cells in a sample by that of (-) DUB with no DUB inhibitor for control cells. Statistical analysis was performed using the two-tailed Student’s t-test using GraphPad Prism 10 software, with calculated p values are represented as follows: *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. Samples with p value representations above their respective bars reflect comparisons between the control and that sample; all other p value notations compare those specific samples. H Nano LC-MS/MS analysis of total proteins collected from HEK293T cells co-transfected with plasmids encoding β-catenin-sfGFP and either βcat_SnP_7-L2-OTUB1 or polyG-L2-OTUB1. Data was log2-transformed, and a t-test was performed to generate a volcano plot of differentially abundant proteins. Most notably, both exogenous β-catenin-sfGFP (CTNNB1GFP) and endogenous β-catenin (CTNNB1) were among the few proteins that were abundantly present in the β-catenin-stabilizing duAb samples over the non-targeting duAb control. I Overexpressed β-catenin-sfGFP (CTNNB1GFP) abundances comparing non-targeting vs. β-catenin-stabilizing duAb treatment in HEK293T cells. J Endogenous β-catenin (CTNNB1) abundances comparing non-targeting vs. β-catenin-stabilizing duAb treatment in HEK293T cells. For I, J, statistical analysis was performed using the one-tailed Student’s t-test using GraphPad Prism 10 software, with calculated p values are represented as follows: *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. Please refer to source data for numeric p values.

Fig. 3. Programmable target stabilization via language model-derived peptides.

A Programmable target stabilization via language model-derived peptides. Created in BioRender. Chatterjee, P. (2025) https://BioRender.com/h25h541. B FOXP3-mCherry stabilization in HEK293T cells. Cellular mCherry fluorescence was measured by flow cytometry in independent replicates (n = 3). Normalized cell fluorescence was calculated by dividing the percentage of mCherry+ cells in a sample by that of control cells expressing a duAb vector expressing a non-specific poly-glycine (polyG) control peptide sequence. C Stabilization of endogenous WEE1 in protein extracts of HepG2 cells analyzed by immunoblotting. Cells were transiently transfected with a pcDNA3 plasmid encoding a polyG-OTUB1 control or one of the peptide-guided OTUB1 constructs as indicated. Transient transfection with an empty pcDNA3 plasmid served as an additional control. Blots were probed with anti-WEE1 and anti-GAPDH antibodies and are representative of biological replicates (n = 3) and technical replicates (n = 2) with similar results. D Stabilization of endogenous PAX3::FOXO1 in protein extracts of RH4 cells analyzed by immunoblotting. RH4 cells were transiently transfected with a pcDNA3 plasmid encoding one of the candidate duAbs while transfection with a polyG peptide-guided duAb served as a control. Blots were probed with anti-FOXO1 and anti-GAPDH antibodies and are representative of biological replicates (n = 3). For all immunoblots in (C) and (D), an equivalent amount of protein was loaded in each lane. Molecular weight (MW) ladder is indicated at left. Intensity of target protein bands was calculated via densitometry and normalized to intensity of GAPDH loading control and then normalized to polyG-OTUB1 control. Data are the average of biological replicates and technical replicates (n = 3 for WEE1 and PAX3::FOXO1). Statistical analysis for this figure was performed using the two-tailed Student’s t-test using GraphPad Prism 10 software, with calculated p values are represented as follows: *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. The p values above each bar in the fold stabilization and densitometry analyses represent the comparison between the control (polyG-OTUB1, no DUB inhibitor) and the respective sample; all other p value notations compare the specified samples. Please refer to source data for numeric p values. All structures were predicted via the AlphaFold3 server, and the shading was done according to AlphaFold’s confidence metric, plDDT, as follows: Very low (plDDT <50) = Orange, Low (70 > plDDT > 50) = Yellow, Confident (70 > plDDT > 90) = Light Blue, Very high (plDDT > 90) = Light Blue.

Fig. 4. Stabilization of p53 and activation of apoptosis via LNPs encapsulating p53-targeting duAbs.

A Programmable design of p53-targeting duAbs, encapsulation and delivery via mRNA-encapsulated lipid nanoparticles (LNPs), and downstream apoptosis activation. Created in BioRender. Wang, T. (2025) https://BioRender.com/x61p552. B Stabilization of endogenous p53 in protein extracts of HeLa cells analyzed by immunoblotting. HeLa cells were transiently transfected with a pcDNA3 plasmid encoding one of the candidate duAbs while transfection with a polyG peptide-guided duAb served as a control. An equivalent amount of protein was loaded in each lane. Blots were probed with anti-p53 and anti-GAPDH antibodies and are representative of biological replicates (n = 3). C Stabilization of endogenous p53 in protein extracts of HeLa cells after the best p53-stabilizing duAb (p53_pMLM_4-OTUB1) was delivered via LNPs analyzed by immunoblotting. HeLa cells were transiently transfected with LNPs encapsulating p53_pMLM_4-duAbs encoded as mRNA (loaded 1 μg and 2 μg, respectively) while transfection with luciferase-encoding mRNA-LNP served as a control. An equivalent amount of protein was loaded in each lane. Blots were probed with anti-p53 and anti-Vinculin antibodies and are representative of biological replicates (n = 3). D Increase in apoptosis hallmark cleaved-PARP-1 (Cl-PARP-1) in protein extracts of HeLa cells after the best p53-stabilizing duAb (p53_pMLM_4-OTUB1) was delivered via LNPs analyzed by immunoblotting. HeLa cells were transiently transfected with LNPs encapsulating p53_pMLM_4-duAbs encoded as mRNA (loaded 1 μg and 2 μg, respectively) while transfection with luciferase-encoding mRNA-LNP served as a control. An equivalent amount of protein was loaded in each lane. Blots were probed with anti-Cl-PARP-1 and anti-Vinculin antibodies and are representative of biological replicates (n = 3). For all immunoblots in (B–D), an equivalent amount of protein was loaded in each lane. Molecular weight (MW) ladder is indicated at left. Intensity of target protein bands was calculated via densitometry and normalized to intensity of GAPDH and Vinculin loading controls and then normalized to applicable controls (polyG-OTUB1 for (B) and luciferase LNP for (C) and (D)). Data are the average of biological replicates and technical replicates (n = 3 for p53 and Cl-PARP-1). Statistical analysis for this figure was performed using the two-tailed Student’s t-test using GraphPad Prism 10 software, with calculated p values are represented as follows: *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. The p values above each bar in the fold stabilization and densitometry analyses represent the comparison between the control (polyG-OTUB1, luciferase LNP) and the respective sample; all other p value notations compare the specified samples. Please refer to source data for numeric p values. The structure for p53 was predicted via the AlphaFold3 server, and the shading was done according to AlphaFold’s confidence metric, plDDT, as follows: very low (plDDT <50) = orange, low (70 > plDDT > 50) = yellow, confident (70 > plDDT > 90) = light blue, very high (plDDT > 90) = light blue.