Design and Evaluation of PROTACs Targeting Acyl Protein Thioesterase 1

PROTAC linker design remains mostly an empirical task. We employed the PRosettaC computational software in the design of sulfonyl‐fluoride‐based PROTACs targeting acyl protein thioesterase 1 (APT1). The software efficiently generated ternary complex models from empirically‐designed PROTACs and suggested alkyl linkers to be the preferred type of linker to target APT1. Western blotting analysis revealed efficient degradation of APT1 and activity‐based protein profiling showed remarkable selectivity of an alkyl linker‐based PROTAC amongst serine hydrolases. Collectively, our data suggests that combining PRosettaC and chemoproteomics can effectively assist in triaging PROTACs for synthesis and providing early data on their potency and selectivity.

PROTAC linker design remains mostly an empirical task.We employed the PRosettaC computational software in the design of sulfonyl-fluoride-based PROTACs targeting acyl protein thioesterase 1 (APT1).The software efficiently generated ternary complex models from empirically-designed PROTACs and suggested alkyl linkers to be the preferred type of linker to target APT1.Western blotting analysis revealed efficient degra-dation of APT1 and activity-based protein profiling showed remarkable selectivity of an alkyl linker-based PROTAC amongst serine hydrolases.Collectively, our data suggests that combining PRosettaC and chemoproteomics can effectively assist in triaging PROTACs for synthesis and providing early data on their potency and selectivity.
Proteolysis-targeting chimeras (PROTACs) are molecules with heterobifunctional properties that target proteins for proteasomal degradation by bringing them in close proximity to an E3 ligase.This leads to the formation of a ternary complex which facilitates ubiquitination and subsequent degradation of the target protein. [1]PROTAC design is still highly empirical, especially in regards to linker length, composition, flexibility and attachment points. [2]he importance of optimizing PROTAC linkers has been widely demonstrated.Burslem and coworkers successfully modulated the selectivity of a lapatinib-based PROTAC by increasing the linker length by one PEG unit.The longer PROTAC selectively degraded EGFR over HER2, while the shorter version degraded both. [3]Smith and coworkers developed isoform-selective PROTACs for the p28 MAPK family by varying linker length and attachments. [4]Additionally, the first PROTACs to enter clinical trials, ARV-110 and ARV-471, both contain more unusually rigid piperidine/pyrazine-based linkers. [5]The optimal linker is likely impossible to generalize and highly dependent on each protein:E3 ligase pair.
Computational modelling could assist PROTAC design and significantly reduce synthetic efforts, especially concerning linker design, which typically involves synthesis of numerous linkers of various types and lengths. [6]One such software is PRosettaC, [7] which performs a holistic protocol for modelling PROTAC-mediated ternary complexes with a workflow aimed at decreasing the complex conformational search space resulting from protein-protein interactions, PROTAC conformations and its positioning relative to the protein-protein complex. [7]RosettaC has successfully predicted ternary complexes for PROTACs that were then validated experimentally.In addition to typical western blotting assays to study degradation, chemoproteomic techniques such as activity-based protein profiling (ABPP) [8] can provide an accessible outlet for early evaluation of these molecules in an enriched pool of related targets which, while not comprising a full degradome, is still highly valuable data to complement classic western blotting analysis of protein degradation. [9]n this work, we sought to incorporate the use of computational tools and ABPP in the design and evaluation of PROTAC linkers for APT1, a serine hydrolase predominantly localized in the cytoplasm that specifically hydrolyzes fatty acyl groups attached to cysteine residues in proteins. [10]Altered APT1 activity has been associated with cancer, neurodegenerative diseases, and cardiovascular disorders. [11]12b] Our integrated platform combines computational evaluation of PROTACs using the PRosettaC software with chemical synthesis and subsequent evaluation of the PROTACs via western blotting to analyze degradation and ABPP for early data on inhibition potency and selectivity within serine hydrolases (Figure 1).PRosettaC successfully generated ternary complex models and predicted the best type of linker to use, with some limitations when analyzing the size of the best linkers.Our results suggest alkyl linkers are ideal to target APT1 when compared with other linker types, potentially due to the lipidlike characteristics of the alkyl chain.ABPP experiments showed remarkable selectivity for APT1 amongst serine hydrolases.This work provides valuable data on PROTAC design and highlights the importance of linker design in achieving target selectivity amongst closely related enzymes and how challenging the modelling of ternary complexes still remains in the current PROTAC landscape.
PROTAC design took several considerations into account.We used pomalidomide as a binder of the E3 ligase cereblon (CRBN), [13] a sulfonyl fluoride as reactive group and varying types of linkers.While the use of a covalent reactive group limited the possibility of a catalytic mechanism, [14] it allowed us to easily employ ABPP techniques in the study of our PROTACs.The first series of compounds used alkyl chains as the main linker moiety (Figure 2).Attachment to pomalidomide was achieved through a pyrazine linker, which has been suggested to diminish pomalidomide side effects, [15] and through a triazole to the reactive group, resulting from the use of click chemistry.
A straightforward modular synthetic route was designed to allow quick generation of the library of compounds using [3 + 2] azide-alkyne Huesgen cycloadditon and peptide coupling chemistries (Figure 2).Slight variations were employed for specific PROTACs.
These compounds were modelled with APT1 and CRBN using PRosettaC.We started by running the program on the APT1 structure (PDB: 6QGS) with the PROTAC containing the longest alkyl chain (C8; Figure 2) since we assumed the longest PROTAC in the series would have a good probability of success.This run produced 200 ternary complex models, including a promising 1 st cluster, which contained 53 % of the results (Supporting Information (SI, Figure S1).Following up on the original PRosettaC publication methodology, the output of the platform consists of clusters that are ordered by the number of solutions they include, with the 1 st cluster being the most populated one.In the case of the C8 PROTAC, the fact that the resulting models converged mostly in the 1 st cluster was a promising outlook.According to the PRosettaC protocol, this  suggests that the model is a low energy solution with a good probability to translate to protein degradation.
9a] Calculating the minimum pair-wise RMSD of the APT1 backbone carbon α atoms following superposition of cereblon in the models revealed a value of 0.17 Å between the 1 st clusters of C8 and the FP-PROTAC, suggesting they represent the same solution, a promising result given the high degree of similarity between both proteins.
We then ran PRosettaC on the remaining series of PROTACs containing alkyl linkers.PRosettaC successfully generated solutions for PROTACs C4-C7 with their 1 st cluster representing the same solution as the 1 st cluster of C8 (for example C4: 0.23 Å, C5: 0.1 Å, C7: 0.19 Å, calculated in the same manner as described above for the FP-PROTAC) (Figure 3A and S5-8).This suggested PROTACs with an alkyl linker longer than 4 carbons may be able to form a ternary complex without steric clashes between CRBN and APT1.Compounds with an alkyl linker shorter than 4 carbons did not generate ternary complex models with PRosettaC.This could be due to the small size of the linker limiting ternary complex formation or due to limitations of the software due to the simplifications it applies in its protocol.Analogous experiments for compound C8 were done for APT2, generating 70 models, with 14 converging in the 1 st cluster (Figure S30).
We then performed a re-clustering of the main solutions of PROTACs C4, C5, C7, C8 together into master models (SI section I.I.).We assessed the ability of the chosen models to accommodate C1-C8 using RDKit and Rosetta.Surprisingly, the results showed that almost all pairs of PROTAC-Model can produce a ternary complex without significant clashes, suggesting linker size might not be as prohibitive as the PRosettaC modelling suggested and that all PROTACs in the range C1-C8 could fit in the ternary complex models.
We then designed PROTACs with more complex linkers using reclustered PRosettaC models created from this first series of PROTACs (Figure 3D-E and Scheme S2).These PROTACs contained additional chemical moieties selected by screening a custom-made library made from commercially available building blocks.In addition to these we empirically designed PROTACs containing linkers with other chemical moieties like aromatic rings and additional amide bonds or small spacers (P5-7), and a potentially reversible PROTAC based on an APT1 inhibitor (P8).
With the compounds in hand, we sought to study both the inhibition and degradation of APT1 using ABPP and western blotting (Figure 4).To get preliminary insight into the potency and selectivity of the compounds we performed competitive gel-based ABPP.Briefly, the series of PROTACs were incubated with HEK293T cell lysates where APT1 was overexpressed, and then general serine hydrolase labelling was performed with a fluorophosphonate-rhodamine probe (Figure 4B).
The alkyl series of PROTACs generally showed good inhibition of APT1 across the family of compounds.Most compounds displayed an IC 50 within the low micromolar range, with some like C1 and C0 inhibiting APT1 at submicromolar concentrations (Figure 4B, D).Competitive ABPP with compound C8 in whole cells suggested a good degree of selectivity within the serine hydrolase family of enzymes, with only a band at approximately 25 kDa showing reduction of fluorophosphate probe labelling (Figure S35).
Compounds P1-4 were shown to be inactive against APT1, having no significant effect on FP binding, which suggested that the increased rigidity of the compounds or the potential hydrogen bonds with surface aminoacids did not benefit binding of the PROTAC to APT1 (Figure S31, S33).The compounds containing linkers with added chemical elements like phenyl, amide and short alkyl spacers (P5-6) revealed moderate inhibition of APT1 (IC 50 = 8.3 and ~15 μM, respectively).This was also observed for the compound with a piperidine-based semi-rigid linker (P7, ~5 μM IC 50 ).Modulating the flexibility of the linker seems to be an important factor to influence APT1 inhibition and potentially degradation.Analogous experiments were performed for APT2, showing a similar range of inhibition potencies, from submicromolar (C0, C1, C3, etc.) to low micromolar (C-1, C5, C7, etc.) (Figure S36).
Competitive ABPP experiments cannot differentiate between inhibition and degradation, reporting only on the interference with the enzyme's catalytic machinery by measuring reduction of APT1 labelling by the fluorophosphonate probe.To analyse the degradation of endogenous APT1 we used western blot.We started by testing the model compound C8 and observed significant degradation of APT1 (DC 50 = 0.13 μM) (Figure 4 and Figure S32).Importantly, degradation of APT1 was abrogated by pre-treatment with the proteasome inhibitor carfilzomib and the neddylation inhibitor MLN4924, confirming the observed loss of signal occurs due to proteasome-and E3 ligase-mediated degradation of APT1 (Figure 4).Further confirmation of CRBN requirement for degradation was observed using MOLT-4 CRBNÀ /À cells.Whereas treatment of wildtype MOLT-4 with C8 promoted degradation of APT1, exposure of CRBN knockout cells to C8 resulted in no APT1 degradation (SI, Figure S32).2a] Some of the shorter compounds in the alkyl series, like C0, showed little to no degradation of APT1, while having the best APT1 inhibition value measured by competitive ABPP.This result was to be expected since shorter compounds could potentially easily engage APT1, while more likely struggling to find conformations that allow formation of the ternary complex with CRBN.Remarkably, compound C1 was able to efficiently degrade APT1 (DC 50 < 0.1 μM), revealing that the short linker might not be as prohibited for the degradation mechanism as the computational studies suggested.
Compounds P1-4 did not induce degradation of APT1, which is consistent with the observations from competitive ABPP that revealed no inhibition of the enzyme.Degradation was observed for some of the compounds with linker motifs that modulate flexibility (P5 and P7, DC 50 ~1 μM) and the noncovalent compound (P8, DC 50 ~5 μM), but less pronounced than that observed with some alkyl linker compounds.
Overall, selected members of our PROTAC library successfully inhibit APT1 and induce its degradation via a proteasomedependent mechanism.The experimental results mostly agreed with the PRosettaC modelling when it comes to the type of linker, with alkyl compounds providing the best results both in PRosettaC and experimentally, but prediction of the effect of linker length was less straightforward.Short compounds like C1 and C3 were shown to be good APT1 degraders despite not generating PRosettaC models, while some of the longer compounds with good PRosettaC results like C7 performed poorly in terms of APT1 degradation.On the other hand, the computationally designed P1-P4 failed to degrade APT1, consistent with the results of PRosettaC, where no models were generated for these compounds.Successful degradation of a target requires recruitment of that same target, which the competitive ABPP assays showed P1-P4 failed to do.Some of the observed discrepancies could also be explained by the inherent limitations and simplifications related with the PRoset-taC platform.
Following the promising results obtained with the gel-based ABPP and western blotting we were intrigued by the identities of proteins being inhibited/degraded by our PROTACs and the apparent selectivity amongst serine hydrolases.To get a deeper insight into the PROTACs effects on serine hydrolases we used mass-spectrometry-based ABPP against the serine hydrolase probe fluorophosphonate-biotin.Briefly, Raji cells were treated with PROTACs C8, P5 and P8.These compounds were chosen to represent different types of linkers that were used in the study.General serine hydrolase labelling was then performed using fluorophosphonate-biotin.Probe-labelled proteins were enriched and analysed by LC-MS/MS.An inhibition ratio was calculated by comparing the samples with a DMSO-treated sample using a reductive demethylation isotopic labelling technique (Figure 5A).Importantly, while PROTAC C1 showed higher potency than C8 in terms of APT1 degradation, we chose C8 as a representative of the alkyl series of compounds due to the known preference of APT1 for substrates with long alkyl chains.
A very selective profile of serine hydrolase inhibition was observed for compound C8, with APT1 emerging as the main target of the compound.A few other serine hydrolases were labelled to a lesser extent.Interestingly, this group of enzymes, which included ABHD6, APT2, ABHD11 and LYPLAL1, are all annotated as lipid-metabolizing enzymes, suggesting a specific target profile for this compound.As we suggested before, this could be explained by the long alkyl linker mimicking a lipid and being recognized by the substrate-binding pockets in APT1 and a few other lipid-metabolizing serine hydrolases.This is consistent with the loss of selectivity for compounds P5 and P8, which did not show any specific serine hydrolase to be inhibited in a significant way.These results also agree with previous observations where the alkyl linker-containing PRO-TACs performed the best both in competitive ABPP and western blotting and further suggest that the hydrophobic interactions between APT1 and the PROTACs are a more important characteristic in inhibitor design than seeking specific interactions with surface amino acids or modulating rigidity and flexibility.APT1 has been shown to preferably bind substrates with alkyl chains on its inhibitors and show poor results with structures like PEG linkers.9a] The preference of these enzymes for alkyl motifs can also be inferred from the structure of known APT1 inhibitors like palmostatin B, [16] and the previous use of a simple alkyl chain decorated with a sulfonyl fluoride to target several lipid metabolizing enzymes. [17]ollowing these promising results we then tested the ability of C8 to induce cytotoxicity on viable hematological cancers (Figure 5B).C8 had pronounced effect in inhibiting the proliferation of acute lymphoblastic and myeloid leukaemia (MOLT-4 and MOLM-13) with IC 50 values of 2.9 and 9.7 μM, respectively.In acute promyelocytic leukaemia (HL-60) and in Burkitt's lymphoma (Raji) the values were 5.5 μM and 16.8 μM, respectively.Importantly, non-tumoral HEK239T cells were less sensitive to treatment (IC 50 of 35.4 μM).While promising, further studies are warranted to verify that these effects are the result of APT1 or other serine hydrolase inhibition and not related to additional targets of the sulfonyl fluoride moiety or other effects of the compounds.
The present work effectively demonstrates the contribution of software like PRosettaC in PROTAC selection and design, especially in regard to the type of linker, but also reveals their limitations.PRosettaC predictions mostly agreed with results from western blotting and ABPP assays in terms of linker type, with a few exceptions when it comes to determining ideal linker length.The use of ABPP in PROTAC evaluation was invaluable in quickly demonstrating the effectiveness of a long alkyl linker compound in inhibiting APT1 and its selectivity amongst the serine hydrolase family, specifically for lipid- metabolizing serine hydrolases.Gel-based ABPP helped predict some of the best performing compounds, with a few exceptions like compounds with very short linkers, where despite a high rate of enzymatic inhibition, this does not translate into proteasomal-dependent degradation.When combined with proper degradation studies via western blotting it establishes a strong platform for the evaluation of PROTAC candidates.Integrating this work with existing literature and knowledge of PROTACs in clinical trials further highlights the challenge of generalizing linker design, suggesting a target-dependent individual approach supported by computational modelling tools and chemoproteomics could be the path to success in this ever-evolving chemical biology field.

Figure 1 .
Figure 1.Overview of the integrated PROTAC design and evaluation platform.

Figure 2 .
Figure 2. A. Model compound and alkyl series of sulfonyl fluoride PROTACs; B. PROTAC C-1. C. General synthetic route used in the synthesis of the PROTACs.

Figure 3 .
Figure 3. A. PRosettaC results for the alkyl series of PROTACs.B. Reclustering of the main results into ternary complex master models.C. Docking of compounds C1-C8 into the reclustered master models.D. PROTACs with linkers designed from commercial building blocks and fitted into the PRosettaC reclustered models (P1-P4).E. PROTACs designed with additional handles to increase or decrease flexibility (P5-P7); non-covalent PROTAC (P8).

Figure 4 .
Figure 4. Evaluation of inhibition and degradation of APT1 by the PROTACs. A. The evaluation of our molecules involved three main steps: gel-based ABPP (A.1), MS-based ABPP (A.2) and western blot (A.3).B. Gel-based ABPP experiments using the PROTAC library using overexpressed APT1 in HEK293T cells (fluorophosphonate-rhodamine probe used at 1 μM).C. Degradation analysis of endogenous APT1 in Raji cells by western blot.D. Summary table comparing competitive ABPP, western blotting and PRosettaC results.E. Western blot analysis after a 4 h treatment with benzyl sulfonyl fluoride and pomalidomide.F.Degradation of APT1 after a 1 h pretreatment with carfilzomib, followed by a 4 h-treatment with C8.G. Degradation of APT1 after a 1 h pretreatment with MLN4924, followed by a 4 h C8 treatment.

Figure 5 .
Figure 5. A. Mass spectrometry-based ABPP of Raji cells treated with selected PROTACs; serine hydrolases were enriched using a fluorophosphonate-biotin probe.A higher SILAC ratio, depicted in green, represents a higher level of inhibition and/or degradation.B. Cell viability assays using compound C8.