Learning the shape of protein microenvironments with a holographic convolutional neural network.

Proteins play a central role in biology from immune recognition to brain activity. While major advances in machine learning have improved our ability to predict protein structure from sequence, determining protein function from its sequence or structure remains a major challenge. Here, we introduce holographic convolutional neural network (H-CNN) for proteins, which is a physically motivated machine learning approach to model amino acid preferences in protein structures. H-CNN reflects physical interactions in a protein structure and recapitulates the functional information stored in evolutionary data. H-CNN accurately predicts the impact of mutations on protein stability and binding of protein complexes. Our interpretable computational model for protein structure-function maps could guide design of novel proteins with desired function.

The pathway for coenzyme M biosynthesis in bacteria.

Mercaptoethane sulfonate or coenzyme M (CoM) is the smallest known organic cofactor and is most commonly associated with the methane-forming step in all methanogenic archaea but is also associated with the anaerobic oxidation of methane to CO2 in anaerobic methanotrophic archaea and the oxidation of short-chain alkanes in Syntrophoarchaeum species. It has also been found in a small number of bacteria capable of the metabolism of small organics. Although many of the steps for CoM biosynthesis in methanogenic archaea have been elucidated, a complete pathway for the biosynthesis of CoM in archaea or bacteria has not been reported. Here, we present the complete CoM biosynthesis pathway in bacteria, revealing distinct chemical steps relative to CoM biosynthesis in methanogenic archaea. The existence of different pathways represents a profound instance of convergent evolution. The five-step pathway involves the addition of sulfite, the elimination of phosphate, decarboxylation, thiolation, and the reduction to affect the sequential conversion of phosphoenolpyruvate to CoM. The salient features of the pathway demonstrate reactivities for members of large aspartase/fumarase and pyridoxal 5'-phosphate-dependent enzyme families.

Albumin-Binding Lutetium-177-Labeled LLP2A Derivatives as Theranostics for Melanoma.

Very late antigen-4 (VLA-4) is a transmembrane integrin protein that is highly expressed in aggressive forms of metastatic melanoma. A small-molecule peptidomimetic, LLP2A, was found to have a low pM affinity binding to VLA-4. Because LLP2A itself does not inhibit cancer cell proliferation and survival, it is an ideal candidate for the imaging and delivery of therapeutic payloads. An analog of [177Lu]Lu-labeled-LLP2A was previously investigated as a therapeutic agent in melanoma tumor-bearing mice, resulting in only a modest improvement in tumor growth inhibition, likely due to rapid clearance of the agent from the tumor. To improve the pharmacokinetic profile, DOTAGA-PEG4-LLP2A with a 4-(p-iodophenyl)butyric acid (pIBA) albumin binding moiety was synthesized. We demonstrate the feasibility of this albumin binding strategy by comparing in vitro cell binding assays and in vivo biodistribution performance of [177Lu]Lu-DOTAGA-PEG4-LLP2A ([177Lu]Lu-1) to the albumin binding [177Lu]Lu-DOTAGA-pIBA-PEG4-LLP2A ([177Lu]Lu-2). In vitro cell binding assay results for [177Lu]Lu-1 and [177Lu]Lu-2 showed Kd values of 0.40 ± 0.07 and 1.75 ± 0.40 nM, with similar Bmax values of 200 ± 6 and 315 ± 15 fmol/mg, respectively. In vivo biodistribution data for both tracers exhibited specific uptake in the tumor, spleen, thymus, and bone due to endogenous expression of VLA-4. Compound [177Lu]Lu-2 exhibited a much longer blood circulation time compared to [177Lu]Lu-1. The tumor uptake for [177Lu]Lu-1 was highest at 1 h (∼15%ID/g) and that for [177Lu]Lu-2 was highest at 4 h (∼23%ID/g). Significant clearance of [177Lu]Lu-1 from the tumor occurs at 24 h (<5%ID/g) while[177Lu]Lu-2 is retained for greater than 96 h (∼10%ID/g). An efficacy study showed that melanoma tumor-bearing mice receiving compound [177Lu]Lu-2 given in two fractions (2 × 14.8 MBq, 14 days apart) had a greater median survival time than mice administered a single 29.6 MBq dose of compound [177Lu]Lu-1, while a single 29.6 MBq dose of [177Lu]Lu-2 imparted hematopoietic toxicity. The in vitro and in vivo data show addition of pIBA to [177Lu]Lu-DOTAGA-PEG4-LLP2A slows blood clearance for a higher tumor uptake, and there is potential of [177Lu]Lu-2 as a theranostic in fractionated administered doses.

Phosphorus containing analogues of SAHA as inhibitors of HDACs.

Histone deacetylases (HDACs) are a family of enzymes responsible for regulating DNA transcription by modulating its binding to histone proteins. HDACs are overexpressed in several types of cancers and are recognised as drug targets. Vorinostat, or suberanilohydroxamic acid (SAHA), is an histone deacetylase (HDAC) inhibitor with a hydroxamic acid as a zinc-binding group (ZBG), and it has been FDA approved for the treatment of T-cell lymphoma. In this work, phosphorus-based SAHA analogues were synthesised to assess their zinc-binding effectiveness compared to the hydroxamic acid of SAHA. Specifically, we examined phosphate, phosphoramidate and phosphorothiolate groups as isosteres of the canonical hydroxamic acid motif of conventional HDAC inhibitors. The compounds were screened for binding to HDAC enzymes from HeLa cell lysate. The most potent derivatives were then screened against HDAC3 and HDAC8 isoforms. HDAC inhibition assays demonstrated that these phosphorus-based SAHA analogs exhibited slow binding to HDACs but with greater potency than phosphonate SAHA analogs examined previously. All compounds inhibited HDACs, the most potent having an IC50 of 50 µM.

HERMES: Holographic Equivariant neuRal network model for Mutational Effect and Stability prediction.

Predicting the stability and fitness effects of amino acid mutations in proteins is a cornerstone of biological discovery and engineering. Various experimental techniques have been developed to measure mutational effects, providing us with extensive datasets across a diverse range of proteins. By training on these data, traditional computational modeling and more recent machine learning approaches have advanced significantly in predicting mutational effects. Here, we introduce HERMES, a 3D rotationally equivariant structure-based neural network model for mutational effect and stability prediction. Pre-trained to predict amino acid propensity from its surrounding 3D structure, HERMES can be fine-tuned for mutational effects using our open-source code. We present a suite of HERMES models, pre-trained with different strategies, and fine-tuned to predict the stability effect of mutations. Benchmarking against other models shows that HERMES often outperforms or matches their performance in predicting mutational effect on stability, binding, and fitness. HERMES offers versatile tools for evaluating mutational effects and can be fine-tuned for specific predictive objectives.

HERMES: Holographic Equivariant neuRal network model for Mutational Effect and Stability prediction.

Predicting the stability and fitness effects of amino acid mutations in proteins is a cornerstone of biological discovery and engineering. Various experimental techniques have been developed to measure mutational effects, providing us with extensive datasets across a diverse range of proteins. By training on these data, traditional computational modeling and more recent machine learning approaches have advanced significantly in predicting mutational effects. Here, we introduce HERMES, a 3D rotationally equivariant structure-based neural network model for mutational effect and stability prediction. Pre-trained to predict amino acid propensity from its surrounding 3D structure, HERMES can be fine-tuned for mutational effects using our open-source code. We present a suite of HERMES models, pre-trained with different strategies, and fine-tuned to predict the stability effect of mutations. Benchmarking against other models shows that HERMES often outperforms or matches their performance in predicting mutational effect on stability, binding, and fitness. HERMES offers versatile tools for evaluating mutational effects and can be fine-tuned for specific predictive objectives.

Evaluation of a bimodal, matched pair theranostic agent targeting prostate-specific membrane antigen.

BACKGROUND: Prostate cancer affects 1 in 6 men, and it is the second­leading cause of cancer-related death in American men. Surgery is one of the main treatment modalities for prostate cancer, but it often results in incomplete resection margins or complete resection that leads to nerve damage and undesirable side effects. In the present work, we have developed a new bimodal tracer, NODAGA-sCy7.5 PSMAi (prostate-specific membrane antigen inhibitor), labeled with the true matched theranostic pair 64Cu/67Cu and a near-infrared fluorescent dye. This agent could potentially be used for concomitant PET imaging, optical surgical navigation, and targeted radiopharmaceutical therapy. METHODS: A prostate-specific membrane antigen (PSMA)-targeting urea derivative was conjugated to NODAGA for copper radiolabeling and to the near-infrared fluorophore sulfo-Cy7.5 (sCy7.5). Binding studies were performed in PSMA-positive PC-3 PIP cells, as well as uptake and internalization assays in PC-3 PIP cells and PSMA-negative PC-3 wild type cells. Biodistribution studies of the 64Cu-labeled compound were performed in PC-3 PIP- and PC-3 tumor-bearing mice, and 67Cu biodistributions of the agent were obtained in PC-3 PIP tumor-carrying mice. PET imaging and fluorescence imaging were also performed, using the same molar doses, in the two mouse models. RESULTS: The PSMA conjugate bound with high affinity to PSMA-positive prostate cancer cells, as opposed to cells that were PSMA-negative. Uptake and internalization were rapid and PSMA-mediated in PC-3 PIP cells, while only minimal non-specific uptake was observed in PC-3 cells. Biodistribution studies showed specific uptake in PC-3 PIP tumors, while accumulation in PC-3 tumor-bearing mice was low. Furthermore, tumor uptake of the 67Cu-labeled agent in the PC-3 PIP model was statistically equivalent to that of 64Cu. PET and fluorescence imaging at 0.5 nmol per mouse also demonstrated that PC-3 PIP tumors could be clearly detected, while PC-3 tumors showed no tumor accumulation. CONCLUSIONS: NODAGA-sCy7.5-PSMAi was specific and selective in detecting PSMA-positive, as opposed to PSMA-negative, tumors in mouse models of prostate cancer. This bioconjugate could potentially be used for PET staging with 64Cu, targeted radiopharmaceutical therapy with 67Cu, and/or image-guided surgery with sCy7.5.

H-Packer: Holographic Rotationally Equivariant Convolutional Neural Network for Protein Side-Chain Packing.

Accurately modeling protein 3D structure is essential for the design of functional proteins. An important sub-task of structure modeling is protein side-chain packing: predicting the conformation of side-chains (rotamers) given the protein's backbone structure and amino-acid sequence. Conventional approaches for this task rely on expensive sampling procedures over hand-crafted energy functions and rotamer libraries. Recently, several deep learning methods have been developed to tackle the problem in a data-driven way, albeit with vastly different formulations (from image-to-image translation to directly predicting atomic coordinates). Here, we frame the problem as a joint regression over the side-chains' true degrees of freedom: the dihedral χ angles. We carefully study possible objective functions for this task, while accounting for the underlying symmetries of the task. We propose Holographic Packer (H-Packer), a novel two-stage algorithm for side-chain packing built on top of two light-weight rotationally equivariant neural networks. We evaluate our method on CASP13 and CASP14 targets. H-Packer is computationally efficient and shows favorable performance against conventional physics-based algorithms and is competitive against alternative deep learning solutions.

Modular Smart Molecules for PSMA-Targeted Chemotherapy.

New targeted chemotherapeutics are urgently needed to minimize off-target toxicity and reduce the high-mortality rate associated with metastatic prostate cancer. Herein, we report on the modular synthesis, pharmacokinetics, and efficacy of two small-molecule-drug conjugates (SMDC) targeted to prostate-specific membrane antigen (PSMA) incorporating either: (i) a cathepsin-B-cleavable valine-citrulline (Val-Cit), or (ii) an acid-cleavable phosphoramidate linker. Crucial components used in the design of the conjugates include: (i) CTT1298, a nanomolar affinity ligand that binds irreversibly to PSMA and has proven in past studies to rapidly internalize and shuttle payloads into PSMA-expressing prostate cancer cells, (ii) MMAE, a known potent cytotoxic payload, and (iii) an albumin-binder, proven to improve residence time of drug conjugates. At dose of 0.8 mg/kg (∼250 nmol/kg), the two SMDCs showed significant efficacy in a PSMA(+) PC3-PIP mouse model of human prostate cancer compared with controls, without inducing systemic toxicity. Though localization of the SMDCs was observed in tissues apart from the tumor, release of MMAE was observed predominantly in tumor tissue, at levels that were 2-3 orders of magnitude higher than non-target tissues. Furthermore, SMDC2, which incorporated a novel pH-responsive phosporamidate linker, demonstrated significantly improved efficacy over SMDC1 that has a Val-Cit linker, with a 100% survival over 90 days and 4 out of 8 mice showing complete tumor growth inhibition after 6 weekly doses of 0.8 mg/kg (244 nmol/kg). Our findings demonstrate the potential of irreversible PSMA inhibitors combined with pH-responsive linkers as a way to specifically deliver chemotherapeutic drugs to prostate cancer tumors with minimal toxicity.

PET Imaging of VLA-4 in a New BRAFV600E Mouse Model of Melanoma.

PURPOSE: Despite unprecedented responses to immune checkpoint inhibitors and targeted therapy in melanoma, a major subset of patients progresses and have few effective salvage options. We have previously demonstrated robust, selective uptake of the peptidomimetic LLP2A labeled with Cu-64 ([64Cu]-LLP2A) for positron emission tomography (PET) imaging in subcutaneous and metastatic models of B16F10 murine melanoma. LLP2A binds with high affinity to very late antigen-4 (VLA-4, integrin α4ß1), a transmembrane protein overexpressed in melanoma and other cancers that facilitates tumor growth and metastasis. Yet B16F10 fails to faithfully reflect human melanoma biology, as it lacks certain oncogenic driver mutations, including BRAF mutations found in ≥ 50 % of clinical specimens. Here, we evaluated the PET tracer [64Cu]-CB-TE1A1P-PEG4-LLP2A ([64Cu]-LLP2A) in novel, translational BRAFV600E mutant melanoma models differing in VLA-4 expression-BPR (VLA-4-) and BPRα (VLA-4+). PROCEDURES: BPR cells were transduced with α4 (CD49d) to overexpress intact cell surface VLA-4 (BPRα). The binding affinity of [64Cu]-LLP2A to BPR and BPRα cells was determined by saturation binding assays. [64Cu]-LLP2A internalization into B16F10, BPR, and BPRα cells was quantified via a plate-based assay. Tracer biodistribution and PET/CT imaging were evaluated in mice bearing subcutaneous BPR and BPRα tumors. RESULTS: [64Cu]-LLP2A demonstrated high binding affinity to BPRα (Kd = 1.4 nM) but indeterminate binding to BPR cells. VLA-4+ BPRα and B16F10 displayed comparable time-dependent [64Cu]-LLP2A internalization, whereas BPR internalization was undetectable. PET/CT showed increased tracer uptake in BPRα tumors vs. BPR tumors in vivo, which was validated by significantly greater (p < 0.0001) BPRα tumor uptake in biodistribution analyses. CONCLUSIONS: [64Cu]-LLP2A discriminates BPRα (VLA-4+) vs. BPR (VLA-4-) melanomas in vivo, supporting translation of these BRAF-mutated melanoma models via prospective imaging and theranostic studies. These results extend the utility of LLP2A to selectively target clinically relevant and therapy-resistant tumor variants toward its use for therapeutic patient care.