Discrete Protein Research Articles

Hierarchical design of pseudosymmetric protein nanocages.

Discrete protein assemblies ranging from hundreds of kilodaltons to hundreds of megadaltons in size are a ubiquitous feature of biological systems and perform highly specialized functions1,2. Despite remarkable recent progress in accurately designing new self-assembling proteins, the size and complexity of these assemblies has been limited by a reliance on strict symmetry3. Here, inspired by the pseudosymmetry observed in bacterial microcompartments and viral capsids, we developed a hierarchical computational method for designing large pseudosymmetric self-assembling protein nanomaterials. We computationally designed pseudosymmetric heterooligomeric components and used them to create discrete, cage-like protein assemblies with icosahedral symmetry containing 240, 540 and 960 subunits. At 49, 71 and 96 nm diameter, these nanocages are the largest bounded computationally designed protein assemblies generated to date. More broadly, by moving beyond strict symmetry, our work substantially broadens the variety of self-assembling protein architectures that are accessible through design.

Beyond the Bubble: A Debate on microRNA Sorting in Extracellular Vesicles.

Over the past decade, a scientific field has developed demonstrating microRNAs are actively sorted into extracellular vesicles via specific nucleotide motifs that interact with discrete RNA-binding proteins. These microRNAs are preferentially transported into recipient cells where they can regulate specific cellular pathways. This mechanism could have enormous potential in explaining how cells signal and regulate other cells nearby or at a distance. Tens of studies have built this theme of a regulated transport of microRNAs. However, some concerns exist about this field. Taken together, there are concerns of a lack of a consistent motif, RNA-binding protein, or preferential microRNA involved in this process. Here we provide expert and extensive analysis of the field that makes the cases for and against an active sorting mechanism. We provide potential explanations on why there is a lack of agreement. Most importantly, we provide ideas on how to move this field forward with more rigor and reproducibility. It is hoped that engaging in a scientific debate of the pros and cons of this field, more rigorous experiments can be performed to conclusively demonstrate this biological activity.

Harnessing AlphaFold to reveal hERG channel conformational state secrets.

To design safe, selective, and effective new therapies, there must be a deep understanding of the structure and function of the drug target. One of the most difficult problems to solve has been resolution of discrete conformational states of transmembrane ion channel proteins. An example is KV11.1 (hERG), comprising the primary cardiac repolarizing current, I Kr. hERG is a notorious drug anti-target against which all promising drugs are screened to determine potential for arrhythmia. Drug interactions with the hERG inactivated state are linked to elevated arrhythmia risk, and drugs may become trapped during channel closure. However, the structural details of multiple conformational states have remained elusive. Here, we guided AlphaFold2 to predict plausible hERG inactivated and closed conformations, obtaining results consistent with multiple available experimental data. Drug docking simulations demonstrated hERG state-specific drug interactions in good agreement with experimental results, revealing that most drugs bind more effectively in the inactivated state and are trapped in the closed state. Molecular dynamics simulations demonstrated ion conduction for an open but not AlphaFold2 predicted inactivated state that aligned with earlier studies. Finally, we identified key molecular determinants of state transitions by analyzing interaction networks across closed, open, and inactivated states in agreement with earlier mutagenesis studies. Here, we demonstrate a readily generalizable application of AlphaFold2 as an effective and robust method to predict discrete protein conformations, reconcile seemingly disparate data and identify novel linkages from structure to function.

Dynamic sampling of a surveillance state enables DNA proofreading by Cas9.

CRISPR-Cas9 has revolutionized genome engineering applications by programming its single-guide RNA, where high specificity is required. However, the precise molecular mechanism underscoring discrimination between on/off-target DNA sequences, relative to the guide RNA template, remains elusive. Here, using methyl-based NMR to study multiple holoenzymes assembled invitro, we elucidate a discrete protein conformational state which enables recognition of DNA mismatches at the protospacer adjacent motif (PAM)-distal end. Our results delineate an allosteric pathway connecting a dynamic conformational switch at the REC3 domain, with the sampling of a catalytically competent state by the HNH domain. Our NMR data show that HiFi Cas9 (R691A) increases the fidelity of DNA recognition by stabilizing this "surveillance state" for mismatched substrates, shifting the Cas9 conformational equilibrium away from the active state. These results establish a paradigm of substrate recognition through an allosteric protein-based switch, providing unique insights into the molecular mechanism which governs Cas9 selectivity.

Microtubule-associated protein, MAP1B, encodes functionally distinct polypeptides

Microtubule-associated protein, MAP1B, is crucial for neuronal morphogenesis and disruptions in MAP1B function are correlated with neurodevelopmental disorders. MAP1B encodes a single polypeptide that is processed into discrete proteins, a heavy chain (HC) and a light chain (LC); however, it is unclear if these two chains operate individually or as a complex within the cell. In vivo studies have characterized the contribution of MAP1B HC and LC to microtubule and actin-based processes, but their molecular mechanisms of action are unknown. Using in vitro reconstitution with purified proteins, we dissect the biophysical properties of the HC and LC and uncover distinct binding behaviors and functional roles for these MAPs. Our biochemical assays indicate that MAP1B HC and LC do not form a constitutive complex, supporting the hypothesis that these proteins operate independently within cells. Both HC and LC inhibit the microtubule motors, kinesin-3, kinesin-4, and dynein, and differentially affect the severing activity of spastin. Notably, MAP1B LC binds to actin filaments in vitro and can simultaneously bind and crosslink actin filaments and microtubules, a function not observed for MAP1B HC. Phosphorylation of MAP1B HC by DYRK1a negatively regulates its actin-binding activity without significantly affecting its microtubule-binding capacity, suggesting a dynamic contribution of MAP1B HC in cytoskeletal organization. Overall, our study provides new insights into the distinct functional properties of MAP1B HC and LC, underscoring their roles in coordinating cytoskeletal networks during neuronal development.

Treponema pallidum Periplasmic and Membrane Proteins Are Recognized by Circulating and Skin CD4+ T Cells.

Histologic and serologic studies suggest the induction of local and systemic Treponema pallidum-specific CD4+ T-cell responses to T. pallidum infection. We hypothesized that T. pallidum-specific CD4+ T cells are detectable in blood and in the skin rash of secondary syphilis and persist in both compartments after treatment. Peripheral blood mononuclear cells collected from 67 participants were screened by interferon-γ (IFN-γ) ELISPOT response to T. pallidum sonicate. T. pallidum-reactive T-cell lines from blood and skin were probed for responses to 89 recombinant T. pallidum antigens. Peptide epitopes and HLA class II restriction were defined for selected antigens. We detected CD4+ T-cell responses to T. pallidum sonicate ex vivo. Using T. pallidum-reactive T-cell lines we observed recognition of 14 discrete proteins, 13 of which localize to bacterial membranes or the periplasmic space. After therapy, T. pallidum-specific T cells persisted for at least 6 months in skin and 10 years in blood. T. pallidum infection elicits an antigen-specific CD4+ T-cell response in blood and skin. T. pallidum-specific CD4+ T cells persist as memory in both compartments long after curative therapy. The T. pallidum antigenic targets we identified may be high-priority vaccine candidates.

Discrete protein condensation events govern calcium signal dynamics in T cells.

Ca 2+ fluctuations in T cells reflect stochastic protein condensation events triggered by single molecular antigen-TCR binding.

Adult-onset deactivation of autophagy leads to loss of synapse homeostasis and cognitive impairment, with implications for alzheimer disease

ABSTRACT A growing number of studies link dysfunction of macroautophagy/autophagy to the pathogenesis of diseases such as Alzheimer disease (AD). Given the global importance of autophagy for homeostasis, how its dysfunction can lead to specific neurological changes is puzzling. To examine this further, we compared the global deactivation of autophagy in the adult mouse using the atg7iKO with the impact of AD-associated pathogenic changes in autophagic processing of synaptic proteins. Isolated forebrain synaptosomes, rather than total homogenates, from atg7iKO mice demonstrated accumulation of synaptic proteins, suggesting that the synapse might be a vulnerable site for protein homeostasis disruption. Moreover, the deactivation of autophagy resulted in impaired cognitive performance over time, whereas gross locomotor skills remained intact. Despite deactivation of autophagy for 6.5 weeks, changes in cognition were in the absence of cell death or synapse loss. In the symptomatic APP PSEN1 double-transgenic mouse model of AD, we found that the impairment in autophagosome maturation coupled with diminished presence of discrete synaptic proteins in autophagosomes isolated from these mice, leading to the accumulation of one of these proteins in the detergent insoluble protein fraction. This protein, SLC17A7/Vglut, also accumulated in atg7iKO mouse synaptosomes. Taken together, we conclude that synaptic autophagy plays a role in maintaining protein homeostasis, and that while decreasing autophagy interrupts normal cognitive function, the preservation of locomotion suggests that not all circuits are affected similarly. Our data suggest that the disruption of autophagic activity in AD may have relevance for the cognitive impairment in this adult-onset neurodegenerative disease. Abbreviations: 2dRAWM: 2-day radial arm water maze; AD: Alzheimer disease; Aβ: amyloid-beta; AIF1/Iba1: allograft inflammatory factor 1; APP: amyloid beta precursor protein; ATG7: autophagy related 7; AV: autophagic vacuole; CCV: cargo capture value; Ctrl: control; DLG4/PSD-95: discs large MAGUK scaffold protein 4; GFAP: glial fibrillary acidic protein; GRIN2B/NMDAR2b: glutamate ionotropic receptor NMDA type subunit 2B; LTD: long-term depression; MAP1LC3/LC3: microtubule associated protein 1 light chain 3; m/o: months-old; PNS: post-nuclear supernatant; PSEN1/PS1: presenilin 1; SHB: sucrose homogenization buffer; SLC32A1/Vgat: solute carrier family 32 member 1; SLC17A7/Vglut1: solute carrier family 17 member 7; SNAP25: synaptosome associated protein 25; SQSTM1/p62: sequestosome 1; SYN1: synapsin I; SYP: synaptophysin ; SYT1: synaptotagmin 1; Tam: tamoxifen; VAMP2: vesicle associated membrane protein 2; VCL: vinculin; wks: weeks.

A C-terminal motif containing a PKC phosphorylation site regulates γ-Protocadherin-mediated dendrite arborization in the cerebral cortex in vivo.

The Pcdhg gene cluster encodes 22 γ-Protocadherin (γ-Pcdh) cell adhesion molecules that critically regulate multiple aspects of neural development, including neuronal survival, dendritic and axonal arborization, and synapse formation and maturation. Each γ-Pcdh isoform has unique protein domains-a homophilically interacting extracellular domain and a juxtamembrane cytoplasmic domain-as well as a C-terminal cytoplasmic domain shared by all isoforms. The extent to which isoform-specific versus shared domains regulate distinct γ-Pcdh functions remains incompletely understood. Our previous in vitro studies identified protein kinase C (PKC)phosphorylation of a serine residue within a shared C-terminal motif as a mechanism through which γ-Pcdh promotion of dendrite arborization via myristoylated alanine-rich C-kinase substrate (MARCKS) is abrogated. Here, we used CRISPR/Cas9 genome editing to generate two new mouse lines expressing only non-phosphorylatable γ-Pcdhs, due either to a serine-to-alanine mutation (PcdhgS/A) or to a 15-amino acid C-terminal deletion resulting from insertion of an early stop codon (PcdhgCTD). Both lines are viable and fertile, and the density and maturation of dendritic spines remain unchanged in both PcdhgS/A and PcdhgCTD cortex. Dendrite arborization of cortical pyramidal neurons, however, is significantly increased in both lines, as are levels of active MARCKS. Intriguingly, despite having significantly reduced levels of γ-Pcdh proteins, the PcdhgCTD mutation yields the strongest phenotype, with even heterozygous mutants exhibiting increased arborization. The present study confirms that phosphorylation of a shared C-terminal motif is a key γ-Pcdh negative regulation point and contributes to a converging understanding of γ-Pcdh family function in which distinct roles are played by both individual isoforms and discrete protein domains.

Forecasting SARS-CoV-2 spike protein evolution from small data by deep learning and regression

The emergence of SARS-CoV-2 variants during the COVID-19 pandemic caused frequent global outbreaks that confounded public health efforts across many jurisdictions, highlighting the need for better understanding and prediction of viral evolution. Predictive models have been shown to support disease prevention efforts, such as with the seasonal influenza vaccine, but they require abundant data. For emerging viruses of concern, such models should ideally function with relatively sparse data typically encountered at the early stages of a viral outbreak. Conventional discrete approaches have proven difficult to develop due to the spurious and reversible nature of amino acid mutations and the overwhelming number of possible protein sequences adding computational complexity. We hypothesized that these challenges could be addressed by encoding discrete protein sequences into continuous numbers, effectively reducing the data size while enhancing the resolution of evolutionarily relevant differences. To this end, we developed a viral protein evolution prediction model (VPRE), which reduces amino acid sequences into continuous numbers by using an artificial neural network called a variational autoencoder (VAE) and models their most statistically likely evolutionary trajectories over time using Gaussian process (GP) regression. To demonstrate VPRE, we used a small amount of early SARS-CoV-2 spike protein sequences. We show that the VAE can be trained on a synthetic dataset based on this data. To recapitulate evolution along a phylogenetic path, we used only 104 spike protein sequences and trained the GP regression with the numerical variables to project evolution up to 5 months into the future. Our predictions contained novel variants and the most frequent prediction mapped primarily to a sequence that differed by only a single amino acid from the most reported spike protein within the prediction timeframe. Novel variants in the spike receptor binding domain (RBD) were capable of binding human angiotensin-converting enzyme 2 (ACE2) in silico, with comparable or better binding than previously resolved RBD-ACE2 complexes. Together, these results indicate the utility and tractability of combining deep learning and regression to model viral protein evolution with relatively sparse datasets, toward developing more effective medical interventions.

Mutagenetic analysis of the biosynthetic pathway of tetramate bripiodionen bearing 3-(2H-pyran-2-ylidene)pyrrolidine-2,4-dione skeleton

BackgroundNatural tetramates are a family of hybrid polyketides bearing tetramic acid (pyrrolidine-2,4-dione) moiety exhibiting a broad range of bioactivities. Biosynthesis of tetramates in microorganisms is normally directed by hybrid polyketide synthase (PKS) and nonribosomal peptide synthetase (NRPS) machineries, which form the tetramic acid ring by recruiting trans- or cis-acting thioesterase-like Dieckmann cyclase in bacteria. There are a group of tetramates with unique skeleton of 3-(2H-pyran-2-ylidene)pyrrolidine-2,4-dione, which remain to be investigated for their biosynthetic logics.ResultsHerein, the tetramate type compounds bripiodionen (BPD) and its new analog, featuring the rare skeleton of 3-(2H-pyran-2-ylidene)pyrrolidine-2,4-dione, were discovered from the sponge symbiotic bacterial Streptomyces reniochalinae LHW50302. Gene deletion and mutant complementation revealed the production of BPDs being correlated with a PKS-NRPS biosynthetic gene cluster (BGC), in which a Dieckmann cyclase gene bpdE was identified by sit-directed mutations. According to bioinformatic analysis, the tetramic acid moiety of BPDs should be formed on an atypical NRPS module constituted by two discrete proteins, including the C (condensation)-A (adenylation)-T (thiolation) domains of BpdC and the A-T domains of BpdD. Further site-directed mutagenetic analysis confirmed the natural silence of the A domain in BpdC and the functional necessities of the two T domains, therefore suggesting that an unusual aminoacyl transthiolation should occur between the T domains of two NRPS subunits. Additionally, characterization of a LuxR type regulator gene led to seven- to eight-fold increasement of BPDs production. The study presents the first biosynthesis case of the natural molecule with 3-(2H-pyran-2-ylidene)pyrrolidine-2,4-dione skeleton. Genomic mining using BpdD as probe reveals that the aminoacyl transthiolation between separate NRPS subunits should occur in a certain population of NRPSs in nature.

Expression of virus-like particles (VLPs) of foot-and-mouth disease virus (FMDV) using Saccharomyces cerevisiae

We engineered Saccharomyces cerevisiae to express structural proteins of foot-and-mouth disease virus (FMDV) and produce virus-like particles (VLPs). The gene, which encodes four structural capsid proteins (VP0 (VP4 and VP2), VP3, and VP1), followed by a translational “ribosomal skipping” sequence consisting of 2A and protease 3C, was codon-optimized and chemically synthesized. The cloned gene was used to transform S. cerevisiae 2805 strain. Western blot analysis revealed that the polyprotein consisting of VP0, VP3, and VP1 was processed into the discrete capsid proteins. Western blot analysis of 3C confirmed the presence of discrete 3C protein, suggesting that the 2A sequence functioned as a “ribosomal skipping” signal in the yeast for an internal re-initiation of 3C translation from a monocistronic transcript, thereby indicating polyprotein processing by the discrete 3C protease. Moreover, a band corresponding to only VP2, which was known to be non-enzymatically processed from VP0 to both VP4 and VP2 during viral assembly, further validated the assembly of processed capsid proteins into VLPs. Electron microscopy showed the presence of the characteristic icosahedral VLPs. Our results clearly demonstrate that S. cerevisiae processes the viral structural polyprotein using a viral 3C protease and the resulting viral capsid subunits are assembled into virion particles.Key points• Ribosomal skipping by self-cleaving FMDV peptide in S. cerevisiae.• Proteolytic processing of a structural polyprotein from a monocistronic transcript.• Assembly of the processed viral capsid proteins into a virus-like particle.

Live-Cell Fluorescence Imaging for Virus-Host Interactions.

Recent technological advances in microscopy have facilitated novel approaches to investigate host-pathogen interactions. In particular, improvements in both microscope hardware and engineered biosensors have helped to overcome barriers to live-cell imaging with fluorescence microscopy. Live fluorescent microscopy allows for the detection of discrete signaling events and protein localization, improving our ability to assess the effects of pharmacologic agents, microbes, or infection with high temporal resolution. Here we describe a protocol for long-term live-cell fluorescence imaging of virus infected cell lines.

Pan-tissue mitochondrial phenotyping reveals lower OXPHOS expression and function across cancer types

Targeting mitochondrial oxidative phosphorylation (OXPHOS) to treat cancer has been hampered due to serious side-effects potentially arising from the inability to discriminate between non-cancerous and cancerous mitochondria. Herein, comprehensive mitochondrial phenotyping was leveraged to define both the composition and function of OXPHOS across various murine cancers and compared to both matched normal tissues and other organs. When compared to both matched normal tissues, as well as high OXPHOS reliant organs like heart, intrinsic expression of the OXPHOS complexes, as well as OXPHOS flux were discovered to be consistently lower across distinct cancer types. Assuming intrinsic OXPHOS expression/function predicts OXPHOS reliance in vivo, these data suggest that pharmacologic blockade of mitochondrial OXPHOS likely compromises bioenergetic homeostasis in healthy oxidative organs prior to impacting tumor mitochondrial flux in a clinically meaningful way. Although these data caution against the use of indiscriminate mitochondrial inhibitors for cancer treatment, considerable heterogeneity was observed across cancer types with respect to both mitochondrial proteome composition and substrate-specific flux, highlighting the possibility for targeting discrete mitochondrial proteins or pathways unique to a given cancer type.

Correlative CD4 and CD8 T-cell immunodominance in humans and mice: Implications for preclinical testing

Antigen-specific T-cell recognition is restricted by Major Histocompatibility Complex (MHC) molecules, and differences between CD4 and CD8 immunogenicity in humans and animal species used in preclinical vaccine testing are yet to be fully understood. In this study, we addressed this matter by analyzing experimentally identified epitopes based on published data curated in the Immune Epitopes DataBase (IEDB) database. We first analyzed SARS-CoV-2 spike (S) and nucleoprotein (N), which are two common targets of the immune response and well studied in both human and mouse systems. We observed a weak but statistically significant correlation between human and H-2b mouse T-cell responses (CD8 S specific (r = 0.206, p = 1.37 × 10−13); CD4 S specific (r = 0.118, p = 2.63 × 10−5) and N specific (r = 0.179, p = 2.55 × 10−4)). Due to intrinsic differences in MHC molecules across species, we also investigated the association between the immunodominance of common Human Leukocyte Antigen (HLA) alleles for which HLA transgenic mice are available, namely, A*02:01, B*07:02, DRB1*01:01, and DRB1*04:01, and found higher significant correlations for both CD8 and CD4 (maximum r = 0.702, p = 1.36 × 10−31 and r = 0.594, p = 3.04−122, respectively). Our results further indicated that some regions are commonly immunogenic between humans and mice (either H-2b or HLA transgenic) but that others are human specific. Finally, we noted a significant correlation between CD8 and CD4 S- (r = 0.258, p = 7.33 × 1021) and N-specific (r = 0.369, p = 2.43 × 1014) responses, suggesting that discrete protein subregions can be simultaneously recognized by T cells. These findings were confirmed in other viral systems, providing general guidance for the use of murine models to test T-cell immunogenicity of viral antigens destined for human use.

Sesame eliciting and safe doses in a large sesame allergic population.

Sesame is a significant food allergen causing severe and even fatal reactions. Given its increasing prevalence in western diet, sesame is listed as an allergenic food requiring labeling in the United States and EU. However, data on the population reaction doses to sesame are limited. All sesame oral food challenges (OFCs), performed either for diagnosis or for threshold identification before the beginning of sesame oral immunotherapy (OIT) between November 2011 and July 2021 in Shamir medical center were analyzed for reaction threshold distribution. Safe-dose challenges with 90-120 min intervals were also analyzed. Two hundred and fifty patients underwent 338 positive OFCs, and additional 158 safe-dose OFCs were performed. The discrete and cumulative protein amounts estimated to elicit an objective reaction in 1% (ED01) of the entire cohort (n = 250) were 0.8 mg (range 0.3-6.3) and 0.7 mg (range 0.1-7.1), respectively, and those for 5% of the population (ED05) were 3.4 mg (range 1.2-20.6) and 4.5 mg (range 1.2-28.8), respectively. Safe-dose OFCs showed similar values of ED01 (0.8, 0.4-7.5 mg) and ED05 (3.4, 1.2-22.9 mg). While doses of ≤1 mg sesame protein elicited oral pruritus in 11.6% of the patients, no objective reaction was documented to this amount in any of the challenges, including safe-dose OFCs. This study provides data on sesame reaction threshold distribution in the largest population of allergic patients studied, with no right or left censored data, and with validation using a safe-dose OFC. It further supports the current methods for ED determination as appropriate for establishing safety precautions for the food industry.

Formation and Loading of a (2S)-2-Ethylmalonamyl Starter Unit in the Assembly Line of Polyketide-Nonribosomal Peptide Hybrid Sanglifehrin A.

Sanglifehrin A (SFA) is a spirolactam-conjugated, 22-membered macrolide with remarkable immunosuppressive and antiviral activities. This macrolide is a result of a hybrid polyketide synthase (PKS)-nonribosomal peptide synthetase (NRPS) assembly line that utilizes (2S)-2-ethylmalonamyl as a starter unit. Here, we report that the formation and loading of this starter unit in the SFA assembly line involve two unusual enzymatic reactions that occur on a discrete acyl carrier protein (ACP), SfaO. An amide synthetase, SfaP, catalyzes the amidation of (2S)-2-ethylmalonyl in a SfaO-dependent manner. Then, a β-ketoacyl-ACP synthase III-like protein, SfaN, transfers resultant (2S)-2-ethylmalonamyl from SfaO onto the loading ACP domain of the hybrid PKS-NRPS assembly line to prime SFA biosynthesis. Both SfaP and SfaN display promiscuous activities. This study furthers the appreciation of assembly line chemistry, as a new paradigm for unusual building block formation and incorporation is provided.

Formation and Loading of a (2S)‐2‐Ethylmalonamyl Starter Unit in the Assembly Line of Polyketide‐Nonribosomal Peptide Hybrid Sanglifehrin A

AbstractSanglifehrin A (SFA) is a spirolactam‐conjugated, 22‐membered macrolide with remarkable immunosuppressive and antiviral activities. This macrolide is a result of a hybrid polyketide synthase (PKS)‐nonribosomal peptide synthetase (NRPS) assembly line that utilizes (2S)‐2‐ethylmalonamyl as a starter unit. Here, we report that the formation and loading of this starter unit in the SFA assembly line involve two unusual enzymatic reactions that occur on a discrete acyl carrier protein (ACP), SfaO. An amide synthetase, SfaP, catalyzes the amidation of (2S)‐2‐ethylmalonyl in a SfaO‐dependent manner. Then, a β‐ketoacyl‐ACP synthase III‐like protein, SfaN, transfers resultant (2S)‐2‐ethylmalonamyl from SfaO onto the loading ACP domain of the hybrid PKS‐NRPS assembly line to prime SFA biosynthesis. Both SfaP and SfaN display promiscuous activities. This study furthers the appreciation of assembly line chemistry, as a new paradigm for unusual building block formation and incorporation is provided.

Plug & play directed evolution of proteins with gradient-based discrete MCMC

A long-standing goal of machine-learning-based protein engineering is to accelerate the discovery of novel mutations that improve the function of a known protein. We introduce a sampling framework for evolving proteins in silico that supports mixing and matching a variety of unsupervised models, such as protein language models, and supervised models that predict protein function from sequence. By composing these models, we aim to improve our ability to evaluate unseen mutations and constrain search to regions of sequence space likely to contain functional proteins. Our framework achieves this without any model fine-tuning or re-training by constructing a product of experts distribution directly in discrete protein space. Instead of resorting to brute force search or random sampling, which is typical of classic directed evolution, we introduce a fast Markov chain Monte Carlo sampler that uses gradients to propose promising mutations. We conduct in silico directed evolution experiments on wide fitness landscapes and across a range of different pre-trained unsupervised models, including a 650 M parameter protein language model. Our results demonstrate an ability to efficiently discover variants with high evolutionary likelihood as well as estimated activity multiple mutations away from a wild type protein, suggesting our sampler provides a practical and effective new paradigm for machine-learning-based protein engineering.

Visualizing hypoxic modulation of beta cell secretions via a sensor augmented oxygen gradient

One distinct advantage of microfluidic-based cell assays is their scalability for multiple concentrations or gradients. Microfluidic scaling can be extremely powerful when combining multiple parameters and modalities. Moreover, in situ stimulation and detection eliminates variability between individual bioassays. However, conventional microfluidics must combat diffusion, which limits the spatial distance and time for molecules traveling through microchannels. Here, we leveraged a multilayered microfluidic approach to integrate a novel oxygen gradient (0–20%) with an enhanced hydrogel sensor to study pancreatic beta cells. This enabled our microfluidics to achieve spatiotemporal detection that is difficult to achieve with traditional microfluidics. Using this device, we demonstrated the in situ detection of calcium, insulin, and ATP (adenosine triphosphate) in response to glucose and oxygen stimulation. Specifically, insulin was quantified at levels as low as 25 pg/mL using our imaging technique. Furthermore, by analyzing the spatial detection data dynamically over time, we uncovered a new relationship between oxygen and beta cell oscillations. We observed an optimum oxygen level between 10 and 12%, which is neither hypoxic nor normoxic in the conventional cell culture sense. These results provide evidence to support the current islet oscillator model. In future applications, this spatial microfluidic technique can be adapted for discrete protein detection in a robust platform to study numerous oxygen-dependent tissue dysfunctions.