Machine Learning-Driven Multiscale Modeling: Bridging the Scales with a Next-Generation Simulation Infrastructure.

Interdependence across time and length scales is common in biology, where atomic interactions can impact larger-scale phenomenon. Such dependence is especially true for a well-known cancer signaling pathway, where the membrane-bound RAS protein binds an effector protein called RAF. To capture the driving forces that bring RAS and RAF (represented as two domains, RBD and CRD) together on the plasma membrane, simulations with the ability to calculate atomic detail while having long time and large length- scales are needed. The Multiscale Machine-Learned Modeling Infrastructure (MuMMI) is able to resolve RAS/RAF protein-membrane interactions that identify specific lipid-protein fingerprints that enhance protein orientations viable for effector binding. MuMMI is a fully automated, ensemble-based multiscale approach connecting three resolution scales: (1) the coarsest scale is a continuum model able to simulate milliseconds of time for a 1 µm2 membrane, (2) the middle scale is a coarse-grained (CG) Martini bead model to explore protein-lipid interactions, and (3) the finest scale is an all-atom (AA) model capturing specific interactions between lipids and proteins. MuMMI dynamically couples adjacent scales in a pairwise manner using machine learning (ML). The dynamic coupling allows for better sampling of the refined scale from the adjacent coarse scale (forward) and on-the-fly feedback to improve the fidelity of the coarser scale from the adjacent refined scale (backward). MuMMI operates efficiently at any scale, from a few compute nodes to the largest supercomputers in the world, and is generalizable to simulate different systems. As computing resources continue to increase and multiscale methods continue to advance, fully automated multiscale simulations (like MuMMI) will be commonly used to address complex science questions.

Lifetime of actin-dependent protein nanoclusters.

Protein nanoclusters (PNCs) are dynamic collections of a few proteins that spatially organize in nanometer-length clusters. PNCs are one of the principal forms of spatial organization of membrane proteins, and they have been shown or hypothesized to be important in various cellular processes, including cell signaling. PNCs show remarkable diversity in size, shape, and lifetime. In particular, the lifetime of PNCs can vary over a wide range of timescales. The diversity in size and shape can be explained by the interaction of the clustering proteins with the actin cytoskeleton or the lipid membrane, but very little is known about the processes that determine the lifetime of the nanoclusters. In this paper, using mathematical modeling of the cluster dynamics, we model the biophysical processes that determine the lifetime of actin-dependent PNCs. In particular, we investigated the role of actin aster fragmentation, which had been suggested to be a key determinant of the PNC lifetime, and we found that it is important only for a small class of PNCs. A simple extension of our model allowed us to investigate the kinetics of protein-ligand interaction near PNCs. We found an anomalous increase in the lifetime of ligands near PNCs, which agrees remarkably well with experimental data on RAS-RAF kinetics. In particular, analysis of the RAS-RAF data through our model provides falsifiable predictions and novel hypotheses that will not only shed light on the role of RAS-RAF kinetics in various cancers, but also will be useful in studying membrane protein clustering in general.

Machine learning-driven multiscale modeling reveals lipid-dependent dynamics of RAS signaling proteins.

RAS is a signaling protein associated with the cell membrane that is mutated in up to 30% of human cancers. RAS signaling has been proposed to be regulated by dynamic heterogeneity of the cell membrane. Investigating such a mechanism requires near-atomistic detail at macroscopic temporal and spatial scales, which is not possible with conventional computational or experimental techniques. We demonstrate here a multiscale simulation infrastructure that uses machine learning to create a scale-bridging ensemble of over 100,000 simulations of active wild-type KRAS on a complex, asymmetric membrane. Initialized and validated with experimental data (including a new structure of active wild-type KRAS), these simulations represent a substantial advance in the ability to characterize RAS-membrane biology. We report distinctive patterns of local lipid composition that correlate with interfacially promiscuous RAS multimerization. These lipid fingerprints are coupled to RAS dynamics, predicted to influence effector binding, and therefore may be a mechanism for regulating cell signaling cascades.

Membrane interactions of the globular domain and the hypervariable region of KRAS4b define its unique diffusion behavior.

The RAS proteins are GTP-dependent switches that regulate signaling pathways and are frequently mutated in cancer. RAS proteins concentrate in the plasma membrane via lipid-tethers and hypervariable region side-chain interactions in distinct nano-domains. However, little is known about RAS membrane dynamics and the details of RAS activation of downstream signaling. Here, we characterize RAS in live human and mouse cells using single-molecule-tracking methods and estimate RAS mobility parameters. KRAS4b exhibits confined mobility with three diffusive states distinct from the other RAS isoforms (KRAS4a, NRAS, and HRAS); and although most of the amino acid differences between RAS isoforms lie within the hypervariable region, the additional confinement of KRAS4b is largely determined by the protein's globular domain. To understand the altered mobility of an oncogenic KRAS4b, we used complementary experimental and molecular dynamics simulation approaches to reveal a detailed mechanism.

Undermining Glutaminolysis Bolsters Chemotherapy While NRF2 Promotes Chemoresistance in KRAS-Driven Pancreatic Cancers.

Pancreatic cancer is a disease with limited therapeutic options. Resistance to chemotherapies poses a significant clinical challenge for patients with pancreatic cancer and contributes to a high rate of recurrence. Oncogenic KRAS, a critical driver of pancreatic cancer, promotes metabolic reprogramming and upregulates NRF2, a master regulator of the antioxidant network. Here, we show that NRF2 contributed to chemoresistance and was associated with a poor prognosis in patients with pancreatic cancer. NRF2 activation metabolically rewired and elevated pathways involved in glutamine metabolism. This curbed chemoresistance in KRAS-mutant pancreatic cancers. In addition, manipulating glutamine metabolism restrained the assembly of stress granules, an indicator of chemoresistance. Glutaminase inhibitors sensitized chemoresistant pancreatic cancer cells to gemcitabine, thereby improving the effectiveness of chemotherapy. This therapeutic approach holds promise as a novel therapy for patients with pancreatic cancer harboring KRAS mutation. SIGNIFICANCE: These findings illuminate the mechanistic features of KRAS-mediated chemoresistance and provide a rationale for exploiting metabolic reprogramming in pancreatic cancer cells to confer therapeutic opportunities that could be translated into clinical trials. GRAPHICAL ABSTRACT: http://cancerres.aacrjournals.org/content/canres/80/8/1630/F1.large.jpg.

Deletion of LRRTM1 and LRRTM2 in adult mice impairs basal AMPA receptor transmission and LTP in hippocampal CA1 pyramidal neurons.

Leucine-rich repeat transmembrane (LRRTM) proteins are synaptic cell adhesion molecules that influence synapse formation and function. They are genetically associated with neuropsychiatric disorders, and via their synaptic actions likely regulate the establishment and function of neural circuits in the mammalian brain. Here, we take advantage of the generation of a LRRTM1 and LRRTM2 double conditional knockout mouse (LRRTM1,2 cKO) to examine the role of LRRTM1,2 at mature excitatory synapses in hippocampal CA1 pyramidal neurons. Genetic deletion of LRRTM1,2 in vivo in CA1 neurons using Cre recombinase-expressing lentiviruses dramatically impaired long-term potentiation (LTP), an impairment that was rescued by simultaneous expression of LRRTM2, but not LRRTM4. Mutation or deletion of the intracellular tail of LRRTM2 did not affect its ability to rescue LTP, while point mutations designed to impair its binding to presynaptic neurexins prevented rescue of LTP. In contrast to previous work using shRNA-mediated knockdown of LRRTM1,2, KO of these proteins at mature synapses also caused a decrease in AMPA receptor-mediated, but not NMDA receptor-mediated, synaptic transmission and had no detectable effect on presynaptic function. Imaging of recombinant photoactivatable AMPA receptor subunit GluA1 in the dendritic spines of cultured neurons revealed that it was less stable in the absence of LRRTM1,2. These results illustrate the advantages of conditional genetic deletion experiments for elucidating the function of endogenous synaptic proteins and suggest that LRRTM1,2 proteins help stabilize synaptic AMPA receptors at mature spines during basal synaptic transmission and LTP.

The Retromer Supports AMPA Receptor Trafficking During LTP.

Alterations in the function of the retromer, a multisubunit protein complex that plays a specialized role in endosomal sorting, have been linked to Alzheimer's and Parkinson's diseases, yet little is known about the retromer's role in the mature brain. Using in vivo knockdown of the critical retromer component VPS35, we demonstrate a specific role for this endosomal sorting complex in the trafficking of AMPA receptors during NMDA-receptor-dependent LTP at mature hippocampal synapses. The impairment of LTP due to VPS35 knockdown was mechanistically independent of any role of the retromer in the production of Aß from APP. Finally, we find surprising differences between Alzheimer's- and Parkinson's-disease-linked VPS35 mutations in supporting this pathway. These findings demonstrate a key role for the retromer in LTP and provide insights into how retromer malfunction in the mature brain may contribute to symptoms of common neurodegenerative diseases. VIDEO ABSTRACT.

Postsynaptic synaptotagmins mediate AMPA receptor exocytosis during LTP.

Strengthening of synaptic connections by NMDA (N-methyl-d-aspartate) receptor-dependent long-term potentiation (LTP) shapes neural circuits and mediates learning and memory. During the induction of NMDA-receptor-dependent LTP, Ca2+ influx stimulates recruitment of synaptic AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptors, thereby strengthening synapses. How Ca2+ induces the recruitment of AMPA receptors remains unclear. Here we show that, in the pyramidal neurons of the hippocampal CA1 region in mice, blocking postsynaptic expression of both synaptotagmin-1 (Syt1) and synaptotagmin-7 (Syt7), but not of either alone, abolished LTP. LTP was restored by expression of wild-type Syt7 but not of a Ca2+-binding-deficient mutant Syt7. Blocking postsynaptic expression of Syt1 and Syt7 did not impair basal synaptic transmission, reduce levels of synaptic or extrasynaptic AMPA receptors, or alter other AMPA receptor trafficking events. Moreover, expression of dominant-negative mutant Syt1 which inhibits Ca2+-dependent presynaptic vesicle exocytosis, also blocked Ca2+-dependent postsynaptic AMPA receptor exocytosis, thereby abolishing LTP. Our results suggest that postsynaptic Syt1 and Syt7 act as redundant Ca2+-sensors for Ca2+-dependent exocytosis of AMPA receptors during LTP, and thereby delineate a simple mechanism for the recruitment of AMPA receptors that mediates LTP.

Modulation of excitation on parvalbumin interneurons by neuroligin-3 regulates the hippocampal network.

Hippocampal network activity is generated by a complex interplay between excitatory pyramidal cells and inhibitory interneurons. Although much is known about the molecular properties of excitatory synapses on pyramidal cells, comparatively little is known about excitatory synapses on interneurons. Here we show that conditional deletion of the postsynaptic cell adhesion molecule neuroligin-3 in parvalbumin interneurons causes a decrease in NMDA-receptor-mediated postsynaptic currents and an increase in presynaptic glutamate release probability by selectively impairing the inhibition of glutamate release by presynaptic Group III metabotropic glutamate receptors. As a result, the neuroligin-3 deletion altered network activity by reducing gamma oscillations and sharp wave ripples, changes associated with a decrease in extinction of contextual fear memories. These results demonstrate that neuroligin-3 specifies the properties of excitatory synapses on parvalbumin-containing interneurons by a retrograde trans-synaptic mechanism and suggest a molecular pathway whereby neuroligin-3 mutations contribute to neuropsychiatric disorders.

LTP requires a unique postsynaptic SNARE fusion machinery.

Membrane fusion during exocytosis is mediated by assemblies of SNARE (soluble NSF-attachment protein receptor) and SM (Sec1/Munc18-like) proteins. The SNARE/SM proteins involved in vesicle fusion during neurotransmitter release are well understood, whereas little is known about the protein machinery that mediates activity-dependent AMPA receptor (AMPAR) exocytosis during long-term potentiation (LTP). Using direct measurements of LTP in acute hippocampal slices and an in vitro LTP model of stimulated AMPAR exocytosis, we demonstrate that the Q-SNARE proteins syntaxin-3 and SNAP-47 are required for regulated AMPAR exocytosis during LTP but not for constitutive basal AMPAR exocytosis. In contrast, the R-SNARE protein synaptobrevin-2/VAMP2 contributes to both regulated and constitutive AMPAR exocytosis. Both the central complexin-binding and the N-terminal Munc18-binding sites of syntaxin-3 are essential for its postsynaptic role in LTP. Thus, postsynaptic exocytosis of AMPARs during LTP is mediated by a unique fusion machinery that is distinct from that used during presynaptic neurotransmitter release.

Dynamic imaging of homo-FRET in live cells by fluorescence anisotropy microscopy.

Multiple lipid and protein components of the plasma membrane of a living cell are organized, both compositionally and functionally, at different spatial and temporal scales. For instance, Rab protein domains in membranes the clathrin-coated pit, or the immunological synapse are exquisite examples of functional compartmentalization in cell membranes. These assemblies consist in part of nanoscale complexes of lipids and proteins and are necessary to facilitate some specific sorting and signaling functions. It is evident that cellular functions require a regulated spatiotemporal organization of components at the nanoscale, often comprising of countable number of molecular species. Here, we describe multiple homo-FRET-based imaging methods that provide information about nanoscale interactions between fluorescently tagged molecules in live cells, at optically resolved spatial resolution.

Postsynaptic complexin controls AMPA receptor exocytosis during LTP.

Long-term potentiation (LTP) is a compelling synaptic correlate of learning and memory. LTP induction requires NMDA receptor (NMDAR) activation, which triggers SNARE-dependent exocytosis of AMPA receptors (AMPARs). However, the molecular mechanisms mediating AMPAR exocytosis induced by NMDAR activation remain largely unknown. Here, we show that complexin, a protein that regulates neurotransmitter release via binding to SNARE complexes, is essential for AMPAR exocytosis during LTP but not for the constitutive AMPAR exocytosis that maintains basal synaptic strength. The regulated postsynaptic AMPAR exocytosis during LTP requires binding of complexin to SNARE complexes. In hippocampal neurons, presynaptic complexin acts together with synaptotagmin-1 to mediate neurotransmitter release. However, postsynaptic synaptotagmin-1 is not required for complexin-dependent AMPAR exocytosis during LTP. These results suggest a complexin-dependent molecular mechanism for regulating AMPAR delivery to synapses, a mechanism that is surprisingly similar to presynaptic exocytosis but controlled by regulators other than synaptotagmin-1.

A DNA nanomachine that maps spatial and temporal pH changes inside living cells.

DNA nanomachines are synthetic assemblies that switch between defined molecular conformations upon stimulation by external triggers. Previously, the performance of DNA devices has been limited to in vitro applications. Here we report the construction of a DNA nanomachine called the I-switch, which is triggered by protons and functions as a pH sensor based on fluorescence resonance energy transfer (FRET) inside living cells. It is an efficient reporter of pH from pH 5.5 to 6.8, with a high dynamic range between pH 5.8 and 7. To demonstrate its ability to function inside living cells we use the I-switch to map spatial and temporal pH changes associated with endosome maturation. The performance of our DNA nanodevices inside living systems illustrates the potential of DNA scaffolds responsive to more complex triggers in sensing, diagnostics and targeted therapies in living systems.

Nanoclusters of GPI-anchored proteins are formed by cortical actin-driven activity.

Several cell-surface lipid-tethered proteins exhibit a concentration-independent, cholesterol-sensitive organization of nanoscale clusters and monomers. To understand the mechanism of formation of these clusters, we investigate the spatial distribution and steady-state dynamics of fluorescently tagged GPI-anchored protein nanoclusters using high-spatial and temporal resolution FRET microscopy. These studies reveal a nonrandom spatial distribution of nanoclusters, concentrated in optically resolvable domains. Monitoring the dynamics of recovery of fluorescence intensity and anisotropy, we find that nanoclusters are immobile, and the dynamics of interconversion between nanoclusters and monomers, over a range of temperatures, is spatially heterogeneous and non-Arrhenius, with a sharp crossover coinciding with a reduction in the activity of cortical actin. Cholesterol depletion perturbs cortical actin and the spatial scale and interconversion dynamics of nanoclusters. Direct perturbations of cortical actin activity also affect the construction, dynamics, and spatial organization of nanoclusters. These results suggest a unique mechanism of complexation of cell-surface molecules regulated by cortical actin activity.

Nanoscale organization of hedgehog is essential for long-range signaling.

Hedgehog (Hh) plays crucial roles in tissue-patterning and activates signaling in Patched (Ptc)-expressing cells. Paracrine signaling requires release and transport over many cell diameters away by a process that requires interaction with heparan sulfate proteoglycans (HSPGs). Here, we examine the organization of functional, fluorescently tagged variants in living cells by using optical imaging, FRET microscopy, and mutational studies guided by bioinformatics prediction. We find that cell-surface Hh forms suboptical oligomers, further concentrated in visible clusters colocalized with HSPGs. Mutation of a conserved Lys in a predicted Hh-protomer interaction interface results in an autocrine signaling-competent Hh isoform--incapable of forming dense nanoscale oligomers, interacting with HSPGs, or paracrine signaling. Thus, Hh exhibits a hierarchical organization from the nanoscale to visible clusters with distinct functions.

Precise positioning of myosin VI on endocytic vesicles in vivo.

Myosin VI has been studied in both a monomeric and a dimeric form in vitro. Because the functional characteristics of the motor are dramatically different for these two forms, it is important to understand whether myosin VI heavy chains are brought together on endocytic vesicles. We have used fluorescence anisotropy measurements to detect fluorescence resonance energy transfer between identical fluorophores (homoFRET) resulting from myosin VI heavy chains being brought into close proximity. We observed that, when associated with clathrin-mediated endocytic vesicles, myosin VI heavy chains are precisely positioned to bring their tail domains in close proximity. Our data show that on endocytic vesicles, myosin VI heavy chains are brought together in an orientation that previous in vitro studies have shown causes dimerization of the motor. Our results are therefore consistent with vesicle-associated myosin VI existing as a processive dimer, capable of its known trafficking function.