The most abundant isoform (HPLC-6) of type I antifreeze protein (AFP 1 ) in winter flounder is a 37-amino-acid-long, alanine-rich, R-helical peptide, containing four Thr spaced 11 amino acids apart. It is generally assumed that HPLC-6 binds ice through a hydrogen-bonding match between the Thr and neighboring Asx residues to oxygens atoms on the {202 h1} plane of the ice lattice. The result is a lowering of the nonequilibrium freezing point below the melting point (thermal hysteresis). HPLC-6, and two variants in which the central two Thr were replaced with either Ser or Val, were synthesized. The Ser variant was virtually inactive, while only a minor loss of activity was observed in the Val variant. CD, ultracentrifugation, and NMR studies indicated no significant structural changes or aggregation of the variants compared to HPLC-6. These results call into question the role of hydrogen bonds and suggest a much more significant role for entropic effects and van der Waals interactions in binding AFP to ice.Type I AFP 1 is the smallest and arguably the simplest of the four macromolecular antifreeze types characterized to date (1). It is in effect a single, long R-helix and therefore lacks tertiary structure (2). The most abundant isoform of this AFP (HPLC-6) from winter flounder (Pleuronectes americanus) is 37 amino acids long, contains three complete 11-amino-acid repeats of Thr-X 2 -Asx-X 7 , where X is generally alanine, and ends with the start of a fourth repeat. This helical periodicity that places the Thr and Asx residues on the same face of the helix, suggested a mechanism for adsorption of the AFP to ice in which these regularly spaced hydrophilic groups would hydrogen bond to oxygen atoms in the ice lattice (3). Adsorption leads to inhibition of ice crystal growth (4) because ice is forced to grow with a surface curvature between the bound AFP, which in turn results in a lowering of the nonequilibrium freezing point below the melting point (5, 6). The difference in these two temperatures is termed thermal hysteresis and is used as a measure of antifreeze activity.At very low concentrations, AFP bind to ice but do not stop its growth. Under these conditions bound AFP is frozen into the ice rather than excluded by the advancing ice front. The protein binding planes in these crystals have been made visable by sublimation (ice etching) and determined to be the {202 h1} pyramidal plane of hexagonal ice (I h ) for type I AFP (5). Moreover, because this antifreeze is a nonglobular, extended molecule it was possible to establish a direction 〈011 h2〉 of binding on the plane. An elegant proof of this resulted from the synthesis of an all D-type I AFP, which was shown by the ice etching method to bind to the same plane but in the mirror image direction (7). This information was used to suggest a hydrogen-bonding match between the i, i + 11 threonines spaced 16.5 Å apart along the helix and accessible ice lattice oxygens spaced 16.7 Å apart along the 〈011 h2〉 direction of the {202 h1} binding plane.On the ba...

No abstract

Pretrained general-purpose language models can achieve state-of-the-art accuracies in various natural language processing domains by adapting to downstream tasks via zero-shot, few-shot and finetuning techniques. Because of their success, the size of these models has increased rapidly, requiring high-performance hardware, software, and algorithmic techniques to enable training such large models. As the result of a joint effort between Microsoft and NVIDIA, we present details on the training of the largest monolithic transformer based language model, Megatron-Turing NLG 530B (MT-NLG), with 530 billion parameters. In this paper, we first focus on the infrastructure as well as the 3D parallelism methodology used to train this model using DeepSpeed and Megatron. Next, we detail the training process, the design of our training corpus, and our data curation techniques, which we believe is a key ingredient to the success of the model. Finally, we discuss various evaluation results, as well as other interesting observations and new properties exhibited by MT-NLG. We demonstrate that MT-NLG achieves superior zero-, one-, and few-shot learning accuracies on several NLP benchmarks and establishes new state-of-the-art results. We believe that our contributions will help further the development of large-scale training infrastructures, large-scale language models, and natural language generations.

Increased aerobic production of ATP at the onset of exercise could be limited by availability of metabolic substrates independent of O2, or interaction between O2 and metabolic substrates. We point out the importance of feedback control to match O2 supply to demand and discuss metabolic control at the onset of exercise.

To investigate biochemical, histochemical and contractile properties associated with strength training and detraining, six adult males were studied during and after 10 weeks of dynamic strength training for the quadriceps muscle group of one leg, as well as during and after a subsequent 12 weeks of detraining. Peak torque outputs at the velocities tested (0-270 degrees X s-1) were increased (p less than 0.05) by 39-60% and 12-37% after training for the trained and untrained legs, respectively. No significant changes in peak torques were observed in six control subjects tested at the same times. Significant decreases in strength performance of the trained leg (16-21%) and untrained leg (10-15%) were observed only after 12 weeks of detraining. Training resulted in an increase (p less than 0.05) in the area of FTa (21%) and FTb (18%) fibres, while detraining was associated with a 12% decrease in FTb fibre cross-sectional area. However, fibre area changes were only noted in the trained leg. Neither training nor detaining had any significant effect on the specific activity of magnesium-activated myofibrillar ATPase or on the activities of enzymes of phosphagen, glycolytic or oxidative metabolism in serial muscle biopsy samples from both legs. In the absence of any changes in muscle enzyme activities and with only modest changes in FT fibre areas in the trained leg, the significant alterations in peak torque outputs with both legs suggest that neural adaptations play a prominent role in strength performance with training and detraining.

A series of 14 residue amphipathic alpha-helical peptides, in which the sidechains of glutamic acid and lysine have been covalently joined, was synthesized in order to determine the effect of spacing, position and orientation of these lactam bridges. It was found that although an (i, i+3) spacing would position the lactam bridge on the same face of the helix, these lactams with 18-member rings were actually helix-destabilizing regardless of position or location. On the other hand, (i, i+4) lactams with 21-member rings were helix-stabilizing but this was dependent on orientation. Glutamic acid-lysine lactams increased the helical content of the peptide when compared with their linear homologue in benign conditions (50 mM KH2PO4, 100 mM KCl, pH 7). Two Glu-Lys (i, i+4) lactams located at the N- and C-termini gave rise to a peptide with greater than 99% helical content in benign conditions. Peptides with Lys-Glu oriented lactams were random structures in benign conditions but in the presence of 50% TFE could be induced into a helical conformation. The stability of the single-stranded alpha-helices, as measured by thermal denaturations in 25% TFE indicated that Glu-Lys oriented lactam bridges stabilized the helical conformation relative to the linear unbridged peptide. One Glu-Lys lactam in the middle of the peptide was more effective at stabilizing helical structure than two Glu-Lys lactams positioned one at each end of the molecule. The lactams with the Lys-Glu orientation were destabilizing relative to the unbridged peptide. This study demonstrates that correct orientation and position of a lactam bridge is critical in order to design peptides with high helical content in aqueous media.

The surface plasmon resonance (SPR) technique was used to study the formation kinetics of a de novo designed coiled-coil (E/K coil). The E/K coil is made up of two distinct peptides (E and K) each with five heptad (g-a-b-c-d-e-f) repeats. The E peptide's heptad sequence is E-V-S-A-L-E-K, and the K peptide's heptad sequence is K-V-S-A-L-K-E. A linker C-nL-G-G-G (nL = norleucine) is present at the C-terminus of the E peptide and at the N-terminus of the K peptide for the SPR studies. Heterodimer formation involves both electrostatic and hydrophobic interactions at the dimer interface. Under conditions that favor the heterodimer formation, the CD signal ([theta]222) varied as a function of peptide concentration. The estimated dissociation constant (Kd) was 2.45 +/- 0.71 nM. Denaturation studies with guanidine-HCI (GdnHC11/2 = 3.9 M) suggested a value of 3.53 +/- 0.48 nM. For the SPR investigation, the peptides were biotinylated and linked to streptavidin in order to increase their effective molecular weight and consequently enhance the signal intensity. Biotinylation in itself did not impede coiled-coil formation based on CD measurements. The biosensor study revealed a slow dissociation rate constant for the heterodimer (kd approximately 2 x 10(-4) s-1) and a moderately fast association rate constant [ka approximately (4.27-4.53) x 10(5) M-1 s-1). This gives a calculated Kd of 0.47-0.50 nM, which agrees reasonably well with the equilibrium CD studies. Therefore, based on the SPR data, the preference for heterodimer formation is due to a combination of moderately fast association and slow dissociation rates.

We describe a novel, two-nanoparticle mRNA delivery system and show that it is highly effective as a means of intracellular enzyme replacement therapy (i-ERT) using a murine model of ornithine transcarbamylase deficiency (OTCD). Our Hybrid mRNA Technology delivery system (HMT) comprises an inert lipid nanoparticle that protects the mRNA from nucleases in the blood as it distributes to the liver and a polymer micelle that targets hepatocytes and triggers endosomal release of mRNA. This results in high-level synthesis of the desired protein specifically in the liver. HMT delivery of human OTC mRNA normalizes plasma ammonia and urinary orotic acid levels, and leads to a prolonged survival benefit in the murine OTCD model. HMT represents a unique, non-viral mRNA delivery method that allows multi-dose, systemic administration for treatment of single-gene inherited metabolic diseases.