Chemical language models for de novo drug design: Challenges and opportunities

Generative deep learning is accelerating de novo drug design, by allowing the generation of molecules with desired properties on demand. Chemical language models – which generate new molecules in the form of strings using deep learning – have been particularly successful in this endeavour. Thanks to advances in natural language processing methods and interdisciplinary collaborations, chemical language models are expected to become increasingly relevant in drug discovery. This minireview provides an overview of the current state-of-the-art of chemical language models for de novo design, and analyses current limitations, challenges, and advantages. Finally, a perspective on future opportunities is provided.     

Integrating structure-based approaches in generative molecular design

Generative molecular design for drug discovery and development has seen a recent resurgence promising to improve the efficiency of the design-make-test-analyse cycle; by computationally exploring much larger chemical spaces than traditional virtual screening techniques. However, most generative models thus far have only utilized small-molecule information to train and condition de novo molecule generators. Here, we instead focus on recent approaches that incorporate protein structure into de novo molecule optimization in an attempt to maximize the predicted on-target binding affinity of generated molecules. We summarize these structure integration principles into either distribution learning or goal-directed optimization and for each case whether the approach is protein structure-explicit or implicit with respect to the generative model. We discuss recent approaches in the context of this categorization and provide our perspective on the future direction of the field.     

Deep generative modeling for volume reconstruction in cryo-electron microscopy

Advances in cryo-electron microscopy (cryo-EM) for high-resolution imaging of biomolecules in solution have provided new challenges and opportunities for algorithm development for 3D reconstruction. Next-generation volume reconstruction algorithms that combine generative modelling with end-to-end unsupervised deep learning techniques have shown promise, but many technical and theoretical hurdles remain, especially when applied to experimental cryo-EM images. In light of the proliferation of such methods, we propose here a critical review of recent advances in the field of deep generative modelling for cryo-EM reconstruction. The present review aims to (i) provide a unified statistical framework using terminology familiar to machine learning researchers with no specific background in cryo-EM, (ii) review the current methods in this framework, and (iii) outline outstanding bottlenecks and avenues for improvements in the field.     

Spatiotemporal expression of sonic hedgehog signalling molecules in the embryonic mesencephalic dopaminergic neurons

Midbrain dopaminergic neurons (mDA) play an important role in controlling the voluntary motor movement, reward, and emotion-based behaviour. Differentiation of mDA neurons from progenitors depends on several secreted proteins, such as sonic hedgehog (SHH). The present study attempted to elucidate the possible role(s) of some SHH signaling components (Ptch1, Gli1, Gli2 and Gli3) in the spatiotemporal development of mDA neurons along the rostrocaudal axis of the midbrain and their possible roles in differentiation and survival of mDA neurons and the significance of using in vitro models for studying the development of mDA neurons. At E12 and E14, only Ptch1 and Gli1 were expressed in ventrolateral midbrain domains. All examined SHH signalling molecules were not detected in mDA area. Whereas, in MN9D cells, many SHH signalling molecules were expressed and co-localized with the dopaminergic marker; tyrosine hydroxylase (TH), and their expression were upregulated with SHH treatment of the MN9D cells. These results suggest that mDA neurons differentiation and survival might be independent of SHH in the late developmental stages (E12-18). Besides, MN9D cell line is not the ideal in vitro model for investigating the differentiation of mDA and hence, the ventral midbrain primary culture might be favored over MN9D line.     

Cellular cholesterol loss by DHCR24 knockdown leads to Aβ production by changing APP intracellular localization

For the past 20 years, the majority of cell culture studies reported that increasing cholesterol level increases amyloid-β (Aβ) production. Conversely, other studies and genetic evidences support that cellular cholesterol loss leads to Aβ generation. As a highly controversial issue in Alzheimer’s disease pathogenesis, the apparent contradiction prompted us to again explore the role of cellular cholesterol in Aβ production. Here, we adopted new neuronal and astrocytic cell models induced by 3β-hydroxysterol-Δ24 reductase (DHCR24), which obviously differ from the widely used cell models with overexpressing amyloid precursor protein (APP) in the majority of previous studies. In neuronal and astrocytic cell model, we found that deficiency of cellular cholesterol by DHCR24 knockdown obviously increased intracellular and extracellular Aβ generation. Importantly, in cell models with overexpressing APP, we found that APP overexpression could disrupt cellular cholesterol homeostasis and affect function of cells, coupled with the increase of APP β-cleavage product, 99-residue transmembrane C-terminal domain. Therefore, we suppose the results derived from the APP knockin models will need to be re-evaluated. One rational explanation for the discrepancy between our outcomes and the previous studies could be attributed to the two different cell models. Mechanistically, we showed that cellular cholesterol loss obviously altered APP intracellular localization by affecting cholesterol-related trafficking protein of APP. Therefore, our outcomes strongly support cellular cholesterol loss by DHCR24 knockdown leads to Aβ production.     

Relationship between hepatic and mitochondrial ceramides: a novel in vivo method to track ceramide synthesis

Ceramides (CERs) are key intermediate sphingolipids implicated in contributing to mitochondrial dysfunction and the development of multiple metabolic conditions. Despite the growing evidence of CER role in disease risk, kinetic methods to measure CER turnover are lacking, particularly using in vivo models. The utility of orally administered ¹³C₃, ¹⁵N l-serine, dissolved in drinking water, was tested to quantify CER 18:1/16:0 synthesis in 10-week-old male and female C57Bl/6 mice. To generate isotopic labeling curves, animals consumed either a control diet or high-fat diet (HFD; n = 24/diet) for 2 weeks and varied in the duration of the consumption of serine-labeled water (0, 1, 2, 4, 7, or 12 days; n = 4 animals/day/diet). Unlabeled and labeled hepatic and mitochondrial CERs were quantified using liquid chromatography tandem MS. Total hepatic CER content did not differ between the two diet groups, whereas total mitochondrial CERs increased with HFD feeding (60%, P < 0.001). Within hepatic and mitochondrial pools, HFD induced greater saturated CER concentrations (P < 0.05) and significantly elevated absolute turnover of 16:0 mitochondrial CER (mitochondria: 59%, P < 0.001 vs. liver: 15%, P = 0.256). The data suggest cellular redistribution of CERs because of the HFD. These data demonstrate that a 2-week HFD alters the turnover and content of mitochondrial CERs. Given the growing data on CERs contributing to hepatic mitochondrial dysfunction and the progression of multiple metabolic diseases, this method may now be used to investigate how CER turnover is altered in these conditions.     

Glutamine-Fructose-6-Phosphate Transaminase 2 (GFPT2) Is Upregulated in Breast Epithelial–Mesenchymal Transition and Responds to Oxidative Stress

Breast cancer cells that have undergone partial epithelial–mesenchymal transition (EMT) are believed to be more invasive than cells that have completed EMT. To study metabolic reprogramming in different mesenchymal states, we analyzed protein expression following EMT in the breast epithelial cell model D492 with single-shot LFQ supported by a SILAC proteomics approach. The D492 EMT cell model contains three cell lines: the epithelial D492 cells, the mesenchymal D492M cells, and a partial mesenchymal, tumorigenic variant of D492 that overexpresses the oncogene HER2. The analysis classified the D492 and D492M cells as basal-like and D492HER2 as claudin-low. Comparative analysis of D492 and D492M to tumorigenic D492HER2 differentiated metabolic markers of migration from those of invasion. Glutamine-fructose-6-phosphate transaminase 2 (GFPT2) was one of the top dysregulated enzymes in D492HER2. Gene expression analysis of the cancer genome atlas showed that GFPT2 expression was a characteristic of claudin-low breast cancer. siRNA-mediated knockdown of GFPT2 influenced the EMT marker vimentin and both cell growth and invasion in vitro and was accompanied by lowered metabolic flux through the hexosamine biosynthesis pathway (HBP). Knockdown of GFPT2 decreased cystathionine and sulfide:quinone oxidoreductase (SQOR) in the transsulfuration pathway that regulates H₂S production and mitochondrial homeostasis. Moreover, GFPT2 was within the regulation network of insulin and EGF, and its expression was regulated by reduced glutathione (GSH) and suppressed by the oxidative stress regulator GSK3-β. Our results demonstrate that GFPT2 controls growth and invasion in the D492 EMT model, is a marker for oxidative stress, and associated with poor prognosis in claudin-low breast cancer.     

Racial and Ethnic Cardiometabolic Risk Disparities in the Type 1 Diabetes Exchange Clinic Registry Cohort

ObjectiveTo determine whether individuals from a historically underrepresented racial group have a higher cardiometabolic risk than historically represented individuals with type 1 diabetes (T1D) considering socioeconomic deprivation.MethodsWe used the multivariable logistic and linear regression models to examine socioeconomic deprivation (upper 10th percentile) by race/ethnicity interaction for each cardiometabolic risk factor and cardiometabolic risk burden score, respectively, across 6320 zip code tabulation areas. We also determined the age-adjusted prevalence of low, moderate, and high cardiometabolic risks defined as 0, 1 to 2, and 3 or more risk factors for hypertension, obesity, dyslipidemia, and off-target glycemia for non-Hispanic White (n = 15 746), non-Hispanic Black (n = 1019), Hispanic (n = 1115), and other (n = 887), respectively.ResultsThe sample comprised 18 767 adolescents and adults with T1D. Those identifying as non-Hispanic Black were more likely to have a high cardiometabolic risk profile, including a 4.5-fold increase in the odds of off-target glycemia, a twofold increase in the odds of systolic hypertension, and 0.29 (unadjusted) and 0.46 (adjusted) increases in a higher cardiometabolic risk burden compared with non-Hispanic White individuals (P < .01). Those identifying as Hispanic had a 3.4-fold increase in the odds of off-target glycemia but were less likely to be overweight/obese or have systolic hypertension compared with non-Hispanic White. However, the lower likelihood of overweight/obesity and hypertension did not persist after considering covariates.ConclusionThere is a need to investigate additional determinants of racially/ethnically underrepresented cardiometabolic health, including structural racism and implicit bias in cardiometabolic care for individuals with T1D.     

Development and Internal Validation of A Prediction Tool To Assist Clinicians Selecting Second-Line Therapy Following Metformin Monotherapy For Type 2 Diabetes

ObjectiveAdults with type 2 diabetes (T2D) face increased risk of many long-term adverse outcomes. While managing patients with T2D, clinicians are challenged to stay informed regarding all new therapies and must consider potential risks and benefits resultant to their use. Metformin (MET) is typically prescribed as first-line therapy, but a second line is often needed, given MET can be insufficient for maintaining long-term glycemic control. Our objective was to develop a predictive decision-making tool to help clinicians use an outcome-based approach to select second-line therapies for patients when MET monotherapy is insufficient for glycemic control.MethodsElectronic health records of 19 277 adults with T2D on MET monotherapy and ≥3 months of either GLP-1RA, DPP-4i, Insulin, SGLT-2i, SFU, or TZD therapy were reviewed at Cleveland Clinic from patient visits occurring between 2005 and 2019. Separate models were developed to predict likelihood of each main outcome measure (stroke, myocardial infarction, worsening hypertension, renal failure, and death). Discrimination and calibration were assessed with bootstrapping.ResultsThe median follow-up time for those without an event was 3.6 years (interquartile range 1.9, 6.3). Model discrimination ability was evaluated by concordance indices (goodness of fit metric with values ranging between 0 and 1: 1 indicates perfect discrimination ability; 0.5 reflects same discrimination ability as chance) demonstrating strong discrimination ability, with concordance index values for outcomes as follows: myocardial infarction (0.786), stroke (0.805), worsening hypertension (0.855), renal failure (0.808), and death (0.827).ConclusionA decision-making tool has been developed that may afford clinicians a more objective and individualized approach to choosing a second-line therapy to control glycemia for persons with T2D.     

Duodenal Mucosa: A New Target for the Treatment of Type 2 Diabetes

ObjectiveAfter a high-fat and high-sugar diet, the duodenal mucosa of rodents proliferate and trigger the signal of insulin resistance, which may be the cause of type 2 diabetes (T2D). In response to this phenomenon, researchers have designed the duodenal mucosal resurfacing (DMR) procedure, mainly through the hydrothermal ablation procedure, to restore the normal mucosal surface, thereby correcting this abnormal metabolic signal. This article aims to understand the changes in duodenum before and after the onset or treatment of T2D, and the potential mechanisms of DMR procedure.MethodsA literature search of PubMed and Web of Science was conducted using appropriate keywords.ResultsBoth animal and clinical studies have shown that the villus thickness, intestinal cells, glucose transporters, enteric nerves, and gut microbiota and their metabolites in the duodenum undergo corresponding changes before and after the onset or treatment of T2D. These changes may be related to the pathogenesis of T2D. DMR procedure may produce beneficial glycemic and hepatic metabolic effects by regulating these changes.ConclusionThe duodenum is an important metabolic signaling center, and limiting nutrient exposure to this critical region will have powerful metabolic benefits. The DMR procedure may regulate glycemic and hepatic parameters through various mechanisms, which needs to be further confirmed by a large number of animal and clinical studies.     

Exploitation of carbohydrate processing enzymes in biocatalysis

Exploitation of enzymes in biocatalytic processes provides scope both in the synthesis and degradation of molecules. Enzymes have power not only in their catalytic efficiency, but their chemoselectivity, regioselectivity, and stereoselectivity means the reactions they catalyze are precise and reproducible. Focusing on carbohydrate processing enzymes, this review covers advances in biocatalysis involving carbohydrates over the last 2–3 years. Given the notorious difficulties in the chemical synthesis of carbohydrates, the use of enzymes for synthesis has potential for significant impact in the future. The use of catabolic enzymes in the degradation of biomass, which can be exploited in the production of biofuels to provide a sustainable and greener source of energy, and the synthesis of molecules that have a range of applications including in the pharmaceutical and food industries will be explored.     

Delta-6 desaturase (Fads2) deficiency alters triacylglycerol/fatty acid cycling in murine white adipose tissue

The Δ-6 desaturase (D6D) enzyme is not only critical for the synthesis of eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA) from α-linolenic acid (ALA), but recent evidence suggests that it also plays a role in adipocyte lipid metabolism and body weight; however, the mechanisms remain largely unexplored. The goal of this study was to investigate if a D6D deficiency would inhibit triacylglycerol storage and alter lipolytic and lipogenic pathways in mouse white adipose tissue (WAT) depots due to a disruption in EPA and DHA production. Male C57BL/6J D6D knockout (KO) and wild-type (WT) mice were fed either a 7% w/w lard or flax (ALA rich) diet for 21 weeks. Energy expenditure, physical activity, and substrate utilization were measured with metabolic caging. Inguinal and epididymal WAT depots were analyzed for changes in tissue weight, fatty acid composition, adipocyte size, and markers of lipogenesis, lipolysis, and insulin signaling. KO mice had lower body weight, higher serum nonesterified fatty acids, smaller WAT depots, and reduced adipocyte size compared to WT mice without altered food intake, energy expenditure, or physical activity, regardless of the diet. Markers of lipogenesis and lipolysis were more highly expressed in KO mice compared to WT mice in both depots, regardless of the diet. These changes were concomitant with lower basal insulin signaling in WAT. Collectively, a D6D deficiency alters triacylglycerol/fatty acid cycling in WAT by promoting lipolysis and reducing fatty acid re-esterification, which may be partially attributed to a reduction in WAT insulin signaling.     

Nucleation-dependent Aggregation Kinetics of Yeast Sup35 Fragment GNNQQNY

An N-terminal hepta-peptide sequence of yeast prion protein Sup35 with the sequence GNNQQNY is widely used as a model system for amyloid fibril formation. In this study, we used a reproducible solubilisation protocol that allows the generation of a homogenous monomeric solution of GNNQQNY to uncover the molecular details of its self-assembly mechanism. The aggregation kinetics data show that the GNNQQNY sequence follows nucleation-dependent aggregation kinetics with a critical nucleus of size ~7 monomers and that the efficiency of nucleation were found to be inversely related to the reaction temperature. The nucleus reduces the thermodynamic energy barrier by acting as a template for further self-assembly and results in highly ordered amyloid fibrils. The fibers grown at different temperatures showed similar Thioflavin T fluorescence, Congo-red binding and β-sheet rich structures displaying a characteristic cross-β diffraction pattern. These aggregates also share morphological and structural identity with those reported earlier. The mature GNNQQNY fibers did not exert significant oxidative stress or cytotoxicity upon incubating with differentiated SHSY5Y cells. To our knowledge, this is the first study to experimentally validate previous nucleus size predictions based on theoretical and molecular dynamics simulations. These findings provide the basis for understanding the kinetics and thermodynamics of amyloid nucleation and elongation of amyloidogenic proteins/peptides associated with many systemic and neurodegenerative diseases.     

Vascular lipid droplets formed in response to TNF,hypoxia, or OA: biochemical composition and prostacyclin generation

Biogenesis of lipid droplets (LDs) in various cells plays an important role in various physiological and pathological processes. However, the function of LDs in endothelial physiology and pathology is not well understood. In the present work, we investigated the formation of LDs and prostacyclin (PGI₂) generation in the vascular tissue of isolated murine aortas following activation by proinflammatory factors: tumor necrosis factor (TNF), lipopolysaccharides (LPS), angiotensin II (AngII), hypoxic conditions, or oleic acid (OA). The abundance, size, and biochemical composition of LDs were characterized based on Raman spectroscopy and fluorescence imaging. We found that blockade of lipolysis by the adipose triglyceride lipase (ATGL) delayed LDs degradation and simultaneously blunted PGI₂ generation in aorta treated with all tested proinflammatory stimuli. Furthermore, the analysis of Raman spectra of LDs in the isolated vessels stimulated by TNF, LPS, AngII, or hypoxia uncovered that these LDs were all rich in highly unsaturated lipids and had a negligible content of phospholipids and cholesterols. Additionally, by comparing the Raman signature of endothelial LDs under hypoxic or OA-overload conditions in the presence or absence of ATGL inhibitor, atglistatin (Atgl), we show that Atgl does not affect the biochemical composition of LDs. Altogether, independent of whether LDs were induced by pro-inflammatory stimuli, hypoxia, or OA and of whether they were composed of highly unsaturated or less unsaturated lipids, we observed LDs formation invariably associated with ATGL-dependent PGI₂ generation. In conclusion, vascular LDs formation and ATGL-dependent PGI₂ generation represent a universal response to vascular proinflammatory insult.     

Freezing of cell sheets using a 3D freezer produces high cell viability after thawing

In cell therapy, transplanting an appropriate number of cells to the target site is crucial. One way to achieve this is to transplant cell sheets. Transplantation of cell sheets has already been utilized for various diseases in clinical practice. However, reducing the cost of cell sheet utilization is essential so as to facilitate the spread of regenerative medicine. Several ways to reduce costs are available, one of which is the use of allogenic cells. Another alternative is the use of cell sheets, which necessitates the development of methods for freezing cell sheets. This is the first study to report the use of a 3D Freezer for freezing cells. 3D Freezers have been used in the field of food processing and technology for a long time. The 3D Freezer freezes objects using cold air at a uniform temperature from all directions. In this study, we analyzed the cooling speed of human fibroblast sheets in 11 cell preservation solutions using a 3D Freezer and a Program Freezer. The cooling speed was ?2 °C per min in the 3D Freezer. Supercooling in 10 cell preservation solutions was lower in the 3D Freezer than in the Program Freezer. Cell viability after freeze–thaw of the cell sheets using 3D Freezer was more than 70% in five cell preservation solutions. The levels of hepatocyte growth factor and transforming growth factor-β1 were the same not only in the fibroblast sheets frozen using the five cell preservation solutions but also in the non-frozen fibroblast sheets. These results suggest that the 3D Freezer can freeze implantable cell sheets immediately after thawing.     

Characteristic fragmentation of polyunsaturated fatty acids with allylic vicinal diols in positive-ion LC/ESI-MS/MS

A characteristic fragmentation was observed for PUFAs that contain allylic vicinal diol groups (resolvin D1, D2, D4, E3, lipoxin A4, B4, and maresin 2), which were derivatized with N,N-dimethylethylenediamine (DMED), in positive-ion ESI-MS/MS. The findings indicate that when these compounds contain an allylic hydroxyl group that is located distal to the terminal DMED moiety in the case of resolvin D1, D4, and lipoxin A4, an aldehyde (-CH=O) is predominately formed, which arises from the breakdown in between vicinal diols, whereas, in the case of an allylic hydroxyl group that is located proximal to the DMED moiety, as in resolvin D2, E3, lipoxin B4, and maresin 2, an allylic carbene (-CH=CH-CH:) is formed. These specific fragmentations could be used as diagnostic ions for characterizing the above seven PUFAs. As a result, it was possible to detect resolvin D1, D2, E3, lipoxin A4, and B4 in sera (20 μl) obtained from healthy volunteers by multiple-reaction monitoring using LC/ESI-MS/MS.     

Towards the Molecular Mechanism of Pulmonary Surfactant Protein SP-B: At the Crossroad of Membrane Permeability and Interfacial Lipid Transfer

Pulmonary surfactant is a lipid-protein complex that coats the alveolar air-liquid interface, enabling the proper functioning of lung mechanics. The hydrophobic surfactant protein SP-B, in particular, plays an indispensable role in promoting the rapid adsorption of phospholipids into the interface. For this, formation of SP-B ring-shaped assemblies seems to be important, as oligomerization could be required for the ability of the protein to generate membrane contacts and to mediate lipid transfer among surfactant structures. SP-B, together with the other hydrophobic surfactant protein SP-C, also promotes permeability of surfactant membranes to polar molecules although the molecular mechanisms underlying this property, as well as its relevance for the surface activity of the protein, remain undefined. In this work, the contribution of SP-B and SP-C to surfactant membrane permeability has been further investigated, by evaluation of the ability of differently-sized fluorescent polar probes to permeate through giant vesicles with different lipid/protein composition. Our results are consistent with the generation by SP-B of pores with defined size in surfactant membranes. Furthermore, incubation of surfactant with an anti-SP-B antibody not only blocked membrane permeability but also affected lipid transfer into the air-water interface, as observed in a captive bubble surfactometer device. Our findings include the identification of SP-C and anionic phospholipids as modulators required for maintaining native-like permeability features in pulmonary surfactant membranes. Proper permeability through membrane assemblies could be crucial to complement the overall role of surfactant in maintaining alveolar equilibrium, beyond its biophysical function in stabilizing the respiratory air-liquid interface.