Self-supervised feature extraction from image time series in plant phenotyping using triplet networks.

MOTIVATION: Image-based profiling combines high-throughput screening with multiparametric feature analysis to capture the effect of perturbations on biological systems. This technology has attracted increasing interest in the field of plant phenotyping, promising to accelerate the discovery of novel herbicides. However, the extraction of meaningful features from unlabeled plant images remains a big challenge. RESULTS: We describe a novel data-driven approach to find feature representations from plant time-series images in a self-supervised manner by using time as a proxy for image similarity. In the spirit of transfer learning, we first apply an ImageNet-pretrained architecture as a base feature extractor. Then, we extend this architecture with a triplet network to refine and reduce the dimensionality of extracted features by ranking relative similarities between consecutive and non-consecutive time points. Without using any labels, we produce compact, organized representations of plant phenotypes and demonstrate their superior applicability to clustering, image retrieval and classification tasks. Besides time, our approach could be applied using other surrogate measures of phenotype similarity, thus providing a versatile method of general interest to the phenotypic profiling community. AVAILABILITY AND IMPLEMENTATION: Source code is provided in https://github.com/bayer-science-for-a-better-life/plant-triplet-net. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

A deterministic method for estimating free energy genetic network landscapes with applications to cell commitment and reprogramming paths.

Depicting developmental processes as movements in free energy genetic landscapes is an illustrative tool. However, exploring such landscapes to obtain quantitative or even qualitative predictions is hampered by the lack of free energy functions corresponding to the biochemical Michaelis-Menten or Hill rate equations for the dynamics. Being armed with energy landscapes defined by a network and its interactions would open up the possibility of swiftly identifying cell states and computing optimal paths, including those of cell reprogramming, thereby avoiding exhaustive trial-and-error simulations with rate equations for different parameter sets. It turns out that sigmoidal rate equations do have approximate free energy associations. With this replacement of rate equations, we develop a deterministic method for estimating the free energy surfaces of systems of interacting genes at different noise levels or temperatures. Once such free energy landscape estimates have been established, we adapt a shortest path algorithm to determine optimal routes in the landscapes. We explore the method on three circuits for haematopoiesis and embryonic stem cell development for commitment and reprogramming scenarios and illustrate how the method can be used to determine sequential steps for onsets of external factors, essential for efficient reprogramming.

Irreversibility of T-Cell Specification: Insights from Computational Modelling of a Minimal Network Architecture.

BACKGROUND/OBJECTIVES: A cascade of gene activations under the control of Notch signalling is required during T-cell specification, when T-cell precursors gradually lose the potential to undertake other fates and become fully committed to the T-cell lineage. We elucidate how the gene/protein dynamics for a core transcriptional module governs this important process by computational means. METHODS: We first assembled existing knowledge about transcription factors known to be important for T-cell specification to form a minimal core module consisting of TCF-1, GATA-3, BCL11B, and PU.1 aiming at dynamical modeling. Model architecture was based on published experimental measurements of the effects on each factor when each of the others is perturbed. While several studies provided gene expression measurements at different stages of T-cell development, pure time series are not available, thus precluding a straightforward study of the dynamical interactions among these genes. We therefore translate stage dependent data into time series. A feed-forward motif with multiple positive feed-backs can account for the observed delay between BCL11B versus TCF-1 and GATA-3 activation by Notch signalling. With a novel computational approach, all 32 possible interactions among Notch signalling, TCF-1, and GATA-3 are explored by translating combinatorial logic expressions into differential equations for BCL11B production rate. RESULTS: Our analysis reveals that only 3 of 32 possible configurations, where GATA-3 works as a dimer, are able to explain not only the time delay, but very importantly, also give rise to irreversibility. The winning models explain the data within the 95% confidence region and are consistent with regard to decay rates. CONCLUSIONS: This first generation model for early T-cell specification has relatively few players. Yet it explains the gradual transition into a committed state with no return. Encoding logics in a rate equation setting allows determination of binding properties beyond what is possible in a Boolean network.

Effects of Liraglutide Monotherapy on Beta Cell Function and Pancreatic Enzymes Compared with Metformin in Japanese Overweight/Obese Patients with Type 2 Diabetes Mellitus: A Subpopulation Analysis of the KIND-LM Randomized Trial.

BACKGROUND AND OBJECTIVE: Glucagon-like peptide-1 receptor agonists (GLP-1RAs) are becoming one of the major therapeutic options for the treatment of type 2 diabetes mellitus (T2DM). This study was conducted as an exploratory analysis to clarify the effects of liraglutide, a GLP-1RA, on beta cell function, fat distribution and pancreas volume compared with metformin in Japanese overweight/obese patients with T2DM. METHODS: A subpopulation of the Keio study for Initial treatment of type 2 Diabetes with Liraglutide versus Metformin (KIND-LM) study participants (n = 20, 10 in oral metformin group and 10 in subcutaneous liraglutide group) who were enrolled at Keio University Hospital and underwent frequently sampled mixed meal tolerance test (MTT) and abdominal computed tomography (CT) at weeks 0 and 24 were included in this analysis. The patients were treated with either metformin or liraglutide throughout the 24-week study period. RESULTS: Changes in glycemic parameters such as glycated hemoglobulin (HbA1c), glycated albumin and 1,5-anhydroglucitol at week 24 were comparable between the groups. An oral minimal model based on MTT revealed that static-phase beta cell responsiveness (Φ s) and static-phase disposition index were significantly increased at week 24 in the liraglutide group but not in the metformin group. There was no significant change in fat distribution as well as body weight at week 24 in either group. Serum amylase and lipase levels modestly but significantly increased in the liraglutide group during the study; however, there was no incidence of pancreatitis and pancreas volume was not changed in the liraglutide group. CONCLUSION: Liraglutide monotherapy for 24 weeks improved beta cell responsiveness with no change in either body weight or fat distribution. Further investigation is needed to clarify the mechanism by which liraglutide increases serum pancreatic enzymes. TRIAL REGISTRATION: The University Hospital Medical Information Network (UMIN) Clinical Trials Registry ( http://www.umin.ac.jp/ctr/ ); UMIN000004243.

Decreased VLDL-Apo B 100 Fractional Synthesis Rate Despite Hypertriglyceridemia in Subjects With Type 2 Diabetes and Nephropathy.

CONTEXT: Subjects with type 2 diabetes mellitus (T2DM) and diabetic nephropathy (DN) often exhibit hypertriglyceridemia. The mechanism(s) of such an increase are poorly known. OBJECTIVE: We investigated very low-density lipoprotein (VLDL)-Apo B 100 kinetics in T2DM subjects with and without DN, and in healthy controls. DESIGN: Stable isotope (13)C-leucine infusion and modeling analysis of tracer-to-tracee ratio dynamics in the protein product pool in the 6-8-h period following tracer infusion were employed. SETTING: Male subjects affected by T2DM, either with (n = 9) or without (n = 5) DN, and healthy male controls (n = 6), were studied under spontaneous glycemic levels in the post-absorptive state. RESULTS: In the T2DM patients with DN, plasma triglyceride (TG) (mean ± SD; 2.2 ± 0.8 mmol/L) and VLDL-Apo B 100 (17.4 ± 10.4 mg/dL) concentrations, and VLDL-Apo B 100 pool (0.56 ± 0.29 g), were ∼60-80% greater (P < .05 or less) than those of the T2DM subjects without DN (TG, 1.4 ± 0.5 mmol/L; VLDL-Apo B 100, 9.9 ± 2.5 mg/dL; VLDL-Apo B 100 pool, 0.36 ± 0.09 g), and ∼80-110% greater (P < .04 or less) than those of nondiabetic controls (TG, 1.2 ± 0.4 mmol/L; VLDL-Apo B 100, 8.2 ± 1.7 mg/dL; VLDL-Apo B 100, 0.32 ± 0.09 g). In sharp contrast however, in the subjects with T2DM and DN, VLDL-Apo B 100 fractional synthesis rate was ≥50% lower (4.8 ± 2.2 pools/d) than that of either the T2DM subjects without DN (9.9 ± 4.3 pools/d; P < .025) or the control subjects (12.5 ± 9.1 pools/d; P < .04). CONCLUSIONS: The hypertriglyceridemia of T2DM patients with DN is not due to hepatic VLDL-Apo B 100 overproduction, which is decreased, but it should be attributed to decreased apolipoprotein removal.

REDEMPTION: reduced dimension ensemble modeling and parameter estimation.

UNLABELLED: Here, we present REDEMPTION ( RE: duced D: imension E: nsemble M: odeling and P: arameter estima TION: ), a toolbox for parameter estimation and ensemble modeling of ordinary differential equations (ODEs) using time-series data. For models with more reactions than measured species, a common scenario in biological modeling, the parameter estimation is formulated as a nested optimization problem based on incremental parameter estimation strategy. REDEMPTION also includes a tool for the identification of an ensemble of parameter combinations that provide satisfactory goodness-of-fit to the data. The functionalities of REDEMPTION are accessible through a MATLAB user interface (UI), as well as through programming script. For computational speed-up, REDEMPTION provides a numerical parallelization option using MATLAB Parallel Computing toolbox. AVAILABILITY AND IMPLEMENTATION: REDEMPTION can be downloaded from http://www.cabsel.ethz.ch/tools/redemption. CONTACT: rudi.gunawan@chem.ethz.ch.

ß-cell mass and turnover in humans: effects of obesity and aging.

OBJECTIVE: We sought to establish ß-cell mass, ß-cell apoptosis, and ß-cell replication in humans in response to obesity and advanced age. RESEARCH DESIGN AND METHODS: We examined human autopsy pancreas from 167 nondiabetic individuals 20-102 years of age. The effect of obesity on ß-cell mass was examined in 53 lean and 61 obese subjects, and the effect of aging was examined in 106 lean subjects. RESULTS: ß-Cell mass is increased by ~50% with obesity (from 0.8 to 1.2 g). With advanced aging, the exocrine pancreas undergoes atrophy but ß-cell mass is remarkably preserved. There is minimal ß-cell replication or apoptosis in lean humans throughout life with no detectable changes with obesity or advanced age. CONCLUSIONS: ß-Cell mass in human obesity increases by ~50% by an increase in ß-cell number, the source of which is unknown. ß-Cell mass is well preserved in humans with advanced aging.

Computational modelling of T-cell formation kinetics: output regulated by initial proliferation-linked deferral of developmental competence.

Bone-marrow-derived progenitors must continually enter the thymus of an adult mouse to sustain T-cell homeostasis, yet only a few input cells per day are sufficient to support a yield of 5 × 10(7) immature T-cells per day and an eventual output of 1-2 × 10(6) mature cells per day. While substantial progress has been made to delineate the developmental pathway of T-cell lineage commitment, still little is known about the relationship between differentiation competence and the remarkable expansion of the earliest (DN1 stage) T-cell progenitors. To address this question, we developed computational models where the probability to progress to the next stage (DN2) is related to division number. To satisfy differentiation kinetics and overall cell yield data, our models require that adult DN1 cells divide multiple times before becoming competent to progress into DN2 stage. Our findings were subsequently tested by in vitro experiments, where putative early and later-stage DN1 progenitors from the thymus were purified and their progression into DN2 was measured. These experiments showed that the two DN1 sub-populations divided with similar rates, but progressed to the DN2 stage with different rates, thus providing experimental evidence that DN1 cells increase their commitment probability in a cell-intrinsic manner as they undergo cell division. Proliferation-linked shifts in eligibility of DN1 cells to undergo specification thus control kinetics of T-cell generation.

Dynamical modelling of haematopoiesis: an integrated view over the system in homeostasis and under perturbation.

A very high number of different types of blood cells must be generated daily through a process called haematopoiesis in order to meet the physiological requirements of the organism. All blood cells originate from a population of relatively few haematopoietic stem cells residing in the bone marrow, which give rise to specific progenitors through different lineages. Steady-state dynamics are governed by cell division and commitment rates as well as by population sizes, while feedback components guarantee the restoration of steady-state conditions. In this study, all parameters governing these processes were estimated in a computational model to describe the haematopoietic hierarchy in adult mice. The model consisted of ordinary differential equations and included negative feedback regulation. A combination of literature data, a novel divide et impera approach for steady-state calculations and stochastic optimization allowed one to reduce possible configurations of the system. The model was able to recapitulate the fundamental steady-state features of haematopoiesis and simulate the re-establishment of steady-state conditions after haemorrhage and bone marrow transplantation. This computational approach to the haematopoietic system is novel and provides insight into the dynamics and the nature of possible solutions, with potential applications in both fundamental and clinical research.

Shortened {beta}-cell lifespan leads to {beta}-cell deficit in a rodent model of type 2 diabetes.

Since the fundamental defect in both type 1 and type 2 diabetes is ß-cell failure, there is increasing interest in the capacity, if any, for ß-cell regeneration. Insights into typical ß-cell age and lifespan during normal development and how these are influenced in diabetes is desirable to realistically establish the prospects for ß-cell regeneration as means to reverse the deficit in ß-cell mass in diabetes. We assessed the mean ß-cell age and lifespan by the classical McKendrick-von Foester equation that describes the age-based heterogeneity of ß-cells in terms of the time-varying ß-cell formation and loss estimated by a ß-cell turnover model. This modeling approach was applied to evaluate ß-cell lifespan in a rodent model of type 2 diabetes in comparison with nondiabetic controls. When rats were 10 mo old, mean ß-cell lifespan was 1 mo vs. 6 mo in rats with type 2 diabetes vs. controls. A shortened ß-cell lifespan in a rat model of type 2 diabetes results in a decrease in mean ß-cell age and thus contributes to decreased ß-cell mass.

Ongoing beta-cell turnover in adult nonhuman primates is not adaptively increased in streptozotocin-induced diabetes.

OBJECTIVE: ß-Cell turnover and its potential to permit ß-cell regeneration in adult primates are unknown. Our aims were 1) to measure ß-cell turnover in adult nonhuman primates; 2) to establish the relative contribution of ß-cell replication and formation of new ß-cells from other precursors (defined thus as ß-cell neogenesis); and 3) to establish whether there is an adaptive increase in ß-cell formation (attempted regeneration) in streptozotocin (STZ)-induced diabetes in adult nonhuman primates. RESEARCH DESIGN AND METHODS: Adult (aged 7 years) vervet monkeys were administered STZ (45-55 mg/kg, n = 7) or saline (n = 9). Pancreas was obtained from each animal twice, first by open surgical biopsy and then by euthanasia. ß-Cell turnover was evaluated by applying a mathematic model to measured replication and apoptosis rates. RESULTS: ß-Cell turnover is present in adult nonhuman primates (3.3 ± 0.9 mg/month), mostly (~80%) derived from ß-cell neogenesis. ß-Cell formation was minimal in STZ-induced diabetes. Despite marked hyperglycemia, ß-cell apoptosis was not increased in monkeys administered STZ. CONCLUSIONS: There is ongoing ß-cell turnover in adult nonhuman primates that cannot be accounted for by ß-cell replication. There is no evidence of ß-cell regeneration in monkeys administered STZ. Hyperglycemia does not induce ß-cell apoptosis in nonhuman primates in vivo.

Modeling nonsteady-state metabolism from arteriovenous data.

The use of arteriovenous (AV) concentration differences to measure the production of a substance at organ/tissue level by Fick principle is limited to steady state. Out of steady state, there is the need, as originally proposed by Zierler, to account for the nonnegligible transit time of the substance through the system. Based on this theory, we propose a modeling approach that adopts a parametric description for production and transit time. Once the unknown parameters are estimated on AV data, the transition time of the substance can be assessed and production can be reconstructed. As a case study, we discuss the estimation of pancreatic insulin secretion during a meal from C-peptide concentrations measured in femoral artery and hepatic vein in 12 subjects. Results support the importance of accounting for nonnegligible transit times, even if C-peptide mean transit time across the splanchnic bed is rather limited (3.3 ± 1.3 min), it affects the estimation of pancreatic insulin secretion which shows a significantly different profile in the early portion of the postprandial period when estimated either with the novel modeling approach or with the simplified steady state equation.

Strength exercise improves muscle mass and hepatic insulin sensitivity in obese youth.

INTRODUCTION: Data on the metabolic effects of resistance exercise (strength training) in adolescents are limited. PURPOSE: The objective of this study was to determine whether a controlled resistance exercise program without dietary intervention or weight loss reduces body fat accumulation, increases lean body mass, and improves insulin sensitivity and glucose metabolism in sedentary obese Hispanic adolescents. METHODS: Twelve obese adolescents (age = 15.5 ± 0.5 yr, body mass index = 35.3 ± 0.8 kg·m; 40.8% ± 1.5% body fat) completed a 12-wk resistance exercise program (two times 1 h·wk, exercising all major muscle groups). At baseline and on completion of the program, body composition was measured by dual-energy x-ray absorptiometry, abdominal fat distribution was measured by magnetic resonance imaging, hepatic and intramyocellular fat was measured by magnetic resonance spectroscopy, peripheral insulin sensitivity was measured by the stable-label intravenous glucose tolerance test, and hepatic insulin sensitivity was measured by the hepatic insulin sensitivity index = 1000/(GPR × fasting insulin). Glucose production rate (GPR), gluconeogenesis, and glycogenolysis were quantified using stable isotope gas chromatography/mass spectrometry techniques. RESULTS: All participants were normoglycemic. The exercise program resulted in significant strength gain in both upper and lower body muscle groups. Body weight increased from 97.0 ± 3.8 to 99.6 ± 4.2 kg (P < 0.01). The major part (∼80%) was accounted for by increased lean body mass (55.7 ± 2.8 to 57.9 ± 3.0 kg, P ≤ 0.01). Total, visceral, hepatic, and intramyocellular fat contents remained unchanged. Hepatic insulin sensitivity increased by 24% ± 9% (P < 0.05), whereas peripheral insulin sensitivity did not change significantly. GPR decreased by 8% ± 1% (P < 0.01) because of a 12% ± 5% decrease in glycogenolysis (P < 0.05). CONCLUSIONS: We conclude that a controlled resistance exercise program without weight loss increases strength and lean body mass, improves hepatic insulin sensitivity, and decreases GPR without affecting total fat mass or visceral, hepatic, and intramyocellular fat contents.

Aerobic exercise increases peripheral and hepatic insulin sensitivity in sedentary adolescents.

CONTEXT: Data are limited on the effects of controlled aerobic exercise programs (without weight loss) on insulin sensitivity and glucose metabolism in children and adolescents. OBJECTIVE: To determine whether a controlled aerobic exercise program (without weight loss) improves peripheral and hepatic insulin sensitivity and affects glucose production (GPR), gluconeogenesis and glycogenolysis in sedentary lean and obese Hispanic adolescents. PATIENTS AND DESIGN: Twenty-nine post-pubertal adolescents (14 lean: 15.1 +/- 0.3 y; 20.6 +/- 0.8 kg/m(2); 18.9+/-1.5% body fat and 15 obese: 15.6 +/- 0.4 y; 33.2 +/- 0.9 kg/m(2); 38.4 +/- 1.4% body fat) (mean +/- SE), completed a 12 wk aerobic exercise program (4 x 30 min/week at >or=70% of VO(2) peak). Peripheral and hepatic insulin sensitivity and glucose kinetics were quantified using GCMS pre- and post-exercise. RESULTS: No weight loss occurred. Lean and obese participants complied well with the program ( approximately 90% of the exercise sessions attended, resulting in approximately 15% increase in fitness in both groups). Peripheral and hepatic insulin sensitivity were higher in lean than obese adolescents but increased in both groups; peripheral insulin sensitivity by 35 +/- 14% (lean) (p < 0.05) and 59 +/- 19% (obese) (p < 0.01) and hepatic insulin sensitivity by 19 +/- 7% (lean) (p < 0.05) and 23 +/- 4% (obese) (p < 0.01). GPR, gluconeogenesis and glycogenolysis did not differ between the groups. GPR decreased slightly, 3 +/- 1% (lean) (p < 0.05) and 4 +/- 1% (obese) (p < 0.01). Gluconeogenesis remained unchanged, while glycogenolysis decreased slightly in the obese group (p < 0.01). CONCLUSION: This well accepted aerobic exercise program, without weight loss, is a promising strategy to improve peripheral and hepatic insulin sensitivity in lean and obese sedentary adolescents. The small decrease in GPR is probably of limited clinical relevance.

Dynamics of beta-cell turnover: evidence for beta-cell turnover and regeneration from sources of beta-cells other than beta-cell replication in the HIP rat.

Type 2 diabetes is characterized by hyperglycemia, a deficit in beta-cells, increased beta-cell apoptosis, and islet amyloid derived from islet amyloid polypeptide (IAPP). These characteristics are recapitulated in the human IAPP transgenic (HIP) rat. We developed a mathematical model to quantify beta-cell turnover and applied it to nondiabetic wild type (WT) vs. HIP rats from age 2 days to 10 mo to establish 1) whether beta-cell formation is derived exclusively from beta-cell replication, or whether other sources of beta-cells (OSB) are present, and 2) to what extent, if any, there is attempted beta-cell regeneration in the HIP rat and if this is through beta-cell replication or OSB. We conclude that formation and maintenance of adult beta-cells depends largely ( approximately 80%) on formation of beta-cells independent from beta-cell duplication. Moreover, this source adaptively increases in the HIP rat, implying attempted beta-cell regeneration that substantially slows loss of beta-cell mass.

Development of factors to convert frequency to rate for beta-cell replication and apoptosis quantified by time-lapse video microscopy and immunohistochemistry.

An obstacle to development of methods to quantify beta-cell turnover from pancreas tissue is the lack of conversion factors for the frequency of beta-cell replication or apoptosis detected by immunohistochemistry to rates of replication or apoptosis. We addressed this obstacle in islets from 1-mo-old rats by quantifying the relationship between the rate of beta-cell replication observed directly by time-lapse video microscopy (TLVM) and the frequency of beta-cell replication in the same islets detected by immunohistochemistry using antibodies against Ki67 and insulin in the same islets fixed immediately after TLVM. Similarly, we quantified the rate of beta-cell apoptosis by TLVM and then the frequency of apoptosis in the same islets using TdT-mediated dUTP nick-end labeling and insulin. Conversion factors were developed by regression analysis. The conversion factor from Ki67 labeling frequency (%) to actual replication rate (%events/h) is 0.025 +/- 0.003 h(-1). The conversion factor from TdT-mediated dUTP nick-end labeling frequency (%) to actual apoptosis rate (%events/h) is 0.41 +/- 0.05 h(-1). These conversion factors will permit development of models to evaluate beta-cell turnover in fixed pancreas tissue.