Cells use signaling pathways to sense and respond to their environments. The transforming growth factor-β (TGF-β) pathway produces context-specific responses. Here, we combined modeling and ...experimental analysis to study the dependence of the output of the TGF-β pathway on the abundance of signaling molecules in the pathway. We showed that the TGF-β pathway processes the variation of TGF-β receptor abundance using Liebig's law of the minimum, meaning that the output-modifying factor is the signaling protein that is most limited, to determine signaling responses across cell types and in single cells. We found that the abundance of either the type I (TGFBR1) or type II (TGFBR2) TGF-β receptor determined the responses of cancer cell lines, such that the receptor with relatively low abundance dictates the response. Furthermore, nuclear SMAD2 signaling correlated with the abundance of TGF-β receptor in single cells depending on the relative expression levels of TGFBR1 and TGFBR2. A similar control principle could govern the heterogeneity of signaling responses in other signaling pathways.
Cancer is a life-threatening disease that affects one in three people. Although most cases are sporadic, cancer risk can be increased by genetic factors. It remains unknown why certain genes ...predispose for specific forms of cancer only, such as checkpoint protein 2 (CHK2), in which gene mutations convey up to twofold higher risk for breast cancer but do not increase lung cancer risk. We have investigated the role of CHK2 and the related kinase checkpoint protein 1 (CHK1) in cell cycle regulation in primary breast and lung primary epithelial cells. At the molecular level, CHK1 activity was higher in lung cells, whereas CHK2 was more active in breast cells. Inhibition of CHK1 profoundly disrupted the cell cycle profile in both lung and breast cells, whereas breast cells were more sensitive toward inhibition of CHK2. Finally, we provide evidence that breast cells require CHK2 to induce a G2-M cell cycle arrest in response of DNA damage, whereas lung cells can partially compensate for the loss of CHK2. Our results provide an explanation as to why CHK2 germline mutations predispose for breast cancer but not for lung cancer.
Cell signaling governs the basic functions of cells by molecular interactions that involve of many proteins. The abundance of signaling proteins can directly influence cellular responses to external ...signal, contributing to cellular heterogeneity. Absolute quantification of proteins is important for modeling and understanding the complex signaling network. Here, we introduce how to measure the amount of TGF-β signaling proteins using quantitative immunoblotting. In addition, we discuss how to convert the measurements of protein abundance to the quantities of absolute molecules per cell. This method is generally applicable to the absolute quantification of other proteins.
The DNA damage response (DDR) protects cells against genomic instability. Surprisingly, little is known about the differences in DDR across tissues, which may affect cancer evolutionary trajectories ...and chemotherapy response. Using mathematical modeling and quantitative experiments, we found that the DDR is regulated differently in human breast and lung primary cells. Equal levels of cisplatin-DNA lesions caused stronger Chk1 activation in lung cells, leading to resistance. In contrast, breast cells were more resistant and showed more Chk2 activation in response to doxorubicin. Further analyses indicate that Chk1 activity played a regulatory role in p53 phosphorylation, whereas Chk2 activity was essential for p53 activation and p21 expression. We propose a novel "friction model," in which the balance of p53 and p21 levels contributes to the apoptotic response in different tissues. Our results suggest that modulating the balance of p53 and p21 dynamics could optimize the response to chemotherapy.
Most machine learning algorithms are configured by a set of hyperparameters whose values must be carefully chosen and which often considerably impact performance. To avoid a time‐consuming and ...irreproducible manual process of trial‐and‐error to find well‐performing hyperparameter configurations, various automatic hyperparameter optimization (HPO) methods—for example, based on resampling error estimation for supervised machine learning—can be employed. After introducing HPO from a general perspective, this paper reviews important HPO methods, from simple techniques such as grid or random search to more advanced methods like evolution strategies, Bayesian optimization, Hyperband, and racing. This work gives practical recommendations regarding important choices to be made when conducting HPO, including the HPO algorithms themselves, performance evaluation, how to combine HPO with machine learning pipelines, runtime improvements, and parallelization.
This article is categorized under:
Algorithmic Development > Statistics
Technologies > Machine Learning
Technologies > Prediction
After a general introduction of hyperparameter optimization, we review important HPO methods such as grid or random search, evolutionary algorithms, Bayesian optimization, Hyperband and racing. We include many practical recommendations w.r.t. performance evaluation, how to combine HPO with ML pipelines, runtime improvements and parallelization.
Cells use signaling pathways to sense and respond to their environments. The transforming growth factor-beta (TGF-beta) pathway produces context-specific responses. Here, we combined modeling and ...experimental analysis to study the dependence of the output of the TGF-beta pathway on the abundance of signaling molecules in the pathway. We showed that the TGF-beta pathway processes the variation of TGF-beta receptor abundance using Liebig's law of the minimum, meaning that the output-modifying factor is the signaling protein that is most limited, to determine signaling responses across cell types and in single cells. We found that the abundance of either the type I (TGFBR1) or type II (TGFBR2) TGF-beta receptor determined the responses of cancer cell lines, such that the receptor with relatively low abundance dictates the response. Furthermore, nuclear SMAD2 signaling correlated with the abundance of TGF-beta receptor in single cells depending on the relative expression levels of TGFBR1 and TGFBR2. A similar control principle could govern the heterogeneity of signaling responses in other signaling pathways.
Mammalian cells can decode the concentration of extracellular transforming growth factor‐β (TGF‐β) and transduce this cue into appropriate cell fate decisions. How variable TGF‐β ligand doses ...quantitatively control intracellular signaling dynamics and how continuous ligand doses are translated into discontinuous cellular fate decisions remain poorly understood. Using a combined experimental and mathematical modeling approach, we discovered that cells respond differently to continuous and pulsating TGF‐β stimulation. The TGF‐β pathway elicits a transient signaling response to a single pulse of TGF‐β stimulation, whereas it is capable of integrating repeated pulses of ligand stimulation at short time interval, resulting in sustained phospho‐Smad2 and transcriptional responses. Additionally, the TGF‐β pathway displays different sensitivities to ligand doses at different time scales. While ligand‐induced short‐term Smad2 phosphorylation is graded, long‐term Smad2 phosphorylation is switch‐like to a small change in TGF‐β levels. Correspondingly, the short‐term Smad7 gene expression is graded, while long‐term PAI‐1 gene expression is switch‐like, as is the long‐term growth inhibitory response. Our results suggest that long‐term switch‐like signaling responses in the TGF‐β pathway might be critical for cell fate determination.
Synopsis
The transforming growth factor‐β (TGF‐β) pathway is a prominent signaling pathway that regulates diverse aspects of cellular homeostasis, including proliferation, differentiation, migration, and death (Massague, 1998). Remarkably, the pleiotropic biological effects of TGF‐β are mediated by a relatively simple signaling module (Clarke and Liu, 2008). An interesting question is how such an apparently straightforward and simple cascade can generate a wide array of biological responses depending on the cellular context.
Members of the TGF‐β superfamily are frequently used as morphogens in early embryo development (Green, 2002). The best‐studied examples include Dpp in Drosophila and Activin in Xenopus (Gurdon and Bourillot, 2001; Lander, 2007). In the developmental context, cells can respond to a graded ligand concentration and produce discrete biological responses (e.g., transcription of certain genes, proliferation, or differentiation; Green, 2002). To convert continuous morphogen stimulation into discrete responses, mechanisms must exist to provide a threshold for the cellular response. How variable TGF‐β ligand doses quantitatively control intracellular signaling dynamics and how continuous ligand doses are translated into discontinuous cellular fate decisions remains poorly understood.
We have previously reported that ligand molecules per cell is the input variable to which the cells respond, and ligand number per cell is the best predictor of signaling responses (Zi and Klipp, 2007a; Clarke et al, 2009). Here, we developed an improved mathematical model to predict TGF‐β signaling responses by calibrating the model with various experimental data sets from different TGF‐β stimulations. Using a combined experimental and mathematical modeling approach, we showed that TGF‐β pulse stimulation results in transient activation of the pathway while repeated short pulses at short time intervals lead to a sustained activation similar to persistent ligand exposure.
We next investigate the system response to variable doses of TGF‐β in HaCaT cells. Our mathematical model predicts that the short‐term Smad2 phosphorylation (after 45 min of TGF‐β stimulation) is a graded response, while long‐term Smad2 activation (after 24 h of TGF‐β stimulation) is a switch‐like response (Figure 5A and B). As shown in Figure 5A–D, both short‐ and long‐term Smad2 phosphorylation can be saturated but doses of TGF‐β that cause maximum response are quite different. Additionally, the shapes of response curves were different. The short‐term Smad2 activation was a graded (Michaelis–Menten‐like) response with a very low apparent Hill coefficient of about 0.8 (Figure 5A and C) while the long‐term Smad2 activation (P‐Smad2 at 24 h) yielded a switch‐like response with an apparent Hill coefficient of about 4.5 (Figure 5B and D). Thus, the Smad2 response is initially graded and sharpens over time to become ultrasensitive. To address whether TGF‐β‐inducible gene expression responses are graded or switch‐like in the short and long term, we measured mRNA levels of Smad7, an early responsive gene of TGF‐β and protein levels of p21 and PAI‐1 whose inductions are delayed and late, respectively. The experimental data show that Smad7 induction exhibits a graded response with corresponding Hill coefficients of about 1.3 (Figure 5E), which is consistent with the graded P‐Smad2 response at 45 min (Figure 5A and C). PAI‐1 induction in response to variable doses of TGF‐β for 24 h is highly ultrasensitive with an apparent Hill coefficient of ∼5.3. Compared with Smad7 and PAI‐1, p21 induction is only modest ultrasensitive (nHill≈2) (Figure 5G). These results suggest short‐term gene induction by TGF‐β appears to be graded while long‐term targets are more switch‐like. Finally, we measured the growth inhibitory response of HaCaT cells to variable doses of TGF‐β. The level of BrdU incorporation is also ultrasensitive with an apparent Hill coefficient of about 4.3 (Figure 5H). Therefore, the long‐term TGF‐β growth inhibitory response also shows a switch‐like behavior. Finally, we show that TGF‐β depletion affects long‐term Smad phosphorylation and switch‐like response of TGF‐β signaling system. These findings shed new light on how continuous ligand doses are translated into discontinuous cell fate decisions in biological systems.
In summary, we have shown that the dose and time course of TGF‐β stimulation have profound effects on Smad signaling dynamics. The rate of ligand depletion controls the duration of Smad2 phosphorylation. Cells can respond to a short pulse of TGF‐β stimulation, and periodic short ligand exposures are sufficient to generate long‐term signaling responses. Short‐term TGF‐β stimulation causes only transient pathway activation and can be terminated by ligand depletion. TGF‐β‐induced Smad2 phosphorylation is graded in the short‐term but ultrasensitive (switch‐like) in the long‐term (Figure 7). Additionally, cell growth arrest in response to TGF‐β shows switch‐like rather than graded behavior. Our modeling and experimental analyses suggest that ligand depletion is likely to be involved in sharpening a graded response into a switch‐like response.
Cells respond in real time to the absolute number of TGF‐β molecules in their environment.
A single pulse of TGF‐β stimulation results in transient SMAD activation whereas repeated short pulses of stimulation result in sustained SMAD activation.
Ligand‐induced short‐term TGF‐β/SMAD signaling activation is graded while long‐term signaling response is switch‐like or ultrasensitive.
TGF‐β ligand depletion is a major cause of conversion from graded short‐term responses to ultrasensitive long‐term responses.
Cells employ signaling pathways to make decisions in response to changes in their immediate environment. Transforming growth factor beta (TGF-β) is an important growth factor that regulates many ...cellular functions in development and disease. Although the molecular mechanisms of TGF-β signaling have been well studied, our understanding of this pathway is limited by the lack of tools that allow the control of TGF-β signaling with high spatiotemporal resolution. Here, we developed an optogenetic system (optoTGFBRs) that enables the precise control of TGF-β signaling in time and space. Using the optoTGFBRs system, we show that TGF-β signaling can be selectively and sequentially activated in single cells through the modulation of the pattern of light stimulations. By simultaneously monitoring the subcellular localization of TGF-β receptor and Smad2 proteins, we characterized the dynamics of TGF-β signaling in response to different patterns of blue light stimulations. The spatial and temporal precision of light control will make the optoTGFBRs system as a powerful tool for quantitative analyses of TGF-β signaling at the single cell level.
The rapid development of time series forecasting research has brought many deep learning-based modules in this field. However, despite the increasing amount of new forecasting architectures, it is ...still unclear if we have leveraged the full potential of these existing modules within a properly designed architecture. In this work, we propose a novel hierarchical neural architecture search approach for time series forecasting tasks. With the design of a hierarchical search space, we incorporate many architecture types designed for forecasting tasks and allow for the efficient combination of different forecasting architecture modules. Results on long-term-time-series-forecasting tasks show that our approach can search for lightweight high-performing forecasting architectures across different forecasting tasks.
Because of its sample efficiency, Bayesian optimization (BO) has become a popular approach dealing with expensive black-box optimization problems, such as hyperparameter optimization (HPO). Recent ...empirical experiments showed that the loss landscapes of HPO problems tend to be more benign than previously assumed, i.e. in the best case uni-modal and convex, such that a BO framework could be more efficient if it can focus on those promising local regions. In this paper, we propose BOinG, a two-stage approach that is tailored toward mid-sized configuration spaces, as one encounters in many HPO problems. In the first stage, we build a scalable global surrogate model with a random forest to describe the overall landscape structure. Further, we choose a promising subregion via a bottom-up approach on the upper-level tree structure. In the second stage, a local model in this subregion is utilized to suggest the point to be evaluated next. Empirical experiments show that BOinG is able to exploit the structure of typical HPO problems and performs particularly well on mid-sized problems from synthetic functions and HPO.