Complex Cellular Phenotype Analysis

LOO Lit Hsin
Senior Principal Investigator

James A. MILLER, Sreetama BASU, Su Su HTWE
Post-Doctoral Research Fellows

Joey Lee Jia Ying
Senior Research Officer

Carmen Kong Jia Wen, Oscar Fu Shufeng
Research Officers

Paul E. Cain
Software Engineer

LOO Lit Hsin
Senior Principal Investigator

GROUP MEMBERS:

James A. MILLER, Sreetama BASU, Su Su HTWE
Post-Doctoral Research Fellows

Joey Lee Jia Ying
Senior Research Officer

Carmen Kong Jia Wen, Oscar Fu Shufeng
Research Officers

Paul E. Cain
Software Engineer

We are a computational biology research group with members from different scientific disciplines, including chemistry, cell biology, computer science, and bioinformatics.

The overall goal of our research is to understand the modes of action (MoAs) of xenobiotics, and predict their human toxicity and/or efficacy. We develop and use novel phenotypic and molecular profiling methods to elucidate the MoAs of xenobiotics, and build computational models that can predict in vivo effects based on these MoAs.

We are part of the Innovations in Food and Chemical Safety (IFCS) Programme in A*STAR. We also collaborate with different academic, industrial, and governmental research groups, including the Institute of Bioengineering and Nanotechnology (IBN), Institute of Molecular and Cell Biology (IMCB), and Molecular Engineering Lab (MEL) from A*STAR; and the United States Environmental Protection Agency (EPA) and the Netherlands National Institute for Public Health and the Environment (RIVM). In 2016, Dr. Loo, the lead Principal Investigator of our group, was awarded the Lush Prize, an international award for animal-free toxicology research.

Our current research is focused on three major areas, namely toxicodynamics of xenobiotics, pulmonary effects of xenobiotics, and phenotypic profiling and computational biology (Fig. 1).

Figure 1
Figure 1: Our current research areas

I. Toxicodynamics of xenobiotics

Many xenobiotics have unknown and/or non-specific intracellular targets. To study the toxicodynamics of these chemicals, unbiased approaches that do not require prior information about the targets or mechanisms of the chemicals are required. Our goal is to elucidate the MoAs of xenobiotics in major target cell types using advanced phenotypic, signaling, and genomic profiling methods.

Phenotypic responses: Toxic xenobiotics often impair cellular functions and lead to changes in cellular phenotypes, such as reorganization of subcellular structures, up/down-regulation of biomolecules, or other phenotypes (Fig. 2). Therefore, quantitative readouts based on changes in cellular phenotypes (Bougen-Zhukov et al., 2017) may be used as surrogate markers for predicting the toxicity of these chemicals. We have developed the first high-throughput and predictive in vitro nephrotoxicity assay (Loo et al., 2017a; Su et al., 2016). We identified a small set of chromatin and cytoskeletal features that can predict the human in vivo PTC toxicity of 44 reference compounds with ~82 % (primary PTCs) or 89 % (immortalized PTCs) accuracies. Our results suggest that a DNA damage response is commonly induced by different PTC toxicants that have diverse chemical structures and injury mechanisms.

Figure 2
Figure 2: Immunofluorescence microscopy images of human lung cells showing different phenotypic responses to non-toxic (blue) and toxic (red) chemicals.

Signaling responses: Besides cell injury, toxic xenobiotics may also induce signaling or inflammatory responses in their targeted cell types. We have developed a rapid, signaling-based cytotoxicity assay that may be used to predict cellular sensitivity to a cytotoxic agent, or identify co-treatments that may sensitize or desensitize cells to the agent (Loo et al., 2017b). We show that early spatiotemporal-specific changes in the phosphorylation levels of intermediate signaling proteins are sufficient to represent the complex cellular responses to a cytotoxic agent and other co-treatments. We have also developed a predictive nephrotoxicity assay based on the RNA expression levels of two pro-inflammatory cytokines, namely interleukin (IL)-6 and -8 (Kandasamy et at., 2015; Su et al., 2014). Although the assay only has two measurements, it can accurately predict the toxicity of xenobiotics in both primary human PTCs and induced pluripotent stem cells (iPSC)-derived PTC-like cells. These results suggest that inflammation is a general response of PTCs to PTC-toxic compounds.

Transcriptomic responses: The expressions of genes involved in key toxicity responses may be up- or down-regulated in response to toxic xenobiotics. Recent advances in transcriptomics technologies have enabled us to quantify these changes at the genome-wide level. In collaboration with Dr. Hoon from MEL, we are developing high-throughput transcriptomic methods to study concentration-dependent changes in key toxicity pathways.

Toxicity Mode-of-Action Discovery (ToxMAD) Platform:Together with four other research institutes in A*STAR, we are using various new molecular and phenotypic profiling technologies developed in A*STAR to elucidate the protein targets and MoAs of xenobiotics with high human exposure or safety concerns. Our focus is to study chemical analogs with related structures but differential cellular effects, and develop fit-for-purpose assays that will be used by regulatory agencies and industrial research laboratories to assess chemical safety.

II. Pulmonary effects of xenobiotics

Human lungs are exposed to inhaled or blood-borne soluble xenobiotics that may originate from the environment, food, consumer products, and/or pharmaceuticals. We are broadly interested in the understanding the biological targets and pathways affected by these chemicals in the lung cells.

In vitro toxicity models: We have recently developed a high-throughput and predictive in vitro pulmonary toxicity assay (Fig 2; Lee et al., 2018). We found that the resulting assay based on two phenotypic features of a human bronchial epithelial cell line, BEAS-2B, can accurately classify 33 reference chemicals with human pulmonotoxicity information (88.8% balance accuracy, 84.6% sensitivity, and 93.0% specificity). In comparison, the predictivity of a standard cell-viability assay on the same set of chemicals is much lower (77.1% balanced accuracy, 84.6% sensitivity, and 69.5% specificity). We also used the assay to evaluate 17 additional test chemicals with unknown/unclear human pulmonotoxicity, and experimentally confirmed that many of the pulmonotoxic reference and predicted-positive test chemicals induce DNA strand breaks and/or activation of the DNA-damage response (DDR) pathway.

Xenobiotic metabolism: In the lungs, bronchial and alveolar epithelial cells are major sites of xenobiotic metabolism, and thus are susceptible to the toxicity induced by xenobiotics that interfere with this process. In collaboration with Dr. Hao Fan from BII, we are studying the mechanisms of xenobiotics that can inhibit Cytochrome P450 family 1 member A1 (CYP1A1), a main extra-hepatic phase I metabolism enzyme highly expressed in the lungs and placenta. We have developed molecular docking models that can be used to predict potential CYP1A1 inhibitors.

III. Phenotypic profiling and computational biology

To extract biological information from the large amount of collected data, new and better methods and tools for image and data analysis are required. Most of our projects are based on the HIPPTox Platform and the cellXpress software developed by us. Our group also develops new methodologies for concentration response modeling, artificial intelligence, and assay automation.

High-throughput In-vitro Phenotypic Profiling (HIPPTox):Phenotypic profiling is a computational procedure to construct quantitative and compact representations of cellular phenotypes based on the cellular images collected in high-content imaging (HCI) experiments (Bougen-Zhukov et al., 2017). We have developed several computational methods for phenotypic profiling, which include the Drug-Profiling ("D-profiling") algorithm (Loo et al., 2007) and the Protein-localization Profiling ("P-profiling") algorithm (Loo et al., 2014). We have used the phenotypic profiles constructed using these methods to classify the effects of small molecules (Loo et al., 2007, 2009), compare spatial and functional divergence of proteins (Loo et al., 2014), or predict toxicity effects of xenobiotic compounds (Su et al., 2016). The High-throughput In-vitro Phenotypic Profiling for Toxicity Prediction (HIPPTox) Platform implement many of these methods, and can be used to detect in vitro bioactivity of chemicals and build predictive in vitro toxicity assays (Fig. 3). The core of the platform is a user-friendly and high-performance phenotypic profiling software called "cellXpress" (Fig. 4; Laksameethanasan et al., 2013), which can handle terabytes of image data and quantify millions of individual cells under different experimental conditions. We have applied HIPPTox to build predictive lung (Lee et al., 2018), kidney (Su et al., 2016), and liver toxicity assays.

Figure 3
Figure 3: High-throughput In-vitro Phenotypic Profiling for Toxicity prediction (HIPPTox) Platform.
Figure 4
Figure 4: The cellXpress software is designed for fast and high-throughput analysis of cellular phenotypes based on microscopy images.

Concentration response modeling: A concentration response curve (CRC) is commonly used to model the relationship between the concentration and effect of a perturbagen. However, for automated perturbagen classification based on quantitative phenotypic features from HCI, it is unclear if commonly used CRC metrics, such as the "half-maximal effective concentration" (EC50) that reports perturbagen potency, are still optimal. We have performed a systematic study on the performances of different CRC metrics in classifying four HCI datasets that consist of phenotypic features from different cell and feature types. Our results suggest that efficacy metrics, especially at higher concentration values, are more likely to provide the most useful information for perturbagen classification. Therefore, HCI experiments should include measurements at high perturbagen concentrations, and efficacy metrics should always be analyzed when building supervised classifiers based on phenotypic features.

Complex Cellular Phenotype Analysis Members

Dr. LOO Lit Hsin
Principal Investigator
 
  Biography Details
  Lab Website
NameTitle
Dr. LOO Lit HsinPrincipal Investigator
Dr. HTWE Su SuPostdoctoral Fellow
Dr. BASU SreetamaPostdoctoral Fellow
Dr. MILLER James AlastairPostdoctoral Fellow
Ms. LEE Jia Ying JoeyResearch Officer
Ms. KONG Jia WenResearch Officer
Mr. FU Shufeng OscarResearch Officer
Mr. CAIN Paul EdwardSoftware Engineer
No Publications

We are a computational biology research group with members from different scientific disciplines, including chemistry, cell biology, computer science, and bioinformatics.

The overall goal of our research is to understand the modes of action (MoAs) of xenobiotics, and predict their human toxicity and/or efficacy. We develop and use novel phenotypic and molecular profiling methods to elucidate the MoAs of xenobiotics, and build computational models that can predict in vivo effects based on these MoAs.

We are part of the Innovations in Food and Chemical Safety (IFCS) Programme in A*STAR. We also collaborate with different academic, industrial, and governmental research groups, including the Institute of Bioengineering and Nanotechnology (IBN), Institute of Molecular and Cell Biology (IMCB), and Molecular Engineering Lab (MEL) from A*STAR; and the United States Environmental Protection Agency (EPA) and the Netherlands National Institute for Public Health and the Environment (RIVM). In 2016, Dr. Loo, the lead Principal Investigator of our group, was awarded the Lush Prize, an international award for animal-free toxicology research.

Our current research is focused on three major areas, namely toxicodynamics of xenobiotics, pulmonary effects of xenobiotics, and phenotypic profiling and computational biology (Fig. 1).

Figure 1
Figure 1: Our current research areas

I. Toxicodynamics of xenobiotics

Many xenobiotics have unknown and/or non-specific intracellular targets. To study the toxicodynamics of these chemicals, unbiased approaches that do not require prior information about the targets or mechanisms of the chemicals are required. Our goal is to elucidate the MoAs of xenobiotics in major target cell types using advanced phenotypic, signaling, and genomic profiling methods.

Phenotypic responses: Toxic xenobiotics often impair cellular functions and lead to changes in cellular phenotypes, such as reorganization of subcellular structures, up/down-regulation of biomolecules, or other phenotypes (Fig. 2). Therefore, quantitative readouts based on changes in cellular phenotypes (Bougen-Zhukov et al., 2017) may be used as surrogate markers for predicting the toxicity of these chemicals. We have developed the first high-throughput and predictive in vitro nephrotoxicity assay (Loo et al., 2017a; Su et al., 2016). We identified a small set of chromatin and cytoskeletal features that can predict the human in vivo PTC toxicity of 44 reference compounds with ~82 % (primary PTCs) or 89 % (immortalized PTCs) accuracies. Our results suggest that a DNA damage response is commonly induced by different PTC toxicants that have diverse chemical structures and injury mechanisms.

Figure 2
Figure 2: Immunofluorescence microscopy images of human lung cells showing different phenotypic responses to non-toxic (blue) and toxic (red) chemicals.

Signaling responses: Besides cell injury, toxic xenobiotics may also induce signaling or inflammatory responses in their targeted cell types. We have developed a rapid, signaling-based cytotoxicity assay that may be used to predict cellular sensitivity to a cytotoxic agent, or identify co-treatments that may sensitize or desensitize cells to the agent (Loo et al., 2017b). We show that early spatiotemporal-specific changes in the phosphorylation levels of intermediate signaling proteins are sufficient to represent the complex cellular responses to a cytotoxic agent and other co-treatments. We have also developed a predictive nephrotoxicity assay based on the RNA expression levels of two pro-inflammatory cytokines, namely interleukin (IL)-6 and -8 (Kandasamy et at., 2015; Su et al., 2014). Although the assay only has two measurements, it can accurately predict the toxicity of xenobiotics in both primary human PTCs and induced pluripotent stem cells (iPSC)-derived PTC-like cells. These results suggest that inflammation is a general response of PTCs to PTC-toxic compounds.

Transcriptomic responses: The expressions of genes involved in key toxicity responses may be up- or down-regulated in response to toxic xenobiotics. Recent advances in transcriptomics technologies have enabled us to quantify these changes at the genome-wide level. In collaboration with Dr. Hoon from MEL, we are developing high-throughput transcriptomic methods to study concentration-dependent changes in key toxicity pathways.

Toxicity Mode-of-Action Discovery (ToxMAD) Platform:Together with four other research institutes in A*STAR, we are using various new molecular and phenotypic profiling technologies developed in A*STAR to elucidate the protein targets and MoAs of xenobiotics with high human exposure or safety concerns. Our focus is to study chemical analogs with related structures but differential cellular effects, and develop fit-for-purpose assays that will be used by regulatory agencies and industrial research laboratories to assess chemical safety.

II. Pulmonary effects of xenobiotics

Human lungs are exposed to inhaled or blood-borne soluble xenobiotics that may originate from the environment, food, consumer products, and/or pharmaceuticals. We are broadly interested in the understanding the biological targets and pathways affected by these chemicals in the lung cells.

In vitro toxicity models: We have recently developed a high-throughput and predictive in vitro pulmonary toxicity assay (Fig 2; Lee et al., 2018). We found that the resulting assay based on two phenotypic features of a human bronchial epithelial cell line, BEAS-2B, can accurately classify 33 reference chemicals with human pulmonotoxicity information (88.8% balance accuracy, 84.6% sensitivity, and 93.0% specificity). In comparison, the predictivity of a standard cell-viability assay on the same set of chemicals is much lower (77.1% balanced accuracy, 84.6% sensitivity, and 69.5% specificity). We also used the assay to evaluate 17 additional test chemicals with unknown/unclear human pulmonotoxicity, and experimentally confirmed that many of the pulmonotoxic reference and predicted-positive test chemicals induce DNA strand breaks and/or activation of the DNA-damage response (DDR) pathway.

Xenobiotic metabolism: In the lungs, bronchial and alveolar epithelial cells are major sites of xenobiotic metabolism, and thus are susceptible to the toxicity induced by xenobiotics that interfere with this process. In collaboration with Dr. Hao Fan from BII, we are studying the mechanisms of xenobiotics that can inhibit Cytochrome P450 family 1 member A1 (CYP1A1), a main extra-hepatic phase I metabolism enzyme highly expressed in the lungs and placenta. We have developed molecular docking models that can be used to predict potential CYP1A1 inhibitors.

III. Phenotypic profiling and computational biology

To extract biological information from the large amount of collected data, new and better methods and tools for image and data analysis are required. Most of our projects are based on the HIPPTox Platform and the cellXpress software developed by us. Our group also develops new methodologies for concentration response modeling, artificial intelligence, and assay automation.

High-throughput In-vitro Phenotypic Profiling (HIPPTox):Phenotypic profiling is a computational procedure to construct quantitative and compact representations of cellular phenotypes based on the cellular images collected in high-content imaging (HCI) experiments (Bougen-Zhukov et al., 2017). We have developed several computational methods for phenotypic profiling, which include the Drug-Profiling ("D-profiling") algorithm (Loo et al., 2007) and the Protein-localization Profiling ("P-profiling") algorithm (Loo et al., 2014). We have used the phenotypic profiles constructed using these methods to classify the effects of small molecules (Loo et al., 2007, 2009), compare spatial and functional divergence of proteins (Loo et al., 2014), or predict toxicity effects of xenobiotic compounds (Su et al., 2016). The High-throughput In-vitro Phenotypic Profiling for Toxicity Prediction (HIPPTox) Platform implement many of these methods, and can be used to detect in vitro bioactivity of chemicals and build predictive in vitro toxicity assays (Fig. 3). The core of the platform is a user-friendly and high-performance phenotypic profiling software called "cellXpress" (Fig. 4; Laksameethanasan et al., 2013), which can handle terabytes of image data and quantify millions of individual cells under different experimental conditions. We have applied HIPPTox to build predictive lung (Lee et al., 2018), kidney (Su et al., 2016), and liver toxicity assays.

Figure 3
Figure 3: High-throughput In-vitro Phenotypic Profiling for Toxicity prediction (HIPPTox) Platform.
Figure 4
Figure 4: The cellXpress software is designed for fast and high-throughput analysis of cellular phenotypes based on microscopy images.

Concentration response modeling: A concentration response curve (CRC) is commonly used to model the relationship between the concentration and effect of a perturbagen. However, for automated perturbagen classification based on quantitative phenotypic features from HCI, it is unclear if commonly used CRC metrics, such as the "half-maximal effective concentration" (EC50) that reports perturbagen potency, are still optimal. We have performed a systematic study on the performances of different CRC metrics in classifying four HCI datasets that consist of phenotypic features from different cell and feature types. Our results suggest that efficacy metrics, especially at higher concentration values, are more likely to provide the most useful information for perturbagen classification. Therefore, HCI experiments should include measurements at high perturbagen concentrations, and efficacy metrics should always be analyzed when building supervised classifiers based on phenotypic features.

Complex Cellular Phenotype Analysis Members

Dr. LOO Lit Hsin
Senior Principal Investigator
 
  Biography Details
NameTitle
Dr. LOO Lit HsinPrincipal Investigator
Dr. HTWE Su SuPostdoctoral Fellow
Dr. BASU SreetamaPostdoctoral Fellow
Dr. MILLER James AlastairPostdoctoral Fellow
Ms. LEE Jia Ying JoeyResearch Officer
Ms. KONG Jia WenResearch Officer
Mr. FU Shufeng OscarResearch Officer
Mr. CAIN Paul EdwardSoftware Engineer
No Publications