Improving the predictive value of interventional animal models data

Chemical entity normalization for successful translational development of Alzheimer's disease and dementia therapeutics

BACKGROUND

Identifying chemical mentions within the Alzheimer's and dementia literature can provide a powerful tool to further therapeutic research. Leveraging the Chemical Entities of Biological Interest (ChEBI) ontology, which is rich in hierarchical and other relationship types, for entity normalization can provide an advantage for future downstream applications. We provide a reproducible hybrid approach that combines an ontology-enhanced PubMedBERT model for disambiguation with a dictionary-based method for candidate selection.

RESULTS

There were 56,553 chemical mentions in the titles of 44,812 unique PubMed article abstracts. Based on our gold standard, our method of disambiguation improved entity normalization by 25.3 percentage points compared to using only the dictionary-based approach with fuzzy-string matching for disambiguation. For the CRAFT corpus, our method outperformed baselines (maximum 78.4%) with a 91.17% accuracy. For our Alzheimer's and dementia cohort, we were able to add 47.1% more potential mappings between MeSH and ChEBI when compared to BioPortal.

CONCLUSION

Use of natural language models like PubMedBERT and resources such as ChEBI and PubChem provide a beneficial way to link entity mentions to ontology terms, while further supporting downstream tasks like filtering ChEBI mentions based on roles and assertions to find beneficial therapies for Alzheimer's and dementia.

Chemical Entity Normalization for Successful Translational Development of Alzheimer's Disease and Dementia Therapeutics

Background

Identifying chemical mentions within the Alzheimer's and dementia literature can provide a powerful tool to further therapeutic research. Leveraging the Chemical Entities of Biological Interest (ChEBI) ontology, which is rich in hierarchical and other relationship types, for entity normalization can provide an advantage for future downstream applications. We provide a reproducible hybrid approach that combines an ontology-enhanced PubMedBERT model for disambiguation with a dictionary-based method for candidate selection.

Results

There were 56,553 chemical mentions in the titles of 44,812 unique PubMed article abstracts. Based on our gold standard, our method of disambiguation improved entity normalization by 25.3 percentage points compared to using only the dictionary-based approach with fuzzy-string matching for disambiguation. For our Alzheimer's and dementia cohort, we were able to add 47.1% more potential mappings between MeSH and ChEBI when compared to BioPortal.

Conclusion

Use of natural language models like PubMedBERT and resources such as ChEBI and PubChem provide a beneficial way to link entity mentions to ontology terms, while further supporting downstream tasks like filtering ChEBI mentions based on roles and assertions to find beneficial therapies for Alzheimer's and dementia.

Applications of Induced Pluripotent Stem Cell-Derived Glia in Brain Disease Research and Treatment

Glia are integral components of neural networks and are crucial in both physiological functions and pathological processes of the brain. Many brain diseases involve glial abnormalities, including inflammatory changes, mitochondrial damage, calcium signaling disturbance, hemichannel opening, and loss of glutamate transporters. Induced pluripotent stem cell (iPSC)-derived glia provide opportunities to study the contributions of glia in human brain diseases. These cells have been used for human disease modeling as well as generating new therapies. This chapter introduces glial involvement in brain diseases, then summarizes different methods of generating iPSC-derived glia disease models of these cells. Finally, strategies for treating disease using iPSC-derived glia are discussed. The goal of this chapter is to provide an overview and shed light on the applications of iPSC-derived glia in brain disease research and treatment.

Utilising Induced Pluripotent Stem Cells in Neurodegenerative Disease Research: Focus on Glia

Induced pluripotent stem cells (iPSCs) are a self-renewable pool of cells derived from an organism's somatic cells. These can then be programmed to other cell types, including neurons. Use of iPSCs in research has been two-fold as they have been used for human disease modelling as well as for the possibility to generate new therapies. Particularly in complex human diseases, such as neurodegenerative diseases, iPSCs can give advantages over traditional animal models in that they more accurately represent the human genome. Additionally, patient-derived cells can be modified using gene editing technology and further transplanted to the brain. Glial cells have recently become important avenues of research in the field of neurodegenerative diseases, for example, in Alzheimer's disease and Parkinson's disease. This review focuses on using glial cells (astrocytes, microglia, and oligodendrocytes) derived from human iPSCs in order to give a better understanding of how these cells contribute to neurodegenerative disease pathology. Using glia iPSCs in in vitro cell culture, cerebral organoids, and intracranial transplantation may give us future insight into both more accurate models and disease-modifying therapies.

Mechanistic new insights of flavonols on neurodegenerative diseases

With a large and increasing elderly population, neurodegenerative diseases such as Parkinson's disease (PD), Huntington disease (HD), Alzheimer's disease (AD), Amyotrophic lateral sclerosis (ALS) and Multiple sclerosis (MS) have become a major and growing health problem. During the past few decades, the elderly population has grown 2.5 % every year. Unfortunately, there are no specific therapeutic remedies available to slow the onset or development of these diseases. An aging brain causes many pathophysiological changes and is the major risk factor for most of the neurodegenerative disorders. Polyphenolic compounds such as flavonols have shown therapeutic potential and can contribute to the treatment of these diseases. In this review, evidence for the beneficial neuroprotective effect of multiple flavonols is discussed and their multifactorial cellular pathways for the progressions of age-associated brain changes are identified. Moreover, the animal models of these diseases support the neuroprotective effect and target the potential of flavonols in the treatment of neurodegenerative diseases.

Levelling the Translational Gap for Animal to Human Efficacy Data

Reports of a reproducibility crisis combined with a high attrition rate in the pharmaceutical industry have put animal research increasingly under scrutiny in the past decade. Many researchers and the general public now question whether there is still a justification for conducting animal studies. While criticism of the current modus operandi in preclinical research is certainly warranted, the data on which these discussions are based are often unreliable. Several initiatives to address the internal validity and reporting quality of animal studies (e.g., Animals in Research: Reporting In Vivo Experiments (ARRIVE) and Planning Research and Experimental Procedures on Animals: Recommendations for Excellence (PREPARE) guidelines) have been introduced but seldom implemented. As for external validity, progress has been virtually absent. Nonetheless, the selection of optimal animal models of disease may prevent the conducting of clinical trials, based on unreliable preclinical data. Here, we discuss three contributions to tackle the evaluation of the predictive value of animal models of disease themselves. First, we developed the Framework to Identify Models of Disease (FIMD), the first step to standardise the assessment, validation and comparison of disease models. FIMD allows the identification of which aspects of the human disease are replicated in the animals, facilitating the selection of disease models more likely to predict human response. Second, we show an example of how systematic reviews and meta-analyses can provide another strategy to discriminate between disease models quantitatively. Third, we explore whether external validity is a factor in animal model selection in the Investigator's Brochure (IB), and we use the IB-derisk tool to integrate preclinical pharmacokinetic and pharmacodynamic data in early clinical development. Through these contributions, we show how we can address external validity to evaluate the translatability and scientific value of animal models in drug development. However, while these methods have potential, it is the extent of their adoption by the scientific community that will define their impact. By promoting and adopting high quality study design and reporting, as well as a thorough assessment of the translatability of drug efficacy of animal models of disease, we will have robust data to challenge and improve the current animal research paradigm.

Overview of Transgenic Mouse Models for Alzheimer's Disease

This review describes several transgenic mouse models of Alzheimer's disease (AD), a devastating neurodegenerative disorder that causes progressive cognitive decline and is diagnosed postmortem by the presence of extracellular amyloid-β (Aβ) plaques and intraneuronal tau neurofibrillary tangles in the cerebral cortex. Currently there is no intervention that cures, prevents, or even slows disease progression. Its complex etiology and pathology pose significant challenges for animal model development, and there is no single model that faithfully recapitulates both the pathological aspects and behavioral phenotypes of AD. Nearly 200 transgenic rodent models of AD have been generated primarily based on mutations linked to Aβ protein misprocessing in the familial form of the disease. More recent models incorporate mutations in tau protein, as well as mutations associated with the sporadic form of the disease. The salient features, strengths, limitations, and key differentiators for the most commonly used and best characterized of these models are considered here. While the translational utility of many of these models to assess the potential of novel therapeutics is in dispute, knowledge of the different models available and a detailed understanding of their features can aid in the selection of the optimal model to explore disease mechanisms or evaluate candidate medications. We comment on the predictive utility of these models considering recent clinical trial failures and discuss trends and future directions in the development of models for AD based on the plethora of clinical data that have been generated over the last decade. © 2019 by John Wiley & Sons, Inc.

Evolving challenges to model human diseases for translational research

Animal models are a significant component of biomedical research and play an important role in translational studies. Traditionally, rodent models have been the mainstay and principal choice of researchers but in recent years, there have been significant changes in the landscape of animal modeling. For example, newer techniques have greatly expanded the use and successful application of large animal models such as pigs for translational studies. The evolving types and species of animal models can influence the research landscape in terms of facilities, expertise, reproducibility and funding streams, which creates new challenges for research studies. It is also important that investigators are prepared to address the necessity of their animal model research and capable to educate the public regarding its value.

Menagerie: A text-mining tool to support animal-human translation in neurodegeneration research

Discovery studies in animals constitute a cornerstone of biomedical research, but suffer from lack of generalizability to human populations. We propose that large-scale interrogation of these data could reveal patterns of animal use that could narrow the translational divide. We describe a text-mining approach that extracts translationally useful data from PubMed abstracts. These comprise six modules: species, model, genes, interventions/disease modifiers, overall outcome and functional outcome measures. Existing National Library of Medicine natural language processing tools (SemRep, GNormPlus and the Chemical annotator) underpin the program and are further augmented by various rules, term lists, and machine learning models. Evaluation of the program using a 98-abstract test set achieved F1 scores ranging from 0.75-0.95 across all modules, and exceeded F1 scores obtained from comparable baseline programs. Next, the program was applied to a larger 14,481 abstract data set (2008-2017). Expected and previously identified patterns of species and model use for the field were obtained. As previously noted, the majority of studies reported promising outcomes. Longitudinal patterns of intervention type or gene mentions were demonstrated, and patterns of animal model use characteristic of the Parkinson's disease field were confirmed. The primary function of the program is to overcome low external validity of animal model systems by aggregating evidence across a diversity of models that capture different aspects of a multifaceted cellular process. Some aspects of the tool are generalizable, whereas others are field-specific. In the initial version presented here, we demonstrate proof of concept within a single disease area, Parkinson's disease. However, the program can be expanded in modular fashion to support a wider range of neurodegenerative diseases.

Are some animal models more equal than others? A case study on the translational value of animal models of efficacy for Alzheimer's disease

Clinical trial failures (>99%) in Alzheimer's disease are in stark contrast to positive efficacy data in animals. We evaluated the correlation between animal and clinical efficacy outcomes (cognition) in Alzheimer's disease using data from registered drugs as well as interventions tested in phase II or III clinical trials for Alzheimer's disease. We identified 20 interventions, which were tested in 208 animal studies in 63 different animal models. Clinical outcome was correlated with animal results in 58% of cases. But, individual animal models showed divergent results across interventions, individual interventions showed divergent results across animal models, and animal model outcomes were determined with 16 different methods. This result is unsurprising due to poor external validity (what do we model) of the animal models. Although the animal models all share Alzheimer's disease symptoms, none represents the whole syndrome. Investigators did not motivate why one model was chosen over another, and did not consider the ways the disease phenomena were generated (spontaneous, (experimentally) induced or by genetic modification), or the species characteristics, which determine the outcomes. The explanation for the lack of correlation between animal and human outcomes can be manifold: the pathogenesis of Alzheimer's disease is not reflected in the animal model or the outcomes are not comparable. Our conclusion is that currently no animal models exist which are predictive for the efficacy of interventions for Alzheimer's disease.

Preclinical Models of Alzheimer's Disease: Relevance and Translational Validity

The only drugs currently approved for the treatment of Alzheimer's Disease (AD) are four acetylcholinesterase inhibitors and the NMDA antagonist memantine. Apart from these drugs, which have minimal to no clinical benefit, the 40-year search for effective therapeutics to treat AD has resulted in a clinical failure rate of 100% not only for compounds that prevent brain amyloid deposition or remove existing amyloid plaques but also those acting by a variety of other putative disease-associated mechanisms. This indicates that the preclinical data generated from current AD targets to support the selection, optimization, and translation of new chemical entities (NCEs) and biologics to clinical trials is seriously compromised. While many of these failures reflect flawed hypotheses or a lack of adequate characterization of the preclinical pharmacodynamic and pharmacokinetic (PD/PK) properties of lead NCEs-including their bioavailability and toxicity-the conceptualization, validation, and interrogation of the current animal models of AD represent key limitations. The overwhelming majority of these AD models are transgenic, based on aspects of the amyloid hypothesis and the genetics of the familial form of the disease. As a result, these generally lack construct and predictive validity for the sporadic form of the human disease. The 170 or so transgenic models, perhaps the largest number ever focused on a single disease, use rodents, mainly mice, and in addition to amyloid also address aspects of tau causality with more complex multigene models including other presumed causative factors together with amyloid. This overview discusses the current animal models of AD in the context of both the controversies surrounding the causative role of amyloid in the disease and the need to develop validated models of cognitive function/dysfunction that more appropriately reflect the phenotype(s) of human aged-related dementias. © 2019 by John Wiley & Sons, Inc.

From Reproducibility to Translation in Neurodegenerative Disease

Despite tremendous investment and preclinical success in neurodegenerative disease, effective disease-altering treatments for patients have remained elusive. One highly cited reason for this discrepancy is flawed animal study design and reporting. If this can be broadly remedied, reproducibility of preclinical studies will improve. However, without concurrent efforts to improve generalizability, these improvements may not translate effectively from animal experiments to more complex human neurodegenerative diseases. Mechanistic and phenotypic variability of neurodegenerative disease is such that most models are only able to interrogate individual aspects of complex phenomena. One approach is to consider animals as models of individual targets rather than as models of individual diseases and to migrate the concept of predictive validity from the individual model to the body of experiments that demonstrate translatability of a target. Both exploratory and therapeutic preclinical studies are dependent upon study design methods that promote rigor and reproducibility. However, the body of evidence that is needed to demonstrate efficacy in therapeutic studies is substantially broader than that needed for exploratory studies. In addition to requiring rigor within individual experiments, convincing evidence for therapeutic potential must assess the relationships between model choice, intended goal of the intervention, pharmacologic criteria, and integration of biomarker data with outcome measures that are clinically relevant to humans. It is conceivable that proof-of-concept studies will migrate to cell-based systems and that animal systems will be increasingly reserved for more distal translational purposes. If this occurs, it is likely to prompt reexamination of what the term "translational" truly means.

Established patterns of animal study design undermine translation of disease-modifying therapies for Parkinson's disease

Translation of disease-modifying therapies in neurodegenerative disease has been disappointing. Parkinson's disease (PD) was used to compare patterns of preclinical study design for symptomatic and potentially disease-modifying interventions. We examined the relationship of model, intervention type and timing, outcomes and outcome measures in 543 animal and human studies (1973-2015) across a contemporary cohort of animal and human interventional studies (n = 445), animal studies for approved interventions (n = 28), animal and human studies for those that failed to translate (n = 70). Detailed study design data were collected for 216 studies in non-human primate (NHP) and rodent toxin-induced models. Species-specific patterns of study design prevailed regardless of whether interventions were symptomatic or potentially disease-modifying. In humans and NHPs, interventions were typically given to both sexes well after the PD phenotype was established, and clinical outcome measures were collected at single (symptomatic) or multiple (disease-modifying) time-points. In rodents, interventions often preceded induction of the model, acute toxic protocols were common, usually given to young males, clinical outcome measures were used less commonly, and outcomes were less commonly assessed at multiple time points. These patterns were more prevalent in mice than rats. In contrast, study design factors such as randomization and blinding did not differ appreciably across symptomatic and disease-modifying intervention categories. The translational gap for potentially disease-modifying interventions in PD in part results from study designs, particularly in mice, that fail to model the progressive nature and relatively late intervention characteristic of PD, or that anchor mechanistic and neuropathologic data to longitudinal clinical outcomes. Even if measures to improve reproducibility are broadly adopted, perpetuation of these norms will continue to impede effective translation.