MPS World Summit 2025
June 9–13, 2025 | Brussels, Belgium
Join AxoSim at MPS World Summit – Booth # 117
Join AxoSim at MPS World Summit. We will be presenting three posters showcasing recent advancements across our CNS and PNS product and service portfolio. Be sure to book time ahead of the meeting to speak with one of our experts or stop by booth #117 to say hello.
Peripheral nerve microphysiological system for screening neuropathy of small molecule and PROTAC chemotherapeutics
Abstract
Eva Schmidt1, Megan Morris1, Megan Terral1, Lowry Curley1, Michael Moore1, 2, 3, Corey Rountree1
1AxoSim Inc., New Orleans, LA, USA, 2 Department of Biomedical Engineering, Tulane University, New Orleans, LA, USA, 3Brain Institute, Tulane University, New Orleans, LA, USA
Microphysiological systems (MPS) have the potential to better inform preclinical stages of drug development by enabling toxicity screening with systems that mimic in vivo physiology. These systems are attracting attention from the pharmaceutical industry in the hope they will curb attrition rates, lower costs, and reduce reliance on animal models. AxoSim has developed an innovative MPS, the NerveSim® platform, for screening neurotoxic compounds using an embedded electrode array (EEA) to record compound action potentials (CAPs) from peripheral nerve cultures. This model comprises a coculture of human iPSC-derived sensory neurons and human primary Schwann cells in a 24 well plate format with each well containing 10 electrodes allowing longitudinal stimulation and recording of CAPs along a 9 mm 3D nerve-fiber.
In this work, we demonstrate NerveSim®’s value as a neurotoxic screening platform by conducting dose response analyses of 6 small molecule and PROTAC (proteolysis targeting chimera) chemotherapeutics known to cause chemotherapy-induced peripheral neuropathy (CIPN): Bortezomib, MMAE, Oxaliplatin, Vincristine, Paclitaxel, and Thalidomide plus Acetaminophen as a negative control. All compounds were administered for 7 days with electrophysiological and brightfield imaging endpoints recorded before and after dosing on the same samples. Electrophysiological data were analyzed to extract conduction velocity (CV), peak response amplitude (AMP), and the velocity density index (VDI), which sums the area-under-the-curve for all electrophysiological responses in a well to create a single metric combining CV and AMP. Imaging metrics were calculated from z-stack brightfield images consisting of fiber length, fiber count, and degeneration index (neurite fragmentation). IC50 curves were generated for each metric after normalizing to the baseline conditions before compound dosing on a sample-by-sample basis. Dosing with higher concentrations of each small molecule resulted in significant deficits in both electrophysiology and imaging metrics compared to the vehicle and negative control, consistent with peripheral neuropathy. Additionally, the PROTAC Thalidomide demonstrated peripheral neuropathy in this system, which no other in vitro or animal model has been able to recapitulate. This comprehensive suite of tools provides a novel approach for pre-clinical assessment of functional and morphological neurotoxicity, neuroprotection, pain, and their respective mechanisms of action.
Development of an AI-based target identification model for anticonvulsant drug discovery
Abstract
Andrew S. LaCroix1, Thomas Luechtefeld2, Nicholas S. Coungeris1, Victoria K. Alstat1, Corey Rountree3
1AxoSim, Inc. Maple Grove, MN, USA, 2InSilica.co Santa Clara, CA, USA, and 3AxoSim, Inc. New Orleans, LA, USA
Epilepsy is a prevalent neurological disorder affecting approximately 79 million people worldwide. Despite advances in treatment, about 30-40% of these patients experience poorly controlled seizures or significant adverse effects from current medications1. The development of novel anticonvulsants has largely been empiric, with few therapies addressing the underlying pathophysiology. Consequently, there is a critical and unmet need for antiepileptic drugs that target the biological mechanisms of the disease rather than merely its symptoms2. In response, we have developed an AI-based target identification algorithm trained on functional screening measurements from neural organoids. This approach enables the prediction of biological targets that could potentially reverse epilepsy phenotypes in vitro. We screened 880 neuroactive compounds in healthy human iPSC-derived neural organoids to build a training dataset of functional perturbations across a diverse set of biological targets. We supplemented this dataset with protein information from UniProt to construct a transformer neural network AI model that relates functional waveform changes to protein perturbations. Our evaluation of the AI-predicted anticonvulsant targets involved several methods. Initially, we noted a 90% cosine similarity between the predicted and actual waveforms across the dataset. An unsupervised clustering approach was also employed to determine the likelihood of predicted waveforms resembling genuine anticonvulsant waveforms, achieving 63% accuracy. Moreover, a small machine learning-based classifier trained on the raw waveforms of 43 known anticonvulsant drugs demonstrated 70% accuracy when applied to AI-predicted waveforms. These tests collectively highlight the utility of our AI target identification model in predicting known anticonvulsants and underscore its potential to discover anticonvulsants with novel mechanisms through an unbiased reliance on protein information and functional human organoid data. With further refinement and broader training incorporating disease-specific organoid biology, this model could pave the way for innovative target identification across various neurodevelopmental and neurodegenerative disorders.
References
1 Ngugi, A.K., Bottomley, C., Kleinschmidt, I., Wagner, R.G., Kakooza-Mwesige, A., Ae-Ngibise, K., Owusu-Agyei, S., Masanja, H., Kamuyu, G., Odhiambo, R. and Chengo, E. (2013). The Lancet Neurology, 12(3), 253-263. doi: 10.1016/S1474-4422(13)70003-6
2 Hakami, T. (2021). Therapeutic Advances in Neurological Disorders. 14, 17562864211037430. doi: 10.1177/17562864211037430
Improving the stability and reproducibility of clinical neurotoxicity predictions from a high-throughput compatible neural organoid platform.
Abstract
Andrew S. LaCroix1, Victoria K. Alstat1, Nicholas S. Coungeris1, David Gallegos2, J Lowry Curley3
1AxoSim Inc., Maple Grove, MN, USA. 2Takeda Pharmaceuticals, Cambridge, MA, USA, 3AxoSim Inc., New Orleans, LA, USA
The drug development process frequently encounters failures due to issues related to safety or efficacy, with neurotoxicity being a primary cause of clinical setbacks. Recent advances in complex in vitro models (CIVM) derived from human tissues have enhanced our ability to recapitulate biological complexity, thus improving the translation from preclinical findings to clinical outcomes. These models, when adopted early in the drug development pipeline, have shown potential in increasing the success rates of new therapeutics (Kang, et al., Biofabrication 2024). Our development of a human induced pluripotent stem cell (iPSC)-derived cortical organoid platform epitomizes this advancement. This platform integrates functional endpoints suitable for high-throughput screening and features reproducible "waveform" activity that can be modulated by excitatory and inhibitory compounds. This confirms the successful integration of glutamatergic and GABAergic circuitry, providing a viable alternative to traditional animal-based preclinical models. In 2022, we utilized this organoid platform to devise a predictive model for clinical neurotoxicity, achieving over 90% specificity and greater than 50% sensitivity. This model represents an ideal method for preliminary screening before standard 2-species animal testing (Wang, et al., ALTEX 2022). We have rigorously tested the stability and reproducibility of our predictions, refining our processes for neurotoxicity score calculation through advances in peak detection, waveform feature extraction, and automated potency calculations. We further validated the stability of neurotoxicity prediction through consistent performance in independent experiments conducted at different sites over several years. This robustness, evidenced by maintained high specificity, ensures that the model does not prematurely exclude potentially effective drugs. These findings strongly support the integration of CIVMs into preclinical settings, promising to streamline the drug development process and reduce the incidence of costly clinical failures due to neurotoxicity.
Contact Us
Interested in scheduling a meeting with one of our experts while at the meeting? We would love to discuss your research needs and how AxoSim can help.