Machine learning interpretable models of cell mechanics from protein images.

Cellular form and function emerge from complex mechanochemical systems within the cytoplasm. Currently, no systematic strategy exists to infer large-scale physical properties of a cell from its molecular components. This is an obstacle to understanding processes such as cell adhesion and migration. Here, we develop a data-driven modeling pipeline to learn the mechanical behavior of adherent cells. We first train neural networks to predict cellular forces from images of cytoskeletal proteins. Strikingly, experimental images of a single focal adhesion (FA) protein, such as zyxin, are sufficient to predict forces and can generalize to unseen biological regimes. Using this observation, we develop two approaches-one constrained by physics and the other agnostic-to construct data-driven continuum models of cellular forces. Both reveal how cellular forces are encoded by two distinct length scales. Beyond adherent cell mechanics, our work serves as a case study for integrating neural networks into predictive models for cell biology.

Zyxin is all you need: machine learning adherent cell mechanics.

Cellular form and function emerge from complex mechanochemical systems within the cytoplasm. No systematic strategy currently exists to infer large-scale physical properties of a cell from its many molecular components. This is a significant obstacle to understanding biophysical processes such as cell adhesion and migration. Here, we develop a data-driven biophysical modeling approach to learn the mechanical behavior of adherent cells. We first train neural networks to predict forces generated by adherent cells from images of cytoskeletal proteins. Strikingly, experimental images of a single focal adhesion protein, such as zyxin, are sufficient to predict forces and generalize to unseen biological regimes. This protein field alone contains enough information to yield accurate predictions even if forces themselves are generated by many interacting proteins. We next develop two approaches - one explicitly constrained by physics, the other more agnostic - that help construct data-driven continuum models of cellular forces using this single focal adhesion field. Both strategies consistently reveal that cellular forces are encoded by two different length scales in adhesion protein distributions. Beyond adherent cell mechanics, our work serves as a case study for how to integrate neural networks in the construction of predictive phenomenological models in cell biology, even when little knowledge of the underlying microscopic mechanisms exist.

Temperature Affects the Biological Control of Dinoflagellates by the Generalist Parasitoid Parvilucifera rostrata.

The increase in emerging harmful algal blooms in the last decades has led to an extensive concern in understanding the mechanisms behind these events. In this paper, we assessed the growth of two blooming dinoflagellates (Alexandrium minutum and Heterocapsa triquetra) and their susceptibility to infection by the generalist parasitoid Parvilucifera rostrata under a temperature gradient. The growth of the two dinoflagellates differed across a range of temperatures representative of the Penzé Estuary (13 to 22 °C) in early summer. A. minutum growth increased across this range and was the highest at 19 and 22 °C, whereas H. triquetra growth was maximal at intermediate temperatures (15-18 °C). Interestingly, the effect of temperature on the parasitoid infectivity changed depending on which host dinoflagellate was infected with the dinoflagellate responses to temperature following a positive trend in A. minutum (higher infections at 20-22 °C) and a unimodal trend in H. triquetra (higher infections at 18 °C). Low temperatures negatively affected parasitoid infections in both hosts (i.e., "thermal refuge"). These results demonstrate how temperature shifts may not only affect bloom development in microalgal species but also their control by parasitoids.

Development of a high cell density transient CHO platform yielding mAb titers greater than 2 g/L in only 7 days.

We developed a simple transient Chinese Hamster Ovary expression platform. Titers for a random panel of 20 clinical monoclonal antibodies (mAbs) ranged from 0.6 to 2.7 g/L after 7 days. Two factors were the key in obtaining these high titers. First, we utilized an extremely high starting cell density (20 million cells/ml), and then arrested further cell growth by employing mild hypothermic conditions (32°C). Second, we performed a 6-variable Design of Experiments to find optimal concentrations of plasmid DNA (coding DNA), boost DNA (DNA encoding the XBP1S transcription factor), transfection reagent (polyethylenimine [PEI]), and nutrient feed amounts. High coding DNA concentrations (12.5 mg/L) were found to be optimal. We therefore diluted expensive coding DNA with inexpensive inert filler DNA (herring sperm DNA). Reducing the coding DNA concentration by 70% from 12.5 to 3.75 mg/L did not meaningfully reduce mAb titers. Titers for the same panel of 20 clinical mAbs ranged from 0.7 to 2.2 g/L after reducing the coding DNA concentration to 3.75 mg/L. Finally, we found that titer and product quality attributes were similar for a clinical mAb (rituximab) expressed at very different scales (volumes ranging from 3 ml to 2 L).

Application of Kaizen Principles to a Large Cochlear Implant Practice: A Continuous Quality Improvement Initiative at Mayo Clinic.

OBJECTIVE: To remove barriers and improve access for patients seeking cochlear implantation. STUDY DESIGN: Prospective quality improvement study at a large tertiary academic care center. METHODS: A Kaizen quality improvement model was applied over the course of a year. Four weeklong meetings were used to identify areas for improvement and remediation. Data were collected at baseline, 90-days, and 1-year postcompletion of the project. Outcome measures included lead times, defined as the wait time between first contact with the clinic and the first appointment, and the wait time between surgery and activation, and cycle times defined as the total test time needed to determine cochlear implant candidacy, and total time needed to complete initial activation. The total inventory kept as clinic stock was also calculated RESULTS:: Kaizen team members collected data for each outcome measure. After the Kaizen principles were applied, the following outcomes were observed: median lead times between first contact with the clinic to candidacy testing, candidacy testing to surgery, and surgery to activation of the implant remained stable from baseline to 1-year follow-up. Median cycle time for candidacy testing decreased from 7.3 hours at baseline to 3.0 hours at 1-year follow-up. Cycle times for initial activation of the device did not change over time. The total inventory of clinic stock was reduced by 31%. CONCLUSIONS: Though outcomes for lead and cycle times varied, implementation of Kaizen principles was found to be an effective method for completing this quality improvement project at a large cochlear implant program overall. LEVEL OF EVIDENCE: 3a.

Polymer-mediated flocculation of transient CHO cultures as a simple, high throughput method to facilitate antibody discovery.

Most biopharmaceutical drugs, especially monoclonal antibodies (mAbs), bispecific antibodies (BsAbs) and Fc-fusion proteins, are expressed using Chinese Hamster Ovary (CHO) cell lines. CHO cells typically yield high product titers and high product quality. Unfortunately, CHO cell lines also generate high molecular weight (HMW) aggregates of the desired product during cell culture along with CHO host cell protein (HCP) and CHO DNA. These immunogenic species, co-purified during Protein A purification, must be removed in a multi-step purification process. Our colleagues have reported the use of a novel polymer-mediated flocculation step to simultaneously reduce HMW, HCP and DNA from stable CHO cell cultures prior to Protein A purification. The objective of this study was to evaluate this novel "smart polymer" (SmP) in a high throughput antibody discovery workflow using transiently transfected CHO cultures. SmP treatment of 19 different molecules from four distinct molecular categories (human mAbs, murine mAbs, BsAbs and Fabs) with 0.1% SmP and 25 mM stimulus resulted in minimal loss of monomeric protein. Treatment with SmP also demonstrated a variable, concentration-dependent removal of HMW aggregates after Protein A purification. SmP treatment also effectively reduced HCP levels at each step of mAb purification with final HCP levels being several fold lower than the untreated control. Interestingly, SmP treatment was able to significantly reduce high concentrations of artificially spiked levels of endotoxin in the cultures. In summary, adding a simple flocculation step to our existing transient CHO process reduced the downstream purification burden to remove impurities and improved final product quality. © 2017 American Institute of Chemical Engineers Biotechnol. Prog., 33:1393-1400, 2017.

Evaluation of piggyBac-mediated CHO pools to enable material generation to support GLP toxicology studies.

Generating purified protein for GLP toxicology studies (GLP-Tox) represents an important and often rate limiting step in the biopharmaceutical drug development process. Toxicity testing requires large amounts of therapeutic protein (>100 g), typically produced in a single 500-2,500 L bioreactor, using the final CHO clonally derived cell line (CDCL). One approach currently used to save time is to manufacture GLP-Tox material using pools of high-producing CHO CDCLs instead of waiting for the final CDCL. Recently, we reported CHO pools producing mAb titers >7 g/L using piggyBac-mediated gene integration (PB CHO pools). In this study, we wanted to leverage high titer PB CHO pools to produce GLP-Tox material. A detailed product quality attribute (PQA) assessment was conducted comparing PB CHO pools to pooled Top4 CDCLs. Four mAbs were evaluated. First, we found that PB CHO pools expressed all four mAbs at high titers (2.8-4.4 g/L in shake flasks). Second, all four PB CHO pools were aged to 55 generations (Gen). All four PB CHO Pools were found to be suitable over 55 Gen. Finally, we performed bioreactor scale-up. PB CHO pool titers (3.7-4.8 g/L) were similar or higher than the pooled Top 4 CDCLs in 5 L bioreactors (2.4-4.1 g/L). The PQAs of protein derived from PB CHO pools were very similar to pooled Top 4 CHO CDCLs according to multiple orthogonal techniques including peptide mapping analysis. Taken together, these results demonstrate the technical feasibility of using PB CHO pools to manufacture protein for GLP-Tox. © 2017 American Institute of Chemical Engineers Biotechnol. Prog., 33:1436-1448, 2017.

Transcriptional and post-transcriptional targeting for enhanced transient gene expression in CHO cells.

OBJECTIVE: To develop a simple approach to increase titers of transient gene expression in CHO cells without relying on host cell line engineering as recent reports suggest that for PEI-mediated transfections, under optimized conditions, DNA delivery into cells and nuclei is not the limiting factor. RESULTS: N, N-Dimethyl acetamide (DMA) was utilized to enhance transcription. To target post-transcriptional events, we evaluated the co-expression of various genes involved in the unfolded protein response, namely XBP1S, ATF4, CHOP and HSPA5. XBP1S overexpression led to a 15-85 % increase in titer for multiple therapeutic proteins. Mechanistic studies confirmed that addition of 0.125 % DMA increased transgene mRNA levels as expected. However, overexpression of XBP1S had no effect on transgene mRNA levels, indicating that it influenced post-transcriptional events. Since DMA and XBP1S targeted different pathways, the combination of the two approaches led to an additive improvement in protein titer (150-250 % titer increase). CONCLUSION: Transcriptional and post-transcriptional pathways of transient gene expression can be targeted to increase titers without resorting to host cell line engineering in a simple, short, 7 day production process.

Pharmacological characterisation of ligand- and voltage-gated ion channels expressed in human iPSC-derived forebrain neurons.

INTRODUCTION: Genetic causes, or predisposition, are increasingly accepted to be part of the ethiopathogenesis of many neuropsychiatric diseases. While genes can be studied in any type of cells, their physiological function in human brain cells is difficult to evaluate, particularly in living subjects. METHODS: As a first step towards the characterisation of human inducible pluripotent stem cell (iPSC)-derived neurons from autism spectrum disorder (ASD) patients, we used gene expression and functional studies to define the regional identity of the typical forebrain differentiation, demonstrate expression patterns of genes of interest in ASD and understand the properties of 'control' iPSC-derived neurons (iCell-Neurons™), with a focus on receptors and ion channels that play a central role in synaptic physio-pathology. RESULTS AND DISCUSSION: The gene expression profile of the iCell-Neurons™ closely resembled that observed in neonatal prefrontal cortex tissues. Functional studies, performed mainly using calcium flux assays, demonstrated the presence of ionotropic glutamate (α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid and N-methyl-D-aspartate) and gamma-aminobutyric acid type A receptors. Voltage-gated sodium and calcium channels were also identified using similar techniques. CONCLUSIONS: Overall, the results reported here suggest that iCell-Neurons™ are a good cellular model of a relatively immature forebrain human neuron population that can be used both as a control in comparison to patients cells, and as host cells in which mutations, insertions and deletions can be used in order to study the molecular mechanisms of ASD and other neurological disorders in an isogenic cellular background.

Bacterial cyclic AMP-phosphodiesterase activity coordinates biofilm formation.

Biofilm-related infections are a major contributor to human disease, and the capacity for surface attachment and biofilm formation are key attributes for the pathogenesis of microbes. Serratia marcescens type I fimbriae-dependent biofilms are coordinated by the adenylate cyclase, CyaA, and the cyclic 3',5'-adenosine monophosphate (cAMP)-cAMP receptor protein (CRP) complex. This study uses S. marcescens as a model system to test the role of cAMP-phosphodiesterase activity in controlling biofilm formation. Herein we describe the characterization of a putative S. marcescens cAMP-phosphodiesterase gene (SMA3506), designated as cpdS, and demonstrated to be a functional cAMP-phosphodiesterase both in vitro and in vivo. Deletion of cpdS resulted in defective biofilm formation and reduced type I fimbriae production, whereas multicopy expression of cpdS conferred a type I fimbriae-dependent hyper-biofilm. Together, these results support a model in which bacterial cAMP-phosphodiesterase activity modulates biofilm formation.