A multi-laboratory evaluation of microelectrode array-based measurements of neural network activity for acute neurotoxicity testing

Accepted Manuscript Title: A multi-laboratory evaluation of microelectrode array-based measurements of neural network activity for acute neurotoxicity testing Author: Andrea Vassallo Michela Chiappalone Ricardo De Camargos Lopes Bibiana Scelfo Antonio Novellino Enrico Defranchi Taina Palosaari Timo Weisschu Tzutzuy Ramirez Sergio Martinoia Andrew F.M. Johnstone Cina M. Mack Robert Landsiedel Maurice Whealan Anna Bal-Price Timothy J. Shafer PII: DOI: Reference:

S0161-813X(16)30041-9 http://dx.doi.org/doi:10.1016/j.neuro.2016.03.019 NEUTOX 1968

To appear in:

NEUTOX

Received date: Revised date: Accepted date:

17-11-2015 25-2-2016 28-3-2016

Please cite this article as: Vassallo Andrea, Chiappalone Michela, De Camargos Lopes Ricardo, Scelfo Bibiana, Novellino Antonio, Defranchi Enrico, Palosaari Taina, Weisschu Timo, Ramirez Tzutzuy, Martinoia Sergio, Johnstone Andrew FM, Mack Cina M, Landsiedel Robert, Whealan Maurice, Bal-Price Anna, Shafer Timothy J.A multi-laboratory evaluation of microelectrode array-based measurements of neural network activity for acute neurotoxicity testing.Neurotoxicology http://dx.doi.org/10.1016/j.neuro.2016.03.019 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

A multi-laboratory evaluation of microelectrode array-based measurements of neural network activity for acute neurotoxicity testing† Andrea Vassallo1,2*, Michela Chiappalone1*, Ricardo De Camargos Lopes1,3, Bibiana Scelfo4, Antonio Novellino5, Enrico Defranchi5, Taina Palosaari4, Timo Weisschu6 Tzutzuy Ramirez6, Sergio Martinoia2, Andrew FM Johnstone7, Cina M Mack7, Robert Landsiedel6, Maurice Whealan4, Anna Bal-Price4 and Timothy J Shafer7 Affiliations 1

Department of Neuroscience and Brain Technologies, Istituto Italiano di Tecnologia, Via Morego 30, 16163, Genova (Italy) 2

Department of Infomatics Bioengineering, Robotics, SystemEngeneering, University of Genova, Genova, Italy. 3

Department of Clinical Engineering, University Hospital of Santa Maria, Av. Roraima, 1000 - Predio 22, Bairro Camobi, Santa Maria - RS – Brazil, CEP: 97105-900 4

Institute for Health and Consumer Protection, European Commission Joint Research Centre, Ispra, Italy 5

Alternative Toxicity Service Unit - ETT SpA, via Sestri 37, 16154, Genova (Italy)

6

Experimental Toxicology and Ecology, BASF, Carl Bosch-Strasse, 67056 Ludwigshafen am Rhein, Germany 7

National Health and Environmental Effects Research Laboratory, U.S. Environmental Protection Agency, Research Triangle Park, NC, USA *These authors contributed equally Corresponding Author Timothy J Shafer Integrated Systems Toxicology Division, MD-B105-03 Office of Research and Development U.S. Environmental Protection Agency Research Triangle Park, NC 27711 Phone: 919-541-0647 Fax: 919-541-4849 Email: [email protected] †

This manuscript has been reviewed by the National Health and Environmental Effects Research Laboratory, U.S. Environmental Protection Agency, and approved for publication. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.

1

Highlights     

Four laboratories tested 6 compounds for neuroactivity using microelectrode arrays All four laboratories were able to correctly identify the three neurotoxic compounds Three non-neuroactive compounds were correctly identified 10/12 times Despite methodological differences, results were consistent across laboratories These results support use of microelectrode arrays for neurotoxicity screening

Abstract There is a need for methods to screen and prioritize chemicals for potential hazard, including neurotoxicity. Microelectrode array (MEA) systems enable simultaneous extracellular recordings from multiple sites in neural networks in real time and thereby providing a robust measure of network activity. In this study, spontaneous activity measurements from primary neuronal cultures treated with three neurotoxic or three non-neurotoxic compounds was evaluated across four different laboratories. All four individual laboratories correctly identifed the neurotoxic compounds chlorpyrifos oxon (an organophosphate insecticide), deltamethrin (a pyrethroid insecticide) and domoic acid (an excitotoxicant). By contrast, the other three compounds (glyphosate, dimethyl phthalate and acetaminophen) considered to be nonneurotoxic (“negative controls”), produced only sporadic changes of the measured parameters. The results were consistent across the different laboratories, as all three neurotoxic compounds caused concentration-dependent inhibition of mean firing rate (MFR). Further, MFR appeared to be the most sensitive parameter for effects of neurotoxic compounds, as changes in electrical activity measured by mean frequency intra burst (MFIB), and mean burst duration (MBD) did not result in concentration-response relationships for some of the positive compounds, or required higher concentrations for an effect to be observed. However, greater numbers of compounds need to be tested to confirm this. The results obtained indicate that measurement of spontaneous electrical activity using MEAs provides a robust assessment of compound effects on neural network function.

Keywords: screening; microelectrode array; cross-laboratory comparison

2

Introduction Interest in developing medium- and high-throughput screening approaches for predictive toxicity testing has been increasing since the publication of the National Academy’s report on Toxicity Testing in the 21st Century (NRC 2007) and the implementation of the regulation concerning the Registration, Evaluation, Authorisation and Restriction of chemicals (REACH) (EU, 2006) in the European Union. These and other reports (Judson et al., 2009; Kavlock et al., 2009) highlighted the lack of toxicity information on thousands of chemicals and proposed that higher throughput, in silico and in vitro models based on human biology would be required in order to generate toxicity information on these chemicals in a timely manner. The present neurotoxicity regulatory guidelines (OECD Test Guidelines TG418, TG419, TG424 and TG426 and EPA Guidelines for Neurotoxicity Risk Assessment: FRL 6011-3) are costly and low throughput. These standard testing approaches for adult and developmental neurotoxicity evaluation are based on animal models and the neurotoxic potency of compounds is mainly determined by neurobehavioral and neuropathological effects in vivo. Using these in vivo approaches, only a small fraction of chemicals have been adequately evaluated for neurotoxicity. Further, effects observed in animals often provide little mechanistic information and are not always predictive of human toxicity, which were among the reasons underlying the proposal of a new toxicity testing paradigm (NRC 2007). Since the publication of the International Program on Chemical Safety document on "Principles and Methods for the Assessment of Neurotoxicity Associated with Exposure to Chemicals," (WHO, 1986) basic research in neurobiology has significantly improved the ability to assess how chemicals may adversely affect the nervous system. Cell cultures derived from nervous tissue have proven to be powerful tools for elucidating cellular and molecular mechanisms of nervous system development and function (Bal-Price et al., 2012), and the throughput needed for screening large numbers of chemicals can be achieved using in vitro approaches. Neuronal electrical activity is a fundamental function of the nervous system and, for this reason, its analysis could be used to evaluate the potential neurotoxic effects of test substances. Primary cultures of neurons and glia plated on grids of planar microelectrodes (i.e. Microelectrode Arrays (MEAs)) form networks of interconnected neurons in culture that exhibit spontaneous electrical activity. MEAs allow for the extracellular recording of this spontaneous electrical activity in the form of action potential “spikes” and groups of spikes (“bursts”; for review, see Pine 2006; Nam and Wheeler, 2011). The use of neural networks on MEAs has been proposed as a screening approach for identification of potential neuroactive or neurotoxic effects of test substances (Johnstone et al., 2010; Defranchi et al., 2011; McConnell et al., 2012; Schultz et al., 2015). Moreover, recent reports have proposed that the use of multiwell MEA plates could enhance the throughput of the assay (Valdivia et al., 2014; Nicolas et al., 2014). Compared to conventional, in vivo assays, electrophysiological evaluation could provide an early functional readout for in vitro neurotoxicity screening. A first step toward the potential application of the MEA data for neurotoxicity assessment is demonstration of the robustness of results across different laboratories. Previously, a multi-laboratory study demonstrated that assessments of the potency of three pharmacological agents were remarkably consistent across six different laboratories (Novellino et al., 2011). However, this previous study only examined neuroactive pharmaceutical agents (fluoxetine, verapamil and muscimol) and did not include any compounds that were not expected to disrupt neural network function (e.g. “negative controls”). The present study was therefore designed as a follow-on to the initial cross-laboratory study. Six test compounds were selected for evaluation by four different laboratories. Three of the compounds were well characterized neurotoxicants; chlorpyrifos oxon (CHO; an organophosphate insecticide), deltamethrin (DEL; a pyrethroid insecticide) and domoic acid (DA; a marine excitotoxicant). 3

By contrast, the other three compounds (glyphosate (GLY), dimethyl phthalate (DMP), acetaminophen (ACE)) are generally recognized to not be neurotoxic, and thus were expected to not cause significant effects on neural network function (“negative controls”). Each of the participating laboratories evaluated the concentration-response relationship of these six compounds and provided the data to a common laboratory for analysis and curve fitting. The focus of these initial studies was the ability of the participating laboratories to identify and separate the neurotoxic and non-neurotoxic compounds.

Materials and Methods Cell Culture All experimental procedures utilizing animals were approved by the institutional animal use board of the respective institutions and were conducted according to ethical guidelines. Experiments were conducted in four independent laboratories using four different cortical cell (Figure 1A) culture models, all of which are mixed cultures of neurons and glia (Björklund et al., 2010; Hogberg et al., 2011; see also supplemental material in Cotterill et al., 2016). Specifically: primary cultures from embryonic (E16-18) rat cortex (Labs 1 and 3), primary cortical cultures from newborn (0-24h) rats (Lab 2) and cryopreserved embryonic (E14-15) mouse cortex (Lab 4). Each laboratory used different cell densities and sources according to their established protocols (see Table 1 for details). Once obtained, cell suspensions were plated on MEA “chips”, medium was added and the chip was kept in a Petri dish (100mm) and placed in a humidified incubator (37oC and 5% CO2 or 10% CO2; depending on the specific laboratory) in order to allow neuronal network maturation, usually at 3-4 weeks (Novellino and Zaldívar 2010; Chiappalone et al., 2006). Each laboratory maintained cultures (e.g. changed medium) according to its own protocol. Important parameters associated with the cultures are provided in Table 1, and complete culture details are provided in the Supplemental Materials. Test Compounds The goal of the present study was to evaluate the ability of multiple laboratories to correctly identify neurotoxic compounds using neural networks grown on MEAs to screen chemicals for potential neurotoxicity. To this end, six chemicals were used; three compounds are well documented to be neurotoxic in vivo, and three that are generally recognized not to be neurotoxic. The active compounds were: 

Deltamethrin, CAS 52918-63-5 (Sigma D9315; ChemService PS2071; 99.5% purity). Deltamethrin is a pyrethroid insecticide known to cause neurotoxicity via disruption of voltage-gated sodium channel function (for review, see Soderlund et al., 2002). Deltamethrin reduces spontaneous activity in postnatal rat neural networks coupled to MEAs (Shafer et al., 2008; McConnell et al., 2012).

 Domoic Acid, CAS 14277-97-5 (Sigma D6152). Domoic acid (DA) is a neurotoxic biotoxin that mimics kainic acid and selectively activates AMPA/kainate receptors (Debonnel et al., 1989; Watanabe et al., 2011). Acute treatment with DA has been demonstrated to decrease spontaneous activity in cortical cultures coupled to MEAs (Hogberg et al., 2011; Nicolas et al., 2014; Wallace et al., 2015).  Chlorpyrifos oxon, CAS 5598-15-2 (ChemService MET-674B; 98.6% purity). CHO is a neurotoxic organophosphate insecticide and is a potent inhibitor of 4

acetylcholinesterase (Moser et al., 2005; Clegg and van Gemert 1999). Recent publications have demonstrated that CHO inhibits spontaneous network activity measured using MEAs (McConnell et al., 2012). The non-neurotoxic compounds were: 

Dimethyl Phthalate, CAS 131-11-3 (Aldrich W525081, >99% purity). DMP can be used as a general insect repellent (Metcalf, 2012). It has been reported to cause reproductive problems in male rats (Howdeshell et al. 2008). However, it is not generally recognized to be neurotoxic, nor a developmental toxicant (Brown 2002).

 Glyphosate, CAS 1071-83-6 (ChemService PS1050, 99.3% purity). GLY is an herbicide acting on metabolic pathways present in plants but not mammals, and for which there is no conclusive evidence of mammalian toxicity (for review, see Williams et al., 2012). GLY has been utilized as a negative control compound in several different in vitro neurotoxicity assays (Breier et al., 2008; Radio et al., 2008; Culbreth et al., 2012) and is without effect on spontaneous activity on neural networks on MEAs (McConnell et al., 2012). 

Acetaminophen, CAS 103-90-2 (Sigma A7085, 99.0% purity). ACE (also known as paracetamol) is a widely used antipyretic with analgesic action. Within normal therapeutic ranges it is not toxic, although acute overexposure (1-2mM) causes cerebral damage (Posadas et al., 2010). ACE has also been used as a negative control in the development of in vitro assays for neurotoxicity (Breier et al., 2008; Radio et al., 2008; Culbreth et al., 2012) and is without effect on spontaneous activity in neural networks on MEAs (McConnell et al., 2012).

During the study each participating laboratory adopted similar operating procedures and good practices for cell culture and data processing. Common acceptance criteria for neuronal network morphology and spontaneous electrophysiology were defined a priori and used to select more than 100 independent experiments (~700 hr of recordings) that were included in this analysis. Prior to being used in experiments, a simple “pre-screening” procedure was utilized by all labs to ensure that only healthy cultures were used for recordings. These acceptance criteria, although largely qualitative, were based on the observations of an experienced investigator. First, only well differentiated neuronal cultures (e.g. healthy appearing cells with extensive neurites) were used for MEA experiments. On a regular basis during the culture time and prior to MEA recording, each chip was observed under an optical microscope for morphological evaluation and to exclude the presence of bacterial/fungal contamination. Further, the network was considered acceptable for the electrophysiological study if the following general parameters were present: • Activity recorded from a minimum of 10 electrodes/well on a single well MEA (59 or 60 electrodes) or 4 electrodes/well for the 6-well chips. • Presence of synchronized burst patterns (e.g. observation of contemporaneous bursts across several or all of the active electrodes, see Supplemental Figure 1), a dynamical behavior very common in primary cultures of neural cells (Chiappalone et al., 2006; Wagenaar et al., 2006; Ham et al., 2008). • A minimum Mean Firing Rate (MFR) of 5 spikes/min for each electrode that was considered active.

5

More rigorous quantitative criteria were applied for burst analysis and curve fitting (see below). Analysis The NeuroTech Lab at the Istituto Italiano di Tecnologia (IIT) was in charge of the collation and analysis of all the experiments performed in this project. IC50 values for data from Lab 2 have been reported previously (Mack et al., 2014), as part of a larger study. It should also be noted that, in the present study, there are slight differences in the dataset evaluated and in how IC50 values were determined, compared to the previously published values (Mack et al., 2014). Chip handling and preparation Standard 59 or 60 electrode MEAs and 6-wells MEAs with 30m diameter electrodes, 150-200m inter-electrode spacing (Multichannel Systems GmbH, Reutlingen, Germany) were employed (Figure 1B). The MEA chips were provided with either an integrated ground electrode, or with an additional electrode which was connected by the user to the ground signal. Prior to plating cells, MEA chips were sterilized and coated with Laminin (Sigma L2020) and Poly-Lysine (Sigma P6407 (PDL) or P2636 (PLL)). The order and amount of substrate deposition varied among laboratories (See supplemental material for details).

Experimental set-up and recording system Table 2 reports the MEA recording systems used, data sampling frequency, applied filters and recording software.

Experimental protocol Experiments were generally conducted between DIV 15 and 57. Each well (an individual single-well or each well of a 6-well MEA chip) was considered as a separate experiment. The recordings (see Table 2 for details) were performed on a minimum of three MEA chips from at least two different culture preparations. In some laboratories (Labs 1 and 4) a 50% change of the medium was conducted immediately prior (10-15 min) to initiating the experiment. Once in the amplifier, all laboratories allowed 15-30 min for the activity to stabilize before recording. In all laboratories, recordings were conducted at 37 ºC, and reference activity was recorded for 30 - 60 min before the first administration of test chemical (control condition). Each laboratory made stock solutions of the compounds in appropriate solvents (Supplemental Table 1). The final levels of DMSO and other solvents used here have been investigated previously (see Mack et al., 2014; Defranchi et al., 2012; Wallace et al., 2015; Alloisio et al., 2015) and were below those that alter mean firing rate. Reagents were typically introduced by the following pipetting procedure to ensure proper mixing: 100 – 200 l of medium was removed from the medium bath covering the networks, mixed with a small volume (2-20l) of the reagent dilution and carefully returned to the medium bath in order to minimize any osmotic or hydrodynamic stress). Specific details for each laboratory are provided in the supplemental methods. Concentration-response relationships were determined in a cumulative manner, in which the concentration of compound present in the medium was increased in a stepwise manner in log or half-log units. The cumulative concentration-response approach was selected due to the low throughput of single- to six-well MEAs. Each laboratory 6

had to administer at least 5 different compound concentrations in each experiment. From the concentrations indicated below: 1pM, 100pM, 300pM, 1nM, 10nM, 100nM, 300nM, 1M, 10M, 100M The typical adopted sequence was 100pM, 1nM, 10nM, 100nM, 1M, 10M, 100M In order to conduct curve-fitting of each individual experiment, a minimum of 5 concentrations of compound must have been tested. Experiments that tested less than 5 concentrations were excluded from the analysis. In some cases, experiments were excluded due to corrupt or missing data files. Signal Processing Data organization, spike and burst detection Raw signals recorded from MEAs contain meaningful information (extracellularly recorded action potential “spikes”) and the background noise. Spike detection is required (Figure 1C) to separate the meaningful information from the background noise. Spikes are detected as alterations in electrical signals that exceed a set threshold. The method of spike detection by each laboratory is provided in Table 2 and was performed online, during the data recording. For the present experiments, only the timestamps of the spikes (not spike amplitude or shape) were considered in the analysis of spike rates and burst properties. For analysis of concentration-response, a 10 min period was selected at the end of the recording period in each concentration of compound, and compared to a 10 min period at the end of the baseline recording period. In order to obtain homogeneous results, firing rate and burst data from three of the four labs were analyzed by IIT and UNIGE personnel using the same software. For this reason, a standard data format was required from these three labs. Because Lab 2 had already completed an analysis of its data (see Mack et al., 2014), experimental values for each parameter and each experiment were used from its analysis for the determination of IC50 values by IIT/UNIGE. The parameters were equivalent (burst detection algorithm was basically the same) but the analysis packages used to determine those parameters did differ for the Lab 2 data (see Mack et al., 2014 for details). The data provided by the three laboratories were organized in comma-separated values (csv) files (one for each experiment). These files contained two columns: the temporal location of spikes as sample number and the corresponding channel number. In order to analyze these data using Matlab (The Mathworks, Natick, MA, USA), a graphical user interface (ConvertCSVFiles v2.0) was developed. This tool required, as input, two different files: the comma-separated values files of the experiment and a template table with the temporal intervals that described different experimental phases (administration of different drug concentrations). As result of the processing, data were subdivided into experimental phases and organized in spike trains. To perform burst detection, a previously published algorithm was used that identified a burst as an ensemble of at least 5 spikes spaced less than 100 ms apart (Chiappalone et al., 2005). After the application of this algorithm to the spike trains, burst trains were generated (Figure 1C). The parameters used to describe the behavior of neural network activity in the presence of compounds were computed from spike and burst trains and plotted as concentrationresponse curves. These parameters were the mean firing rate (MFR, mean number of spikes

7

per second), the mean burst duration (MBD) and the mean frequency intra burst (MFIB, mean frequency of spikes recruited in burst). NeuroPharma NeuroPharma is a graphical user interface custom-written in the Matlab environment by the IIT and UNIGE laboratories, and specifically implemented for neuropharmacological experiments (Figure 2B). This software will be made available to all interested people upon request to M Chiappalone. The strength of this software is the semiautomatic approach that allows a quick and efficient analysis of data acquired using MEA systems. Using NeuroPharma (whose main function and main window are shown in panels A and B of Figure 2, respectively), it is possible to perform conversion from the proprietary format of the recording system to Matlab format. Moreover it allows a series of processing on the raw data, such as threshold computation, spike detection and burst detection. Further analysis of the statistical parameters can then be conducted using two approaches:  Multiple Experiments Analysis performs the analysis of one or more experiments at the same time. Statistical parameters are computed for each phase of experiment. It is possible to plot the parameters’ profile during the experiment and export results in Excel (Microsoft Corporation, Redmond, WA, USA) tables.  Group Analysis allows pooling of several experiments for simultaneous analysis. The result consists of the computation of the average value for each selected parameter. The results are then exported in Excel tables.

Given that different labs provided the spike trains of the experiments, the NeuroPharma spike detection function was not used. The software was used to perform the burst detection and the computation of described statistical parameters (MFR, MBD, MFIB). IC50 computation The final step of analysis was the computation of half maximal inhibitory concentration (IC50) values for MFR, MBD and MFIB in experiments conducted using active compounds (DEL, DA, CHO). This analysis was applied to the data from all 4 labs, so that consistent approaches were used to determine IC50 values from all labs (this in part also accounts for differences in the values published in Mack et al., 2014 by lab 2). In order to obtain these results, the concentration-response curves were fitted using three different functions depending on the trend of the MFR value of each experiment. Fitting and IC50 computation were performed using Origin (OriginLab Corporation). The fitting functions used are described below: DoseResp Function: 𝑦 = 𝐴1 +

𝐴2 −𝐴1 1+10(𝐿𝑂𝐺𝑥0−𝑥) 𝑝

A1 = bottom asymptote, A2 = top asymptote, LOGx0 = center, p = Hill slope. Logistic Function: 𝑦 =

𝐴1 −𝐴2

𝑥 1+(𝑥 )𝑝

+ 𝐴2

0

A1 = initial value, A2 = final value, x0 = center, p = power. Some experiments (especially those performed using DA) showed a biphasic behavior. Given that Origin doesn't offer fitting functions for this type of behavior, a new function was implemented (Beckon et al., 2008) 8

Biphasic Function: 𝑦 = 𝑀 (

1 𝜀𝑈𝑝 𝛽𝑈𝑝 1+( ) 𝑥

)(

1 𝛽𝐷𝑛 𝜀 1+( 𝐷𝑛 ) 𝑥

) 𝑤𝑖𝑡ℎ (𝛽𝑈𝑝 > 0, 𝛽𝑈𝑝 < 0)

Where βUp is the steepness of the rising (positive) slope; εUp, the dose at the midpoint of the rising slope; βDn, the steepness of the falling (negative) slope; εDn, the dose at the midpoint of the falling slope; and M, the multiplicative coefficient. The original formula (Beckon et al., 2008) didn't include the multiplicative coefficient M and the fitting performed was confined between 0 and 1. To overcome this problem the coefficient M was added. Additional information about the concentration-response curves fitting and IC50 computation are reported in the Supplementary Materials and Supplemental Figure 2. In order to perform a comparison between different labs, IC50 values were computed for each individual experiment from each lab. The mean and standard deviation of IC50 values for each lab were then determined in order to compare the IC50 values across the laboratories. Results Mean baseline values for the mean firing rate (MFR), mean burst rate (MBR), mean burst duration (MBD) and mean firing rate in burst (MFIB) across all experiments were generally consistent across the different laboratories (Table 3). There were differences between the individual labs in some parameters (e.g. Lab 2 had the shortest burst duration, while Lab 3 had the highest burst rate), but it should also be noted that for all of these parameters, there was high variability within each lab, as indicated by the standard deviations of most parameters. MFIB and MBD were among the most consistent parameters within each lab, as indicated by the relatively lower standard deviations.

The effects of the six compounds (chlorpyrifos oxon, deltamethrin, domoic acid, acetaminophen, dimethyl phthalate and glyphosate) on spontaneous neural network activity were evaluated based on three parameters: Mean Firing Rate (MFR), Mean Burst Duration (MBD) and Mean Frequency Intra Burst (MFIB). Results for MFR are reported in Figure 3. In general, active compounds decreased MFR in a concentration-dependent manner (Figure 3A-C). For CHO, concentration-response curves of the four labs were well described by a sigmoidal trend (Figure 3A). Concentration-response curves for DEL were also sigmoidal, except for Lab1, where a biphasic curve provided a better fit of the data (Figure 3B). Concentration-response curves from the four labs for DA show a biphasic trend, except for Lab2 where a sigmoidal response was observed (Figure 3C). The negative control compounds (Figure 3D-F) in general did not have concentrationdependent effects on MFR. The exceptions were DMP results from Lab2 (Figure 3E) in which MFR values decrease at higher concentrations, and Gly results from Lab 1, wherein MFR values increased with increasing concentrations. In order to determine overall variability across the different labs, the data were binned into three concentration ranges; low, 1pM-100pM; medium, 300pM-100nM; and high, 300nM100μM. The cross laboratory coefficient of variation (CV) in each bin was determined by taking the standard deviation of the values from all laboratories and dividing by the mean of these values in each bin. These values are reported in Table 4. In general, in the low concentration range where no laboratory reported effects of chemicals, the CVs range from 715%. In the middle concentration range, CV values begin to increase, especially for the positive compounds CHO, DEL and DA (CVs range from 32-35%). This is driven by the differences in response potency and the shape of the concentration-response curves between the different 9

laboratories. In some cases, large CVs were driven primarily by the divergent results from a single laboratory (e.g. Lab1 for DEL and Lab2 for CHO). At the highest concentration range, the CVs are not reported for the positive chemicals, as the values for MFR trend towards zero for all labs. For the negative control compounds, CVs in the medium and high concentration ranges increased, largely being driven by the divergent responses for DMP and GLY from labs2 and 1, respectively. IC50 values for each individual experiment were determined and the mean IC50 value for each of the active chemicals was determined for each of the four laboratories (Tables 5, 6 and 7). For deltamethrin and domoic acid, the IC50 values were within tenfold across all 4 labs, with 3 of the 4 labs being within twofold. A similar pattern was observed for CHO, wherein 3 of the 4 labs obtained IC50 values within 4 fold, while one lab reported a value that was ~650 fold more potent. Mean IC50 values of the active compounds are plotted for all the four labs in supplemental figure 3.

Results for MBD are reported in Figure 4. In general active compounds decreased MBD at higher concentrations (Figure 4A-C), while negative control compounds (Figure 4D-F) were generally without effect on this parameter. Concentration-response curves for CHO effects on MBD were biphasic (Lab 4), sigmoidal (Lab 1) and decreasing (Lab 3) (Figure 4A). For Lab 2, MBD was not altered at concentrations below those that caused a cessation in firing, as indicated by the relatively flat concentration-response curve from this lab. Concentration-response curves for DEL from three of the four labs show a sigmoidal trend (Figure 4B), while DA concentration-response curves are biphasic (Lab 1) or sigmoidal (Labs 3 and 4) (Figure 5C). For Lab 2, due to the cessation of firing (and hence bursting) at higher concentrations, the trend is flat, but only includes lower concentrations where effects on MFR were smaller or not present. Results MFIB suggest that active compounds decreased MFIB at higher concentrations (Figure 5A-C). Negative control compounds (Figure 5D-F) do not cause this response: the MFIB values were not altered by exposure to any of the negative control compounds. Concentration-response curves for CHO from the four labs are either flat (Labs 1 and 2) or decreasing (Labs 3 and 4; Figure 5A). Concentration-response curves for DEL were sigmoidal (Lab 1 and 3), flat (Lab 2) or increasing (Lab 4; Figure 5B). Concentration-response curves for DA were biphasic (Lab 4), flat (Labs 1 and 2) or decreasing (Lab 3) (Figure 5C). Curves for ACE (Figure 5D), DMP (Figure 5E) and GLY (Figure 5F) indicate no concentration-related response. Discussion The present study demonstrates that assessment of neural networks using primary rodent cortical cells cultured on MEAs provides information about the potential acute neurotoxicity hazard of test compounds that is qualitatively similar across different laboratories. Data from the four laboratories that participated in the study all were able to identify DEL, CHO and DA as modulators of neural network electrophysiology, while ACE, DMP and GLY did not alter neural network physiology in most instances. Importantly, these results were obtained despite substantial differences in both the methods employed and model systems used, which is an indicator of the robustness of the system to identify compounds

10

causing electrophysiological changes. These data are consistent with a previous study in which neuroactive pharmaceutical compounds were examined (Novellino et al., 2011). I. Correct identification of neurotoxic and non-neurotoxic compounds based on MFR From a qualitative standpoint, the consistency of results across the different laboratories was good, with all 4 laboratories obtaining concentration-dependent changes in MFR for all 3 positive compounds. With respect to the negative compounds, a defined concentrationresponse relationship for effects on MFR was not observed by all 4 laboratories for ACE, and by 3 of 4 laboratories for DMP and GLY. One of four labs found that DMP (Lab 2) and GLY (Lab 1) inhibited or increased MFR, respectively. Thus, correct identification of the positive compounds across the laboratories was 100% (12 observed of 12 expected, and correct identification of the negative controls across the laboratories was 83% (10 observed of 12 expected). All of the neurotoxic compounds reduced MFR at the highest concentrations tested. While cell viability was not directly assessed in the current study, previous studies (McConnell et al., 2012; Wallace et al., 2015) indicate that acetaminophen, glyphosate, deltamethrin, chlorpyrifos oxon and domoic acid are not cytotoxic at concentrations up to 50 µM following a 1 hr exposure, suggesting that it is unlikely that cytotoxicity contributed significantly to the current results. From a quantitative perspective, there was more variability between the different laboratories. First, the types of concentration-response curves that best fit each data set did sometimes differ between the different laboratories. One of four labs observed a biphasic response for DEL, while two of four labs observed a biphasic response to DA, consistent with Frega and co-workers (2012). Interestingly, while Lab 2 observed a sigmoidal response to DA both here and in a previous publication (Wallace et al., 2015), it observed a biphasic response to NMDA and glutamate (Lantz et al., 2014). Thus, lack of an increased response to low levels of glutamatergic activation likely does not explain this result in Lab 2. With respect to IC50 values, the results were generally, within one order of magnitude for all instances except for the response to CHO from Lab 2. In this case, an IC50 value ~2.5 orders of magnitude more potent was determined. For DM, the IC50 values ranged from 0.2 – 1.5 µM, which is consistent with previously reported values in the cortex (3.0 µM, Alloisio et al., 2015; 0.13 µM, Shafer et al., 2008; 1.30 µM, Scelfo et al., 2012). For DA, others have reported IC50 values of 0.4 µM (Nicolas et al., 2014) and (estimated) between 0.1 and 0.5 µM (Hogberg et al., 2011), which is consistent with the values reported here. The consistency of the estimated IC50 values within this study and in comparison to previously published studies corroborates the previous crosslaboratory study of the ability of MEAs to assess pharmaceutical compounds (Novellino et al., 2011), and supports the robustness of this approach for screening chemicals for neurotoxicity. II. Effects of compounds on Burst Parameters In a previous study, the network MFR was selected as the electrophysiological parameter to evaluate compound effects because it was sensitive, simple to extract, and robust (Novellino et al., 2011). In fact, MFR is a commonly used metric for both drug and chemical effects on neural network function. Here, two additional parameters (MBD and MFIB) were evaluated in addition to the MFR. Only with MFR was there 100% correct identification of the positive compounds. Using MBD and MFIB, there were instances where there was no concentration-response relationship observed for some of the positive compounds. Thus, in the present study, MFR was the most sensitive parameter to neurotoxicant-induced changes. By contrast, for both MBD and MFIB, none of the labs observed any effect of the negative compounds. Thus, it is difficult to determine whether these parameters provide a better parameter for evaluation than MFR, as the various parameters differed in how well they correctly identified either positive versus negative compounds. Further clarification will 11

require testing much larger chemical sets containing both positive and negative control compounds. Testing of larger numbers of compounds will also provide additional data to evaluate whether some combination of different parameters provides the best overall correct identification of both positive and negative compounds, such as in the approaches taken in Mack et al., 2014 and Alloisio et al., 2015. III. Challenges and caveats of this multilab study. Although the present study was a coordinated effort by several different laboratories, it is not a “formal validation” study that utilized a strict experimental protocol. Thus there were a number of challenges associated with conducting the study and caveats that should be considered regarding the resultant data. One of the first challenges was standardizing approaches. Each participating laboratory had experience and had its own established protocol for neural network culturing and recording using MEAs. Rather than force each participating laboratory to utilize the same protocol, it was agreed upon that by having each lab use the preparations and chemical handling protocols that it was accustomed to, the ability of each laboratory to detect alterations would be maximized. It was agreed to have common basic acceptance criteria for utilizing a neural network, a general common experimental approach in terms of concentration-ranges tested, and a common analysis approach for burst analysis and curve fitting. Some attempts at standardization failed. For example, although the use of common chemical stocks (same vendor and lot#) was planned, difficulties in getting the chemicals shipped through international customs prevented this, and while most of the data was analyzed by a common protocol, difficulties with transferring files in a timely and highfidelity manner resulted in Lab 2 conducting a separate analysis. One other aspect that was standardized across the labs was that all laboratories conducted the concentration-response experiments in using a cumulative design. In this design, receptor (de)sensitization may influence the response to treatment with higher concentrations of compound. However, since all labs used this approach, this would only contribute to differences if there were differences between the labs in the degree of (de)sensitization. Because of the differences between methods and models, there are a number of caveats that should be considered when interpreting the data. Factors such as differences in culture age and type (e.g. different developmental stages or ratios of neurons/glia), culture conditions (including coating and the type of MEA chip used), and media composition (FBS was present in the media of Lab2 and albumin was present in the B27 used by lab 3) may all give rise to differences in the basic properties of the cultures and/or how they responded to treatment. Lab 2 was the only lab in the study that conducted recordings in the presence of serum (10% FBS). While the presence of FBS might be expected to bind compounds and decrease their bioavailability, no consistent right-shift of the concentration-response curves were observed in data from this lab, indicating that the presence of FBS did not have such an effect. Further, chemical handling protocols (e.g. differences in solvents) and recording protocols (e.g. differences in spike thresholds) all could contribute to both qualitative and quantitative differences in results observed between the laboratories. Because of the differences between methods and models used in this study, attempting to quantify different factors that contributed to the variability of the data across the laboratories would be confounded, and thus were not a focus of the present study. Finally, there are two additional caveats that should be noted with respect to these results. First, the study was not conducted in a blinded manner. Second, although these results support a high degree of sensitivity and selectivity, the number of compounds assessed was small. Thus, as the numbers of compounds assessed by this method increases, the possibility of divergent results between laboratories increases. 12

IV. Potential application/acceptance of the MEA measurements for neurotoxicants screening and prioritization This study adds to a growing number of studies that have demonstrated the usefulness of assessing neural networks on MEAs as a screening tool for acute neurotoxicity. Here, testing of several compounds across different laboratories yielded qualitatively similar results despite significant methodological differences. Previous studies with larger numbers of compounds have indicated that this approach has sensitivity (correct identification of positives) of >75%, and selectivity (correct rejection of negative compounds) nearing 100% (Defranchi et al., 2011; McConnell et al., 2012). While fewer studies have incorporated negative controls, glyphosate has been reported to be without effects on MFR (McConnell et al., 2012; Valdivia et al., 2014; Alloisio et al., 2015) and several studies have reported mepiquat to be without effects (Defranchi et al., 2011; Valdivia et al., 2014; Alloisio et al., 2015). The studies cited above include both within- and between-laboratory replications, further supporting that this approach is robust. Recent publications using higher-throughput 48 well MEA plates (Valdivia et al., 2014; Nicholas et al., 2014) also demonstrate inter-laboratory consistency (diphenhydramine, amiodarone and domoic acid) and indicate that the throughput capabilities of MEAs are sufficient for screening and prioritization. Neural networks on MEAs respond to both pharmaceutical as well as neurotoxic compounds. Thus, altered responses on MEAs are not in and of themselves indicative of a neurotoxic response. However, for the purposes of screening and prioritization of environmental chemicals, a general assumption is that they are not specifically intended to have an impact on neural activity. Thus, a positive response in an MEA-based screen could raise concern for potential neurotoxicity and prompt consideration for additional hazard characterization. Here, determination of IC50 values provided a means of identifying neuroactive vs non-neuroactive compounds. However, what level of change is necessary to classify a compound as neuroactive may not always be as clear and has to be established, especially if screening a large number of compounds at a single concentration (e.g. in McConnell et al., 2012; Valdivia et al., 2014). Once a determination of activity has been established, additional factors can be considered that may aid in prioritization. These include but are not limited to: 1) the potency of the compound relative to other known neurotoxicants, 2) the potency of the compound for effects on network function relative to the potency for cytotoxicity, 3) the potency of other known toxic effects; 4) estimated or known exposure levels, or 5) estimated or known target tissue concentrations. It should be noted that either increases or decreases in MFR beyond a predetermined threshold should be considered when screening chemicals with MEAs, as several known neurotoxicants have been reported to increase (lindane, RDX, fipronil etc) or decrease (carbaryl, deltamethrin, domoic acid) this parameter. Neural networks grown on MEAs also have been demonstrated to be amenable to integrated testing strategies. A recent study demonstrated that MEAs provided a complimentary addition to ion channel binding assays included in the ToxCastTM program (Valdivia et al., 2014). In addition, this approach was recently utilized, in conjunction with cell-based assays (for proliferation, differentiation and neurite outgrowth) and alternative in vivo models (zebrafish and C. elegans) to provide comparisons of toxicity between “replacement” organophosphate flame retardants and conventional, halogenated flame retardants for which there are significant safety concerns (Behl et al., 2015). Finally, it should be emphasized here that the initial step in establishing an assay using neural networks on MEAs will be to demonstrate its utility for hazard screening and prioritization of compounds for additional testing for hazard characterization. Several 13

challenges exist for using data from any in vitro assay for prediction of human risk. For example, the concentration of a potentially neurotoxic substance in the target tissue (e.g. CNS) after oral, dermal or inhalation exposure is often not known. Likewise the metabolites formed from the substance in the body are often not known. Hence, the test concentrations in in vitro systems (including MEAs) may exceed the relevant range or not test relevant metabolites. Another significant challenge is the lack of human in vivo reference data for the assessment of the predictivity of the system. These limitations are important to consider when using in vitro test results for the assessment of potential neurotoxic effects of a substance on humans, and it is likely that a combination of approaches will be needed to address these issues.

IV. Future directions. In conjunction with the previous results for pharmaceutical compounds (Novellino et al., 2011), the present results indicate that cross-laboratory compound effects on spontaneous network activity can be consistently identified. With respect to future directions, several goals must be simultaneously accomplished to move this approach towards a formal “validation”. Future studies of this nature should consider adopting a uniform protocol, with standardized culture models and protocols, experimental protocols, and analysis protocols. Furthermore, a much larger number of compounds (20-50), including both positive and negative controls, covering a wider range of chemical classes should be evaluated. This almost certainly will require the use of more recently available multiwell MEA (mwMEA) platforms that allow for 12-96 separate networks to be evaluated on a single plate. The capacity of mwMEAs to screen large numbers of chemicals has been demonstrated recently (Valdivia et al., 2014; Nicolas et al., 2014), thus such a study is feasible. Testing larger numbers of compounds, and recording from multiple wells will also provide a much larger dataset from which to conduct a more quantitative evaluation of laboratory reproducibility. For example, a more formal analysis could be conducted using Random Forest Analysis and Support Vector Machines (e.g. as in Mack et al., 2014) or other approaches (Alloisio et al., 2014) to determine which variables have the greatest ability to separate the active from inactive compounds, and if a model is built on these variables, how often it correctly identifies the compound as active or inactive. In addition, while rodent models have been utilized here and are still the standard model for study of network function using MEAs, incorporation of human neural network models derived from stem cells (either embryonic or induced, although the latter has fewer ethical concerns) would be a desired improvement and would meet the recommendation of the Toxicity testing in the 21st Century (NRC, 2007) report to utilize human cells whenever possible. However, publications utilizing human stem cell-derived neurons on MEAs are limited (Ylä-Outinen et al., 2010), and it is not clear that a suitable, standardized human model is currently available for distribution to multiple laboratories. Acknowledgements The authors appreciate the European Union’s Joint Research Center in Ispra, Italy for initiation and organization of this project, and thank Ms Susanna Aloisio, ETT, for her assistance with data management and analysis at that site. The authors also thank Dr William Boyes at the US EPA and Dr. Pam Lein at University of California, Davis, for comments on a previous version of this manuscript.

14

References Alloisio S, Nobile M, Novellino A. Multiparametric characterisation of neuronal network activity for in vitro agrochemical neurotoxicity assessment. Neurotoxicology. 2015. 48, 152-65. Bal-Price AK, Coecke S, Costa L, Crofton KM, Fritsche E, Goldberg A, Grandjean P, Lein PJ, Li A, Lucchini R, Mundy WR, Padilla S, Persico AM, Seiler AE, Kreysa J. Advancing the science of developmental neurotoxicity (DNT): testing for better safety evaluation. ALTEX. 2012. 29, 202-15. Beckon W N, Parkins C, Maximovich A, Beckon AV. A general approach to modeling biphasic relationships. Environmental Science & Technology 2008. 42, 1308-14. Behl M, Hsieh J-H, Shafer TJ, Mundy WR, Rice JR, Boyd WA, Freedman JH, Hunter ES, Jarema K, Padilla S and Tice R. Use of alternative assays to identify and prioritize organophosphorous flame retardants for potential developmental neurotoxicity. Neurotoxicology and Teratology 2015. 49, 79-85. Björklund, U.; Persson, M.; Rönnbäck, L.; et al. Primary cultures from cerebral cortex and hippocampus enriched in glutamatergic and GABAergic neurons. Neurochem. Res. 2010. 35, 1733–42. Breier J, Radio N, Mundy WR, and Shafer TJ. Development of a high-throughput assay for assessing anti-proliferative compounds using human neuroprogenitor cells. Tox. Sci. 2008. 105, 119-133. Brown NA. Selection of test chemicals for the ECVAM international validation study on in vitro embryotoxicity tests. European Centre for the Validation of Alternative Methods. Altern Lab Anim. 2002. 30, 177-98. Chiappalone M, Novellino A, Vajda I, Vato A, Martinoia S, and van Pelt J. Burst detection algorithms for the analysis of spatio-temporal patterns in cortical networks of neurons. Neurocomputing. 2005. 65-66, 653-662. Chiappalone M1, Bove M, Vato A, Tedesco M, Martinoia S. Dissociated cortical networks show spontaneously correlated activity patterns during in vitro development. Brain Res. 2006. 1093, 41-53. Clegg DJ, van Gemert M. Determination of the reference dose for chlorpyrifos: proceedings of an expert panel. J Toxicol Environ Health B Crit Rev. 1999. 2, 211-55. Cotterill E, Hall D, Wallace K, Mundy WR, Eglen S and Shafer TJ. Characterization of Early Cortical Neural Network Development in Multiwell Microelectrode Array Plates. Revision Submitted to J. Biomolec Screen 1/26/15. Culbreth ME, Harrill JA, Freudenrich TM, Mundy WR, and Shafer TJ. Comparison of Chemical-induced Changes in Proliferation and Apoptosis in Mouse and Human Neuroprogenitor Cells. Neurotoxicology 2012. 33, 1499-1510.

15

Defranchi E, Novellino A, Whelan M, Vogel S, Ramirez T, van Ravenzwaay B, Landsiedel R. Feasibility assessment of mirco-electrode chip assay as a method of detecting neurotoxicity in vitro. Frontiers in Neuroengineering. 2011. 4, 1-12. Debonnel G, Beauchesne L, de Montigny C. Domoic acid, the alleged "mussel toxin," might produce its neurotoxic effect through kainate receptor activation: an electrophysiological study in the dorsal hippocampus. Can J Physiol Pharmacol 1989. 67, 29-33. European Union (2006). Regulation (EC) No 1907/2006 of the European Parliament and the Council of 18 December 2006 on the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH), Frega M, Pasquale V, Tedesco M, Marcoli M, Contestabile A, Nanni M, et al. Cortical cultures coupled to Micro-Electrode Arrays: A novel approach to perform in vitro excitotoxicity testing. Neurotoxicology and Teratology. 2012. 34, 116–27. Ham MI, Bettencourt LM, McDaniel FD, Gross GW. Spontaneous coordinated activity in cultured networks: analysis of multiple ignition sites, primary circuits, and burst phase delay distributions. J Comput Neurosci. 2008. 24, 346-57. Hogberg HT, Sobanski T, Novellino A, Whelan M, Weiss DG, Bal-Price AK. Application of micro-electrode arrays (MEAs) as an emerging technology for developmental neurotoxicity: evaluation of domoic acid-induced effects in primary cultures of rat cortical neurons. Neurotoxicology. 2011. 32, 158-68. Howdeshell KL, Wilson VS, Furr J, Lambright CR, Rider CV, Blystone CR, Hotchkiss AK, Gray LE Jr. A mixture of five phthalate esters inhibits fetal testicular testosterone production in the sprague-dawley rat in a cumulative, dose-additive manner. Toxicol Sci. 2008. 105, 153-65. Johnstone AF, Gross GW, Weiss DG, Schroeder OH, Gramowski A, Shafer TJ. Microelectrode arrays: a physiologically based neurotoxicity testing platform for the 21st century. Neurotoxicology. 2010. 31, 331-50. Judson RS, Richard A, Dix, DJ. et al., The toxicity data landscape for environmental chemicals. Environ. Health Perpect. 2008. 117, 685-695. Kavlock RJ, Austin CP, and Tice RR. Toxicity testing in the 21st century: Implications for human health risk assessment. Risk Anal. 2009. 29, 485-487. Lantz SR, Mack CM, Wallace K, Keys EF, Shafer TJ, and Casida JE. Glufosinate excitotoxicity mediated in part by N-Methyl-D-aspartate receptor activation. Neurotoxicology 2014. 45, 38-47. Mack CM, Lin B, Turner J, Johnstone AFM, Burgoon L and Shafer TJ. Burst and principal components analysis of MEA data separates chemicals by class. Neurotoxicology. 2014. 40, 75-85

16

McConnell ER, McClain MA, Ross J, Lefew WR, Shafer TJ. Evaluation of multi-well microelectrode arrays for neurotoxicity screening using a chemical training set. Neurotoxicology. 2012. 33, 1048–57. Metcalf RL. “Insect Control” in Ullmann’s Encyclopedia of Industrial Chemistry” WileyVCH, Weinheim, 2002 Moser VC, Phillips PM, McDaniel KL, Marshall RS, Hunter DL, Padilla S. Neurobehavioral effects of chronic dietary and repeated high-level spike exposure to chlorpyrifos in rats. Toxicol Sci. 2005. 86, 375-86. Nam Y, and Wheeler BC. In vitro microelectrode array technology and neural recordings. Crit Rev Biomed Eng 2011. 39, 45-62. Nicolas J, Hendriksen PJM, van Kleef RGDM, de Groot A, Bovee TFH, Rietjens IMCM, Westerink RHS. Detection of marine neurotoxins in food safety testing using a multielectrode array. Mol Nutr Food Res. 2014. 58, 2369-78. Novellino A, Zaldívar JM. Recurrence quantification analysis of spontaneous electrophysiological activity during development: Characterization of in vitro neuronal networks cultured on multi electrode array chips Advances in Artificial Intelligence. 2010. Article ID 209254. http://dx.doi.org/10.1155/2010/209254 Novellino A, Scelfo B, Palosaari T, Price A, Sobanski T, Shafer T, Johnstone A, Gross G, Gramowski A, Schroeder O, Jugelt K, Chiappalone M, Benfenati F, Martinoia S, Tedesco M, Defranchi E, D’Angelo P, Whelan M. Development of micro-electrode array based tests for neurotoxicity: assessment of interlaboratory reproducibility with neuroactive chemicals. Frontiers Neuroeng. 2011. 4, 4. NRC. Toxicity Testing in the Twenty-First Century: A Vision and a Strategy. Washington, D.C: The National Academies Press, 2007. Pine J (2006). A history of MEA development. In: Advances in network electrophysiology (M. Taketani and M Baudry, eds). pp 3-23. Springer. Posadas I, Santos P, Blanco A, Muñoz-Fernández M, Ceña V. Acetaminophen induces apoptosis in rat cortical neurons. PLoS One. 2010 5, e15360. Radio N, Breier J, Shafer TJ, and Mundy WR. Development of a high-throughput assay for assessing neurite outgrowth. Tox. Sci. 2008. 105, 106-118. Scelfo B1, Politi M, Reniero F, Palosaari T, Whelan M, Zaldívar JM. Application of multielectrode array (MEA) chips for the evaluation of mixtures neurotoxicity. Toxicology. 2012. 299, 172-83. Soderlund DM, Clark JM, Sheets LP, Mullin LS, Piccirillo VJ, Sargent D, Stevens JT, Weiner ML. Mechanisms of pyrethroid neurotoxicity: implications for cumulative risk assessment. Toxicology. 2002. 171, 3-59.

17

Shafer TJ, Rijal SO, and Gross GW. Complete inhibition of spontaneous activity in neuronal networks in vitro by deltamethrin and permethrin. Neurotoxicology. 2008. 29, 203-12. Schultz L., Zurich M, Culot M, da Costa A, Landry C, Bellwon P, Kristl T, Hörmann K, Ruzek S, Aiche S, Reinert K, Bielow C, Gosselet F, Cecchelli R, Huber CG, Schroeder OH, Gramowski-Voss A, Weiss DG, Anna Bal-Price A. Evaluation of drug-induced neurotoxicity based on metabolomics, proteomics and electrical activity measurements in complementary CNS in vitro models. Toxicology in Vitro 2015. 30, 138-65. Valdivia P, Martin MT, Houck K, Lefew WR, Ross J and Shafer TJ. Multi-well microelectrode array recordings detect neuroactivity of ToxCast compounds. Neurotoxicology 2014. 44, 204–17. Wagenaar DA, Pine J, Potter SM. An extremely rich repertoire of bursting patterns during the development of cortical cultures. BMC Neurosci. 2006. 7, 7:11. Watanabe KH, Andersen ME, Basu N, Carvan MJ 3rd, Crofton KM, King KA, Suñol C, Tiffany-Castiglioni E, Schultz IR. Defining and modeling known adverse outcome pathways: Domoic acid and neuronal signaling as a case study. Environ Toxicol Chem. 2011 30, 9-21. Wallace K, Strickland JD, Valdivia P, Mundy WR and Shafer TJ. A multiplexed method for the determination of compound effects on network function and viability in MEAs. Neurotoxicology, 2015. 49, 79-85 Williams AL, Watson RE, DeSesso JM. Developmental and reproductive outcomes in humans and animals after glyphosate exposure: a critical analysis. J Toxicol Environ Health B Crit Rev. 2012. 15, 39-96. World Health Organization. Principles and Methods for the Assessment of Neurotoxicity Associated with Exposure to Chemicals. (Environmental Health Criteria No 60.). 1986. 180 pages. Ylä-Outinen L, Heikkilä J, Skottman H, Suuronen R, Äänismaa R and Narkilahti S. Human cell-based micro electrode array platform for studying neurotoxicity. Frontiers in Neuroengineering, September 2010. Volume 3. Article 111.

18

Figure 1. Primary rat cortical cell cultures plated on MEAs exhibit spontaneous electrical activity. A 21 day in vitro (DIV) culture of cortical primary neurons on top of a planar 60-electrode array (from Lab 3). B Layouts of the two MEAs used in this study: a standard array (top panel) and a 6-well array (bottom panel). A standard MEA device has 60 electrodes arranged over an 8 by 8 square grid, with the four corners missing. One of the electrodes can be replaced by one ground reference, allowing recording from the remaining 59 electrodes. A 6-well MEA device is constituted by six independent culture chambers, divided by a makrolon separator. Inside each well, nine electrodes and one internal reference electrode allow recording of electrophysiological activity from a dissociated neural culture. C Sample trace recorded from a single microelectrode. The top panel illustrates a typical raw cortical signal characterized by the presence of spiking and bursting activity. The red dotted line constitutes a typical threshold for detecting spikes (calculated as -5σ, where σ represents the standard deviation of the basal noise). The middle panel shows the result of the spike detection procedure obtained with the red threshold depicted in the upper panel: the “Spike Train” provides a record of the temporal pattern of spikes without reference to the amplitude of those events. For this reason the peak amplitude is equal to 1. The lower panel depicts the result of the burst detection. As with the spike trains, the burst train is a temporal measure and does not consider amplitude.

19

Figure 2. Block diagram of the signal processing procedure and Graphical User Interface of NeuroPharma. A NeuroPharma main functions block diagram. The software computes a threshold for spike detection, and a subsequent burst detection. Spike and burst trains are binned and statistical parameters are computed for each bin. Computed parameters can be plotted as function of time (instantaneous rate plot) or as function of drug concentration (dose-response curve). Functions outside the red box were not used for our analysis. B NeuroPharma main window. Data analysis and parameters selection and plotting are available from different panels. The plotted graph represents the dose response curves for four experiments.

20

Figure 3. Graphical representation of the Mean Firing Rate (MFR) for positive and negative compounds. A Chlorpyrifos Oxon. B Deltamethrin. C Domoic Acid. D Acetaminophen. E Dimethyl Phthalate. F Glyphosate. Data from different labs (Lab1 in black, Lab2 in red, Lab3 in green and Lab4 in blue) are plotted in a semi-logarithmic manner and are represented as means ± SEMs of 3-7 individual experiments. Values are normalized and expressed as percentage respect to the first experimental phase (control phase) that corresponds to 100%. Positive compounds are reported in the left column (A, B, C); negative compounds are reported in the right column (D, E, F).

21

Figure 4. Graphical representation of the Mean Burst Duration (MBD) for positive and negative compounds. A Chlorpyrifos Oxon. B Deltamethrin. C Domoic Acid. D Acetaminophen. E Dimethyl Phthalate. F Glyphosate. Data from different labs (Lab1 in black, Lab2 in red, Lab3 in green and Lab4 in blue) are plotted in a semi-logarithmic manner and are represented as means ± SEMs of 3-7 individual experiments.. Values are normalized and expressed as percentage respect to the first experimental phase (control phase) that corresponds to 100%. Positive compounds are reported in the left column (A, B, C); negative compounds are reported in the right column (D, E, F).

22

Figure 5. Graphical representation of the Mean Frequency intra Burst (MFIB) for positive and negative compounds. A Chlorpyrifos Oxon. B Deltamethrin. C Domoic Acid. D Acetaminophen. E Dimethyl Phthalate. F Glyphosate. Data from different labs (Lab1 in black, Lab2 in red, Lab3 in green and Lab4 in blue) are plotted in a semi-logarithmic manner and are represented as means ± SEMs of 37 individual experiments.. Values are normalized and expressed as percentage respect to the first experimental phase (control phase) that corresponds to 100%. Positive compounds are reported in the left column (A, B, C); negative compounds are reported in the right column (D, E, F).

23

Table 1. Methods for neuronal culture among different laboratories: BASF (Lab 1)

EPA (Lab 2)

ETT(Lab 3)

JRC (Lab 4)

rat cortex E18

new born (0-24hr) rat frontal cortex

rat cortex E18

(cryopreserved) mouse cortex E1415

Dissociation

Trypsin+ DNase

mechanical (+filter)

trypsin

no

Chip Coating

Laminin + PDL

PLL + Laminin

Laminin + PDL

Laminin + PDL

Cell density

30K

250K

50K

50K

Culture medium

Lonza Medium (Bullet kit CC4461)

Neurobasal A + 10% FBS NBA / FBS+Glutamate

Neurobasal +2% B27 and 1% Glutamax 1

Lonza, (CC-4461, growth medium + LGlutamine + +NSF1)

Fraction and frequency of the medium change (MC)

50% MC, 1st after 2/3 days, then 2 times a weeks

100%MC, 1st after 24 hrs to FBS + Glutamine (1%). 2nd 3 days later to FBS, then once a week thereon with FBS

50% MC, 1st after 2/3 days, then 2 times a weeks

50%MC, once a week for the first 3 weeks, then 50%MC 2 times a week there on.

Antibiotics in cell medium

Gentamycin and Amphotericin B (0.2%)

Penicillin (100 units/mL), and Streptomycin (0.1 mg/mL)

NONE

Gentamycin and Amphotericin B (0.2%)

Cell culture Reference

See supplemental material

Mack et al., 2014*

Alloisio 2015

Model culture

–

cell

et

al.,

Novellino 2011

et

al.,

*Lab 2 has changed its culture methods since these data were collected. See Wallace et al., 2015 for the most recent methods used in Lab 2.

24

Table 2. Recording conditions across participating laboratories

Recording DIV Type MEA

of

Recording atmosphere (CO2, pH, ...)

BASF (Lab 1)

EPA (Lab 2)

ETT (Lab 3)

JRC (Lab 4)

24-57

15-28

26-42

19-44

Single well; 59 electrodes; internal ground

Single well; 59 electrodes; internal ground

6-well, electrodes/well

cap and water vapor enriched with

FEP membrane + 10 µl water per

cap and water vapor enriched with

cap and water vapor enriched with

administration

5%CO2+20%O2+N2

5%CO2+20%O2+N2

NBA Media (no B27) + 10% FBS

Neurobasal +2% B27 and 1% Glutamax 1

Lonza, (CC-4461, growth medium + LGlutamine + Gentamycin + AmphotericinB +NSF1)

7%CO2+19%O2+N2

Media kit CC-

9

Single well; 60 electrodes; external ground

Recording Media

Lonza (Bullet 4461)

Presence of serum while recording

NO

YES

NO

NO

Recording System

Multi Channel Systems 1060BC preamplifier

Multi Channel Systems 1060BC preamplifier

Multi Channel Systems 1060BC preamplifier

Multi Channel Systems 1060BC preamplifier

Sampling Frequency

10kHz

25kHz

10kHz

10kHz

Acquisition Software

MC_Rack v4.0

MC_Rack v4.0

MC_Rack v4.0

MC_Rack v4.0

Spike Detection Threshold

±14.7 µV Threshold (=~5.5x rms noise)

±15 µV Threshold (~5x rms noise)

5.5x rms noise (=~25 µV)

5.5x rms noise (~12.5 µV)

Other information

Amplifier Gain 1000x. Band pass digital filter: 604000Hz

Amplifier Gain 1200x. High pass digital filter; cutoff 200 Hz

Amplifier Gain 1000x. Band pass digital filter: 60-4000Hz

Amplifier Gain 1000x. Band pass digital filter: 60-4000Hz

25

Table 3: Baseline values of spike and burst parameters* Parameter

Lab 1 (BASF)

Lab 2 (EPA)

Lab 3 (ETT)

Lab 4 (JRC)

MFR

1.45 ± 1.03

3.54 ± 3.12

3.37 ± 1.73

2.13 ± 1.59

MBR

4.52 ± 2.78

4.90 ± 4.57

11.59 ± 6.27

6.88 ± 4.41

MFIB

52.7 ± 9.0

194.7 ± 23.3

101.3 ± 25.3

129.3 ± 47.7

MBD

0.12 ± 0.06

0.04 ± 0.01

0.14 ± 0.05

0.07 ± 0.05

% Active Electrodes

87 ± 18

62 ± 22

97 ± 7

82 ± 19

# of experiments

25

25

22

22

*Mean ± s.d. MFR = Mean firing rate (spikes/sec) MBR = Mean Burst rate (bursts/min) MFIB = Mean firing rate in burst (spikes/sec) MBD = Mean burst duration (sec)

26

Table 4 Cross-laboratory CV values for compound effects on MFR* Low (1-100 pM) Medium (0.3-100 nM) High (0.3-100 µM) Chlorpyrifos oxon 8.2 35.1 ND Deltamethrin 17.3 31.2 ND Domoic Acid 10.5 32.5 ND Acetaminophen 14 13.3 18.5 Glyphosate 12.5 18.8 20.4 Dimethyl phthalate 10.3 12.4 43.5 *

Coefficients of Variation were determined by dividing the standard deviation of all values in a range by the mean. ND=Not determined

27

Table 5. IC50 values for chlorpyrifos oxon Lab

1

2

3

4

Exp number

IC50 (µM)

Exp2

2.99

Exp3

3.46

Exp4

2.36

Exp1

0.013

Exp2

0.016

Exp3

0.036

Exp4

0.012

Exp5

0.003

Exp1

0.68

Exp3

0.05

Exp4

0.58

Exp5

0.31

Exp1

11.6

Exp2

14.0

Exp3

6.3

IC50 Mean ± SE (µM) 2.9 ± 0.3

0.016 ± 0.005

0.41 ± 0.14

10.6 ± 2.3

28

Table 6. IC50 values for deltamethrin Lab

1

2

Deltamethrin

3

4

Exp number

IC50 (µM)

Exp1

2.39

Exp2

1.27

Exp3

0.97

Exp1

0.15

Exp2

0.06

Exp3

0.11

Exp6

0.10

Exp7

0.17

Exp8

0.70

Exp9

0.47

Exp1

0.27

Exp2

0.11

Exp3

0.89

Exp4

0.50

Exp1

0.87

Exp2

0.38

Exp3

0.16

Exp4

0.10

IC50 (µM)

Mean±SE

1.5 ± 0.4

0.3 ± 0.09

0.4 ± 0.2

0.4 ± 0.2

29

Table 7. IC50 values for domoic acid Lab

1

2

Domoic Acid

3

4

Exp number

IC50 (µM)

Exp1

0.41

Exp2

0.30

Exp3

0.35

Exp4

0.31

Exp5

0.36

Exp1

0.18

Exp2

0.11

Exp3

0.13

Exp1

1.4

Exp2

3.4

Exp3

1.1

Exp4

0.4

Exp1

0.19

Exp2

0.26

Exp3

0.14

Exp4

0.23

IC50 (µM)

Mean±SE

0.3 ± 0.02

0.1 ± 0.05

1.6 ± 0.6

0.2 ± 0.03

30