Exposure

Wild-type adult zebrafish, originally received from OBI petshop (Leipzig, Germany) were kept in 120 L fish tanks containing carbon-filtered tap water at 26 °C under a 10h:14h dark-light cycle. Eggs were collected approximately 1 hour after light onset and inspected using a light microscope. Fertilized eggs were incubated at 26 °C, while unfertilized eggs were discarded. For lethality experiments, which were conducted to find adequate exposure concentrations for transcriptome experiments, three embryos with an age of 24 hours post fertilization (hpf) were distributed into a 7.5 mL GC vials, each containing 6 mL exposure or control medium. Regarding treatment conditions three technical replicates and regarding control conditions six technical replicates were analysed. Vials were sealed and incubated in a climate chamber shaking (climate chamber: Vötsch 1514, Vötsch Industrietechnik GmBH (Balingen-Frommern); shaker: Edmund Bühler SM-30 Control, 26 °C, 75 rpm, 12h:12h light:dark) till 96 hpf. At 48 hpf, 72 hpf and 96 hpf sublethal and lethal effects were recorded. For transcriptome experiments 20 embryos per replicate were transferred to two 20 mL GC vials at 24 hpf. A volume of 18 mL of exposure medium was added to each vial and vials were sealed and incubated in a climate chamber (same settings as above) until sampling. Embryos were exposed to a range of 5 different concentrations of the toxicant and RNA was extracted at 3, 6, 12, 24, 48, 72 hours after start of exposure (hpe). Three substances were used for exposure experiments: diuron (CAS:330-54-1 ; purity: 99.6%; charge: #SZBB265XV ; supplier: Fluka ), diclofenac sodium salt (CAS:15307-79-6 ; purity: n.A.; charge: #BCBP9916V ; supplier: Sigma ), and naproxen sodium salt (CAS: 26159-34-2; purity: 98-102%; charge: #MKBV4690V ; supplier: Sigma). Exposure medium for all experiments was prepared by dissolving the substance in freshly prepared ISO-water (ISO 7346-3: 79.99 mM CaCl2·2H2O, 20.00 mM MgSO4·7H2O, 30.83 mM NaHCO3, 3.09 mM KCl; pH 7.4, oxygenized) (diclofenac, naproxen) or in 0.1 % methanol (CAS:67-56-1 ; purity: HPLC Grade, charge:, supplier: J.T. Baker) as solubilising agent (diuron). For the diuron exposure we used 0.1 % methanol as solvent control.

RNA extraction and hybridization

At observation time points, two vials of embryos (20 in total) were transferred into Eppendorf tubes. RNA was extracted by addition of Trizol and homogenization using a T10 basic Ultra-Turrax (IKA, Werke GmbH & Co. KG, Germany) for 20 sec at maximum speed. RNA isolation was performed using a pipetting robot (Microlab Star, Hamilton Life Science Robotics, Germany) following the manual provided for Total RNA Extraction Kit MagMAX 96 for microarrays and conducted in a 96-well plate. Quality of isolated RNA was assessed using a Bioanalyzer (Agilent 2100 Technologies, Waldborn) and the Agilent RNA 6000 Nano Kit. RNA samples were used for further processing if RIN values derived from ribosomal RNA absorption adopted values > 7 and calculated concentrations exceeded 25 ng/µL. All RNA samples were diluted to a concentration level of 25 ng/µL(10 µL in total) by addition of RNAse free water. Thereof, 2.3 µL (57 ng RNA) were utilised as starting amount of RNA for spike mix preparation.

Transcript abundance was measured with microarray analysis using Oaklabs ArrayXS Zebrafish microarray slides (XS-200104, oak-labs.com, Germany). Microarray experiments were performed using the Agilent Low Input Quick Amp WT Labeling Kit according to the Agilent One-Color Microarray-Based Exon Analysis Protocol (V2.0, August 2015). This protocol included introduction of spike-in RNA, RNA transcription and amplification into complementary DNA, and cDNA transcription and amplification into cRNA with simultaneous incorporation of Cy3 (fluorescently labelled cytidine nucleotide). The cRNA was fragmented and hybridized to Oaklabs ArrayXS Zebrafish microarray slides using the Agilent Hybridization kit and protocol as well as Agilent hybridization oven and chambers. Subsequently, microarray slides were washed and scanned with the Agilent High-Resolution Microarray Scanner according to the Agilent protocol. Intensity values were extracted from captured images using Agilent Feature Extraction Software (Version 11.5.1.1). RNA samples in the Diclofenac experiment were partly processed by Oaklabs GmbH (Henningsdorf, Germany). Potential batch effect, caused by handling the samples in two different laboratories was accounted for during data analysis.

Toxicokinetics

For the determination of internal chemical concentrations 3 replicates of 9 embryos per replicate were exposed and harvested after the different respective exposure durations. Embryos that were not hatched at the time point of harvest were not dechorionated for this experiment. Embryos were transferred in a 2 mL safe lock Eppendorf tube, washed with medium and dried from remaining water using a microliter pipette. The embryos were shock-frozen with liquid nitrogen and stored at -80°C until extraction. In dependence on the substance either acetonitrile, methanol-water-mix or pure water (see table below) was added to each sample respectively after homogenizing with a glass rod. The homogenate underwent 5 minutes sonication in an ultrasonic bath (Sonorex RK 512 H ; Bandelin Electronic) and 4 hours incubation at 20°C and 1000 rpm (Thriller; Peqlab) and were finally centrifuged 10 minutes at 10000 rpm and 4°C (Model 2K15; Sigma). The supernatants were separated from the matrix and used for the chemical analysis of the internal chemical concentrations. The chemical analyses were performed either on a HPLC system (Merck Hitachi – La Chrom) with diode array (DAD; model L7450) or ultraviolet-detector (UV; model L-7400). A calibration curve was acquired in the respective medium. No additional spike-in or internal standard was used for calibration.

library("dplyr")

for (substance in c("diuron", "diclofenac", "naproxen")) {
    exposurec <- read.csv(file = paste0("./data/", substance, "/tkData/exposure_concentrations_", 
        substance, ".csv"), header = T, sep = "\t", quote = "", as.is = T)
    
    exposurec_mean <- aggregate.data.frame(exposurec$concentration_umol_l, by = list(time_hpe = exposurec$time_hpe, 
        nominal_concentration_value = exposurec$nominal_concentration_value, 
        experiment_date = exposurec$experiment_date, comment = exposurec$comment), 
        FUN = mean, na.rm = T)
    
    
    
    colnames(exposurec_mean)[5] <- "concentration_umol_l"
    
    exposurec_se <- dplyr::summarise(group_by(exposurec_mean[!is.na(exposurec_mean$concentration_umol_l), 
        ], time_hpe, nominal_concentration_value), mean = mean(concentration_umol_l), 
        sd = sd(concentration_umol_l), N = length(concentration_umol_l), se = sd(concentration_umol_l)/sqrt(length(concentration_umol_l)))
    
    
    assign(x = paste0("exposurec_", substance, "_se"), value = exposurec_se)
    
    assign(x = paste0("exposurec_", substance), value = exposurec_mean)
    
    
}

library("cowplot")

exposurec_plot_diuron <- ggplot(data = exposurec_diuron[exposurec_diuron$nominal_concentration_value == 
    20, ]) + geom_smooth(aes(x = time_hpe, y = concentration_umol_l), se = F) + 
    geom_point(data = exposurec_diuron_se[exposurec_diuron_se$nominal_concentration_value == 
        20, ], aes(x = time_hpe, y = mean)) + geom_errorbar(data = exposurec_diuron_se[exposurec_diuron_se$nominal_concentration_value == 
    20, ], aes(x = time_hpe, ymin = mean - se, ymax = mean + se)) + ylim(c(0, 
    20)) + theme(plot.margin = unit(c(20.5, 5.5, 5.5, 5.5), "pt")) + ylab("concentration [µmol/L]") + 
    xlab("time [hpe]")  #+scale_x_log10(breaks = c(3,6,12,24,48,96), limits = c(1.5,96))

exposurec_plot_diclofenac <- ggplot(data = exposurec_diclofenac[exposurec_diclofenac$nominal_concentration_value == 
    6.6, ]) + geom_smooth(aes(x = time_hpe, y = concentration_umol_l), se = F) + 
    geom_point(data = exposurec_diclofenac_se[exposurec_diclofenac_se$nominal_concentration_value == 
        6.6, ], aes(x = time_hpe, y = mean)) + geom_errorbar(data = exposurec_diclofenac_se[exposurec_diclofenac_se$nominal_concentration_value == 
    6.6, ], aes(x = time_hpe, ymin = mean - se, ymax = mean + se)) + ylim(c(0, 
    8)) + theme(plot.margin = unit(c(20.5, 5.5, 5.5, 5.5), "pt")) + ylab("concentration [µmol/L]") + 
    xlab("time [hpe]")

exposurec_plot_naproxen <- ggplot(data = exposurec_naproxen[exposurec_naproxen$nominal_concentration_value == 
    135, ]) + geom_smooth(aes(x = time_hpe, y = concentration_umol_l), se = F) + 
    geom_point(data = exposurec_naproxen_se[exposurec_naproxen_se$nominal_concentration_value == 
        135, ], aes(x = time_hpe, y = mean)) + geom_errorbar(data = exposurec_naproxen_se[exposurec_naproxen_se$nominal_concentration_value == 
    135, ], aes(x = time_hpe, ymin = mean - se, ymax = mean + se)) + ylim(c(0, 
    160)) + # xlim(c(0,74)) +
theme(plot.margin = unit(c(20.5, 5.5, 5.5, 5.5), "pt")) + ylab("concentration [µmol/L]") + 
    xlab("time [hpe]")

plot_grid(exposurec_plot_diuron, exposurec_plot_diclofenac, exposurec_plot_naproxen, 
    labels = c("Diuron", "Diclofenac", "Naproxen"), ncol = 3)

Figure 4: Concentrations measured in exposure medium

Normalisation

Next, we normalise the data between the arrays. Here we normalize all three datasets together, using cyclic-loess normalisation. However normalisation of single datasets is also possible.

dslist <- list(data_diuron, data_diclofenac, data_naproxen)
dslist_norm <- toxprofileR::normalizeBatch(dslist = dslist)

Figure 8: Boxplots before and after normalization

Further data preprocessing

In the next step some further preprocessing is applied to the data. Measurements of duplicate probes are averaged (at the same time removing probes which are flagged of poor quality), a recent annotation of the array is added and batch correction is applied, if necessary. In our example, this was necessary for the diclofenac experiment, since the arrays were processed in two different labs.

data_proc_diuron <- toxprofileR::preprocess(dslist_norm[[1]], batchcorrect = F)

batch_diclofenac <- as.factor(as.numeric(dslist_norm[[2]]$targets$comment == 
    "Oaklabscan"))
data_proc_diclofenac <- toxprofileR::preprocess(dslist_norm[[2]], batchcorrect = T, 
    batch = batch_diclofenac)

## Found2batches

## Adjusting for35covariate(s) or covariate level(s)

## Standardizing Data across genes

## Fitting L/S model and finding priors

## Finding parametric adjustments

## Adjusting the Data

data_proc_naproxen <- toxprofileR::preprocess(dslist_norm[[3]], batchcorrect = F)

save(data_proc_diuron, file = "./data/diuron/data_proc_diuron.Rd")
save(data_proc_diclofenac, file = "./data/diclofenac/data_proc_diclofenac.Rd")
save(data_proc_naproxen, file = "./data/naproxen/data_proc_naproxen.Rd")

Calculate logFC

In our experiment gene expression was measured at several time points. Since the embryo develops during that time there is a lot of change in gene expression due to development, we are not primarily interested in here. Therefore we normalize the expression values to the controls of each time point, therefore retrieving logFCs for each timepoint.

for (substance in c("diuron", "diclofenac", "naproxen")) {
    assign(paste("data", substance, "logFC", sep = "_"), value = toxprofileR::calc_logfc(get(paste0("data_proc_", 
        substance))))
    
    save(list = paste("data", substance, "logFC", sep = "_"), file = paste0("./data/", 
        substance, "/data_logfc_", substance, ".Rd"))
}

mobi-CTR model

The mobi-CTR model describes a sustained response on the concentration scale and a temporal response on the time scale. It is based on the “Hill equation”, a 3-parameter non-linear model, originally describing the binding of hemoglobin to oxygen dependent on oxygen saturation (Hill 1910). Due to its flexibility on the one hand, and physiological meaningfulness on the other hand, it was later on used in many applications (reviewed in Goutelle et al. 2008) and also proposed for pharmacological dose response modeling (Wagner 1968). One representation of the Hill-equation is shown below. It is defined by the parameters \(logFC_{max}\), \(slope\), and \(X_{50}\). The parameter \(logFC_{max}\) is the maximum logarithmic fold change observed for the respective transcript, the \(slope\) defines the steepness of the curve and \(X_{50}\) defines the concentration, where the response reaches half-maximum.

We observed that on time scale most responses showed a biphasic response. This progression can be captured by a time dependency of the parameter \(X_{50}\). Empirically, we discovered that the dynamics of the reciprocal of \(X_{50}\) is in many cases accurately captured by the logarithmic gaussian function. We call the reciprocal of \(X_{50}\) “Sensitivity”, since a small \(X_{50}\) indicates a sensitive response (i.e., the half-maximum is already reached at low concentrations). Thus, we get a full regression model describing the time and concentration dependent \(logFC\) after compound exposure: \[ \begin{align} logFC(c) &= \dfrac{logFC_{max}}{1+e^{-slope*(\log(c)-\log(X_{50}))}}\label{logFCc}\\ sensitivity(t) &= \frac{1}{X_{50}(t)} = S_{max}*e^{-0.5*(\frac{log(t)-log(t_{max})}{S_{dur}})^{2}}\label{Sensitivity}\\ logFC(c,t) &=\\ &logFC_{max}/\left[1+exp\left(-slope*(\log(c)-\right.\right.\\ &\log(1/(S_{max}*exp(-0.5*(log(t)-\\ &\left.\left.log(t_{max}))/S_{dur})^{2})))\right)\right]+\epsilon,\\ &\epsilon \sim \mathcal{N}(0,\,\sigma^{2}), \end{align} \] where \(logFC_{max}\) corresponds to the maximum fold change of the respective gene across all conditions, \(S_{max}\) is the maximum sensitivity (\(1/EC_{50}\)) of the gene, \(t_{max}\) is the time-point of maximum sensitivity, and \(S_{dur}\) represents a measure of duration of the sensitivity interval.

Extrema

To get an estimate for the paramter \(logFC_{max}\) in the CTR-model, we load the logFC data of all datasets and find extreme values. For each node the maximum and minimum \(logFC\) across all given experimental treatments is retrieved.

library("toxprofileR")

for (substance in c("diuron", "diclofenac", "naproxen")) {
    
    load(file = paste0("./data/", substance, "/data_logfc_", substance, ".Rd"))
    
}

dslist <- list(data_diuron_logFC, data_diclofenac_logFC, data_naproxen_logFC)

load(file = "./data/all/universe_nodeframe.Rd")

extrema <- toxprofileR::get_extrema_nodes(dslist = dslist, nodeframe = nodeframe)

save(extrema, file = "./data/all/node_extrema.Rd")

Fitting

Parameter estimation is performed using the shuffled complex evolution algorithm implemented in the hydromad package. It is based on the Nelder-Mead-Algorithm, but includes the generation and shuffling of several simplices generated from random points across the parameter space. This makes the algorithm robust and independent from starting values, which is why we selected it for parameter estimation of our toxnode models.

library("cowplot")

for (substance in c("diuron", "diclofenac", "naproxen")) {
    
    tcta_paramframe_som <- toxprofileR::fit_tcta(elist = get(paste("data", substance, 
        "logFC", sep = "_")), extrema = extrema, nodeframe = nodeframe, cluster = F)
    
    p1 <- ggplot(data = data.frame(AICweight = c(unlist(tcta_paramframe_som$AICw_up_hill)[unlist(tcta_paramframe_som$direction_hill == 
        1)], unlist(tcta_paramframe_som$AICw_down_hill)[unlist(tcta_paramframe_som$direction_hill == 
        -1)])), aes(AICweight)) + geom_histogram(col = "black", alpha = 0.2, 
        binwidth = 0.1) + xlab("AIC weight compared to null model") + theme(axis.text = element_text(size = 24), 
        axis.title = element_text(size = 28))
    
    p2 <- ggplot(data = data.frame(AICweight = unlist(tcta_paramframe_som$AICw_vspline_hill)), 
        aes(AICweight)) + geom_histogram(col = "black", alpha = 0.2, binwidth = 0.1) + 
        xlab("AIC weight compared to spline") + theme(axis.text = element_text(size = 24), 
        axis.title = element_text(size = 28))
    
    print(p1)
    print(p2)
    
    assign(x = paste0("tcta_paramframe_som_", substance), value = tcta_paramframe_som)
    save(list = paste0("tcta_paramframe_som_", substance), file = paste0("./data/", 
        substance, "/tcta_", substance, ".Rd"))
}

## 77221.75 sec elapsed

## 68325.344 sec elapsed

## 60162.049 sec elapsed