Cell culture.

The adherent MCF-7 cells (2012, CLS cell lines services GmbH, Germany, catalogue number 300273) were cultivated in Dulbecco’s modified eagle medium (DMEM) with 10% foetal calf serum (FCS) and without antibiotics, in a humidity controlled incubator with 5% CO₂ and at 37°C. They were detached before reaching confluence using short washes with Dulbecco’s phosphate buffered saline (DPBS) and a trypsin substitute TrypLE Express (all solutions from Thermo Fischer Scientific Inc, United States), which was competitively blocked by full medium addition and removed through a centrifugation step (3 min, 1000 rpm).

Measurement settings.

The oxygen was measured in percentage of air saturation, where 100% a.s. (air saturation) corresponds to the oxygen level in the air. The measuring interval was set to 1 min for sensor calibration and to 10 min for the rest of the measurement. The lot number of the OxoDish OD-1309-01 was entered into the SDR’s software (SDR_v38, PreSens Precision Sensing GmbH, Germany) allowing for the general calibration of the plates performed by the company for the whole lot.

Sensor correction.

The read-out of all the 24 sensors of two empty OxoDish was measured in the incubator for at least 10 time points after the read-out was stable (adaptation to temperature, oxygen and humidity conditions in the incubator).

Seeding.

The MCF-7 cells (passage 39) were seeded in a volume of 500 μL at a density of 150 000 cells/well in order to obtain a significant oxygen reduction. Three wells of each plate were filled with medium only in order to control the oxygen level in the incubator. The medium was changed gently 24 h after seeding.

Cisplatin treatment.

Fresh stock solutions of 5 mM and 1 mM Cisplatin (Sigma-Aldrich LLC, United States) were prepared in a 0.9% NaCl solution (B. Braun Melsungen AG, Germany). 24 h after medium change, different negligible volumes (30 μL or less) of the Cisplatin stock solutions were added to each well in triplicates (final concentrations: 300, 200, 100, 90, 80, 70, 60, 50, 40, 30, 20, 10, 0 μM).

Programming environment and packages.

The statistical framework R was chosen as programming environment [8], complemented with R packages for the fitting of sigmoid curves: drc [9] and the generation and customisation of the graphical user interface (GUI): gWidgets [10], gWidgetsRGtk2 [11], RGtk2 [12] and cairoDevice [13], which are dependent on the GTK+ widget library (http://www.gtk.org/). These resources are all freely available and compatible with Windows and Mac computers.

User interface.

Particular effort was put in designing a friendly user interface for the TReCCA Analyser. It was improved after being tested by 25 users for intuitiveness, clarity, and speed. Two screen shots of different tabs of the user interface, the “Analysis options” tab and the “Graph output” tab, are displayed in Fig 1A and 1B respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. TReCCA Analyser user interface and analysis steps.

Consecutive tabs are clicked through for data analysis and plotting. These include the “Analysis options” tab (A), where the analyses to be performed on the data are selected and their corresponding parameters entered, and the “Graph output” tab (B), where the graphs are visualised and can be further customised. These tabs can be seen at a higher resolution in S1 and S2 Figs. All the analysis steps are summarised in the program flow chart (C). The black rectangle encloses a more precise representation of the available analysis options. The black arrows depict the succession of steps that are performed in our example of the analysis of the effect of Cisplatin on MCF-7 cells, the coloured arrows depict alternative analysis flows.

https://doi.org/10.1371/journal.pone.0131233.g001

Analysis flow.

The flow chart in Fig 1C depicts the analysis flow of the TReCCA Analyser. The first step consists in importing the data into the program as a.csv file(s) in the “Data input” tab. The first column has to contain the time points of the experiment and the other columns the time-resolved data. The file import function can be customised to fit the experimental output files. In the “Labels and colours” tab, a template is filled with the experimental layout indicating which condition is to be found in each column of the data and how to plot it (colour, line type, legend). The template is a guide for the analysis, for example it determines which columns should be averaged as technical replicates. The analysis options and corresponding parameters, described more in detail in the next paragraphs, are selected from the “Analysis options” tab. Finally, the graph options are picked out from the “Graph options” and “Graph output” tab allowing the automatic customised creation of the graphs corresponding to each step of the analysis. These graphs, as well as.csv files containing all the analysis results, can then be exported by the user for further use through the “Graph output” and “Data output” tabs respectively. All of the settings can be saved for later reference and similar experiments. All the data can also be saved as an R-data file to allow a new customisation and exportation of the graphs without having to run the analysis again.

Analysis options.

The analysis options available in the TReCCA Analyser are briefly presented in the following paragraph, more details are available in the S1 File (for the equations) and TReCCA Analyser user manual (www.uni-heidelberg.de/fakultaeten/biowissenschaften/ipmb/biologie/woelfl/Research.html).

Basic formatting: The time unit can automatically be converted by indicating the input unit and desired output unit, and time points at the beginning or the end of the measurement can be excluded. Moreover, the time frame of the data can be shifted (to set a different start experiment time) or the whole data can be scaled to another unit by multiplying it by a fixed factor.

Sensor correction: In the challenging case where the calibration relationship follows an exponential equation of the form P + Se^αy (where P is the plateau, S the span and α an unknown factor), a sensor correction can be performed. This is the case with the OxoDish, although a Stern-Volmer relationship would be expected [14], as described in S3 and S4 Figs and S2 File. The data x is first converted to the correct logarithmic form z = ln((x − P)/S). Then the mean of each sensor m_s is calculated over the n time points (Eq A in S1 File), as well as the mean of the whole plate m_p, which is the average of all the sensor means m_s for s number of sensors (Eq B in S1 File). The sensor corrected value x_t,s for each time point t and each sensor s is then corrected and converted back to logarithmic form (Eq 1), followed by the normalisation step of sensor correction (Eq C in S1 File).

(1)

Normalisation: Each data column can be normalised over time to a reference condition set at a target value, for example a non-treated condition being set to 100% of viability. First, the columns with the data of the normalisation condition are automatically detected through the template and their average at each time point m_t is calculated. Then, every single value is divided by its corresponding value m_t and multiplied by the target value M (Eq D in S1 File), thereby setting the normalisation conditions to the target value and shifting the other conditions.

Average and standard deviation: The average and the standard deviation of technical replicates are automatically determined for each time point, according to either the name, number or colour given to each column in the template (Eq E in S1 File).

Smoothing: The data smoothing is performed by replacing a measured data point x_t by the average of n neighbouring data points around that time point (Eq F in S1 File). This transformation leads to the loss of (n − 1)/2 data points at the beginning and at the end of the measurement and n necessarily has to be an uneven number.

Slope calculation: A linear fit function built in R is applied to each smoothed data point and n smoothed data points on either side (Eq G in S1 File) to determine the slope m and the corresponding residual (which is displayed as a grey shade behind the graphs).

Time-resolved IC₅₀: The IC₅₀/EC₅₀ of a compound is determined based on any of the previous time-resolved data. First, the time frame that should be used for the IC₅₀ calculation is set, as well as how regularly the fittings should be performed. The user has to assign concentrations to the labels of the template, and the different concentrations are then fitted automatically and over time. Four different log-logistic model functions, each varying by their number of parameters, are available to perform the sigmoid fit (Eqs H, I and J in S1 File), the four logistic being the most commonly applied one (Eq 2) (2)

Basic formatting.

The MCF-7 and Cisplatin treatment data was extracted from the software provided with the SDR reader (SDR version v38), which is required for recording but does not enable adequate analysis of the data obtained. The TReCCA Analyser provides analysis options and plotting possibilities for the SDR data, but is also suitable for any other type of large time-resolved analytical measurements. The time scale was first converted from seconds to days. The last part of the graph is excluded, as after day 5 the viability of the controls decreases due to the lack of medium change. The raw data obtained after basic data formatting can be seen in Fig 2D. The cells seeded on the first day reduce the level of dissolved oxygen, which is replenished after medium change in the second day. The dissolved oxygen is then reduced again but this time in a more homogeneous way and to a lesser extent as the cell adhesion process does not interfere with the basal cell respiration levels. Starting on the third day, the dissolved oxygen level depends on the Cisplatin concentration: high Cisplatin concentrations lead to cell death and higher dissolved oxygen values. The cells in the non-treated condition reach confluence, the cell number and thereby the total oxygen consumption becomes constant as reflected by the stable dissolved oxygen values.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Medium oxygen level of MCF-7 cells when exposed to different concentrations of Cisplatin over time.

All the sensors are measured empty for sensor calibration between minutes 0 and 136 (A), then a sensor calibration is performed (B), followed by normalisation to the wells containing only medium (C). The whole time frame of the experiment is visible in the raw data (D): MCF-7 cell seeding at 136 min, medium change at day 1.09 and Cisplatin addition at different concentrations at day 2.1. The sensor correction and the normalisation to the conditions containing only medium are applied to all the data (E, F respectively). The time points used for sensor correction/normalisation (A, B, C) are depicted as black rectangles (D, E, F respectively). The triplicates of the sensor corrected and normalised data are averaged (G) and their standard deviation is displayed as a grey shadow around each curve. The legend at the bottom of the figure is valid for all the subfigures, microM stands for micromolar. These graphs are displayed as they are produced by the TReCCA Analyser.

https://doi.org/10.1371/journal.pone.0131233.g002

The data gathered on the first two days serves as quality control for the experiment: the basal cell respiration is as expected and reflects on the health of the cells; there are no outlier curves due to experimental errors or bacterial/yeast contaminations (that would have caused a sharp decrease in dissolved oxygen levels), the equipment is running correctly, the time points were respected exactly.

Standard data analysis.

Sensor correction (recalibration and normalisation) was performed on the raw data as detailed in the S2 File, using 91.3% a.s. as the target value (Fig 2E), thereby homogenising the initial output as can be seen more precisely on a smaller oxygen scale (Fig 2A and 2B). For experiments only taking place on such a scale, the sensor correction can be critical for the distinction of different conditions (S5 Fig).

As the level of dissolved oxygen measured in wells containing only medium slightly increases over the course of the experiment, the sensor corrected data was normalised by setting the average of the control wells to 91.3% a.s. (Fig 2F). The media curves are consequently straight horizontal lines and the rest of the data is slightly shifted accordingly. On a smaller scale, the normalisation does not affect the stabilised calibration values (Fig 2C). The technical replicates of the normalised data are then averaged (Fig 2G). The corresponding standard deviations are displayed as grey shades, which is usually challenging for highly resolved data sets using common spread sheet applications.

Smoothing and slope calculation.

For the next analysis steps, we focus on the data after Cisplatin treatment between day 2.2 and 5 (Fig 3A). The time frame could have been shifted to start at the time of treatment but was left for clarity. In order to determine the slope of the curves over time, the averaged data is first smoothened for n = 11 (Fig 3C) as can be seen more precisely at a bigger scale (Fig 3B and 3D). The slope of the smoothed data is then determined for n = 15 (Fig 3E). It reveals that the amount of dissolved oxygen increases first and fastest for the highest Cisplatin concentrations, its rate of change reaches a maximum and then decreases slowly until stabilising. Lower Cisplatin concentrations follow the same reaction pattern, however the initial speed of change in respiratory behaviour is slower. The non-treated condition and the lowest Cisplatin concentrations have a constant oxygen consumption during the course of the experiment, as the slope is close to zero. The slope calculation allows to pin-point the exact time when the change of respiration is highest or constant; it also leads to a more accurate comparison of the different Cisplatin concentrations.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Smoothing and slope calculation.

The oxygen levels in the media of MCF-7 cells exposed to different Cisplatin concentrations (as depicted in Fig 2G) between day 2.2 and 5 (A) are smoothed (C) by replacing each time point by the average of an 11 time point-neighbourhood. The smoothing can be seen more precisely on a bigger scale as it is the case for 75–80% a.s. and day 3–3.5 by comparing the raw data (B) to the smoothed data (D). The slope of each time point (E) is then obtained by performing a linear fit of each point and 15 points on either side of it. The residual error of the fit is displayed as a grey shadow around each curve (very small in this case). The legend in Fig 3E is valid for all the subfigures, microM stands for micromolar.

https://doi.org/10.1371/journal.pone.0131233.g003

Time-resolved IC₅₀ calculation.

The four-parameter log-logistic model (Eq 2) is fitted to the averaged data for every time point between day 2.5 and 4.7 (Fig 4A). The parameter e, which is the IC₅₀ value, is then plotted over time to display the time-resolved IC₅₀ of Cisplatin (Fig 4B) which ranges from 80 μM at day 2.5, to 30μM at day 4.7. These values are in accordance with previously published data [15, 16] taking into account diverging protocol details (seeding density, Cisplatin solvant, viability test, time points). The longer the exposure to Cisplatin, the lower the concentration needed to reach 50% control viability. This decrease begins almost immediately, it is fast for short exposure times but levels out progressively, almost reaching a steady-state. Cisplatin has a rapid effect on cell respiration combined with a high cell permeability, its lack of stability over time could explain the levelling out.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Time-resolved IC₅₀ of MCF-7 cells exposed to Cisplatin.

The oxygen level in the media of MCF-7 cells treated with different Cisplatin concentrations between days 2.2 and 5 (as depicted in Fig 3A) are fitted for each time point, as exemplified for day 2.50 (green), 2.89 (turquoise), 3.28 (blue), 3.66 (pink), 4.05 (red), 4.43 (yellow) (A). MicroM stands for micromolar. The IC₅₀ over time (B) is then determined from the values of all the fits between days 2.5 and 4.7. The residual error of the fit is displayed as a grey shadow around each curve.

https://doi.org/10.1371/journal.pone.0131233.g004