Mammalian cell growth characterisation by a non-invasive plate reader assay

Automated and non-invasive mammalian cell analysis is currently lagging behind due to a lack of methods suitable for a variety of cell lines and applications. Here, we report the development of a high throughput non-invasive method for tracking mammalian cell growth and performance based on plate reader measurements. We show the method to be suitable for both suspension and adhesion cell lines, and we demonstrate it can be adopted when cells are grown under different environmental conditions. We establish that the method is suitable to inform on effective drug treatments to be used depending on the cell line considered, and that it can support characterisation of engineered mammalian cells over time. This work provides the scientific community with an innovative approach to mammalian cell screening, also contributing to the current efforts towards high throughput and automated mammalian cell engineering.


Supplementary Information
. Numbers of biological repeats for each sample are reported in Table S3.Data analysis is described in the Methods section and in Supplementary Note 1. Source data are provided as a Source Data file.S3.For cell counts, data are presented as mean values +/-SD.f) Linear relation between ln(GI) and ln(C) for HT1080 cells when cultured in DMEM containing 5mg/L (left) or 15mg/L (right).The red line is the best fit of data within the linear region.Coefficient of determinations are indicated (r 2 =0.914 for 5mg/L and r 2 =0.953 for 15mg/L).e) Bar plot of CF for HT1080 cells cultured in DMEM containing either 5mg/L or 15mg/L phenol  S2.Numbers of biological repeats for each sample are reported in Table S3.Data analysis is described in the Methods section and in Supplementary Note 1. Source data are provided as a Source Data file.

Figure S6. Reproducibility test of plate reader growth characterisation for K562 and Jurkat cells grown in different conditions.
To test the automatable capability of our plate reader assay, the data set obtained for each growing condition was split.CF from one subset were averaged together (Table S5 S3.Data analysis is described in the methods section and in Supplementary Note 1. Exact p values and source data are provided as a Source Data file.Data were normalised to 100% growth rate in the 0-24 hours window.The height of the bar represents the mean value of the single replicates shown as dots.After colchicine addition, µc is decreased by ~100%, and µp is decreased by ~75%.Growth rates are reported in Supplementary data file 2 while all replicates can be found in Supplementary data file 1. Source data are provided as a Source Data file.S2. f) Bar plot of CF for HEK293TLP-EBFP cells compared to their HEK293T parental cell line.The height of the bar represents the average value of the single replicates shown as black dots.Numbers of biological repeats for each sample are reported in Table S3.Data analysis is described in the Methods section and in Supplementary Note 1. Source data are provided as a Source Data file.A two-sided t-test indicated that normalised EBFP levels increase significantly each day (**: p<0.01).Numbers of biological repeats for each sample are reported in Table S3.Data and statistical analysis are described in the Methods section.Exact p values and source data are provided as a Source Data file.Tables S4.Average CF +/-SEM for cells grown at 37°C with Glu.

Supplementary tables
Our method is based on the three algorithms briefly described below (pseudo-codes in the following Tables I-III).
ALGORITHM 1 allows to select the time window corresponding to the fastest growth in manual counts growth curves and to compute the growth rate in such time-window.To do this, the algorithm: i) first performs linear fits of at least 3 consecutive points in sliding windows that start from time 0 and slide until the end of the growth curve (lines 1 -10 in Table I).ii) Then, the algorithm selects the range of data points whose linear fit gives the maximum slope (line 12).In this way, the lower, the upper data limits, and the number of points to fit are optimized in the regression.The slope of the linear fit of these data points is the growth rate of the exponential phase (µ c ).This procedure is performed on curves obtained with manual counts and expressed in log scale, normalized to their initial value.The choice of having at least 3 data points has been based on the interest in estimating the average growth rate in the exponential phase and not the maximum growth rate.The latter, given the resolution of our manual counts, could be more easily biased by eventual experimental variability in cell counts.However, the supplied script for data analysis allows the user to choose the minimum number of data points on which to perform the fit.
In our analysis, such number (i.e., the parameter minNdata in Table I-III) has been set to 3 for all the growth conditions at 37°C, except for i) Jurkat Man 37°C where it was set to 4 and ii) Jurkat and K562 growing at 33°C where it was set to 5. The choice of the two values at points i) and ii) was driven by a longer duration of the exponential phase in the analysed conditions with respect to the other conditions.Concerning data treated with doxo, the described procedure was applied to the data treated with all concentrations of doxo (by starting with at least 2, 3 or 4 data points, depending on the condition) except for HT1080 + doxo 1 µM.In this latter case, since the cells were almost notgrowing, the automatic procedure was biased by fluctuations in cell counts.For this reason, we manually estimated the time window for calculating the growth rates as including the first 3 data points of each curve.
ALGORITHM 2 aims at computing the CF.For each growth curve measured with both C and GI, it relies on the outputs of ALGORITHM 1 to i) first selecting the time points in GI growth curve that correspond to the fastest growth in C growth curve (lines 1,2 in Table II), then ii) it computes the growth rate in GI growth curve (µ p ) as the slope of the linear fit among such points (line 3), and finally iii) calculates the conversion factor CF = µ p /µ c (lines 4-5).
ALGORITHM 3 quantifies the linear relation between ln(GI) and ln(C) in their region of fastest growth.To do this, as ALGORITHM 2 it i) first uses the output of ALGORITHM 1 to identify the values of GI corresponding to the region of fastest growth in C growth curve (lines 1-6 in Table III) and then ii) computes the linear fit between ln(GI) and ln(C) (lines 7,8).
Pseudo-codes of algorithms developed for data analysis.

Outputs:
1. idxµC = positions of time-points with the maximum slope

Figure S1 .
Figure S1.GI calculation for suspension cells from phenol red individual plate reader absorbance values.a) Representative duplicate variations of Abs430 (yellow) and Abs560 (pink) over time (left) for RPMI medium when K562 cells are cultured.Resulting Abs430 over Abs560 ratios (right, blue) over time.b) Average background and error of the mean for Abs430/Abs560 ratio of RPMI over time (n=8) in control wells containing no cells.c) Representative duplicate GI profiles over time for K562 cells, resulting from Abs430/Abs560 ratio normalised to RPMI background.d) Representative duplicates of Abs430 (yellow) and Abs560 (pink) over time (left) for RPMI medium over cultured Jurkat cells.Resulting Abs430 over Abs560 ratios (right, blue) over time.e) Background and error of the mean for Abs430/Abs560 ratio of RPMI over time (n=8) in wells containing no cells.f) Representative duplicate GI profiles over time for Jurkat cells, resulting from Abs430/Abs560 ratio normalised to RPMI background.Source data are provided as a Source Data file.

Figure S2 .
Figure S2.GI calculation for adhesion cells from phenol red individual plate reader absorbance values.Abs430 (yellow) and Abs560 (pink) over time (left) for DMEM medium when HT1080 (a) and HEK293T (b) cells are cultured.Resulting Abs430 over Abs560 ratio (right, blue) over time (n=4).Average background and error of the mean for Abs430/Abs560 ratio of DMEM over time (n=3) in control wells containing no cells (c, d) GI profiles over time for HT1080 (e) and HEK293T (f) cells, resulting from Abs430/Abs560 ratio normalised to DMEM background (n=4).Source data are provided as a Source Data file.

Figure S3 .
Figure S3.Linear relation of ln(GI) and ln(C) for Jurkat, HT1080 and HEK293T cells grown at 37 o C with Glu.Graphs representing the relation between GI and C for a) Jurkat, b) HT1080 and c) HEK293T cells in logarithmic scale.The red solid line is the best fit of the data in linear region for which r 2 values are indicated.Linear fit equations are reported in TableS2.Numbers of biological repeats for each sample are reported in TableS3.Data analysis is described in the Methods section and in Supplementary Note 1. Source data are provided as a Source Data file.

Figure S4 .
Figure S4.Effect of phenol red concentration on GI curves for HT1080 cells.a) Individual growth curves resulting from counting HT1080 cells cultured with 5mg/L (light orange) or 15mg/L (dark orange) phenol red (n=3).b) Overtime measure of Abs430 (green and yellow lines) and Abs560 (purple and pink lines) for HT1080 cells cultured in DMEM containing 5mg/L or 15mg/L phenol red (n=5).c) Overtime Abs430/Abs560 ratio for DMEM background in control wells containing no cells in the presence of 5mg/L (light blue) or 15mg/L (dark blue) phenol red (n=3).d) GI profiles over time for HT1080 cells cultured in DMEM containing 5mg/L (light blue) or 15mg/L (dark blue) of phenol red (n=5).e) Representative growth curves resulting from phenol red acidification (GI, blue, left vertical axis) and averaged cell counts (C, orange, right vertical axis) of HT1080 cells cultured with 5mg/L (light blue and light orange) or 15mg/L (dark blue and dark orange) of phenol red (n=3).Numbers of biological repeats for each sample are reported in TableS3.For cell counts, data are presented as mean values +/-SD.f) Linear relation between ln(GI) and ln(C) for HT1080 cells when cultured in DMEM containing 5mg/L (left) or 15mg/L (right).The red line is the best fit of data within the linear region.Coefficient of determinations are indicated (r 2 =0.914 for 5mg/L and r 2 =0.953 for 15mg/L).e) Bar plot of CF for HT1080 cells cultured in DMEM containing either 5mg/L or 15mg/L phenol

Figure S5 .
Figure S5.Growth of Jurkat suspension cells under different conditions can be characterised by a plate reader assay.Representative GI (blue, left vertical axis) and C (orange, right vertical axis) profiles over time for Jurkat cells grown with glucose (Glu) at 33 o C (a) or mannose (Man) at either 37 o C (b) or 33 o C (c).All biological replicates can be found in Supplementary data file 1. Relation between GI and C for Jurkat cells grown in presence of Glu at 33 o C (d) or Man at either 37 o C (e) or 33 o C (f) in logarithmic scale.The red solid line is the linear fit of the data in the linear region for which r 2 values are indicated.Linear fit equations are reported in TableS2.Numbers of biological repeats for each sample are reported in TableS3.Data analysis is described in the Methods section and in Supplementary Note 1. Source data are provided as a Source Data file.
Figure S6.Reproducibility test of plate reader growth characterisation for K562 and Jurkat cells grown in different conditions.To test the automatable capability of our plate reader assay, the data set obtained for each growing condition was split.CF from one subset were averaged together (TableS5) to convert µp into µcp in the remaining data subset.a) Bar plots of all (grey) or a subset (sub, blue) of CF for K562 (top) and Jurkat (bottom) cells grown in presence of either glucose (Glu) or mannose (Man) at 37 o C or 33 o C. b) Bar plots of µc (orange) and µcp (grey), calculated as µp x CF average, for K562 (top) and Jurkat (bottom) cells grown in presence of either Glu or Man at 37 o C or 33 o C are shown.A two-sided t-test indicated that average CF of all or one subset are compatible (p>0.05) and that µcp are compatible with actual µc (ns, non-significant: p>0.05).The height of the bars represents the average value of the single replicates shown as dots.Numbers of biological repeats for each sample are reported in TableS3.Data analysis is described in the methods section and in Supplementary Note 1. Exact p values and source data are provided as a Source Data file.

Figure S8 .
Figure S8.Jurkat cells treated with colchicine.Analysis of growth rate changes in Jurkat cells untreated (0ug/ml) or treated (0.025ug/ml) with colchicine added at 24h after the start of the assay (black arrow).a) Growth curves resulting from phenol red acidification (GI, blue, left vertical axis) and cell counts (C, orange, right vertical axis) of Jurkat cells cultured with colchicine 0ug/ml (dark blue and dark orange) or 0.025ug/ml (light blue and light orange) (n=4).Data are presented as mean values +/-SEM.b) Histograms of relative µc (orange, left) and µp (blue, right) values of Jurkat treated cells.Data show µc and µp for the 0-24 hours time window (preceding colchicine addition and for the 24-48 hours time window (after colchicine addition).Data were normalised to 100% growth rate in the 0-24 hours window.The height of the bar represents the mean value of the single replicates shown as dots.After colchicine addition, µc is decreased by ~100%, and µp is decreased by ~75%.Growth rates are reported in Supplementary data file 2 while all replicates can be found in Supplementary data file 1. Source data are provided as a Source Data file.

Figure S9 .
Figure S9.Growth of HEK293TLP-EBFP cells can be characterised with the plate reader-based assay.a) Abs430 (yellow) and Abs560 (pink) over time (left) for DMEM medium when HEK293TLP-EBFP cells are cultured.Resulting Abs430 over Abs560 ratio (right, blue) over time (n=4).b) Average background and error of the mean for Abs430/Abs560 ratio of DMEM over time (n=3) in control wells containing no cells.c) GI profiles over time for HEK293TLP-EBFP cells, resulting from Abs430/Abs560 ratio normalised to DMEM background (n=4).d) Representative growth curves resulting from phenol red acidification (GI, blue) and cell counts (C, orange).All biological replicates can be found in Supplementary data file 1. e) Relation between GI and C in logarithmic scale.Red solid line is the linear fit of the data in linear region for which r 2 =0.916.The linear fit equations are reported in TableS2.f) Bar plot of CF for HEK293TLP-EBFP cells compared to their HEK293T parental cell line.The height of the bar represents the average value of the single replicates shown as black dots.Numbers of biological repeats for each sample are reported in TableS3.Data analysis is described in the Methods section and in Supplementary Note 1. Source data are provided as a Source Data file.

Figure S10 .
Figure S10.Fluorescence readout for mCherry and EBFP in presence and absence of phenol red in the medium.Cells were grown in the incubator with and without induction with 2ng/µl dox for 2 days (a) (mCherry, n=3) and 3 days (b) (EBFP, n=2).Cells were then resuspended in medium with (DMEM) and without (FluoroBrite) phenol red and fluorescence was measured at the plate reader.Cell counts were performed in parallel.A two-sided t-test indicated that average fluorescence level in DMEM and FluoroBrite are compatible (ns, non-significant: p>0.05).Data and statistical analysis are described in the Methods section.Raw values are reported in the source data file.Source data are provided as a Source Data file.

Figure S11 .
Figure S11.EBFP expression level in HEK293TLP-EBFP cells after dox induction.Bar plot of normalised fluorescence per cell (calculated as EBFP/10 5 C) for HEK293TLP-EBFP cells induced with 2ng/µl dox over three days.Data are presented as mean values + SD with single replicates shown as dots.A two-sided t-test indicated that normalised EBFP levels increase significantly each day (**: p<0.01).Numbers of biological repeats for each sample are reported in TableS3.Data and statistical analysis are described in the Methods section.Exact p values and source data are provided as a Source Data file.

2.
µC = value of the maximum slope 3. bestFit = best linear fit giving the maximum slope Initialization of variables: assign 0 to variable i Initialization of matrices: slopes = matrix of size (ndata) x (ndata -ndataToadd); R2s = matrix of size (ndata) x (ndata -ndataToadd); indexesfits = matrix of size (ndata) x (ndata -ndataToadd); # this will store the position of the first and the last data in the growth curve used for each fit.dataC at positions from i to (i + ndataToadd) 6 linear fit of dataToFit; 7 save the slope of the linear fit in the matrix slopes; 8 save the R2 of the linear fit in the matrix R2s; the indices in indexesfits corresponding to the fit with the maximum slope 13 return: IdxµC, µC

TABLE S1 .
Intervals of linearity for each condition: region of fastest growth for C growth curves and corresponding values for GI growth curves.

TABLE S2 .
Results of the best fits for determining the relation between ln(GI) and ln(C) for all growing conditions with corresponding r 2 and mean parameter values +/-SEM.

Table I :
Pseudo-code for selecting the timepoints corresponding to the region of fastest growth in manual counts (C) growth curves and computing µ c .