Defining rules for cancer cell proliferation in TRAIL stimulation

Bioinformatics

Spatiotemporal CA model (model A)

We developed a CA model (Methods), initially using rules and related parameters from Conway’s game of life:

  1. i.

    Any cell with less than two live neighbors dies, caused by under-population.

  2. ii.

    Any cell with two or three live neighbors lives on to the next generation.

  3. iii.

    Any cell with more than three neighbors dies, caused by overcrowding.

  4. iv.

    Any dead/empty cell with three live neighbors becomes live cell as by reproduction (division).

Although the rules, owing to its abstractness or oversimplification, may not be sufficient to model cancer self-organization successfully, nevertheless, they can generate complex self-organizing spatiotemporal patterns that have been explored in numerous scientific fields.20 We simulated to see how the automata will evolve in a simulated dish for three experimental conditions: (i) untreated HT1080 (WT), (ii) HT1080 treated with TRAIL (TRAIL), and (iii) HT1080 treated with TRAIL and PKC inhibitor BIS I (TR + BIS) based on the actual cell numbers after initial treatment. These conditions were taken from our previous experimental work in a dish,11 where we estimated the actual cell numbers for each condition to initialize the CA model. WT had an average of 1 × 105 cells, TRAIL had 6 × 104 and TR + BIS had 5 × 103 cells after 24 h treatment.

In other words, we run the CA model with Conway’s game of life in three initial conditions determined by the number of cancer cells in unstimulated (WT), after 24 h treatment using TRAIL alone (TRAIL), and TRAIL plus BIS I (TR + BIS). The model was simulated for 15 discrete arbitrary time steps and repeated 30 times to check whether the random orientation of cells for each run at initial time ( = 0) will have any significant variable outcomes (Fig. 1). For WT (high initial cell numbers), we observed the number of HT1080 cells decreased significantly for the first time-step, owing to overcrowding (Rule iii), but quickly recovered and increased rapidly with time (Fig. 1a, b, blue). For TRAIL (moderate initial cell numbers), the simulations, although fluctuating, shows a general increase with time and converged to similar cell numbers with WT by 15 time steps (Fig. 1a, b, red). Note that the variability between each of the 30 simulations were low for both WT and TRAIL (Fig. 1c, blue & red). Contrary, TR + BIS (low initial cell numbers) showed stable population of cells in time, despite relatively high variability between the simulations (Fig. 1a–c, green). Note that in some runs, the actual cell numbers are even decreased from time 0 (Fig. 1b, c, green). In other words, the CA Model simulations indicate that, if all the initial cell numbers are low (TR + BIS) and fixed in their location (immovable), the cell proliferation capacity will remain low. However, if the initial numbers are moderate (TRAIL) or high (WT), the cells will continue to proliferate over time. Thus, according to our simulations, TR + BIS may successfully control the progression of HT1080 in cell cultures over time.

Fig. 1

Simulations of spatiotemporal evolution of HT1080 cells using CA model A. a 2-D top view of the simulations for 15 time steps, b the total numbers of cells in time, c the number of survival cells at end time ( = 15) for 30 simulations. Untreated (WT), TRAIL stimulated (TRAIL), and TRAIL and BIS I treated (TR + BIS)

CA model with distinguishing E & M cells (model B)

It is now generally appreciated that EMT occurs during cancer cell progression and metastasis.2,3 The EMT is a central process of embryogenesis or normal cell development, in which epithelial (E) cells reach fibroblast-like mesenchymal (M) state. In brief, the fairly round shaped E cells are connected by cell–cell adhesion, facilitating cell layer formation and remain largely static.21,22 Unlike the E cells, the thin and long shaped M cells are not bounded to other cells and are free to move across a medium.23 It is also known that, under stressed conditions, E cells can transform into M cells, a process that is referred to as EMT.24

It is notable to mention here that HT1080 are fibrosarcoma cell line derived from a mesenchymal lineage, and hence, often considered not to exist in the epithelial stage. However, recent works have indicated the occurrence of mesenchymal-to epithelial transition in fibroblast cells.25 To investigate this, we performed experiments on HT1080 cells, in 48 plated wells (Methods), left untreated, stimulated with 200 ng/mL of TRAIL, and pre-treated with 10 μm of BIS I prior to TRAIL stimulation (200 ng/mL). Notably, the experiments reproduced previous observations of ~ 60% and 5% cancer cell survival (compared with unstimulated) after 24 h for the TRAIL and TR + BIS treatments, respectively (Fig. 2a–c, top panels).

Fig. 2
Fig. 2

Microscope and immunofluorescence images of HT1080. Cells were imaged by microscope (top) and Cytell imaging system using dyed nucleus in blue (bottom). Based on the cell morphology and the intensity of green anti-Vimentin antibody fluorescence (Methods), cells can be classified as epithelial or mesenchymal (see insert). a WT, b TRAIL, and c TR + BIS

When the morphology of the HT1080 cells was checked on the microscope, they mainly appeared to be made up of E cells for the unstimulated and TRAIL-stimulated condition, whereas TR + BIS treatment showed visible M cells (Fig. 2a–c, top). To confirm this observation, we used anti-Vimentin antibody with our HT1080 cell culture to track its fluorescence (Vimentin, an intermediate filament protein specifically expressed in M cells, is often used as a marker for mesenchymal cells26,27,28). We noticed that untreated HT1080 cell culture did not show any Vimentin fluorescence, indicating large majority of E cells (Fig. 2a, bottom). However, when we investigated the fluorescence for TRAIL, and TR + BIS-treated cells, we observed fluorescence for certain cells for the former and majority for the latter (Fig. 2b, c, bottom). Based on the fluorescence intensity, we estimated that the number of M cells in HT1080 was low for untreated (~ 2%) and TRAIL treated (~ 10%), but significantly higher for TR + BIS treated (~ 95%) (Fig. S1A). Thus, from these data, we gather that TR + BIS treated remnant cells have largely undergone EMT in a plate.

It is now conceivable that the rules for the automaton in our model should be able to reflect the generalized cancer cell proliferation, considering the EMT. Thus, rules (ii) and (iv) were modified and two additional rules (v and vi, see below) to distinguish E and M cells in the HT1080 population were included. The overall rules are as follows:

  1. i.

    Any E cell with less than two live neighbors dies, caused by under-population.

  2. ii.

    Any E cell with two or three live neighbors become M cells on to the next generation.

  3. iii.

    Any E cell with more than 3 neighbors dies, caused by overcrowding.

  4. iv.

    Any dead/empty cell with three live neighbors (E or M) becomes live cell (E or M) as by division.

  5. v.

    Any M cell is able to move randomly to an empty cell on to the next generation.

  6. vi.

    Any M cell that is unable to move becomes an E cell on to the next generation.

The rationale for choosing rule (v) is based on the fact that M cells are mobile,21 hence, they are allowed to move anywhere in the lattice randomly. For rule (vi), it is now known the reverse of EMT, the mesenchymal-to-epithelial or MET, is also a possibility when cells become crowded.29

We simulated the revised CA model with actual E and M cells for the initial proportions of cells identified at time 0. For both WT and TRAIL, the resultant simulations show a general increase in the overall HT1080 population survival levels compared with the previous model without E and M cell distinction (Fig. 3a, red & blue). For TR + BIS, where each cell has a low probability of contact with another cell due to large between space, the populations remain little changed when all are E cells as observed from Fig. 1. Introducing M cells here increases the probability of contact by the moving cells. This feature allows the proliferation capacity of isolated cells through rules (ii) and (v), which this time resulted in significant cell proliferation for the TR + BIS (Fig. 3a, green).

Fig. 3
Fig. 3

Simulations of spatiotemporal evolution of HT1080 cells using revised EMT CA model B. ac Similar layout to Fig. 1. Note, epithelial and mesenchymal cells are represented in green and red, respectively. d Time evolution of isolated and clustered cells for TR + BIS Simulation. (bottom panel) Close-up views of the simulations, Blue arrow: isolated cell remains largely unaffected. Yellow arrow: cell-to-cell contact of clustered cells invokes proliferation according to CA rules

Notably, the simulations show that at early time steps, cells that form small initial clusters through cell-cell interaction tend to grow more rapidly in time, unlike isolated cells (Fig. 3b, yellow and blue arrows). Thus, the simulations predict that even with a highly effective treatment strategy of TRAIL with BIS I for HT1080 cells, with the increased proportion of M cells relative to E cells, in the longer term, the treatment could become ineffective due to the spatial-temporal re-organization of the clustered cancers.

Experimental investigation

To scrutinize our CA predictions, we prepared time course experiments that can be directly compared with the simulation results. To monitor the proliferation of remnant HT1080 cells, we removed the existing media from the remaining cells at 24 h and replaced it with a new media without any treatment for all three conditions (Fig. 4a and Methods). The cells were monitored using live microscopy and counted at 0, 6, 24, 48, 96, and 144 h after the media were replaced.

Fig. 4
Fig. 4

Experimental evolution of HT1080 cells. a Schematic experimental design. HT1080 cells were plated in 48-well plate prior to 48 h of medium change. After 24 h of plating cells, cells were treated following conditions (i) untreated (n = 3), (ii) TRAIL stimulated (200 ng/ml) (n = 3), (iii) TRAIL + BIS treated (10 μm of BIS was pre-treated 30 min prior to 200 ng/ml of TRAIL stimulation) (n = 3). b Average plot (with S.D.) of three independently repeated experiments (Fig. S3) for cell numbers against time at 0, 6, 24, 48, 96, and 144 h. Untreated in blue (WT), TRAIL-stimulated in red (TRAIL), and TRAIL + BIS in green (TR + BIS). Note that the S.D within the replicates are low (Fig. S3), but between independent runs are larger due to variable initial (t = 0) cancer cell numbers. c The effect of retreating TR + BIS for evolving HT1080 cells. HT1080 cells, treated initially with TR + BIS, were again re-treated with the same dosage of TRAIL (200 ng/ml) and BIS (10 μm) after 144 h (n = 3). Note that about 90% of the evolving cells died for the repeated treatment after 24 h

Notably, in general accordance with our CA simulations, the HT1080 cells in (i) untreated and (ii) TRAIL-treated conditions rapidly grew and filled the entire wells within 24 h (Fig. S2S3 and Fig. 4b). For (iii) TR + BIS treatment, the cells remained stable upto 48 h or 96 h, after which sharp proliferation occurred (Fig. S2S3 and Fig. 4b). These data demonstrate, even after treatment that kills over 95% cells, that the remaining small number of cancer cells are able proliferate quickly after a limited lag time, and highlights the difficulty facing current therapeutics.

To check for the EMT or MET process in our cancer proliferation, we once again targeted Vimentin for fluorescence tracking in the culture with time (Methods). We noticed, for each condition, the proportion of M cells increased in time initially (indicating EMT process), but subsequently decreased when the respective cultures became overcrowded with cells for all three conditions (Fig. S2, bottom), reminiscent of MET process.26 Using cytell cell imaging system, this was verified by the actual counting of E and M cells’ proportions (Table 1 and Fig. S1B–F).

Table 1 Experimental (averaged over three replicates) proportion of epithelial (E) over mesenchymal (M) cells

Finally, to test whether our TR + BIS treatment strategy will still be effective despite the self-organization of cancer cells, we specifically re-treated TR + BIS cell cultures after 144 h. Remarkably, we observed the re-treatment yielded ~ 90% cell death despite the regrowth of cells after the initial treatment (Fig. 4c). These data show that although cancer cells are able to proliferate after treatment, they do not become resistant to the second treatment within the timeframe of the experiments. Thus, we believe the cyclic treatment of TR + BIS can still be an effective long-term strategy, despite the self-organizing or proliferative ability of the remaining treated cells.

Finalizing rules for CA Model with experimental fitting (model C)

Now that the experimental cancer cell proliferations have been tracked, we incorporated a genetic algorithm-based (GA) parameter fitting function into our model B and searched for a best fit using minimization of distances between experimental and simulated points (Methods). We, iteratively, modified the rules manually and allowed GA to automatically fit the cell numbers for each rule. The best result (Fig. 5a) follows:

  1. i.

    Any E cell with less than four live neighbors become M cell on the next generation.

  2. ii.

    Any M cell with more than eight live neighbors become E cell on the next generation.

  3. iii.

    Any dead/empty cell with two to six live neighbors (E or M) becomes live cell (E or M) as by division.

  4. iv.

    Any M cell is able to move randomly to an empty cell on to the next generation.

  5. v.

    Any M cell that is unable to move becomes an E cell on to the next generation.

Fig. 5
Fig. 5

Simulations of spatiotemporal evolution of HT1080 cells using experimentally fitted model C. a The total numbers of cells in time (simulation in solid lines and experimental data in dashed lines with error bars). Note that this time, we kept the simulation time to 12 steps so that we cover at least two time points for every cell cycle of ~ 24 h. That is, each simulation step corresponds to 12 h or half cell cycle. b The total numbers of cells in time with model parameters individually fitted to each condition. The parameters for rules (i) 4 (WT), 1 (TRAIL, TR + BIS), (ii) 9 (WT), 5 (TRAIL, TR + BIS), and (iii) 5–8 (WT), 4 (TRAIL), 2–4 (TR + BIS). Simulation in solid lines and experimental data in dashed lines with error bars. c 2-D top view of the simulations in b. d The proportion of epithelial e over mesenchymal (M) cells in time. Simulation in solid lines and experimental data (Table 1) in dashed lines with error bars

Although the simulations curves captured the general trends, WT and TRAIL cell number simulations were overpredicted for most of the time points (Fig. 5a). We conjectured that this might be owing to each condition possessing different model parameters. Thus, to better match the experimental proliferation numbers, we varied the model C parameters for rules (i) to (iii) individually to match each condition. That is, we kept the rules the same but the parameters within the rules were varied. Notably, this time, we were able to better fit the experimental data according to each condition (Fig. 5c, d).

Articles You May Like

Human gut microbiome physiology can now be studied in vitro using Organ Chip technology
Electrode’s ‘hot edges’ convert carbon dioxide gas into fuels and chemicals
Remote Island Chain Has Few People — But Hundreds Of Millions Of Pieces Of Plastic
New coating could have big implications for lithium batteries
How To Help A Kid Survive Early Puberty

Leave a Reply

Your email address will not be published. Required fields are marked *