Cancer-driven dynamics of immune cells in a microfluidic environment

Abstract

Scope of the present work is to infer the migratory ability of leukocytes by stochastic processes in order to distinguish the spontaneous organization of immune cells against an insult (namely cancer). For this purpose, spleen cells from immunodeficient mice, selectively lacking the transcription factor IRF-8 (IRF-8 knockout; IRF-8 KO), or from immunocompetent animals (wild-type; WT), were allowed to interact, alternatively, with murine B16.F10 melanoma cells in an ad hoc microfluidic environment developed on a LabOnChip technology. In this setting, only WT spleen cells were able to establish physical interactions with melanoma cells. Conversely, IRF-8 KO immune cells exhibited poor dynamical reactivity towards the neoplastic cells. In the present study, we collected data on the motility of these two types of spleen cells and built a complete set of observables that recapitulate the biological complexity of the system in these experiments. With remarkable accuracy, we concluded that the IRF-8 KO cells performed pure uncorrelated random walks, while WT splenocytes were able to make singular drifted random walks that collapsed on a straight ballistic motion for the system as a whole, hence giving rise to a highly coordinate response. These results may provide a useful system to quantitatively analyse the real time cell-cell interactions and to foresee the behavior of immune cells with tumor cells at the tissue level.

Figure 1. Micro-fluidic co-culture immune-cancer system: design and fabrication.

Figure 2. Experimental design and methodologies: melanoma vs WT/KO immune cell interactions.

Figure 3. Differential migratory behavior of KO and WT cells toward melanoma cells.

B16 melanoma cells (green-labeled) and spleen cells (red-labeled) from either WT or KO mice at 24 h (A) and 48 h (B) after loading onto the microfluidic device. (C) Detail of splenocytes interacting with B16 cells.

Figure 4. Upper panel shows examples of real trajectories performed by WT cells, while lower panel depicts the same for KO cells.

WT, but not KO, splenocytes migrate from top-right through bottom-left following a chemokine gradient.

Figure 5. Prospect of the perturbation of B16 cells in microfluidic devices loaded with WT or IRF8 KO immune cells.

(A) Perturbation factor has been calculated for three representative B16 cells (Mel1, Mel2, Mel3) taken in each device by depicting the area fraction of that cell in each timepoint respect to the preceding timepoint. This allowed to obtain the percentage area variation for each given timepoint for the 24–42 h time interval. The examined B16 cells were located in the melanoma compartment of the microfluidic devices. (B) Multiple comparison between B16 cells showed in panel A by Mann-Whitney U test. Each of three B16 cells from device loaded with WT immune cells (WT-Mel1, WT-Mel2, WT-Mel3) was combined with each of three B16 cells from device loaded with IRF8 KO immune cells (KO-Mel1, KO-Mel2, KO-Mel3).

Figure 6. Logarithm of the probability distribution for the step length along the x direction (upper panel) and along the y direction (lower panel).

Experimental data (symbols) with standard errors are fitted by the exponential distribution (solid line) given by Eq. 8. All fits display R² ≈ 0.99. The best fit coefficients are reported in Tab. I, where a comparison between average values from experiments and theoretical description is also provided.

Figure 7. Inset: polar histogram of the turning angle.

The distribution has zero mean, hence, no angular correlation is observed. Main plot: angular correlation function C_KO(τ) of the turning angle θ. This correlation function shows more statistical noise at large τ.

Figure 8. Inset: binned data () with standard errors of mean instantaneous speed and related best fit (solid line).

KO splenocytes do not slow down over the observation time of 350 frames. Thus, for practical purposes, we can consider them as being in a time-independent state. Main plot: mean step length for each KO cell (thin curve) and mean step length averaged over all splenocytes (thick curve) at each time, which is essentially constant.

Figure 9. Mean displacement 〈r(t)〉 (upper panel) and mean squared displacement 〈r²(t)〉 (lower panel) for KO splenocytes.

Experimental (binned) data () with standard errors are compared with best fits (solid line) whose coefficients are properly shown.

Figure 10. Left panel: straightness index S_KO for each KO splenocyte (thin curves) and straightness index averaged over all splenocytes (thick curve) at each time.

Right panel: binned data for () with standard errors for KO splenocytes and best linear fit (solid line).

Figure 11. ψ_x(Δx) of WT-PRE splenocytes along the positive direction (upper panel, notice the semilogarithmic scale) and along the negative direction (lower panel, notice the logarithmic scale).

Analogous plots are obtained for the y direction (not shown). Experimental data of the distributions () with standard errors are compared with best fits (solid line). Note that the distribution is broadened along the negative x direction (and, analogously, along the positive y direction).

Figure 12. ψ_x(Δx) of WT-POST splenocytes along the positive direction (upper panel, notice the semilogarithmic scale) and along the negative direction (lower panel, notice the logarithmic scale).

Analogous plots are obtained for the y direction (not shown). Experimental data of the distributions () with standard errors are compared with best fits (solid line). Note that the distribution is broadened along the negative x direction (and, analogously, along the positive y direction).

Figure 13. Left panels: angular correlation function C_WT(τ) of the turning angle θ of WT-PRE splenocytes (upper panel) and of WT-POST splenocytes (lower panel).

Each turn occours every 4 minutes. In both cases the correlation is mostly positive, meaning that the process has memory. In the inset we show the polar histogram of the angle measured with respect to the horizontal axis. Right panels: distribution of the turning angle for WT-PRE splenocytes (upper panel) and of WT-POST splenocytes (lower panel). The experimental distribution is fitted by a Gaussian peaked at 0 rad, since splenocytes tend to maintain the same direction. Note that the resulting Gaussian shape is in full agreement with the mathematical description of the motion in terms of a (biased and uncorrelated) stochastic process (further an analysis of its variance will be successfully exploited in the last section due to numerical simulations).

Figure 14. Mean step length versus time for WT-PRE and WT-POST splenocytes (left and right panel, respectively).

Results for each splenocyte (thin curves) are compared with the resulting average over all splenocytes (thick curve). In both cases, appears constant in time.

Figure 15. Mean instantaneous speed of WT-PRE splenocytes (upper panel) and of WT-POST splenocytes (lower panel).

Binned data () with standard errors are best fitted by a constant line (solid curve). In both cases, no evident acceleration is observed, but, in the latter case, speed is lower.

Figure 16. 〈r(t)〉 versus t for WT-PRE splenocytes (upper panel) and for WT-POST splenocytes (lower panel).

As expected, the mean displacement grows linearly with time. Binned data () with standard errors are compared with best fit (solid line), whose coefficients are also reported.

Figure 17. (◊) and 〈r²(t)〉 (○) vs t for WT-PRE splenocytes (upper panel) and for WT-POST splenocytes (lower panel).

Symbols represent experimental data with standard errors, while the solid lines represent the best fits. Notice that, for both sets of splenocytes, and 〈r²(t)〉 are nicely overlapped, both growing with the square of time.

Figure 18. Straightness index S for WT-PRE splenocytes (left panel) and for WT-POST splenocytes (right panel); the thick lines represent the related averages over all splenocytes at each time 〈S〉.

As expected for a random walk with bias, it takes values close to 1.

Figure 19. Comparison between experimental data (), their best fit (solid line) and data from numerical simulation (the area highlighted is centered on the average value, where the average is performed over N independent realizations, and spans over its standard deviation).

The KO case (left panel), the WT-PRE case (central panel) and the WT-POST case (right panel) are all considered. Notice that experimental data and best fits are those previously reported in Figs. 9 and 16, respectively.