1.1 Experimental methods

The in vitro cultures that are being compared are described in detail in Kotze et al. [11]. Briefly, suspected TB patients were enrolled in the study. Bronchoalveolar fluid was collected with bronchoscopies performed by qualified clinicians and nursing staff according to international guidelines, and alveolar macrophages were isolated from this fluid [16]. Peripheral blood samples were taken by venipuncture into two 9mL sodium heparinized (NaHep) vacutainers, and PBMCs were isolated using the Ficoll-Paque method described previously [11]. Total CD3+ T cells were isolated from PBMC by MACS Microbead technology, as described previously [9]. First, alveolar macrophages were infected with Mycobacterium bovis Bacille Calmette-Guerin (BCG) at a multiplicity of infection (MOI) of 1 for 4 hours. Antimycotic antibiotic (penicillin/streptomycin/amphotericin B) supplemented media was then used to remove extracellular bacteria for an hour, with washes performed afterward. Alveolar macrophages were then treated with biocompatible NanoShuttle (m3D Biosciences Inc., Greiner Bio-One). The spheroid culture was established by the addition of a magnetic neodymium drive to levitate the alveolar macrophages, allowing the formation of an initial spheroid. Alveolar macrophages without a levitation drive were used to establish a traditional monolayer culture (Fig 1A). After 48 hours, autologous T cells were added to both the monolayer and 3D spheroid cultures. In the spheroid culture, an organized structure developed, with T cells forming a cuff around the macrophages (Fig 1B). In traditional monolayers, macrophages and T cells randomly settled to the bottom of the culture well. Thus, both traditional and spheroid cultures were created using the same patient cells, ratio of cells, bacterial strain, MOI, and timing of T cell addition. After 6 days of culture, spheroid cultures were mechanically dissociated using gentle pipetting, to ensure single cell suspensions of both spheroid and traditional cultures. Single cells were lysed and serial dilutions were plated on Middlebrook 7H11 agar plates (BD Biosciences). Colonies were manually counted after 21 days of growth. One of the findings of this work was that spheroids were better able to control bacteria than their traditional counterpart at 6 days post infection, which is reflected in the CFU/mL measurements (Fig 1C).

1.2 Computational methods

The computational model established and calibrated in Petrucciani et al. [15] was used to simulate the traditional and spheroid cell cultures. The model code can be found at https://github.itap.purdue.edu/ElsjePienaarGroup/TB-in-vitro release v1.0.1. Briefly, this computational model is an agent-based model (ABM) with bacteria, macrophages, CD4+ and CD8+ T cells represented as agents. Each of these agent types has an established set of biology-based rules to govern their behavior. An overview of these rules can be found in Petrucciani et al. Fig 2. [15] Bacteria grow and divide with rates dependent on whether the bacteria are extracellular or localized within a macrophage. Macrophages, CD4+ and CD8+ T cells are all subtypes of immune cells. These immune cells move probabilistically up a chemokine gradient, secrete cytokines when activated, and age and die with set lifespans. Chemokine secretion is based on the infection and activation status of the immune cell and estimated initial ranges of these values are drawn from literature. [17] These chemokines then diffuse throughout the simulated volume and decay, which is represented with a partial differential equation model [18].

Download:

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

Fig 2.

Example spatial layout of 4 simulation cases: a) spheroid structure with 3D movement, b) spheroid structure with gravity-limited movement, c) no structure with gravity-limited movement seen from the side, and d) no structure with 3D movement seen from the side. All simulations are done with 3 dimensions, but the ones with gravity-limited movement represent experimental “2D” models.

https://doi.org/10.1371/journal.pcbi.1012266.g002

Macrophages move around the environment and phagocytose bacteria with a set probability. If a macrophage internalizes one or more bacteria it becomes an infected macrophage. Infected macrophages no longer move but can kill the intracellular bacteria. If an infected macrophage is unable to control the infection and reaches a certain limit of intracellular bacteria, it bursts, releasing the bacteria into the environment. Both macrophages and infected macrophages can be activated, which increases their phagocytosis and killing abilities. Activation is a two-step process, requiring NF-κB and STAT1 signaling. NF-κB signal can be activated in one of three ways: interacting with an activated CD4+ T cell, a local TNF-α concentration above a threshold, or a high number of local bacteria. STAT1 is activated by IFN-γ, a cytokine secreted by activated T cells.

T cells are subdivided into CD4+ and CD8+ T cells. Both of these cell types follow immune cell rules and can be divided into TB-specific and non-TB-specific classes. TB-specific T cells can be activated and will proliferate when activated. CD4+ T cells are activated when they interact with a macrophage that has interacted with a bacterium. This is the equivalent of antigen presentation on MHC II. When CD4+ T cells are activated, they can activate NF-κB for macrophages through direct interaction or STAT1 though IFN-γ secretion. CD8+ T cells can be activated when they interact with a macrophage that has interacted with a bacterium and is also STAT1 activated. When CD8+ T cells are activated, they secrete IFN-γ like CD4+ T cells and can kill infected macrophages through their cytotoxic functionalities.

The difference in the two in vitro models of interest is simulated by adjusting the movement rules and the initial structure in our simulation. The traditional simulation has “gravity-limited” movement rules. Cells in this environment cannot float and can only move up the z-direction if they are moving on top of another cell. Neighboring cells are checked to ensure that they will meet these criteria. If the criteria are not met, then the grid square is removed from potential locations to move. All unoccupied and valid locations are then used to probabilistically move up the chemokine gradient if the concentration threshold for directed movement is met. In contrast, the spheroid simulation cells are not limited by gravity and can move in three dimensions. This means that all unoccupied grid squares are valid for probabilistic chemotaxis and random movement.

The initial structure of the cells in traditional and spheroid simulations also differs. The spheroid simulation is initiated as a sphere of macrophages in the center of the grid and the movement rules allow cells to move in any direction randomly or in response to chemokine gradients. In contrast, the traditional simulation has macrophages spread out throughout the grid space initially, with the movement rules resulting in the cells ‘sinking’ to the bottom of the grid (representing the bottom of the cell culture plate). When T cells are added to the spheroid simulation, they are placed in a spherical shell around the macrophage core. In the traditional simulation, T cells are distributed throughout the simulation space and, as with macrophages, allowed to ‘sink’ to the bottom.

1.3 Calibration and parameter selection

This hybrid model was parameterized in Petrucciani et al. with experimental data for bacterial fold change, cell count, and cell viability at day 6 from both the spheroid and the traditional cultures [15]. Calibration was done by performing a large initial sweep of the parameter space with 1000 Latin hypercube samples and 7 replicates. High traditional bacterial fold change only occurred in a few simulations, so iterative enrichment of these regions was performed. This was done by selecting enrichment points, narrowing the parameter space around these points, and repeating this sampling and enrichment process until increases in the desired output (i.e.., traditional bacterial fold change) no longer improved. Please refer to Petrucciani et al. 2024 for more details and an uncertainty analysis. [15] The final collection of calibrated parameter sets are then composed of all simulations from all iterations that produced outputs within the experimental ranges for our 6 criteria (spheroid and traditional bacterial fold change, cell counts, and cell viability). A full list of parameters, units, and initial ranges can be found in Table 1. Parameters in the text will be italicized for clarity. Distributions of parameters within various groups of simulations can be seen in S1 Fig.

Download:

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

Table 1. Parameters that are varied during calibration.

Initial ranges are either determined by literature, estimated through preliminary simulations (e), or broadened to the full mathematically possible range (f). The collection of calibrated parameter sets can be found in the supplementary material: parameter descriptions are given in S1 Table. This table is reproduced from Petrucciani et al. [15].

https://doi.org/10.1371/journal.pcbi.1012266.t001

1.4 Virtual knockout simulations

Virtual elimination of individual parameters or mechanisms (i.e., virtual knockout) simulations were performed by setting individual model parameters to zero. Whenever virtual knockouts are mentioned, it refers to simulations. Four different knockout scenarios were simulated. First, IFN-γ secretion from CD4+ T cells was eliminated. This was accomplished by running the simulation with an ActivatedCD4IFNSecretion parameter value of zero. Next, IFN-γ secretion from CD8+ T cells was knocked out. This was similarly accomplished by setting ActvatedCD8IFNSecretion to zero. These two knockouts were also simulated in tandem. This double knockout lacked any IFN-γ, meaning there was no macrophage activation. Finally, CD8+ cytotoxic killing was eliminated by setting CD8KillProbability to zero. These knockouts were compared with each other and the default simulations to quantify the relative contributions of these mechanisms to bacterial control and macrophage activation.

1.5 Movement rule swap

To assess the impact of movement restrictions or initial structure on infection progression, we computationally uncouple these two mechanisms. The default simulations are for spheroid and traditional setups. The spheroid simulation has 3D movement rules and the imposed spheroid structure (Fig 2A), while the traditional simulation has gravity-limited movement rules and no structure (Fig 2C). We simulate 2 new scenarios: one with 3D movement rules and no structure (traditional floating, Fig 2D) and one with gravity-limited movement and imposed structure (spheroid gravity, Fig 2B). These new simulation setups are compared with the two default setups to quantify the impact of movement and structure on bacterial control, macrophage activation, and T cell activation.

2.1 Calibrated simulations predict heterogeneous bacterial dynamics that are consistent with experimental observations

In Petrucciani et al., both spheroid and traditional simulations were initialized to reflect the corresponding experimental conditions as described in Section 2.1 and were calibrated to experimental data at day 6 [15]. A brief description of this calibration process is given in Section 2.3, but full details can be found in the previous work. Parameters are assumed to be the same between traditional and spheroid simulations. Here, calibrated simulations are examined to identify the causative mechanisms of bacterial control in spheroid and traditional simulations. Heterogeneity is observed in bacterial and host immune dynamics in these calibrated simulations because there is 1) a lot of uncertainty between initial and end time points where experimental data are not available and 2) the experimental data at the end time point has a large range/spread. Three qualitatively different behaviors emerge from this collection of simulations.

To separate the qualitatively different behaviors, a point of separation is selected visually (Fig 3A) to separate the dynamic behaviors of bacterial counts: parameter sets where bacterial loads follow similar trajectories between spheroid and traditional simulations (similarly controlled simulations), vs. parameter sets where bacterial load diverges between spheroid and traditional simulations (differentially controlled simulations). Due to the overlap in the experimental ranges of CFU (Fig 3A) between spheroid and traditional data, some of our simulations have similar bacterial fold changes for the traditional and spheroid simulations (Fig 3B). As expected, aggregate cellular dynamics (such as activated macrophages and activated T cells) are also similar for the spheroid and traditional similarly controlled simulations (S2 Fig). Other simulations show that it is possible to have differential control of bacterial growth between the traditional and spheroid conditions (Fig 3D and 3E). The spheroid and traditional CFU counts in these differentially controlled simulations begin diverging around days 3 and 4, and this divergence is only due to differences in initial structure and cell movement (in 3D vs. restricted to the bottom of the culture plate, respectively).

Download:

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

Fig 3.

Spheroid (blue) and traditional (green) simulations have been calibrated in a paired manner to in vitro data. a) Bacterial count dynamics for spheroid and traditional simulations over the 6-day time course. The blue and green bars on the right-hand side represent the ranges of experimental calibration data (calculated from fold changes). The black x is the point used to differentiate between similarly and differentially controlled simulations. We divide the simulations into two groups with traditional CFU greater than 6000 (differentially controlled simulations) or less than or equal to 6000 (similarly controlled simulations) at time step 550 (about day 2.29). b) Due to the overlap in the experimental data for CFU, some spheroid and traditional simulations perform similarly. c) The similarly controlled runs have a higher baseKillingProbability than the differentially controlled runs. Two patterns emerge in the differentially controlled simulations: d) simulations with two major decreases and e) simulations with one major decrease.

https://doi.org/10.1371/journal.pcbi.1012266.g003

These groups (similarly vs differentially controlled) correspond closely to separating the simulations into groups with low and high probability for an internalized bacteria to be killed by a resting macrophage (baseKillingProbability) (Fig 3C). Simulations where bacteria were similarly controlled in traditional and spheroid conditions have higher baseKillingProbability. This implies that in these similarly controlled simulations, resting macrophages are contributing more to controlling bacterial growth, and therefore these simulations are expected to depend less on adaptive immune parameters (such as the fraction of TB specific CD4s). The differentially controlled simulations have lower baseKillingProbability, meaning decreased killing ability of resting macrophages. This suggests that bacterial control is more dependent on the adaptive immune system following the addition of T cells at day 2; and that the added T cells are more effective in the spheroid simulations than the traditional simulations.

The differentially controlled simulations are further subdivided into two groups. Several parameters show bimodal distributions in their estimated values for differentially controlled simulations. The bimodal distributions correspond across these parameters (i.e., for all the simulations that have high parameter 1, they all have low parameter 2 and vice versa) (S1 Fig). Thus, the differentially controlled simulations are split into a group with cd4doublingtime less than 11 hours (differentially controlled group 1, DC-1) and cd4doublingtime greater than or equal to 11 hours (differentially controlled group 2, DC-2) as this is between the two peaks (S1 Fig). These same two groups are also found if dividing between the two peaks based on maximumCD8Generations at 14.8, phagocytosisThreshold at 9.1, ActivatedCD4IFNSecretion at 26, or ActivatedCD4TNFSecretion at 32. When these groups are separated based on cd4doublingtime, they also exhibit different dynamic behaviors. DC-1 simulations have two major decreases in bacterial count on day 3 and day 5 (Fig 3D), while DC-2 simulations have one smoother decrease (Fig 3E).

DC-1 parameter trends include impaired CD8+ T cell behaviors (lower maximum CD8+ T cell generations, lower activated TB-specific CD8+ T cell fraction, lower TB-specific CD8+ T cell fraction, higher CD8+ T cell deactivation probability), improved CD4+ T cell behaviors (higher CD4+ T cell activated lifespan, higher activated TB-specific CD4+ T cell fraction, higher CD4+ T cell IFN and TNF secretion), and some parameters that don’t fit either of these groupings (lower STAT1 span, lower macrophage maximum life span, higher granuloma fraction of diffusion, higher CD8+ kill probability, slightly lower bacterial external doubling time, lower phagocytosis threshold, higher lifespan for activated macrophages). The distributions of the parameters for these two groups can be seen in S1 Fig. Many of these parameters show inverted relationships between CD4+ and CD8+ T cell behaviors, suggesting that the distinction between these groups is which T cell fraction is dominant in count or behavior. These could represent natural variation between individual or different immune statuses. For example, DC-1 could represent people with exhausted CD8+ T cells due to chronic inflammation, and DC-2 could represent people with reduced CD4+ T cell counts due to HIV infection or iatrogenic immunosuppression.

Overall, our results illustrate that many different bacterial dynamic behaviors can be consistent with the experimental data. These diverse bacterial dynamics are driven by diverse parameter ranges, which could represent either a) different host immune characteristics or b) large degrees of freedom in our simulation parameters due to limited time resolution in the experimental data. In the following sections, these two groups (DC-1 simulations and DC-2 simulations) are explored further, through the comparison of spheroid and traditional setups, to predict possible mechanistic reasons for improved bacterial control in the spheroid simulation compared to the traditional simulations.

2.2 The divergence of bacterial load in spheroid and traditional DC-1 simulations is due to increased proportions of activated macrophage killing in spheroid simulations

DC-1 simulations show an initial divergence in bacterial load between day 3 and day 4 (Fig 4A). Although the bacterial counts continue to diverge from day 4 to 6, we examine simulation outputs at the time of this initial divergence to identify the driving mechanisms of the initial divergence.

Download:

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

Fig 4. a) Bacterial dynamics for DC-1 simulations.

The time range of interest (day 3–4) is outlined. These are the same time courses as shown in Fig 3C. Proportions of bacterial killing due to b) activated macrophages, c) base macrophages, and d) CD8+ T cell killing from day 3 to 4. Comparing total counts of e) activated macrophages, f) STAT1 activated macrophages, g) NF-κB activated macrophages, h) activated T cells, i) activated CD4+ T cells, and j) activated CD8+ T cells between the spheroid and traditional simulations at day 4. ** p< = 1e-2, *** p< = 1e-3.

https://doi.org/10.1371/journal.pcbi.1012266.g004

Bacteria in the simulation can be killed in 3 different ways: killed by resting macrophages, killed by activated macrophages, and killed along with infected macrophages by CD8+ T cells. The proportions of total bacterial killing due to each of these death methods are compared between spheroid and traditional simulations (Fig 4B-4D). During our time period of interest (day 3 to 4), spheroid simulations have a significantly higher proportion of bacterial killing by activated macrophages. This corresponds to an equivalent lower proportion of bacterial killing by resting macrophages in the spheroid simulations compared to traditional simulations. No significant differences in cytotoxic CD8+ T cell killing are seen between the two simulation setups, with minimal cytotoxic killing observed in both spheroid and traditional simulations between day 3 and 4.

A higher proportion of activated macrophage killing in the spheroid simulations prompted a closer look at macrophage activation at day 4. The total number of activated macrophages is significantly higher in the spheroid than in the traditional simulation (Fig 4E). This difference in macrophage activation is due to STAT1 rather than NF-κB, as the spheroid has significantly more STAT1 activated macrophages and significantly fewer NF-κB activated macrophages than the traditional simulation (Fig 4F and 4G). This is constant with the previously-performed uncertainty analysis that showed a correlation between higher IFN-γ secretion and lower bacterial counts [15]. The fact that the spheroid has more STAT1 activated macrophages implies that the spheroid culture should have more IFN-γ, which activates STAT1, and more activated T cells, which produce IFN-γ. However, fewer activated T cells are seen in the spheroid simulation (Fig 4H–4J). When these activated T cells are subdivided into CD4+ and CD8+ T cells, less activation is still seen in both CD4+ and CD8+ T cell populations in the spheroid simulations. These trends in macrophage and T cell activation are consistent from day 4 to 6, with the spheroid simulations having more activated macrophages even with fewer activated T cells (S3 Fig). These results suggest that the improved bacterial control in the spheroid simulation is due to increased macrophage activation, specifically STAT1 activation, that cannot be accounted for by bulk differences in T cell activation.

2.3 In DC-1 simulations, fewer activated T cells are able to activate more macrophages due to spatial distribution in the spheroid simulations

In the previous section, we note a contradiction between day 3 and 4 when bacterial loads in the spheroid and traditional simulations start to diverge, where fewer activated T cells are able to activate more macrophages in the spheroid simulations. Because the difference in STAT1 activation of macrophages is not due to bulk numbers of T cells, we next investigate the simulated spatial layout of macrophages and T cells. To determine how close macrophages are to the nearest T cells, the closest T cell neighbor of each macrophage is found. A K-D tree is built using the locations of T cells [37]. This K-D tree is then queried using each macrophage location to find the nearest T cell. These distances are averaged over all macrophages in each simulation. The distribution of these means is compared for spheroid and traditional simulations, representing the average distance from macrophage to T cell (Fig 5). This process is repeated, constructing K-D trees with different subsets of T cells (i.e., activated, activated CD4+, activated CD8+) to find the average distances between macrophages and these subsets. Macrophages in spheroid simulations are farther away from T cells in general (Fig 5A), but are closer to IFN-γ producing activated T cells (Fig 5B). This suggests that although there are fewer activated T cells in the spheroid simulations, the macrophages are more closely colocalized to IFN-γ producing T cells leading to more macrophage activation. This trend holds true for both activated CD4+ and CD8+ T cells (Fig 5C and 5D), suggesting they both contribute to macrophage activation. These trends are consistent across days 3 to 6 (S4 Fig). Thus, our results indicate that increased colocalization of macrophages with activated CD4+ and CD8+ T cells in the spheroid simulations is associated with more macrophage activation and more bacterial killing by activated macrophages. This colocalization can be visualized in a representative spheroid simulation at day 4 (Fig 5E and 5F). The activated T cells are primarily located at the edge of the macrophage core. This proximity leads to more macrophage activation and more bacteria killing. Our simulations therefore suggest that for DC-1 simulations, the spatial layout in spheroids allows for better bacterial control.

Download:

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

Fig 5.

The mean distance from macrophages to nearest a) T cells, b) activated T cells, c) activated CD4+ T cells, and activated CD8+ T cells. Each dot represents an average of all cells in a single simulation. The distribution of the mean distances across the set of simulations is compared between spheroid and traditional at day 4. e) A slice through the center of a representative spheroid simulation from the DC-1 simulations at day 4 colored by cell type. f) The same simulation with macrophages in blue, activated T cells enlarged and in yellow, and all other cells in grey. ** p< = 1e-2, **** p< = 1e-4.

https://doi.org/10.1371/journal.pcbi.1012266.g005

2.4 In DC-2 simulations, spheroid simulations have higher proportions of bacterial death due to higher CD8+ T cell activation, in addition to activated macrophage killing

DC-2 simulations do not have same the initial decrease in bacterial load at day 3 that we observe in DC-1 simulations. Instead, DC-2 simulations have a shallower slope downwards starting at day 3 and continuing for the rest of the simulation (Fig 6A). As the spheroid and traditional simulation initially start to diverge at day 3, the proportions of bacterial killing from day 3 to 4 are analyzed (Fig 6B-6D). Similar to DC-1 simulations, more bacterial killing is found to be due to activated macrophages and less killing due to base macrophages in the spheroid simulation compared to traditional simulations. However, in contrast to DC-1 simulations, DC-2 simulations also show an increased proportion of killing due to CD8+ T cell killing. DC-2 parameter trends include higher TB-specific CD8+ T cell fraction and CD8+ T cell deactivation probability, which are both correlated with increased bacteria killed by CD8+ T cells in our previous uncertainty analysis [15]. As in DC-1 simulations, more activated macrophages are seen in the spheroid simulation with STAT1 as the limiting factor. However, in this case, we do see more activated T cells in spheroid simulations (Fig 6H). When subdivided into CD4+ and CD8+ T cells, we see that higher T cell activation in spheroids is primarily due to higher CD8+ T cell activation (Fig 6I and 6J). Thus, in DC-2 simulations, the bulk difference in T cell activation can account for the increase in macrophage activation and activated macrophage killing, and the bulk difference in CD8+ T cell activation can account for the increase in CD8+ T cell killing. However, it remains unclear why more CD8+ T cells are activated in the spheroid simulations than in the traditional DC-2 simulations.

Download:

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

Fig 6.

a) Bacterial count over time for DC-2 simulations. The initial deviation of spheroid and traditional simulations occurs during day 3 to 4 (black rectangle). These are the same time courses as shown in Fig 3D. Proportions of bacterial killing due to b) activated macrophages, c) base macrophages, and d) CD8+ T cell killing from day 3 to 4. Comparing total counts of e) activated macrophages, f) STAT1 activated macrophages, g) NF-κB activated macrophages, h) activated T cells, i) activated CD4+ T cells, and j) activated CD8+ T cells between the spheroid and traditional simulations at day 4. *** p< = 1e-3.

https://doi.org/10.1371/journal.pcbi.1012266.g006

2.5 Increased activated CD8+ T cells in DC-2 spheroid simulations is due to closer proximity to activating macrophages

The higher proportion of bacteria killed by CD8+ T cells in DC-2 spheroid simulations corresponds to higher counts of activated CD8+ T cells (Fig 6). TB-specific (TBS) CD8+ T cells become activated when they directly interact with a macrophage that has interacted with bacteria and that is also STAT1 activated. As stated previously, STAT1 activation is higher in the spheroid simulation from day 3 onward. Therefore, one would expect that more STAT1 activated macrophages would be able to activate more TB-specific CD8+ T cells in the spheroid simulations compared to the traditional simulations. The relative trend of macrophages that have interacted with bacteria is not consistent over time though; spheroid simulations have fewer on day 2, more on day 3, and fewer again by day 5 (S5 Fig). As before, to assess the likelihood of TB-specific CD8+ T cells coming into contact with Mtb-interacted macrophages, we explored the spatial layout of the cells. To determine how close TB-specific CD8+ T cells are to the nearest macrophage, the closest macrophage neighbor of each TB-specific CD8+ T cell is found. A K-D tree is constructed using all of the macrophage or STAT1 activated macrophage locations [37]. This tree is used to find the closest macrophage to every TB-specific CD8+ T cell. These distances are averaged over all TB-specific CD8+ T cells in each simulation. The distribution of these means is compared for spheroid and traditional simulations (Fig 7). While TB-specific CD8+ T cells are on average further from macrophages in the spheroid simulations (Fig 7A), they are closer to the STAT1 activated macrophages (Fig 7B). This proximity can be visualized in a representative spheroid simulation at day 4 (Fig 7C and 7D). The TB-specific CD8+ T cells are localized in the cuff around the edges of the spheroid, and the STAT1 macrophages are along the edge of the macrophage core. This proximity leads to more TB-specific CD8+ activation and more cytotoxic killing. The closer colocalization of TB-specific CD8+ T cells to STAT1 activated macrophages in spheroids is consistent across the time course of the simulation (S6 Fig). These results together suggest that higher levels of STAT1 activation along with the closer colocalization of STAT1 activated macrophages with TB-specific CD8+ T cells lead to higher CD8+ T cell activation and more cytotoxic killing in the DC-2 spheroid simulations.

Download:

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

Fig 7.

Mean distance between TBS CD8+ T cells and (a) macrophages or (b) STAT1 activated macrophages. Each dot represents the mean distance of all cells for a single simulation. c) A slice through the center of a representative spheroid simulation from the DC-2 simulations at day 4 colored by cell type. d) The same simulation with CD8+ T cells in blue, STAT1 activated macrophages in orange, and all other cells in grey. * p< = 0.05, **** p< = 1e-4.

https://doi.org/10.1371/journal.pcbi.1012266.g007

2.6 DC-1 simulations have a CD4+ T cell dominant response, while DC-2 simulations have a mixed/CD8+ T cell dominant response

In DC-1 simulations, CD4+ and CD8+ T cells both have lower bulk levels of activation, but closer proximity to macrophages. This could suggest that both subsets of T cells play a role in macrophage activation. To further dissect the role that IFN-γ derived from CD4+ T cells and CD8+ T cells plays in macrophage activation, we perform virtual knockouts (KOs) of CD4+ and CD8+ IFN-γ secretion independently and together as control. Macrophage activation is not possible in the double KO, leading to higher bacterial loads and confirming the role of macrophage activation in section 3.2. CD4+ T cell activation should be unaffected as it is upstream of the mechanism being knocked out. All four KOs have similar levels of CD4+ activation to the original simulations (S7 Fig). Macrophage activation is downstream and directly impacted by the amount of IFN-γ secreted. CD8+ T cell activation is indirectly downstream of IFN-γ secretion, as STAT1 activated macrophages activate CD8+ T cells. In DC-1 simulations, the double KO performs similarly to the KO of IFN-γ secretion from CD4+ T cells in terms of bacterial control (Fig 8A). Both of these cases have less macrophage and CD8+ T cell activation (Fig 8D and 8G). This is in contrast to the CD8+ IFN-γ KO which has similar levels of macrophage and CD8+ T cell activation to the original scenario, despite having worse bacterial control. Taken together this suggests CD4+ T cells play a larger role than CD8+ T cells in activating both macrophages and CD8+ T cells, at least in DC-1 simulations. These results are consistent for the spheroid and traditional simulations (S7 Fig). Note that fold-changes of virtual knockout outputs relative to control outputs are not the same in spheroid and traditional simulations as there is an upper limit of bacterial growth. However, these relationships between virtual knockout outputs and control outputs trend in the same direction.

Download:

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

Fig 8. Simulations were run with IFN-γ secretion virtual knockouts from CD4+ T cells, CD8+ T cells, or both.

A simulation knocking out CD8+ T cell killing of infected macrophages was also performed. The original simulation and the 4 knockout scenarios were compared by looking at a-c) bacterial counts at day 6, d-f) activated macrophage counts at day 4, and g-i) activated CD8+ T cell counts at day 4. These outputs are shown for the DC-1 spheroid simulations (n = 118), DC-2 spheroid simulations(n = 71), and DC-2 traditional simulations(n = 71).

https://doi.org/10.1371/journal.pcbi.1012266.g008

Similar KOs were performed for DC-2 simulations for completeness’ sake, with the expectation that CD8+ T cells would be more important based on bulk cell counts. DC-2 simulations show similar bacterial counts between the double KO and the CD8+ T cell IFN-γ secretion KO (Fig 8B). Again, these two cases have significantly less macrophage and CD8+ T cell activation than the default scenario (Fig 8E and 8H). Thus, IFN-γ secreted from CD8+ T cells is the major activator of macrophages and, indirectly, CD8+ T cells in DC-2 simulations. This suggests some autoactivation/feedback loop of CD8+ T cells, where the IFN-γ they secrete overall leads to more CD8+ T cell activation. Local IFN-γ, both autocrine and paracrine, has been shown to enhance CD8+ T cell cytotoxic activity, motility, and survival in the context of viral infection and graft rejection [38,39].

In the DC-2 traditional simulations, CD4+ and CD8+ IFN-γ KO both have higher counts of bacteria (Fig 8C), suggesting that CD4+ T cells play a more important role in the traditional setup than the spheroid, even though the DC-2 traditional simulations have the same parameter values as the DC-2 spheroid simulations. This indicates that, even while controlling for all parameters, changing the structure/movement not only changes bacterial control but also the mechanisms responsible for that control.

CD8+ T cells contribute to bacterial control both directly through cytotoxic killing and indirectly via IFN-γ production leading to macrophage activation. In DC-2 simulations, increases in both of these mechanisms are seen. KOs of CD8+ IFN-γ production and CD8+ cytotoxic killing were performed independently to see which contributes more to bacterial control. In DC-2 simulations, IFN-γ secretion from CD8+ T cells is more important in bacterial control than cytotoxic killing (Fig 8C). Some portion of macrophage and CD8+ T cell activation is dependent on this IFN-γ secretion (Fig 8F and 8I).

Taken together, our results predict a CD4+ T cell dominant response in DC-1 simulations and a CD8+ T cell dominant/mixed response in DC-2 simulations (Table 2). Note that DC-1 simulations are defined by lower CD4+ T cell doubling times, which are associated with increased CD4+ T cell secretions and lower CD8+ T cell generations due to the bimodal parameter distributions. CD8+ T cells are then compensating for reduced CD4+ T cell proliferation and secretion in DC-2 simulations. Not only is this seen in the responses to the KOs, but also in the absolute numbers of these cell types. Activated CD4+ T cells are present in larger numbers in DC-1 simulations (Fig 4I and Fig 6I), and the number of activated CD8+ T cells is larger in DC-2 simulations (Fig 4J nd Fig 6J). Our simulation therefore predicts multiple potential mechanisms of bacterial control between spheroid and traditional cultures: macrophage dominant in similarly controlled simulations, CD4+ dominant in DC-1 simulations, and mixed/CD8+ dominant in DC-2 simulations. This heterogeneity could be an indication of possible biological responses or could inform future data collection (e.g., CD4+ and CD8+ T cell counts at day 4) to further narrow parameter ranges and eliminate non-biological responses.

Download:

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

Table 2. Bacterial count relative to control for DC-1 and -2 spheroid and traditional virtual knockouts and the predicted dominant T cell fraction in these bacterial responses as based on the virtual knockout results.

https://doi.org/10.1371/journal.pcbi.1012266.t002

2.7 The initial structure and movement rules work synergistically to reduce bacterial load

We next aimed to answer whether the proximity of T cells and macrophages is predicted to be due to the initial structure of cells (i.e., layered spheroid vs randomly distributed traditional cultures) or due to the movement rules (i.e., free movement in 3D vs gravity-limited movement at the bottom of the simulation space). To separate these two factors, the movement rules for the simulations are swapped. The spheroid is simulated with gravity-limited movement rules (spheroid gravity), and the traditional setup is simulated with 3D movement rules (traditional floating). Surprisingly, both of these sets of simulations perform more poorly than the original setups in controlling bacteria (Fig 9A). Results are shown for DC-1 simulations only, but the DC-2 simulations show similar trends (S8 Fig). This implies that 3D or gravity-limited movement rules in and of themselves do not dictate the level of bacteria killing, rather they work along with the structure to reduce the bacterial load. To identify the mechanisms behind these impacts of cell movement and culture structure, the proportions of bacterial killing relative to the original sets are explored. The spheroid simulation with gravity-limited movement rules has almost no bacteria killed via activated macrophages and CD8+ cytotoxic cells when compared to the original spheroid (Fig 9B). This corresponds to a decrease in activation of macrophages, CD4+ and CD8+ T cells (Fig 9E-9G). The traditional simulations with floating have a smaller decrease in activated killing, but a larger decrease in CD8+ T cell killing (Fig 9B-9D) when compared to the traditional simulations. This corresponds with a small decrease in CD4+ T cell activation and a large decrease in CD8+ T cell activation (Fig 9F-9G).

Download:

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

Fig 9. The movement rules were swapped for the spheroid and traditional simulations, creating spheroid gravity and traditional floating.

a) Bacterial counts at day 6 in spheroid, spheroid gravity, traditional, and traditional floating scenarios. Proportions of bacterial killing due to b) activated macrophages, c) base macrophages, and d) CD8+ T cells from day 3 to 4. Total counts of e) activated macrophages, f) activated CD4+ T cells, and g) CD8+ T cells on day 4. h) Mean distance from macrophages to activated T cells on day 4. i) Mean distance from CD8+ T cells to STAT1 macrophages on day 4. This analysis is done with DC-1 simulations, the same analysis for DC-2 simulations is in the S8 Fig. * p< = 0.05, ** p< = 1e-2, *** p< = 1e-3.

https://doi.org/10.1371/journal.pcbi.1012266.g009

Again, the spatial distribution of cells is analyzed by looking at mean distances between cells (Fig 9H and 9I). The macrophages are closest to activated T cells in the spheroid, followed by the traditional, floating traditional, and finally the spheroid gravity. This order corresponds to the number of activated macrophages in each simulation (Fig 9E). Macrophages being closer to activated T cells leads to more STAT1 activation, more overall activation, and more activated macrophage killing. Similarly, the distance from CD8+ T cells to STAT1 macrophages is smallest in the spheroid, followed by the traditional, traditional floating, and finally the spheroid with gravity (Fig 9I). Activated CD8+ T cell count in order of highest to lowest is traditional, spheroid, traditional floating, and spheroid gravity. So, proximity of CD8+ T cells to STAT1 macrophages corresponds with higher CD8+ T cell activation and more CD8+ T cell killing.

Traditional floating performs more poorly in terms of bacterial control, since it is hypothesized that the addition of the 3rd dimension as an option for movement effectively reduces the concentration of cells. The traditional floating cells are slightly farther apart than the equivalent cells in the traditional setup (Fig 9H and 9I), which is reflected in the macrophage and CD8+ T cell activation. The spheroid gravity also performs more poorly in terms of bacterial control. The spheroid gravity simulations start with the cells in similar initial conditions. But, as the movement rules are gravity-limited, the cells quickly fall to the bottom of the dish. As the macrophages are added at day 0 and the T cells are added at day 2, the macrophages form a compact dome that the T cells then surround. This conserves the layered structure, but it is forced to be denser. Due to the tight packing of the macrophages in this scenario, the T cells infiltrate less and have less opportunity for interaction (Fig 9H and 9I). This suggests that how the cells colocalize is dependent on both structure and density.

Swapping the movement rules showed that neither layered structure nor 3D movement independently improved bacterial control. Rather these changes predict reduced levels of macrophage and T cell activation, which corresponded to distances to activating cells. So, the layered structure and 3D movement of the cells are hypothesized to work together synergistically to activate immune cells and improve bacterial control.