Fig 1.
Spectrum and heterogeneity of lymph node (LN) condition during Mtb infection.
(A) Five sections of LNs taken from Mtb-infected nonhuman primates (NHPs), arranged by increasing infection severity. The columns are adjacent sections from the same LN stained with different panels of antibodies. The left-most panels present (top) normal arrangements of cell populations including CD3 + T cells, CD20 + B cells, and CD11c+ myeloid cells (macrophages and dendritic cells); and (bottom) normal vasculature architecture. LNs with increasingly severe disease are shown from left to right, culminating in a LN that is completely effaced by granuloma-associated macrophages (right). (B) LNs from the same animal, or even adjacent segments of a single LN, can have substantially different levels of disease. In this LN, B cells (green), T cells (red), and CD11c + DCs and macrophages (blue) are shown in a non-diseased (left) and effaced (right) segments from the same LN.
Table 1.
Summary of multi-scale model outcome metrics. To predict host fates, we measure multiple outcomes from each LN and, for diseased LNs, each LN granuloma. These include predictions of total bacterial burden (CFU), time-to-sterilization, and effacement.
Fig 2.
Multi-scale development of multiple LN and LN granuloma sub-models.
Our model represents three major scales of study: whole host (lung, LNs and blood) (A), individual LN scale (tissue) (B) and cellular scale (C-D). Immune cell types tracked within each LN and blood, respectively, are listed including macrophages and T cells of different subtypes. Created in BioRender. Krupinsky, K. (2025) https://BioRender.com/h16o401.
Fig 3.
Experimental design for virtual infection: Virtual hosts defined and measured over time using a multi-scale model (MSM).
(A, B) Trajectory of antigen presenting cells (APCs) delineating the difference between virtual hosts representing (A) LTBI and (B) active pulmonary infection. These trajectories, generated from our HostSim model of pulmonary infection [39], capture two major motifs of how APCs are sent from lungs to LNs in response to multiple lung granulomas (see S1 Fig). Details of different compartments for the MSM are in Fig 2. (C) Prior to pulmonary infection, multiple virtual lymph nodes within each uninfected host maintain a stable, steady state population of immune cells. (D) Following a simulated pulmonary infection with Mtb at day 1, individual LNs become activated when APCs carrying Mtb antigen from the lungs enter the LN by day 15 and antigen presentation induces clonal expansion of T cells. (E) We examine how infection within a LN impacts outcomes by inducing infection within two LNs for each host (i.e., seed them with viable Mtb at day 21); LN granuloma formation follows in those LNs (referred to as diseased). (C-E) were created in BioRender. Krupinsky, K. (2025) https://BioRender.com/m68b077.
Fig 4.
Evolution of immune cell population dynamics in activated and diseased cases within Multiple-LNs for 1000 virtual LTBI hosts.
Our model is calibrated to capture key dynamics of Mtb-specific T cells (A, C) and total T cells (B, D) within activated (A, B) and diseased (C, D) cases. Activated hosts have five LNs receiving Mtb activated APCs. Diseased hosts have five activated LNs receiving Mtb activated APCs and LN granulomas forming in LN #1 and #2. For diseased LNs, our model captures the dynamics of LN bacterial load (E) and macrophages (F). We simulated 1000 separate virtual hosts for each case, generating a distinct trajectory for each of their LNs based on their parameterization. Lines in each plot show cell populations from the indicated LN within one host. For LN bacterial load (E) and macrophages (F), lines are colored by bacterial load trajectory: growing large (purple lines), stabilization (teal lines), and sterilization (yellow lines). Flow cytometry data from individual NHP LNs taken at necropsy are represented by black dots from [13]. Note that lines are truncated on virtual host death (see Methods, Section 6).
Fig 5.
Granuloma bacterial loads are driven by a balance of macrophage infection and activation.
Granulomas are pooled from 1000 LTBI hosts. (A) Proportion of 2000 virtual LN granulomas by fate: no bacteria present (sterilized), stable bacterial growth (stable), and uncontrolled bacterial growth at 481 days post lung infection (N = 2000). (B) Summary of sensitivity analysis detailing significant parameters driving total bacterial load. PRCCs are binned into 50-day bins for ease of analysis (see Methods). Shading indicates average PRCC value during a time interval t (given a parameter is at least significant for 30 days in t). White boxes indicate no significant correlation for longer than 30 days in t. A blue color indicates a positive correlation, and red color indicates a negative correlation. Significance alpha = 0.01 after Bonferroni correction. Complete model state descriptions (MR, MI, E4, etc.) can be found in Table 2 in Methods and parameter value description found in Tables A, B, and C in S2 Appendix.
Table 2.
State variable symbolic definitions. This table contains symbolic and plain text names of state variables and their corresponding descriptions. Plain text names are referenced in Figs 4, 5, 8, and 9. All cells are counted in units of average cell numbers per population.
Fig 6.
Time-to-sterilization of LN granulomas is driven by macrophage behavior within LTBI hosts.
Time to sterilization for a simulated LN is defined as the first-time post-LN-infection that a LN contains less than 0.5 total bacteria, or one day beyond the end of the simulation did not sterilize (see Methods). (A) Time to sterilization among 308 diseased LNs from 1000 simulated hosts that were sterilized within the 480-day simulation period. (B) Significant PRCC correlates between functional groups of parameters and output of interest, namely time-to-sterilization (significance with alpha = 0.01 after Bonferroni correction). Our analysis used 388 individual diseased LNs with granulomas from 1000 simulated hosts. Complete model state descriptions (MR, MI, E4, etc.) can be found in Table 2 and parameter values in Tables A, B, and C in S2 Appendix.
Fig 7.
Simulation and experimental comparison: LN model reproduces NHP LN effacement distributions.
(A) Proportions of experimentally obtained NHP LNs with greater (and less) than 50% effacement at day 201. (B) Proportions of simulated LNs at 201 days post-infection from 2000 virtual LN granulomas from 1000 simulated hosts.
Fig 8.
Metrics for tracking infection progression and adaptive immune response.
(A) Simulated FDG avidity of each LN over time beginning post-infection (starting at day 1). (B) Number of effluxing effector T cells for hosts with LTBI (left) and active pulmonary infection (right). Solid lines represent the median numbers of effluxing effector T cells from individual LNs (n = 2000 for diseased and n = 3000 for activated, pooled from 1000 virtual hosts). Shaded regions represent interquartile ranges (IQRs).
Fig 9.
Predicting drivers of FDG avidity within a host with LTBI using sensitivity analysis of simulated FDG avidity.
Left panel contains data from 2000 simulated individual diseased LNs and right panel contains data from 3000 individual simulated activated LNs with no granuloma forming. All simulated LNs are taken from the set of 1000 virtual hosts. Shading indicates correlation between parameter and FDG avidity during time interval t (given a parameter is at least significant for 30 days in t). White boxes indicate that the parameter is not significantly correlated with model outcome at any time points in t. A blue color indicates a positive correlation, and red color indicates a negative correlation. Significance alpha = 0.01 after Bonferroni correction. Complete model state descriptions (MR, MI, E4, etc.) can be found in Table 2, and parameters in Tables A, B, and C in S2 Appendix.
Fig 10.
Predicting drivers of Mtb-specific T-cell efflux: comparing LTBI and active hosts using sensitivity analysis of Mtb-specific T-cell efflux.
Left panels contain data from 2000 individual simulated diseased LNs and right panels contain data from 3000 individual simulated uninfected/activated LNs. All simulated LNs are from the same set of 1000 virtual hosts. Shading indicates correlation between parameter and FDG avidity during time interval t (given a parameter is at least significant for 30 days in t). White boxes indicate not significantly correlated at any time points in t. A blue color indicates a positive correlation, and red color indicates a negative correlation. Significance alpha = 0.01 after Bonferroni correction. Complete model state descriptions (MR, MI, E4, etc.) can be found in Table 2 and parameters in Tables A, B, and C in S2 Appendix.