We model the temporal and spatial dynamics of a solid tumour growing in a pre-existing homeostatic non-cancerous tissue, and responding to therapy with an inhibitor drug delivered via the tissue’s vascular system. With our model we set out to explore the implications of the transient and reversible nature of stroma reactivity in the context of intermittent treatment with the inhibitor drug. We first consider a range of intermittent treatment schedules and determine a regime that displays long-term control of tumour burden. We then characterise the residual disease in relation to the temporal evolution of TME and cancer-stroma crosstalk through the course of treatment. Specifically, we focus on co-localisation and density of activated stroma and cancer cells over time. Finally, exploring the diffusion dynamics of the inhibitor drug delivered via the bloodstream and of the molecular signalling in the TME, we establish a link between treatment outcome and local vessel density. Ultimately, our spatial and temporal analysis sheds light on the complex dynamics behind the development of transient resistance observed in tumours undergoing targeted therapy with inhibitor drugs.
Hybrid discrete-continuum model
We adopt a hybrid discrete-continuum model to describe the dynamics of cells acting as individual agents, coupled with the reaction-diffusion dynamics of drug and concentrations of signalling molecules that mediate EMDR (Fig. 1). The modelling framework was first introduced to model nematode movement and chemotaxis50.
Cancer cell, 
, behaviour is dependent on the local concentration of the proliferation signal, 
, with thresholds for death, hd, and proliferation, hp. Cancer cells provide autocrine promotion of local proliferation signal at rate β. TME comprises of both passive, 
, and reactive stroma. Reactive stroma can be in either an activated, 
, or deactivated, 
, state. A targeted inhibitor drug, 
, depletes proliferation signal at rate δ and is removed from the system through vessel sites 
at rate μ. Local concentration of targeted drug above threshold hr triggers activation of reactive stroma cells adjacent to a cancer cell, in turn providing paracrine promotion of the proliferation signal at rate γ. Activated reactive stroma reverts to a deactivated state if the drug concentration falls below hr.
We introduce the proliferation signal, p(x, t), which represents the local net accumulation of pro-growth versus pro-apoptotic signalling within the tissue10. The local signal intensity, sensed by a cancer cell, will determine its viability and proliferative status, based on two thresholds (proliferation above hp, death below hd, and growth-inhibited for intermediate values of p(x, t)).
Modulation of the proliferation signal is provided by the cancer cells, the inhibitor drug, and a subset of non-cancer cells in the TME that are able to engage in crosstalk with the cancer cells when challenged with drug treatment. In our model we refer to all non-cancer cells in the TME as stroma. We designate as reactive the subset of the stroma population that, through cell-to-cell contact, is able to interact with the cancer cells. The experimental counterpart of these cells are CAFs that display some level of differentiation when drug-naive, and a rescue capability to cancer cells in close proximity upon delivery of the drug31. The remaining stroma is considered passive.
We introduce a generic inhibitor drug, d(x, t), that is delivered through the blood circulatory system, diffuses through the tissue, and targets a key driver mutation in the cancer cells. The inhibitor drug inhibits pro-growth signalling and enhances pro-apoptotic signalling, ultimately reducing the viability of cancer cells. We assume that reactive stroma proximal to cancer is activated by the local drug concentration, namely when d(x, t) is above threshold hr. Although targeted therapies do not directly affect CAFs (being mutation specific), CAF activation reflects therapy-triggered wound/stress response from tumour cells. In the first instance, we adopt the simplest description of tumour-stroma crosstalk requiring contact-mediated activation (although we explore the effect of non-contact-mediated crosstalk in Supplementary Information S.1). Given that diffusion of signalling molecules is physically limited by ECM and uptake by local cancer cells, it is reasonable to assume highest levels of crosstalk at the tumour-stroma interface. Once activated, reactive stroma provides paracrine production of the proliferation signal that can rescue cancer cells by returning them to a viable, proliferative state. While there is no definite evidence of tumour microenvironment renormalisation after the cessation of treatment, clinical studies show a renewed response upon re-administration of TKI therapies51,52,53. Furthermore, in-vivo data points at the normalisation of the TME following the cessation of inhibitor drug delivery as a possible explanation for re-sensitisation, which, in turn, allows further response upon the reintroduction of treatment54. Therefore, activation of reactive stroma in our model is transient and contingent on therapy-triggered recruitment cues from cancer cells.
Tuning intermittent treatment to control tumour burden and reduce total treatment days
We explore alternative treatment scenarios of intermittent treatment (Fig. 2). After detection (t = 0) the tumour will continue to grow unless treatment to reduce the tumour is initiated. The tumour burden increases rapidly as cancer invades the homeostatic tissue (Fig. 2c). Alternatively, if the tumour is treated with the inhibitor drug continuously from detection, we observe at first a reduction in tumour burden as the proliferation signal is brought below the proliferative threshold and the bulk of cancer cells become quiescent and die. However, as treatment continues, there is progressive activation of reactive stroma, eventually resulting in overall growth of the tumour (Fig. 2d). Through paracrine promotion of the proliferation signal, reactive stroma is able to rescue a portion of the cancer cells, returning them to a proliferative state. However, the cumulative effect of stroma activation, under continuous treatment conditions, operates on a longer timescale compared to that of proliferation signal depletion. This results in a delay before regrowth is observed, following the initial response. The resulting tissue is composed of a mass of surviving, proliferative cancer cells infiltrated by activated reactive stroma.
a Tumour burden for t ∈ [0, 240] days under no treatment, continuous treatment, and five intermitted treatment schedules (τT = {10, 30, 50} days and τH = 20 days). Individual realisations are shown, with 30 stochastic simulations conducted for each treatment regime. Reduction in tumour burden is observed as the length of the treatment period τT of intermittent treatment increases. Continuous treatment initially displays a very good response to treatment, followed by EMDR-driven relapse. b–h show spatial distribution and drug concentration at representative time points for a single simulation of the regime of interest. b Day 0, the initial condition for all simulations. c Day 51 of no treatment. d Day 181 of continuous treatment regime. e Day 150 of intermittent treatment (τT = 10 days, τH = 20 days) regime. f Day 100 of intermittent treatment (τT = 30 days, τH = 20 days) regime. g Day 150 of intermittent treatment (τT = 50 days, τH = 20 days) regime. h Day 181 of intermittent treatment (τT = 50 days, τH = 20 days) regime. Animations of the proliferation signal, spatial distribution and drug concentration for each treatment regime are available in Supplementary Information S.5.
The transient and reversible nature of stroma activation, triggered by drug treatment, can be exploited by modulating drug delivery through intermittent treatment. Namely, introducing treatment holidays, where delivery of the inhibitor drug is paused, allows us to control the promoting and rescuing action of reactive stroma. However, reducing overall stroma activation with pauses in drug delivery comes at the expense of tumour burden control. During treatment holidays no additional drug enters the domain while the drug already in the domain diffuses out through the vessels. Eventually, reduced drug concentration leads to deactivation of reactive stroma cells, and hence removal of the local paracrine assistance provided to rescue cancer cells. As a result, those cancer cells that are relying on this paracrine assistance can become quiescent and die. However, cancer cells that survive while local drug concentration decreases sufficiently can re-enter a proliferative state. This can lead to a surge in proliferation over a treatment holiday, albeit hindered by spatial competition from stroma cells that had infiltrated the space freed up by bulk cancer death over the previous treatment delivery. Crucially, the overall outcome of an intermittent regime depends on the prevalence and time scales of all of the processes described above. To investigate this further, we consider intermittent treatment schedules with regular alternating periods of drug delivery and drug holiday. We define τT and τH to be the length of the drug delivery and drug holiday periods, respectively.
An example of the resulting tumour burden for varying lengths of drug delivery period τT, with τH = 20 days, is shown in Fig. 2. We observe that with short treatment periods (τT = 10 days) the drug concentration in the domain is not sufficient to cause significant death in the cancer population (Fig. 2e). For τT = 30 days tumour growth is reduced, however the drug concentration in the domain is not sufficient to control the tumour burden (Fig. 2f). Note that, after approximately 150 days, the outcome is comparable to that of the continuous treatment regime. For treatment period lengths greater than 30 days, we observe significant reduction in tumour burden within the first two treatment periods (Supplementary Fig. S2c) and long-term control of tumour burden at low levels (Fig. 2g, h).
To compare the outcomes for different choices of τT we extend our investigation of these regimes to a longer window of time. We consider the tumour burden and the cumulative days of drug delivery, relative to the continuous treatment case, over 590 days of therapy. These measures allow us to quantitatively consider the trade-off between reduction of tumour burden and duration of pharmaceutical intervention. Under the parameter regime adopted here, intermittent treatment with τT ≥ 30 days results in lower tumour burden compared to continuous treatment (Fig. 3a). Exploring the 40 ≤ τT ≤ 100 days range further, and considering outcomes over multiple replicates, we observe a non-linear correspondence between τT and relative tumour burden (Fig. 3b).
a Cumulative days of drug delivery (measured as the sum of drug delivery days over 590 days of therapy) against relative tumour burden (measured as the sum of total cancer cell count over the t ∈ [0, 590] day window normalised to the continuous treatment case) for ({tau }_{T}=left{10,20,30,40,50,60,70,80,90,100right}) days and τH = 20 days. For each schedule, averages over 30 simulations and 95% confidence intervals are shown. The star indicates reference measures for continuous treatment. As τT is increased, relative tumour burden initially decreases and cumulative days of drug delivery increases, but from τT = 50 days, the relative tumour burden increases. b The inset zooms in on measures for schedules with τT ≥ 40 days. The treatment regime τT = 50 days, τH = 20 days is chosen for further analysis.
Despite the reduced toxicity of most inhibitor drugs, the cumulative drug delivery time would be a crucial factor in the context of clinical decision-making over treatment scheduling for a patient. We must therefore consider the trade-off between minimizing relative tumour burden and limiting cumulative treatment days. These considerations suggest treatment regimes with 40 ≤ τT ≤ 60 provide improved tumour burden while also reducing total number of drug delivery days. A more extensive investigation of treatment regimes (τH, τT) can be found in Supplementary Information S.2.
It is important to note that these quantitative results rely on the specific parameter regime adopted (see Table 1 and Supplementary Information S.3 for details of experimentally-informed parameter calibration). In a clinically relevant scenario, some of the model parameters would have to be calibrated against patient-specific measurements.
Henceforth, we will adopt the τT = 50 days and τH = 20 days as the intermittent treatment schedule of choice to investigate how fluctuating environmental conditions modulate transient stroma activation, crosstalk with cancer, and, ultimately, the resulting residual disease.
Tumour-stroma colocation shapes the emergence of EMDR
With the treatment regime τT = 50 and τH = 20 we observe control of tumour burden but not eradication (Fig. 2). Figure 4a shows the cell counts of cancer and activated stroma cell types along with the mean field drug concentration. Once the dynamics settle around an approximately cyclic pattern we observe that surviving cancer (i.e. residual disease) is located in similar regions over consecutive treatment cycles. An example of spatial configurations of residual disease at corresponding times of the three consecutive treatment cycles is shown in Fig. 4b. It can be seen that the regions in the domain where the cancer persists are the same for each of the three time points considered.
a Timecourse of a single representative simulation of treatment schedule τT = 50 and τH = 20 over 590 days. Drug concentration mean field value; cancer cell populations; and activated stroma cell populations. b Spatial distribution of cells at 245, 315 and 385 days, corresponding to lowest total cancer cell population in treatment cycles away from the initial transient (corresponding time points are indicated in a). c, d Longitudinal occupancy of cancer and stroma, respectively, in the domain for t ∈ [150, 590] days (discarding transient) over 30 simulations. Occupancy is measured as fraction of time a lattice location in the domain is occupied by the cell type of interest. e Distributions of average number of cancer cell neighbours of cancer cells and activated stroma in the activation window over the same 30 simulations in (c, d) (discarding transient) with standard error shown. Here the activation window is the last 60% of the 50 days treatment window. Animations of the spatial distribution, proliferation signal and drug concentrations for treatment regime τT = 50 days, τH = 20 days are available in Supplementary Information S.5.
After discarding the transient window t ∈ [0, 150] days, we quantify the longitudinal occupancy of cancer cells over the remainder of the treatment. This measure allows us to determine regions in the domain where surviving cancer is located once the dynamics become approximately cyclic (Fig. 4c). Comparing these regions to those with high longitudinal occupancy of activated stroma (Fig. 4d), we note that they largely overlap. This co-location points to the fact that activated stroma is driving resistance, that is, in these regions we observe tissue-scale EMDR at play. There is, however, a smaller region with remarkably high longitudinal occupancy of cancer which does not correspond to a region of high stroma activation. We will later analyse and compare these distinct regions.
To investigate local cell-to-cell interactions between cancer and stroma, we consider the activation window: the period of the treatment cycle when activated stroma is present. In the treatment regime considered, this corresponds to the end of each drug delivery period, when the inhibitor drug concentration has reached a level sufficient for stroma activation. Moving our analysis from tissue- to cell-scale dynamics, we characterise the makeup of the neighbourhood of cancer cells surviving during the activation window. The distribution of other cancer cells neighbouring a surviving cancer cell is approximately symmetric, with an average of four cancer neighbours in their Moore neighbourhood (Fig. 4e). This represents a shift to the left when compared to the initial distribution of cancer neighbours of cancer cells grown in the homeostatic, drug-free environment (Supplementary Fig. S5). With fewer cancer neighbours providing autocrine signalling, cancer survival depends on paracrine signalling from the TME. Analysing the distribution of activated stroma neighbours around surviving cancer cells during the activation window, we can see that a small number of activated stromal neighbours is sufficient to provide paracrine protection from the effects of the inhibitor drug (Fig. 4e).
Eradication, survival, and persistence niches
Having observed two different patterns of survival, one driven by co-location of reactive stroma, and the other in the absence of it, we move to fully characterise the TME conditions that allow emergence of resistance. The spatial distribution of residual disease reveals three distinct niches in the domain. Examples of an eradication niche, an EMDR-driven survival niche, and a persistence (non EMDR-driven) niche are shown in Fig. 5a. As the inhibitor drug enters the domain, and concentration builds up, a wave of cancer cell death follows (links to animations are provided in Supplementary Information S.5). Both the survival and eradication niches experience these dynamics of bulk death, whereas the persistence niche does not and cancer cells survive throughout the treatment in a quiescent state. However, in the survival niche small clusters of cancer cells escape the effect of treatment for the duration of the drug delivery window. As the wave of cancer death occurs there is the opportunity for stromal cells to infiltrate this newly accessible space (yet within the constraints of contact inhibition).
a Cell distributions at day 526 from a single representative simulation. Zoomed-in insets are examples of a survival niche (S), an eradication niche (E) and a persistence niche (P). b Vessel density measure, ρ(x), over the domain, for the static vessel distribution ({{{mathcal{V}}}}). Here (hat{alpha }=0.1). Boxes tracing the same regions considered in (a), show higher ρ in the survival niche compared to the eradication niche, and lowest ρ in the persistence niche. c Distributions of different cell types of neighbours to cancer cells, over the t ∈ [0, 590] days window, for the same 30 simulations as Fig. 4. Distributions of average cancer, passive stroma and activated stroma neighbours of cancer cells with standard errors in each niche are shown.
To investigate the dual (promoting and competing) nature of cancer-stroma interactions we compare TME conditions in these niches. Firstly, we characterise cancer cell neighbourhoods and find that over the entire treatment window, cancer cells in the survival niche have fewer cancer neighbours and more passive stroma neighbours, when compared to those in the eradication niche (Fig. 5c). These observations point to reduced autocrine assistance and more spatial competition from passive stroma neighbours, respectively. A similar result is observed when considering spatial competition from reactive stroma (irrespective of activation status, Supplementary Fig. S6). Reduced autocrine signalling and increased spatial competition are features that we would intuitively attribute to an eradication niche, rather than a survival niche. However, when analysing the activated stroma neighbours of cancer cells in both niches, we can clearly see that cancer cells in the survival niche experience higher paracrine promotion over the course of treatment. Therefore, we find that it is the paracrine stimulus to proliferation in the survival niche that can shift the modulation of proliferative signal and enable survival and growth, making up for loss of autocrine promotion and enhanced spatial competition. We note that in the persistence niche the distribution of cancer neighbours (hence the autocrine signalling to the average cancer cell) is approximately similar to the one in the survival niche. However, given the complete absence of activated stroma in the neighbourhood of cancer cells, paracrine promotion does not explain survival in the persistence niche. We will next identify other mechanisms at play that can explain survival in the persistence niche.
Since all niches emerge from homogeneous conditions (i.e. the same initial mass of cancer cells immersed in comparably reactive stromal tissue), we look at the inhibitor drug intermittent delivery to identify the source of homogeneity-breaking. Local build up of the drug concentration can induce both cancer cell death, and activation of reactive stroma proximal to cancer. This suggests that the different outcomes depend on the inhibitor drug concentration which, in turn, is determined by vessel distribution and diffusion dynamics. Where the density of vessels is higher, the local concentration of the inhibitor drug will build up quicker compared to where vessel density is lower.
We hypothesise that cancer cells in regions of high vessel density experience more intense effects of the inhibitor drug but are also more likely to be rescued by activated stroma. Figure 5b shows the vessel density measure, ρ, for the given vessel distribution, showing that higher density correlates with survival, when compared to the density in the region identified as the eradication niche. This is consistent with our hypothesis that cancer survival depends on vessel density. For low vessel density the local targeted drug concentration will not reach a level sufficient to kill cancer cells within the drug delivery period. As the vessel density increases the local targeted drug concentration will become sufficient to cause bulk death of cancer cells until, for even higher vessel densities, the local targeted drug concentration will reach the threshold hr, giving more time for stroma activation and assistance over each cycle of drug delivery. This protective action will unfold over the long timescale of consecutive treatment cycles. However, we expect this to be a non-linear effect. Excessively high vessel density will facilitate higher drug concentrations during drug delivery and lead to more cancer cell deaths on a much shorter timescale. On the flip side, extremely low vessel density will result in insufficient build up of drug concentration, and reduced cell death. Notably, vessel density in the persistence niche is smaller than that in the eradication niche. Therefore, while stroma activation was identified as a key mechanism behind residual disease in the survival niche (Fig. 5c), low vessel density can explain residual disease in the persistence niche.
Vessel density driven trade-off in treatment outcome
To further investigate how vessel density affects the two antithetic processes of stroma activation and cancer death during treatment, we consider an experiment where intermittent treatment is applied to domains of varying vessel densities. These domains are obtained by systematically decreasing the spacing of vessels placed on a regular grid; all details of this setup are discussed in Supplementary Information S.6. At corresponding times across simulations we observe different responses to intermittent treatment. At low vessel densities, the targeted drug enters the domain at fewer locations and hence is not able to sufficiently diffuse and build up throughout the domain, resulting in cancer cell survival. We call this type of treatment failure, poor perfusion failure (PPF). Figure 6 shows instances of PPF for low vessel densities (lower mean field ρ insets colour-coded red and Supplementary Fig. S7 for additional time resolution).
Distribution of cells at day 139 from single representative simulations with increasing vessel density. Vessel sites are determined to reflect a target density across the domain (see Supplementary Information S.6 for details). Increasing mean field ρ values are 0.54 × 10−3, 1.63 × 10−3, 2.18 × 10−3, 2.73 × 10−3, 3.28 × 10−3, 3.82 × 10−3, 4.38 × 10−3 and 4.94 × 10−3. Treatment failure due to poor perfusion of the drug (PPF) is evident for low vessel density, while EMDR drives treatment failure for higher vessel densities.
As vessel density increases, the inhibitor drug enters the domain at more locations and is able to diffuse and build up sufficiently to cause bulk death of cancer cells without causing significant activation of stroma. This results in eradication of cancer cells, indicative of treatment success (centre insets colour-coded purple of Fig. 6 and Supplementary Fig. S7 for additional time resolution).
At greater vessel densities, the inhibitor drug enters the domain at more locations and is able to diffuse and build up quickly in the tissue. Reactive stroma is then much more likely to activate and provide the additional paracrine promotion of the proliferation signal required to rescue cancer cells. This results in EMDR (higher mean field ρ insets colour-coded yellow of Fig. 6 and Supplementary Fig. S7 for additional time resolution).
Remarkably, the persistence and survival niches identified in previous simulations with a realistic irregular vessel distribution, display outcomes suggestive of PPF and EMDR, respectively.
Drug dynamics shape distinct niches
Having observed a transition from PPF to EMDR as the vessel density and/or drug delivery period increases, we further investigate conditions of vessel density and treatment scheduling that can modulate resistance. During one cycle of treatment (one drug delivery period followed by one drug holiday period) the drug concentration builds up in the domain with regions of high local vessel density surpassing the threshold hr quicker than regions with low local vessel density. Figure 7a shows a snapshot of the drug concentration field just after halfway through the drug delivery period. At this point, a significant fraction of the survival niche experiences drug concentrations sufficient for stroma activation (d(x, t) ≥ hr), while only a very small portion of the eradication niche and none of the persistence niche experience this condition. Later in the drug delivery period everywhere in the survival niche is now above the threshold, while a significant fraction of the eradication niche and a very small portion of the persistence niche are now at drug concentrations sufficiently high for stroma to become activated (Fig. 7b). We argue that low overall drug concentrations result in residual disease as they are not sufficient to cause cancer cell death and long exposure to above-threshold concentrations over each delivery cycle results in increased stroma activation, tipping the balance between the death-inducing and activation-promoting action of the drug, in favour of the latter.
Drug concentration, d, for the third treatment cycle (drug delivery + holiday period) with yellow highlight for locations where d ≥ hr. Representative eradication (E), survival (S) and persistence (P) niches are the same as in Fig. 5. a Drug concentration in the middle of the treatment period. The survival niche has a larger fraction of above-threshold locations. The eradication niche has a very low fraction of above-threshold locations and the persistence niche has no locations above the threshold. b Drug concentration near the end of the delivery period. Above-threshold locations now cover the entirety of the survival niche, and a large fraction of the eradication niche. There is a very small fraction of locations above the threshold in the persistence niche.
Since a significant fraction of locations in the survival niche experience drug concentrations above hr for a longer time than the eradication niche, the probability of stroma activation is higher in the survival niche than in the eradication niche. This agrees with the results of longitudinal analysis of activated stroma neighbourhood (Fig. 5c) and activated stroma occupancy (Fig. 4d). On the other hand, the much slower build up of drug concentration in the persistence niche results in local drug concentrations that are not sufficient to kill cancer cells within the drug delivery window. This agrees with results of longitudinal cancer occupancy analysis (Fig. 4c). Insufficient drug concentrations in the persistence niche result in less decay of the proliferation signal and cancer cells do not die within the drug delivery window. Without the added paracrine promotion of the proliferation signal from activated stroma cells, cancer cells in the eradication niche are not able to survive. Conversely, the additional promotion of the proliferation signal provided by the paracrine signalling from the activated stroma in the survival niche enables survival of cancer cells, and ultimately the emergence of EMDR.
Dormancy and sustained proliferation as distinct mechanisms for survival
Lastly, we consider the cumulative effects of consecutive rounds of drug delivery periods in shaping the proliferation signal, and ultimately the TME landscape which determines cell fate (survival or death) locally, and residual disease at the larger tissue scale. To do so we analyse the third treatment cycle of the intermittent treatment schedule (Fig. 8).
Proliferation signal, p, during the third treatment cycle with yellow highlight for locations where p ≥ hp (proliferation window) and black highlight for locations where p < hd (death window). Representative eradication (E), survival (S) and persistence (P) niches are the same as in Figs. 5 and 7. a At the start of the drug delivery period all three niches contain locations where the proliferation signal is above hd, forming a bulk region in the survival and persistence niches, and small sparse clusters in the eradication niche. Regions in the proliferative window are only present in the survival niche. b In the middle of the drug delivery period a small number of locations in the survival niche remain in the proliferative window. The proliferation signal in the eradication and persistence niches is significantly depleted, with no locations in the proliferative window. c By the end of the drug delivery period in the survival niche the region in the proliferative window has increased through cancer proliferation and further stroma activation. The eradication niche is almost entirely in the death window, although limited regions in the proliferation window indicate late activation of stroma. There are no such locations in the persistence niche, which is largely in the quiescent window. d Towards the end of the drug holiday period, in the eradication and persistence niches no locations are in the proliferative window. In the survival niche the region in the proliferative window which appeared during drug delivery expands further, while other locations move from the death to the quiescent window.
At the beginning of this drug delivery period residual disease is present in each of the three niches. In the eradication niche there are small clusters of locations where the proliferation signal is above threshold for cancer cell death hd. This is quite different to the proliferation signal landscape in the survival and persistence niches where the locations where the proliferation signal is above hd form a bulk mass.
When treatment commences the drug concentration quickly builds up in the survival niche depleting the proliferation signal. This results both in death at locations away from vessels, and stroma activation followed by a rebound in proliferation signal levels in regions closer to the vessels. This rebound is sufficient to sustain high proliferation signal levels well into the holiday period, despite stroma deactivation. Local paracrine promotion of the proliferation signal is hence the driver for residual disease in the survival niche.
In the eradication niche the drug concentration builds up slowly over the drug delivery window. This translates into a slower depletion of proliferation signals to levels that can trigger cell death, as well as delayed stroma activation. The results of this slower timescale of stroma activation results in sparse cancer cell survival. Residual disease at the beginning of the holiday window is limited and decreases over the following treatment cycles, eventually wiping out the cancer cell population. Lack of paracrine support to proliferation signal is therefore the cause for successful eradication in this region.
In the persistence niche the drug diffusion dynamics are even slower, resulting in drug concentrations that are not sufficient to effectively deplete the proliferation signal. Here, proliferation signal is solely reliant on autocrine signalling produced by clusters of cancer cells. Signal levels remain low but above hd throughout the treatment cycle, allowing cancer cells to survive the treatment in a quiescent state, and build up some reservoir of proliferation signal over the holiday window. Treatment escape by dormancy is therefore the driver for residual disease in the persistence niche.