# The Evolution of Tumour Composition During Fractionated Radiotherapy: Implications for Outcome

## Abstract

Current protocols for delivering radiotherapy are based primarily on tumour stage and nodal and metastases status, even though it is well known that tumours and their microenvironments are highly heterogeneous. It is well established that the local oxygen tension plays an important role in radiation-induced cell death, with hypoxic tumour regions responding poorly to irradiation. Therefore, to improve radiation response, it is important to understand more fully the spatiotemporal distribution of oxygen within a growing tumour before and during fractionated radiation. To this end, we have extended a spatially resolved mathematical model of tumour growth, first proposed by Greenspan (Stud Appl Math 51:317–340, 1972), to investigate the effects of oxygen heterogeneity on radiation-induced cell death. In more detail, cell death due to radiation at each location in the tumour, as determined by the well-known linear-quadratic model, is assumed also to depend on the local oxygen concentration. The oxygen concentration is governed by a reaction-diffusion equation that is coupled to an integro-differential equation that determines the size of the assumed spherically symmetric tumour. We combine numerical and analytical techniques to investigate radiation response of tumours with different intratumoral oxygen distribution profiles. Model simulations reveal a rapid transient increase in hypoxia upon regrowth of the tumour spheroid post-irradiation. We investigate the response to different radiation fractionation schedules and identify a tumour-specific relationship between inter-fraction time and dose per fraction to achieve cure. The rich dynamics exhibited by the model suggest that spatial heterogeneity may be important for predicting tumour response to radiotherapy for clinical applications.

## Keywords

Fractionated radiotherapy Avascular tumours Moving boundaries Hypoxia## 1 Introduction

Cancer remains one of the main causes of mortality worldwide, with an estimated 8.2 million deaths in 2012 (Torre et al. 2015). Of the millions of people diagnosed with some form of cancer each year, about half will receive radiotherapy as part of their treatment (Fowler 2006). Typically, radiation is delivered to the tumour as a series of small doses, or fractions, administered over a period of several days or weeks in order to limit the toxic side effects to healthy cell populations. The conventional fractionation schedule for most tumours, where a dose of approximately 2 Gy (Gray) is delivered once a day Monday to Friday up to a total of 50–70 Gy, has remained the standard of care for many years (Ahmed et al. 2014; Marcu 2010). While altered schemes, such as hyperfractionation, accelerated fractionation and hypofractionation, have been proposed for certain indications, the selection of an optimal fractionation protocol for a particular tumour would clearly benefit from a more individualised approach. In particular, beyond tumour location and stage, patient-specific factors that may be important in determining response to a particular treatment are currently not considered. It is in this regard that mathematical modelling has the potential to play an important role, identifying key factors that determine treatment outcomes and ultimately identifying patient-specific treatment protocols.

Tumours differ in their responses to irradiation, with radiosensitivity being understood to be an intrinsic property of the cell population that could be estimated from molecular analysis of biopsy samples (Eschrich et al. 2009). Other factors that affect a tissue’s response to radiotherapy include repopulation between radiation doses, tissue re-oxygenation and redistribution of cells within the cell cycle (Thrall 1997). For rapidly proliferating tumours in particular, repopulation between fractions of radiation is important. For tumours containing larger regions of hypoxia, re-oxygenation may play a pivotal role. A more detailed description of the underlying radiobiology can be found in Joiner and van der Kogel (2009).

Due to the difficulties of studying tumours in vivo, such effects are often investigated experimentally using in vitro multicellular tumour spheroids. Tumour spheroids represent a controlled environment of intermediate complexity between 2D culture media and in vivo models in which important insight can be gained about the tumour micro-environment and the effect of therapies on a growing tumour (Carlson et al. 2006; Hirschhaeuser et al. 2010; Mueller-Klieser 1987; Sutherland et al. 1981). However, tumour spheroid growth is avascular with a typical diameter of less than 5 mm, so while useful for calibrating mathematical models, parameters derived from tumour spheroids may need adjustment before they can be used to simulate in vivo tumours.

*d*Gy, of radiation is given by

Many different mathematical models and methodologies have been used to study tumour growth. Among the simplest of these are ordinary differential equation (ODE) models that aim to qualitatively capture observed growth dynamics. Tumours of small volume typically grow exponentially, and their growth rates decelerate as microenvironmental factors such as limited space and nutrient availability become important, which leads ultimately to sigmoidal patterns of growth (Sachs et al. 2001; McAneney and O’Rourke 2007). Popular ODE models include the logistic and Gompertz growth models where the volume-saturation limit is represented by the carrying capacity, *K*.

In many existing models of (avascular) tumour growth, the tumour is assumed to be spatially homogeneous. Such phenomenological models contain few parameters and do not account for complex underlying biological interactions. While such models may provide limited biological insight, the ease with which model parameters may be estimated from limited clinical data makes them attractive for making predictions about radiotherapy response and identifying personalised fractionation protocols in the clinic (Prokopiou et al. 2015).

A variety of more complex, spatially resolved, models have been proposed in order to provide more mechanistic insight, often reflecting the heterogeneous nature of growing tumours. In more detail, as a tumour grows, it will typically develop hypoxia and necrosis in regions where oxygen or nutrient supply is inadequate. Multicellular, avascular tumour spheroids are used as in vitro models to study, in a controlled environment, effects observed in vivo (Carlson et al. 2006; Hirschhaeuser et al. 2010; Mueller-Klieser 1987; Sutherland et al. 1981; Folkman and Hochberg 1973). In a similar manner, it is natural to develop mathematical models for avascular tumour spheroids before considering the more complex case of vascular tumour growth.

One of the simplest spatially resolved models of avascular tumour growth was proposed in Greenspan (1972). In the Greenspan model, under the assumption of spherical symmetry, the outer tumour radius evolves in response to a single diffusible, growth-rate-limiting nutrient, commonly taken to be oxygen. Internal free boundaries, decomposing the tumour spheroid into a central necrotic core, an outer proliferating rim and an intermediate hypoxic annulus, are identified as contours of the oxygen profile across the tumour.

Under certain simplifying assumptions, Greenspan’s model can be reduced to a coupled system of ODEs and algebraic equations. Most other spatially resolved continuum models are formulated as systems of partial differential equations (PDEs). Such approaches include frameworks from multiphase modelling and morphoelasticity. Multiphase models consider a mixture of two or more continuum constituents, or phases, which collectively comprise the tumour environment. Systems of PDEs are obtained by applying mass and momentum balances to each phase and making suitable constitutive assumptions about their mechanical properties and interactions (Breward et al. 2003; Byrne et al. 2003; Preziosi and Tosin 2009). Morphoelastic models of tumour growth provide a theoretical framework within which to study biological tissues for which growth and elasticity are inter-related (Araujo and McElwain 2004).

While incorporating more biological detail, these more sophisticated approaches typically involve more parameters that may be difficult to estimate in practice. In this paper, we present a simple spatially resolved model for tumour spheroid growth and response to radiotherapy in order to investigate the effects of spatial heterogeneity. In Sect. 2, we develop the model, extending Greenspan’s original spatial model for tumour spheroid growth (Greenspan 1972) to include radiation effects. In Sect. 3.1, we numerically solve the model equations and discuss key features of the resulting dynamics. We present further numerical simulations and model analysis in order to understand the tumour dynamics exhibited by the model. In Sect. 3.2, we study tumour regrowth following a single dose of radiation, and in Sect. 3.3, we consider the long-term behaviour of a tumour for different fractionation schedules.

## 2 Model Development

In this section, we introduce a spatial model of avascular tumour growth originally presented in Greenspan (1972). Although many spatially resolved models that account for tumour growth and hypoxia have been proposed, a significant portion of these are based upon Greenspan’s original work, and it remains a simple, yet useful starting point for investigating the effect of hypoxia on tumour responses to radiotherapy. Accordingly, in this paper we extend Greenspan’s model to account for the effects of radiation and arrive at a new model for tumour response to radiotherapy.

### 2.1 Summary of Greenspan’s Original Model

*R*(

*t*) and the oxygen concentration at a distance \(0\le r\le R(t)\) from the tumour centre by

*c*(

*r*,

*t*). We suppose that the oxygen concentration is maintained at a constant level, \(c_\infty \), on \(r=R(t)\). Oxygen diffuses on a much shorter timescale than tumour growth, taking \(\sim 10\) s to diffuse 100 \(\upmu \)m by comparison with a tumour growth period measured in days (Greenspan 1972). As such we assume that the oxygen concentration is in a quasi-steady state. Internal free boundaries at \(r=R_H(t)\) and \(r=R_N(t)\) are defined implicitly and denote contours on which the oxygen concentration attains the threshold values \(c_H\) and \(c_N\), respectively. These interfaces decompose a well-developed tumour into a central necrotic core (\(0< r < R_N\)) where \(c \le c_N\), and an outer, oxygen-rich region (\(R_H<r<R\)) in which \(c_H < c\), these regions being separated by a hypoxic annulus (\(R_N<r<R_H\)) in which \(c_N< c < c_H\) (see Fig. 1 for a schematic).

*c*(

*r*,

*t*),

*R*(

*t*), \(R_H(t)\) and \(R_N(t)\) evolve over time and can be written in dimensionless form as follows:

*H*(.) is the Heaviside function [\(H(x)=1\) for \(x>0\), and \(H(x)=0\) otherwise], and \(\varGamma \), \(\lambda _A\), \(\lambda _N\), \(c_\infty \), \(c_H\) and \(c_N\) are positive constants, with \(\varGamma \), \(\lambda _A\) and \(\lambda _N\) corresponding to rates of oxygen consumption, apoptosis and necrosis, respectively. Equations (2)–(8) may be reduced to a system involving a single ODE for

*R*(

*t*) and algebraic equations for \(R_H(t)\) and \(R_N(t)\). Analysis of the resulting equations and details of the non-dimensionalisation may be found in Greenspan (1972) and Byrne (2012), and for further explanation of the underlying model assumptions, see Appendix A.1. Figure 2 shows how the size and composition of the tumour spheroid evolve for a typical simulation of the Greenspan model using the dimensionless parameter values in Table 1. A table of dimensional parameter values is given in Appendix A.1. Where possible, parameter estimates are taken from the literature; however, we note that some values pertain to estimates used in other modelling scenarios. In this paper, specific parameter combinations serve only to highlight the dynamics of the model and, in particular, the resulting model analysis is independent of the choice of parameter values. The influence of some of the parameters on the growth dynamics and steady state tumour composition is explored in Appendix A.2.

### 2.2 Incorporating Radiotherapy into the Greenspan Model

We adapt Greenspan’s model to account for radiotherapy by assuming that when a dose of radiotherapy is applied, it causes an instantaneous dose-dependent death of viable cancer cells and, thus, a change in the size and composition of the tumour. The efficacy of the radiation and subsequent cell death depends on the local oxygen concentration. As such, we determine the radiation-induced cell death in each tumour region separately. We denote by \(R^{\pm }\), \(R_H^{\pm }\) and \(R_N^{\pm }\) the radii immediately before (−) and after (\(+\)) radiotherapy in the proliferating, hypoxic and necrotic tumour compartments.

#### 2.2.1 Radiation-Induced Cell Death and the Linear-Quadratic Model

*d*), immediately after a dose

*d*of radiation is given by

Dimensionless parameter values for tumour spheroid growth model

Parameter | Symbol | Value | Source |
---|---|---|---|

Oxygen consumption rate | \(\varGamma \) | 1.1051 | Grimes et al. (2014) |

Oxygen concentration at tumour boundary | \(c_{\infty }\) | 1 | Grimes et al. (2014) |

Hypoxic oxygen threshold | \(c_H\) | 0.1 | Grimes et al. (2014) |

Anoxic oxygen threshold | \(c_N\) | 0.008 | Grimes et al. (2014) |

Apoptosis constant | \(\lambda _A\) | 0.32 | Frieboes et al. (2007) |

Necrosis constant | \(\lambda _N\) | 0.0061 | Schaller and Meyer-Hermann (2006) |

We assume further that dead material within the necrotic core is unaffected by radiation and so we impose that \(R_N^+ = R_N^-\).

*d*of radiation gives a total volume change of

- (i)
\(c_N<c(R_N^+)<c_H\)- irradiated tumour is a fully developed tumour spheroid with 3 layers;

- (ii)
\(c_N<c_H<c(R_N^+)\)- irradiated tumour has a necrotic core, but no hypoxic annulus.

#### 2.2.2 Reconciling Pre- and Post-radiation Tumour Growth

In the normal growth regime without radiotherapy, \(R_N\) is defined implicitly by the oxygen concentration [see Eq. (5)]. When the tumour volume is reduced due to radiation, the oxygen concentration at the centre of the tumour increases and the location of the interface on which \(c=c_N\) will shift towards the tumour centre or disappear. As the necrotic core is unaffected by radiation, \(c(R_N^+)>c_N\) and the growth of the tumour spheroid immediately post-radiotherapy does not follow the original dynamics in the absence of treatment.

Equations (2)–(4), (6)–(8) and (13) drive growth while \(c(R_N)>c_N\), until the necrotic radius is such that \(c(R_N)=c_N\), at which time the standard Greenspan model holds and growth is driven by Eqs. (2)–(8).

### 2.3 Summary: Statement of Full Model (Dimensionless)

*c*and \(\frac{\partial c}{\partial r}\) across \(r=R_H, R_N\) are also imposed.

## 3 Investigation of Model Behaviour

We now consider the effects of various treatment protocols on a given tumour under this model. In Sect. 3.1, we solve the model numerically and highlight key features of the simulations for different parameter combinations. We explore these features in more detail in the following sections. In Sect. 3.2, we study tumour regrowth following a single dose of radiation, and in Sect. 3.3 we consider the long-term behaviour of a tumour for different fractionation schedules.

### 3.1 Numerical Simulation Results

For each tumour growth regime, the oxygen profile *c*(*r*, *t*) may be solved analytically. Equations (14)–(23) may then be reduced to an ODE for *R* and two algebraic equations for \(R_H\) and \(R_N\), as described in Appendix A.1. We solve the resulting system of equations numerically using a finite difference scheme implemented in MATLAB.

In Fig. 4a, we present results showing a typical tumour response to a conventional fractionation schedule (2 Gy, 5 days / week) simulated for 6 weeks using the parameter values in Tables 1 and 2. Cell death induced by the first fraction of radiation delivered at \(t=1\) results in the loss of the hypoxic annulus. While \(c(R_N) > c_N\), the necrotic core decays exponentially, as defined by Eq. (13). Figure 5a shows how the oxygen concentration in the necrotic core evolves throughout the first week of the radiotherapy protocol. We see the mismatch between \(c(R_N)\) and \(c_N\) following delivery of the first fraction (at \(t=1\)). As the tumour grows between fractions, oxygen concentration in the necrotic core decreases, while each fraction of radiation and the corresponding volume loss results in re-oxygenation.

Figure 4b shows how the system dynamics change when a tumour with a much lower apoptotic rate (\(\lambda _A\)) is exposed to the standard fractionation protocol. The slower rate of apoptosis results in a larger steady state tumour in the absence of radiotherapy. Similarly, a larger tumour volume is supported throughout treatment and a more gradual volume reduction is observed. In this case, the hypoxic annulus reappears during treatment (as observed during the weekend breaks in the protocol) since the necrotic core decays such that \(c(R_N)=c_H\) before the end of the fractionation protocol. A rapid transient increase in \(R_H\) is observed when the hypoxic annulus reappears. We characterise this behaviour in Appendix A.3 where we show that \(R_H\) increases like \(\sqrt{t}\) on this short timescale.

For the simulation in Fig. 4b, a tumour spheroid comprising a necrotic core and transient hypoxia persists at the end of treatment. The tumour volume appears to be evolving towards a periodic orbit and, as such, the model predicts that continuing the same fractionation schedule indefinitely will not yield significant further volume reduction.

Since the tumour composition changes dynamically throughout treatment, the efficacy of each radiation fraction also varies. The survival fraction changes by about 10% from the start to the end of treatment due to the shrinkage of the necrotic core (Fig. 4b). In this case, a characteristic shape within the curve in Fig. 5b is repeated weekly due to the weekly oscillations in tumour composition, with the spike associated with the rapid reappearance of hypoxia. Such dynamic behaviour clearly depends on tumour-specific parameters and highlights the additional details that can be observed when spatial effects are included in a mathematical model (c.f. difference in tumour composition between Fig. 4a, b).

### 3.2 Single Hit Regrowth

*R*such that \(0<R^2<\frac{6}{\varGamma }(c_{\infty }-c_H)\) and \(R_H=R_N=0\), since when \(R=\sqrt{\frac{6}{\varGamma }(c_{\infty }-c_H)}\), \(c(0)=c_H\) and therefore larger tumours will contain a hypoxic region and/or necrotic core. Solving Eq. (14) subject to boundary conditions (18)–(19), we deduce that the oxygen profile for this tumour composition is given by

*R*(

*t*):

*d*to regrow to its initial (i.e. pre-radiotherapy) size. The tumour radius immediately after irradiation is given by \(\gamma (d)\tilde{R}\), where \(\gamma (d) = \mathrm{exp}(-\frac{1}{3}(\alpha d+\beta d^2))\) is the survival fraction given by the linear-quadratic formula. It is straightforward to show that the time, \(\Delta t\), until the tumour regrows to its original size is given by

*d*. This holds for all \(0<\tilde{R}^2<\frac{6}{\varGamma }(c_{\infty }-c_H)\) if the inequality \(3c_\infty -5\lambda _A+2c_H>0\) is satisfied. Then, Eq. (27) represents an increasing family of curves where the minimum bounding curve is given by Eq. (28). That is, for \(0<\tilde{R_1}<\tilde{R_2}<\sqrt{\frac{6}{\varGamma }(c_{\infty }-c_H)}\) and fixed dose

*d*, \(\Delta t(d;R_1)<\Delta t(d;R_2)\).

*d*with an inter-fraction time less than \(\Delta t\) will result initially in a net shrinkage of the tumour throughout treatment (i.e. the tumour volume is smaller at the delivery of each radiation fraction). Conversely, if we wait longer than \(\Delta t\) to re-irradiate, then the tumour will have grown larger than its original size. Thus, Eq. (27) defines a curve in the \((d,\Delta t)\)-plane for treatment protocols which give rise to periodic behaviour for a given initial tumour radius \(\tilde{R}\), since the cell kill induced by radiotherapy is exactly balanced by the regrowth between fractions. This curve is indicated in Fig. 6. We note that the curve describing protocols that yield periodic behaviour becomes concave for large

*R*, and in particular, as

*R*tends to its steady state value, \(R^*\) say, then for a given dose, \(d>0\), \(\Delta t(d) \rightarrow \infty \).

### 3.3 Periodic Surface in ‘Treatment Space’

In the previous section, we investigated the response of tumours of fixed initial radius \(\tilde{R}\) to different fractionation protocols as specified by a dose, *d*, and inter-fraction time, \(\Delta t\). The \((d,\Delta t)\)-plane can be partitioned into regions in which the tumour volume increases or decreases during treatment. However, we note that the location of these regions depends on the tumour radius at the time of irradiation, \(\tilde{R}\). As such, when considering the behaviour of a tumour throughout an entire course of radiotherapy, it is not a line in the \((d,\Delta t)\)-plane that we are interested in, but a surface in ‘treatment space’, \(\mathbb {T} \subset \mathbb {R}^3\), with components \((R,d,\Delta t)\).

For a protocol delivering *n* fractions of radiation, we can identify the response of a given tumour with a discrete trajectory, \((R(t_i),d_i,\Delta t_i) \in \mathbb {T}\), through treatment space. Note, we consider the trajectory as discrete points pre-irradiation so that \(R(t_i)=R(t_i)^-\) in our previous notation. Here, \(t_i\), \(d_i\) and \(\Delta t_i\) are the time, dose and inter-fraction time, respectively, of the *i*th fraction, for \(i=1,\ldots ,n\). The union of the curves in the \((d,\Delta t)\)-plane determining periodicity defines a surface, S, in \(\mathbb {T}\) (see Fig. 7a). Now, as a tumour progresses through the course of radiotherapy, points on its trajectory that lie above this surface indicate net growth of the tumour by the time of delivery of the next radiation dose, while points below the surface result in a more desirable decrease in tumour volume. Therefore, for a given tumour spheroid, S defines a surface that partitions \(\mathbb {T}\) into treatment protocols that cannot halt tumour progression and those that lead to tumour decay.

With this understanding, more general statements about the outcome of different fractionation protocols can be made. We first consider dosing schedules in which the same dose *d* is administered at constant intervals \(\Delta t\). In this case, if a point in \(\mathbb {T}\) lies above S, then the tumour spheroid will grow until it converges to a point on S, at which time it becomes periodic (Fig. 7b).

Below S, the outcome of radiotherapy depends on the boundary curve of S as \(R \rightarrow 0\) [see Eq. (28)]. If the fractionation schedule lies above this curve in the \((d,\Delta t)\)-plane, then the system will eventually converge to the corresponding point on S. However, for *d* and \(\Delta t\) values below this curve the tumour decays to arbitrarily small volumes (Fig. 7b). We note that Eq. (28) still holds for this analysis even though its derivation requires an initial tumour composed entirely of viable cells. This is the case since, in the model, necrotic material is only created via nutrient starvation and not as a result of irradiation. As such, for ongoing treatment protocols that result in a decreasing tumour volume, \(R \gg R_N\) as \(t \rightarrow \infty \).

We can extend this concept to more general treatment schedules. If there exists \(m \in \mathbb {N}\) such that for all \(i>m\) the point \((d_i,\Delta t_i)\) of the *i*th fraction lies below the curve given by Eq. (28), then treatment is sufficient to drive the tumour volume arbitrarily close to 0. However, using Fig. 4b as an illustrative example, we observe a simulation in which the weekend break of the standard fractionation protocol corresponding to \(\Delta t = 3\) days gives rise to the periodic orbit.

When predicting how a tumour may respond to radiotherapy, two quantities of interest are the *potential volume doubling time*, \(T_\mathrm{pot}\), and the *survival fraction after* 2 Gy *of radiation*, \(S_{2\,\mathrm{Gy}}\), of the tumour. We translate these definitions into the model and identify tumour characteristics for which the model would eventually predict regrowth over the weekend.

*R*, we obtain

## 4 Discussion

In vivo tumours are highly heterogeneous cellular entities characterised by high inter-patient variability. Allied to this, the local efficacy of radiotherapy delivered to the tumour site is affected by a number of variables associated with the tumour’s heterogeneous composition and micro-environment. In particular, the local oxygen concentration can significantly influence radiation-induced cell death, with well-oxygenated regions being shown to exhibit up to threefold greater radiosensitivity than hypoxic tumour populations. With this in mind, in this paper we have presented a simple, spatially resolved model in order to investigate the effects of tumour composition on radiotherapy response. Our model builds on the tumour growth model first proposed by Greenspan (1972). We extend this to incorporate the effects of radiotherapy taking into account spatially varying radiosensitivities.

Numerical simulations and mathematical analysis of the model reveal how the tumour’s growth dynamics and spatial composition change throughout treatment. Heterogeneity within the tumour not only affects the initial response to radiotherapy, but also how this response changes throughout the duration of the treatment protocol (see Fig. 5b). For parameter regimes in which hypoxia re-emerges during treatment, a rapid transient increase in the width of the hypoxic annulus is observed. This behaviour arises naturally from the model and the underlying process driving this phenomenon remains to be elucidated. The model more generally also classifies protocols that may result in tumour progression, a nonzero periodic tumour volume, or tumour decay. We identify a surface in ‘treatment space’ dependent on tumour-specific growth and radiosensitivity parameters and determine that successful protocols correspond to those that remain below this surface throughout treatment. The wide variety of dynamics observed suggests that spatial heterogeneity may be important for simulating tumour response to radiotherapy and, in particular, for making clinical predictions.

The model presented in this paper makes numerous assumptions and simplifications about the underlying biology. For radiation-induced cell death, we take the common approach of modelling this process as an instantaneous effect. However, the linear-quadratic model was established to determine long-term clonogenic survival after radiotherapy. Biologically, cell death after radiation may occur via a number of different mechanisms, with many irradiated cells dying only after attempting mitosis one or more times (Joiner and van der Kogel 2009). Consequently, radiation-induced cell death may not elicit the instantaneous volume reduction modelled here. In future work, we will model this process in more detail in order to describe the short term response to radiotherapy and the corresponding spatial changes in tumour composition more accurately. Additionally, a more detailed model will allow us to relax the restrictions of the spherically symmetric geometry associated with the current model, making it more applicable to modelling vascular tumour growth and in vivo responses to radiotherapy.

Radiation biology has a rich history of integrating theoretical and quantitative approaches to advance biological studies and clinical concepts. From the simple and widely used LQ model emerged the concepts of *biologically equivalent dose* (BED) and *tumour control probability* (TCP) to guide the development of clinical fractionation protocols. Different TCP models with increasing complexity, from simple Poissonian TCP to stochastic birth-death processes, have recently been scrutinised, and mathematical analysis suggests that simple models may be best positioned for clinical utility (Gong et al. 2013). Clinical data are often insufficient to inform mathematical models (Chvetsov et al. 2015). Both simple and more complex mathematical models are facing numerous hurdles in the attempt to integrate them into radiation oncology, and more work is needed to fully harness the potential of mathematical modelling for precision radiation oncology (Rockne and Frankel 2017).

However, for a model to be useful in making patient-specific predictions in the clinic, parameters must be identifiable with respect to the limited clinical data typically available. As such, current research in this area often focusses on phenomenological ODE models with few parameters to be estimated. Typically, these models contain no information about the spatial heterogeneity in tumour composition and radiotherapy response. With this in mind, we also propose a comparison of the developed spatially resolved model with these phenomenological approaches as future work. We aim to identify situations in which the spatially resolved and spatially averaged models agree well, and those in which there is a significant difference. The tumour composition changes observed in model simulations suggests that averaged parameter values in simple, phenomenological models may not sufficiently capture the tumour dynamics during treatment for some tumour compositions and parameter combinations.

While more sophisticated models may be difficult to parametrise in practice, they have the potential to increase biological insight and inform further modelling studies. Spatially resolved models, such as the one presented in this paper and the future work proposed to generalise some of the simplifying assumptions made here, may aid in the development of alternative clinically focussed models which capture more of the key features than existing phenomenological models. More complex models incorporating more biological detail may also be used to generate data for *in silico* testing of ODE models and model selection paradigms; comparing the quality of fit and future predictions of a range of simple models against a known ‘ground-truth’.

## Notes

### Acknowledgements

This work was supported by the Engineering and Physical Sciences Research Council [grant number EP/G037280/1]. TL would also like to thank the Moffitt Cancer Center, where some of this work was undertaken, for their hospitality.

## References

- Ahmed KA, Correa CR, Dilling TJ, Rao NG, Shridhar R, Trotti AM, Wilder RB, Caudell JJ (2014) Altered fractionation schedules in radiation treatment: a review. Semin Oncol 41(6):730–750. https://doi.org/10.1053/j.seminoncol.2014.09.012. http://linkinghub.elsevier.com/retrieve/pii/S0093775414002322
- Alper T, Howard-Flanders P (1956) Role of oxygen in modifying the radiosensitivity of
*E. coli*B. Nature 178:978–979CrossRefGoogle Scholar - Araujo RP, McElwain DLS (2004) A history of the study of solid tumour growth: the contribution of mathematical modelling. Bull Math Biol 66(5):1039–1091. https://doi.org/10.1016/j.bulm.2003.11.002 3843079723MathSciNetCrossRefzbMATHGoogle Scholar
- Breward C, Byrne H, Lewis CE (2003) A multiphase model describing vascular tumour growth. Bull Math Biol 65(4):609–640. https://doi.org/10.1016/S0092-8240(03)00027-2 CrossRefzbMATHGoogle Scholar
- Byrne HM (2012) Continuum models of avascular tumour growth. In: Antoniouk AV, Melnik RVN (eds) Mathematics and life sciences, chap 12. De Gruyter, pp 279–312Google Scholar
- Byrne H, Chaplain M (1998) Necrosis and apoptosis: distinct cell loss mechanisms in a mathematical model of avascular tumour growth. J Theor Med 1:223–235. https://doi.org/10.1080/10273669808833021 CrossRefzbMATHGoogle Scholar
- Byrne HM, King JR, McElwain DLS, Preziosi L (2003) A two-phase model of solid tumour growth. Appl Math Lett 16(4):567–573. https://doi.org/10.1016/S0893-9659(03)00038-7 MathSciNetCrossRefzbMATHGoogle Scholar
- Carlson DJ, Stewart RD, Semenenko VA (2006) Effects of oxygen on intrinsic radiation sensitivity: a test of the relationship between aerobic and hypoxic linear-quadratic (LQ) model parameters. Med phys 33(9):3105–3115. https://doi.org/10.1118/1.2229427 CrossRefGoogle Scholar
- Chvetsov AV, Sandison GA, Schwartz JL, Rengan R (2015) Ill-posed problem and regularization in reconstruction of radiobiological parameters from serial tumor imaging data. Phys Med Biol 60(21):8491–8503. https://doi.org/10.1088/0031-9155/60/21/8491
- Enderling H, Chaplain MAJ, Hahnfeldt P (2010) Quantitative modeling of tumor dynamics and radiotherapy. Acta Biotheor 58(4):341–353. https://doi.org/10.1007/s10441-010-9111-z CrossRefGoogle Scholar
- Eschrich SA, Pramana J, Zhang H, Zhao H, Boulware D, Lee JH, Bloom G, Rocha-Lima C, Kelley S, Calvin DP, Yeatman TJ, Begg AC, Torres-Roca JF (2009) A gene expression model of intrinsic tumor radiosensitivity: prediction of response and prognosis after chemoradiation. Int J Radiat Oncol Biol Phys 75(2):489–496. https://doi.org/10.1016/j.ijrobp.2009.06.014 CrossRefGoogle Scholar
- Folkman J, Hochberg M (1973) Self-regulation of growth in three dimensions. J Exp Med 138(4):745–753. https://doi.org/10.1084/jem.138.4.745 CrossRefGoogle Scholar
- Fowler JF (2006) Development of radiobiology for oncology—a personal view. Phys Med Biol 51(13):R263–R286. https://doi.org/10.1088/0031-9155/51/13/R16 CrossRefGoogle Scholar
- Frieboes HB, Lowengrub JS, Wise S, Zheng X, Macklin P, Bearer EL, Cristini V (2007) Computer simulation of glioma growth and morphology. NeuroImage 37:S59–S70. https://doi.org/10.1016/j.neuroimage.2007.03.008. http://linkinghub.elsevier.com/retrieve/pii/S1053811907001784
- Gong J, Dos Santos MM, Finlay C, Hillen T (2013) Are more complicated tumour control probability models better? Math Med Biol 30:1–19. https://doi.org/10.1093/imammb/dqr023 MathSciNetCrossRefzbMATHGoogle Scholar
- Greenspan H (1972) Models for the growth of a solid tumor by diffusion. Stud Appl Math 51:317–340CrossRefzbMATHGoogle Scholar
- Grimes DR, Kelly C, Bloch K, Partridge M (2014) A method for estimating the oxygen consumption rate in multicellular tumour spheroids. J R Soc Interface 11(92):20131124. https://doi.org/10.1098/rsif.2013.1124. http://rsif.royalsocietypublishing.org/content/11/92/20131124.short
- Hirschhaeuser F, Menne H, Dittfeld C, West J, Mueller-Klieser W, Kunz-Schughart LA (2010) Multicellular tumor spheroids: an underestimated tool is catching up again. J Biotechnol 148(1):3–15. https://doi.org/10.1016/j.jbiotec.2010.01.012. http://linkinghub.elsevier.com/retrieve/pii/S0168165610000398
- Joiner M, van der Kogel A (2009) Basic clinical radiobiology, 4th edn. CRC Press, Boca Raton. https://doi.org/10.1201/b13224 Google Scholar
- Marcu LG (2010) Altered fractionation in radiotherapy: from radiobiological rationale to therapeutic gain. Cancer Treat Rev 36(8):606–614. https://doi.org/10.1016/j.ctrv.2010.04.004. http://linkinghub.elsevier.com/retrieve/pii/S0305737210000824
- McAneney H, O’Rourke SFC (2007) Investigation of various growth mechanisms of solid tumour growth within the linear-quadratic model for radiotherapy. Phys Med Biol 52(4):1039–1054. https://doi.org/10.1088/0031-9155/52/4/012 CrossRefGoogle Scholar
- Mueller-Klieser W (1987) Multicellular spheroids—a review on cellular aggregates in cancer research. J Cancer Res Clin Oncol 113(2):101–122. https://doi.org/10.1007/BF00391431 CrossRefGoogle Scholar
- Muriel VP (2006) The impact on oncology of the interaction of radiation therapy and radiobiology. Clin Transl Oncol 8(2):83–93. https://doi.org/10.1007/s12094-006-0163-0 CrossRefGoogle Scholar
- Nilsson P, Thames HD, Joiner MC (1990) A generalized formulation of the ‘incomplete-repair’ model for cell survival and tissue response to fractionated low dose-rate irradiation. Int J Radiat Biol 57(1):127–142. https://doi.org/10.1080/09553009014550401 CrossRefGoogle Scholar
- O’Rourke SFC, McAneney H, Hillen T (2009) Linear quadratic and tumour control probability modelling in external beam radiotherapy. J Math Biol 58(4–5):799–817. https://doi.org/10.1007/s00285-008-0222-y MathSciNetCrossRefzbMATHGoogle Scholar
- Preziosi L, Tosin A (2009) Multiphase modelling of tumour growth and extracellular matrix interaction: mathematical tools and applications. J Math Biol 58(4–5):625–656. https://doi.org/10.1007/s00285-008-0218-7 MathSciNetCrossRefzbMATHGoogle Scholar
- Prokopiou S, Moros EG, Poleszczuk J, Caudell J, Torres-Roca JF, Latifi K, Lee JK, Myerson R, Harrison LB, Enderling H (2015) A proliferation saturation index to predict radiation response and personalize radiotherapy fractionation. Radiat Oncol 10(1):159. https://doi.org/10.1186/s13014-015-0465-x. http://www.ro-journal.com/content/10/1/159
- Rockne RC, Frankel P (2017) Mathematical modeling in radiation oncology. In: Wong JYC, Schultheiss TE, Radany EH (eds) Advances in radiation oncology, Springer International Publishing, chap 12, pp 255–271Google Scholar
- Sachs RK, Hahnfeld P, Brenner DJ (1997) The link between low-LET dose-response relations and the underlying kinetics of damage production/repair/misrepair. Int J Radiat Bioly 72(4):351–374. https://doi.org/10.1080/095530097143149 CrossRefGoogle Scholar
- Sachs RK, Hlatky LR, Hahnfeldt P (2001) Simple ODE models of tumor growth and anti-angiogenic or radiation treatment. Math Comput Model 33(12–13):1297–1305. https://doi.org/10.1016/S0895-7177(00)00316-2 MathSciNetCrossRefzbMATHGoogle Scholar
- Schaller G, Meyer-Hermann M (2006) Continuum versus discrete model: a comparison for multicellular tumour spheroids. Philos Trans R Soc A Math Phys Eng Sci 364(1843):1443–1464. https://doi.org/10.1098/rsta.2006.1780. http://rsta.royalsocietypublishing.org/cgi/doi/10.1098/rsta.2006.1780
- Sutherland R, Carlsson J, Durand R, Yuhas J (1981) Spheroids in cancer research. Cancer Res 41(July):2980–2984. http://cancerres.aacrjournals.org/content/41/7/2980.abstract
- Thrall DE (1997) Biologic basis of radiation therapy. Vet Clin N Am Small Anim Pract 27(1):21–35CrossRefGoogle Scholar
- Torre LA, Bray F, Siegel RL, Ferlay J, Lortet-tieulent J, Jemal A (2015) Global cancer statistics, 2012. CA Cancer J Clin 65(2):87–108. https://doi.org/10.3322/caac.21262 CrossRefGoogle Scholar
- Withers HR (1999) Radiation biology and treatment options in radiation oncology. Cancer Res 59(7 Suppl):1676s–1684sGoogle Scholar

## Copyright information

**Open Access**This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.