Quantifying the Dynamics of Field Cancerization in Tobacco-Related Head and Neck Cancer: a Multiscale Modeling Approach Marc D
Total Page:16
File Type:pdf, Size:1020Kb
Published OnlineFirst October 20, 2016; DOI: 10.1158/0008-5472.CAN-16-1054 Cancer Integrated Systems and Technologies: Mathematical Oncology Research Quantifying the Dynamics of Field Cancerization in Tobacco-Related Head and Neck Cancer: A Multiscale Modeling Approach Marc D. Ryser1, Walter T. Lee2,3, Neal E. Ready4, Kevin Z. Leder5, and Jasmine Foo6 Abstract High rates of local recurrence in tobacco-related head and dence of the local field size on age at diagnosis, with a doubling neck squamous cell carcinoma (HNSCC) are commonly attrib- of the expected field diameter between ages at diagnosis of 50 uted to unresected fields of precancerous tissue. Because they and 90 years, respectively. Similarly, the probability of harbor- are not easily detectable at the time of surgery without addi- ing multiple, clonally unrelated fieldsatthetimeofdiagnosis tional biopsies, there is a need for noninvasive methods to was found to increase substantially with patient age. On the predict the extent and dynamics of these fields. Here, we basis of these findings, we hypothesized a higher recurrence risk developed a spatial stochastic model of tobacco-related in older than in younger patients when treated by surgery alone; HNSCC at the tissue level and calibrated the model using a we successfully tested this hypothesis using age-stratified out- Bayesian framework and population-level incidence data from come data. Further clinical studies are needed to validate the the Surveillance, Epidemiology, and End Results (SEER) regis- model predictions in a patient-specific setting. This work high- try. Probabilistic model analyses were performed to predict the lights the importance of spatial structure in models of epithelial field geometry at time of diagnosis, and model predictions of carcinogenesis and suggests that patient age at diagnosis may be age-specific recurrence risks were tested against outcome data a critical predictor of the size and multiplicity of precancerous from SEER. The calibrated models predicted a strong depen- lesions. Cancer Res; 76(24); 7078–88. Ó2016 AACR. Introduction Major Findings Head and neck squamous cell carcinoma (HNSCC) arises in the Patient age at diagnosis was found to be a critical predictor epithelial lining of the oral cavity, pharynx, and larynx. The annual of the size and multiplicity of precancerous lesions. This incidence rate of HNSCC is estimated to be around 600,000 new finding challenges the current one-size-fits-all approach to cases worldwide (1), and in the United States alone, the death toll surgical excision margins. is approximately 11,500 cases per year (2). While a subgroup of HNSCC, including the oropharynx, is caused by infection with high-risk types of the human papillomavirus (HPV; ref. 3), the majority of HNSCC are HPV-negative and primarily associated with tobacco use and alcohol consumption (1). Despite a growing number of therapeutic strategies, survival in HPV-negative HNSCC has not improved significantly over the past decades, with a low median survival of approximately 20 months (4). 1Duke University, Department of Mathematics, Durham, North Carolina. 2Division Poor prognosis in tobacco-related head and neck cancers is of Head and Neck Surgery & Communication Sciences, Duke University School of commonly attributed to the development of local recurrences and Medicine, Durham, North Carolina. 3Section of Otolaryngology-Head and Neck 4 metastases after removal of the primary tumor (1). Field cancer- Surgery, Durham VA Medical Center, Durham, North Carolina. Division of fi Medical Oncology, Duke University School of Medicine, Durham, North Carolina. ization, or the presence of premalignant elds surrounding the 5Department of Industrial & Systems Engineering, University of Minnesota, primary tumor, has been shown to drive the high rate of local Minneapolis, Minnesota. 6School of Mathematics, University of Minnesota, Min- recurrence (5–11). In fact, molecular studies have shown that the neapolis, Minnesota. majority of HPV-negative HNSCC develop within local fields of Note: Supplementary data for this article are available at Cancer Research premalignant cells that are clonally related to the resected primary Online (http://cancerres.aacrjournals.org/). cancer (7, 9). These fields can be much larger than the actual fi Corresponding Authors: Jasmine Foo, Department of Mathematics, University carcinoma and are generally dif cult to detect without genomic of Minnesota, 240 Vincent Hall, 206 Church St. SE, Minneapolis, MN 55455. analyses due to their visually normal appearance (12). If such an Phone: 612-625-0131; Fax: 612-626-2017; E-mail: [email protected]; and Marc D. invisible premalignant field extends beyond the surgical margins, Ryser, Department of Mathematics, Duke University, 120 Science Drive, 117 the portion of the field left behind after resection of the primary Physics Building, Durham, NC 27708. Phone: 919-669-2847; Fax: 919-660- tumor increases the risk of subsequent recurrence and contributes 2821; E-mail: [email protected] to poor prognosis (11). The occurrence of precancerous fields was doi: 10.1158/0008-5472.CAN-16-1054 first reported by Slaughter and colleagues (5) and has since been Ó2016 American Association for Cancer Research. documented in most epithelial cancers (10, 13–15). 7078 Cancer Res; 76(24) December 15, 2016 Downloaded from cancerres.aacrjournals.org on September 29, 2021. © 2016 American Association for Cancer Research. Published OnlineFirst October 20, 2016; DOI: 10.1158/0008-5472.CAN-16-1054 Dynamics of Field Cancerization Quick Guide to Equations and Assumptions Model Assumptions We developed and calibrated a spatial stochastic model of tobacco-related head and neck squamous cell carcinoma at the tissue level. The major model assumptions are: * (Epi)genetic events lead to mutated cells with increased fitness advantage, and mutant clones can spread through the basal layer of the affected epithelium prior to onset of invasive cancer. * Because of the time scales of carcinogenesis, only long-lived progenitor cells in the basal layer of the epithelium are relevant; this monolayer of progenitor cells is modeled as a two-dimensional lattice, where each node is occupied by a cell. * Wild-type and mutant progenitor cells undergo evolutionary competition in the basal layer; only nearest-neighbor interactions are considered. * Tissue architecture and homeostasis are maintained (constant population size) until the invasive cancer stage. Key Model Parameters Total population size (N); cellular transition rates from normal to precancer (u1) and from precancer to carcinoma in situ (u2), respectively; relative proliferative advantage of precancer (s1) and carcinoma in situ cells (s2); mean sojourn time from preclinical lesion to clinical diagnosis with cancer (1=c). Key Equations We were interested in quantifying the geometry of the field of precancerous cells surrounding the tumor at time of diagnosis (s3). The key equations are: (i) The survival function (probability that cancer has not been diagnosed by time (t) of the model, Z t l ÀÁ StðÞ¼eÀct þ ceÀct exp g 1=3;t3 À lt þ ct dt; 1=3 0 3 2 where g is the incomplete gamma function, l Nu1s1 and u2s2pc2ðs1Þ=3: (ii) Conditioned on diagnosis occurring at time t, the probability density function of the radius of the precancer field that surrounds the primary tumor, Z t 3 = ; 3 clz 2 zðÞr À c2ðÞt À s gðÞ1 3 s PðR ðÞ¼t rj s ¼ tÞ¼ 1 = ðÞr À c ðÞt À s exp À þ l À s À cðÞt À s ds l 3 stÀr c2 2 1=3 ÀS'ðÞt 0 3 3 3 for all r 2½0; c2t, and zero otherwise, where c2 is the radial expansion speed of the precancer field and z 3=c2: (iii) Conditioned on diagnosis occurring at time t, the probability of harboring two or more clonally unrelated precancer fields is Z ÀðÞcþl t t lce Às3 ÀðÞ2lþc s PðMtðÞ> 1j s3 ¼ tÞ¼1 À 1 À e e ds: S'ðÞt 0 Model Calibration The microscopic tissue-level models were calibrated on the basis of population-level incidence data. Using a computational Bayesian framework, we determined the posterior distributions for the identifiable parameters l, ; and c: On the basis of the posterior distributions, we derived model-based predictions of field quantities at time of diagnosis. To date, it remains difficult to account for the phenomenon of tissue in absence of any information on the extent of the suspected field cancerization in clinical practice. The main reason for this field. To overcome this barrier, we synthesized data and knowl- translational barrier is a poor understanding of the dynamics and edge sources from the tissue, clinical and population scales to geometry of these invisible fields. Indeed, it is impossible to develop a quantitative model of HNSCC carcinogenesis that account for the risk factor of an unresected field of precancerous accounts for spatial features of the precancer field. On the basis www.aacrjournals.org Cancer Res; 76(24) December 15, 2016 7079 Downloaded from cancerres.aacrjournals.org on September 29, 2021. © 2016 American Association for Cancer Research. Published OnlineFirst October 20, 2016; DOI: 10.1158/0008-5472.CAN-16-1054 Ryser et al. of this model, we then sought to identify aspects of standard starts expanding, further hits to the EGFR or TGFb pathways can clinical practice that could be improved by means of patient- lead to moderate dysplasia, CIS, and eventually invasive HNSCC. specific modeling tools. Microscopic model of carcinogenesis To capture the spatial dynamics of the above mechanisms of Materials and Methods carcinogenesis, we developed a stochastic Moran model on a Biological mechanism of carcinogenesis regular two-dimensional lattice (27, 28). Initially, all cells on the To model the carcinogenesis of tobacco-related HNSCC, we lattice are normal progenitor cells (type 0) and proliferate at rate fi rst developed a spatial stochastic model of homeostasis in f0. Cell division is stochastic, and when a progenitor cells divides, stratified squamous epithelia of the head and neck. The homeo- one daughter cell replaces the mother cell, and the other daughter static epithelia of this region undergo periodic bottom–up renew- cell replaces one of the nearest neighbor cells on the lattice, chosen al (16), whereby long-lived progenitor cells in the basal layer of uniformly at random.