This investigation reports the results of a prospective cohort study of mechanistic-based genetic abnormalities evaluated as predictors of EA and demonstrates the modulating effect of NSAIDs on EA risk in patients with BE. We hypothesized that a panel of somatic genetic abnormalities involving TP53, CDKN2A,
and DNA content could improve prediction of progression to EA and that NSAID use may modulate EA risk. In this longitudinal study spanning more than a decade, we showed that a combination of 17p LOH, 9p LOH, and DNA content tetraploidy and aneuploidy provide significant, independent EA risk prediction. NSAID use is associated with reduction of EA risk, and the protective effect was highly significant for patients who have multiple high-risk molecular abnormalities at baseline. These analyses include 34 EA endpoints, which is second only to our previous 15-y report of histology and flow cytometry (42 EAs) and substantially larger than most other longitudinal studies of biomarkers in BE from other centers, which have typically reported 12 or fewer incident cancers [41
]. This prospective study has been conducted in a single center with a high-risk cohort; studies in other centers will be required to determine whether our results can be generalized to other patient populations and to validate the results for clinical application. Our results are consistent, however, with previous longitudinal studies of single biomarkers from other centers, including TP53
abnormalities and flow cytometry [41
]. To our knowledge, no previous studies in patients with BE or any other human premalignant condition have prospectively evaluated the contributions of TP53
gene inactivation (methylation, mutation, and LOH) and DNA content abnormalities in combination with candidate interventions to assess their potential utility as biomarkers for future cancer risk and cancer prevention.
The mechanisms by which TP53
regulate the cell cycle under normal and abnormal conditions have been investigated extensively in elegant molecular studies in vitro and in model organisms. Perturbations of these genes and the pathways in which they act have profound, mechanistic associations with human cancer based upon evidence accumulated in numerous laboratories [1
]. In BE, neoplastic progression is characterized by clonal evolution in which genetic instability generates variants on which natural selection acts, resulting in waves of clonal expansion, generation of new variants, and further selection [76
abnormalities typically arise in clones with CDKN2A
], creating a condition permissive for clonal variants, including tetraploid and aneuploid populations, to survive and expand [79
]. Thus, the abnormalities in this biomarker panel assess viable clones that undergo expansion (CDKN2A)
and survive chromosomal instability (TP53,
DNA content). Assessment of multiple stages of clonal evolution may be the basis for improved risk stratification compared to single biomarkers.
We simultaneously measured DNA content abnormalities (tetraploidy and aneuploidy), inactivation of TP53
(mutation and LOH), and inactivation of CDKN2A
(mutation, methylation, and LOH), all of which have been shown to be mechanistically related to neoplastic progression in BE [28
abnormalities have been shown in numerous studies to be predictive of EA [41
]. In the present study, TP53
mutations were strongly associated with 17p LOH and aneuploidy and were not selected in the multivariate analysis. Selection of LOH over methylation or mutation in the Cox model could occur because LOH is a common manifestation of the chromosomal instability that is characteristic of neoplastic progression in BE [29
]. LOH could also be selected as the “second hit” for inactivating TP53
Alternatively, LOH events often span large chromosomal regions that could include other genes such as HIC1
] and p14ARF
] that may confer additional selective advantages over mutational or methylation events that affect only TP53
Recently, Maley et al. reported that clonal diversity measures derived from evolutionary biology retained significant independent EA risk prediction with 17p (TP53)
LOH and abnormal ploidy, but 9p LOH became nonsignificant when incorporating evolutionary variables [88
]. The nonproportional hazard varying with time that we found for 9p LOH in the present study may reflect the genetic background of the CDKN2A
clone. For example, expansion of a CDKN2A
abnormal clone that is otherwise genetically stable may homogenize the neoplasm, minimizing diversity on which natural selection might act to promote progression [76
]. In contrast, expansion of a CDKN2A
abnormal clone predisposed to genetic instability through either environmental or somatic genetic factors would result in increased diversity that could promote progression.
Although this study had a large number of EA endpoints for published studies of BE, the number of cancers was relatively small compared to studies in breast and colon, for example. Therefore, estimates of RR over time showing the 9p LOH nonproportional hazard over time need to be further investigated in future studies. In addition, RR for EA presented in at intermediate follow-up times should be treated with caution. We do not recommend that clinicians manage patients on these data alone. Within the BE research field, there is a paucity of data concerning progression to EA in patients with different molecular abnormalities. We present the intermediate time points to inform design of future intervention and multicenter studies, while maintaining statistical rigor in our analyses.
Identification of host genetic factors, including inherited, highly penetrant mutations in cancer susceptibility for hereditary breast cancer (BRCA1, BRCA2),
familial polyposis coli (APC),
and those predisposing to hereditary non-polyposis colon cancer, among others, in combination with knowledge of environmental factors, have the potential to reduce cancer morbidity and mortality by early detection and prevention [5
]. In contrast to inherited mutations in relatively uncommon susceptibility genes, less is known concerning temporal progression of somatic genetic abnormalities in more prevalent sporadic premalignant conditions. Genetic progression models have been proposed for many types of cancers based largely on cross-sectional data [11
]. TP53, CDKN2A,
and DNA content (tetraploidy and aneuploidy) abnormalities are among the most common abnormalities in cancers and premalignant conditions affecting multiple organs, including head and neck, lung, breast, bladder, and pancreas, among others [1
]. Few data exist as to their ability to predict future cancer or how these lesions can be modulated by chemoprevention efforts. Advances have been made in cancer risk prediction for patients with oral premalignant lesions in multiple retrospective, longitudinal studies [13
]. Lee et al. combined multiple biomarkers and patient characteristics in Cox regression analysis and found that chromosome polyploidy, together with high p53
expression, LOH, and histology was the best predictor of cancer risk in a 10-y study of 70 patients with 22 cancer outcomes [13
]. Thus, genetic progression models may be a rich source of hypotheses for retrospective longitudinal and prospective biomarker validation studies.
In experimental model systems and observational studies, aspirin and other NSAIDs have been reported to inhibit cyclo-oxygenase 2, increase apoptosis, decrease inflammation, decrease proliferation, and inhibit angiogenesis [4
]. It has recently been shown that the absolute size of aneuploid clones and clones with TP53
lesions is a risk factor for progression to EA [79
]. NSAIDs may act by reducing clone size through increasing apoptosis and decreasing angiogenesis and proliferation. By increasing apoptosis, NSAIDs may decrease the generation of viable clones, thus decreasing diversity and limiting the pool of genetic variants on which natural selection may act. In addition, NSAIDs function to reduce inflammation, which may in turn reduce the mutation rate in evolving clones and decrease the number of cellular variants. The cohort analyzed in the present report is a subset of that described in Vaughan et al. [25
]. Of the molecular cohort described in the present study, 65% were current NSAID users—including use in follow-up—and this proportion is comparable with the 63% regular NSAID users (including use in follow-up) in the total cohort. About half of the users took aspirin for reasons concerning cardiovascular health, and the vast majority of the other users primarily took ibuprofen for pain relief. In a previous paper, we reported a strong protective association for EA with current NSAID use (at baseline or during follow-up), a rather rapid diminution of the association among former users, and no evidence of a stronger association with increasing frequency and/or duration of use. Furthermore, we did not find significant differences in risk of EA or aneuploidy according to type of NSAID. Thus for the present report, our patients were classified simply as NSAID user or nonuser (former or never) based on use at least once per week for ≥6 mo any time during follow-up, regardless of daily frequency, duration, or type. By focusing on the salient results from the previous analyses of a larger dataset regarding NSAIDs, we avoided multiple comparisons and could focus on the relationship of NSAID use with the status of molecular abnormalities.
Our results show a protective association between NSAID use and progression to EA in all participants, and particularly among those with multiple somatic genetic abnormalities. The vast majority of patients in this cohort had gastroesophageal reflux disease and were undergoing therapy, predominately using proton pump–inhibitors, to reduce reflux symptoms. It is unclear as to how the frequency or severity of symptoms may affect NSAID use by participants. However, to our knowledge, symptoms from reflux are not associated with intermediate endpoints or cancer in BE, so it is unlikely that symptoms could explain the association of NSAID use with reduced risk of EA. None of the patients had endoscopically visible concomitant conditions of the stomach, such as ulcers, at the baseline endoscopy that could have conceivably altered NSAID use. Given the experimental evidence provided in model systems and in previous studies, we propose that NSAIDs act within the BE tissue to modulate progression to EA. Our study is early translational research, as defined by the National Cancer Institute Translational Research Working Group (http://www.cancer.gov/aboutnci/trwg/presentations
). These results advance our understanding of the molecular mechanisms of neoplastic progression as well as the mechanisms by which aspirin and other NSAIDs may prevent cancer. As such, these results are consistent with the National Institutes of Health goals of prevention, prediction, and personalized medicine. Several other types of research will be essential for these results to reach the clinic: multicenter randomized trials with a mechanistic focus to determine the effects of aspirin or other NSAIDs on progression to EA and intermediate endpoints (high-risk biomarkers), health services research to evaluate the effect of the interventions as well as the cost effectiveness of biomarkers that reduce frequencies of endoscopy and numbers of biopsies, and establishment of reimbursement mechanisms to support dissemination and adoption (National Cancer Institute's translational continuum; http://www.cancer.gov/trwg/TRWG-definition-and-TR-continuum
Our study builds on a body of observational research, a recent meta-analysis, and supportive data in other cancers, most notably colon cancer where several clinical trials have reported a protective association of aspirin and other NSAIDs, although some findings have been inconsistent [9
]. The present study extends these findings by evaluating the association of NSAID use with somatic genetic events that define progression in BE and demonstrates the benefits of NSAID use in patients at high risk of progressing to EA. Our results are consistent with previous results showing a protective effect on the part of NSAIDs in patients with HGD [25
], and with computer models indicating that high-risk patients are the most likely to benefit from NSAID intervention [117
]. The incidence of EA is sufficiently low in BE that designing statistically rigorous, adequately powered prevention studies with cancer as an endpoint is not feasible for most research cohorts. Thus, knowledge of the interaction of NSAIDs and somatic genetic abnormalities will help define entry criteria for randomized intervention trials with a cancer endpoint. The combination of a somatic genetic biomarker panel that identifies those patients with BE who are at high risk of progression to EA, combined with an inexpensive, widely available, and relatively safe means of preventing neoplastic progression in such high-risk patients, could have significant public health and economic benefits.