Our aim was to study the independence of a high risk and a HPV gene expression profile for predicting local recurrence, when analyzed in a model with known clinical predictors in advanced HNSCC patients treated with chemoradiotherapy. A gene expression profile designed by Chung et al. [31] was previously validated to predict locoregional recurrence after chemoradiotherapy on a series of 92 advanced HNSCC patients by Pramana et al. [32]. From this series we analyzed 75 patients to test association of clinical factors and gene expression with local control. The main finding of this study was that the two gene expression profiles had an independent effect on local recurrence in a model with clinical factors and were the most important independent factors in a multivariate model, together with tumor site. This implies that they could in the future be a valuable addition to the clinical factors that are currently used for prediction of local recurrence.
In this study, it was not possible to test for HPV presence in DNA and therefore, gene expression was used to identify patients with a HPV-like profile. As shown in studies that used DNA tests for HPV, patients with a HPV positive profile had a better cure rate [24, 25]. Lassen et al. and van den Broek et al. showed that high p16INK4A expression (immunohistochemistry) independently predicted good treatment response and survival in patients with head and neck cancer treated with conventional (chemo-) radiotherapy [23, 36]. In their most recent paper, Lassen et al. showed that p16 positive patients do not seem to react to hypoxic modification during radiotherapy [37]. P16 (CDKN2A) was also one of the genes we analyzed with the Slebos HPV profile. To our knowledge, our study is the first to show that a HPV gene set can predict local recurrence.
We are not aware of any other externally validated gene expression signature predicting local recurrence in head and neck cancer patients treated with (chemo-) radiotherapy. Other authors have searched for profiles able to predict recurrence in head and neck cancer [27, 28, 29]. Ginos et al. studied 41 surgically treated patients, in which they found genes that correlated with recurrent disease. None of those genes correlated with site, grade or stage [28]. Ganly et al. found 2 genes predictive of locoregional recurrence after chemoradiotherapy in 35 patients, using a 277-gene cDNA array [29]. Dumur et al. found 142 genes predictive of locoregional recurrence in 19 patients treated with radiotherapy with or without chemotherapy [27]. The clinical factors they studied (age, gender, stage and location) were not significant in a univariate analysis and therefore no multivariate analysis was performed.
The Chung and HPV profiles are therefore, to date, the only validated signatures for prediction of local recurrence in HNSCC patients. In addition, the present series is the first to be large enough to test independence of validated signatures from clinical factors in a multivariate model. As can be seen in figure 2.2, a combination of site, Chung expression profile and HPV profile, leads to a subgrouping of patients, where the best group has no local recurrences and the worst group has no cures in it. Although the patient numbers were not very high, these kind of subgroups could be very useful to select patients for therapy. The value and robustness of this combination will need to be confirmed in independent studies.
The present study indicates that gene expression signatures can add valuable additional information to current clinical predictors. In future randomized trials, expression profile measurements can thus be useful in indicating which patients benefit most from the treatment being tested, and thus lead to more rationale and effective application of new therapies.
Conclusion
Gene expression profiles can be useful for predicting local control, independent of clinical factors, after chemoradiotherapy in advanced pharynx and oral cavity tumors. Together with tumor site, the Chung high risk signature and HPV profile status were the most important predictors of local control.