FMISO PET is a noninvasive imaging technique that has been used to image tumor hypoxia in HNC for more than 15 years [2, 12,13,14,15]. However, it remains a research radiotracer requiring custom production and an approved IND (Investigative New Drug) and IRB. Glucose metabolism is regulated by many pathways including hypoxia. The use of FDG PET to infer information of hypoxia is an attractive one, which has been investigated in multiple studies with the ultimate goal of obviating the need for hypoxia-specific imaging probes [5, 16].
Several pre-clinical studies suggested a correlation between glucose metabolism, as measured with FDG, and hypoxia. In cell lines studies, Minn and co-workers showed a mean increase of [3H]-FDG uptake of 120% and 46% under anoxic conditions over that measured at a baseline 20% atmospheric oxygen concentration for two head and neck squamous cell carcinoma cell lines, UT-SCC-5 and UT-SCC-20A, respectively [17]. Pre-clinically, using a Dunning prostate tumor model, Pugachev and co-workers showed a positive correlation between FDG uptake and hypoxia defined by pimonidazole staining [18]. Wyss and colleagues also found similar results [19]. However, in the clinical setting, data are limited and discordant, and most studies showing a moderate to weak correlation between the two radiotracers. For example, Zimny et al. showed FDG uptake to be independent of the tumor oxygenation status which was not correlated with the corresponding pO2-polarography measurements [16]. The same group, however, showed a strong correlation between FMISO tumor-to-muscle ratio and the frequency of pO2 readings less than or equal to 5 mmHg [16]. Likewise, in a study that included 12 HNC patients who underwent FDG and FMISO exams prior to radiotherapy, Thorwarth et al. showed an ambiguous correlation between the two radiotracers [14]. More promising results were reported by Rajendran and co-workers, where a mean correlation of 0.62 between FDG and FMISO concentrations was observed in a study that included 26 HNC patients, based on analyzing primary tumor sites [13].
Due to the wide availability of FDG, it would be attractive to assess whether a correlation between the two radiotracers, i.e. FMISO and FDG, exists, and hence whether the FDG may be used as a surrogate for tumor hypoxia. In this study, we have investigated the correlation between tumor hypoxia and glucose metabolism, as imaged by FMISO and FDG PET, respectively, using a voxel-wise analysis as well as global semi-quantitative parameters (SUVmax and SUVmean). We also reported on the general similarity of the two radiotracers’ distributions within the MTV by means of comparing the corresponding activity volume histograms (AVHs). Nineteen primary tumors and 20 metastatic LNs from a total of 20 HNSCC patients were included in the analysis. Primary tumors and LNs that did not show FDG or FMISO uptake were excluded. The primary tumor sites and the metastatic LNs were analyzed independently in case the correlation between FDG and FMISO differed, as was previously reported by Komar et al. [20]. Quantitatively, the correlation between FDG and FMISO uptakes was also investigated using two criteria; first, this was done using the MTV, and then second, using only the HV which was defined by a tumor-to-blood ratio (T/B) threshold greater than 1.2, as was initially proposed by Rajendran and colleagues [5]. As clarified above, this threshold value of 1.2 was an operational definition and was based on the observation that < 5% of voxels in whole-body FMISO PET images corresponding to normal tissues exceeded this value.
Voxel-by-voxel FDG-FMISO analysis showed a strong correlation R (> 0.7) in ~ 68% of the primary tumors and in 40% of the LNs. However, these correlation coefficients can be significantly impacted by good correspondence between the less relevant low-activity regions, whereas the potential correlation between the high activity FDG and FMISO regions is more important. Restricting the analysis to the HVs defined by a T/B > 1.2 resulted in a decrease in the correlation coefficient values so that only 39% of the primary tumors and 22% of the LNs resulted in R values > 0.7. Inaccuracies in patient setup and consequent image registration based on mutual information of the CT component of the FMISO1 and FMISO2 PET/CT studies also affect the strength of the correlation [21]. The error in patient setup was previously estimated by Hong et al [22]. in head and neck cancer patients to be in the order of 6.97 mm (i.e., approximately two PET pixels).
Although FDG uptake is related to expression and activity of specific GLUT and glycolytic activity in the tumor, it is also affected by several other factors such as radioresistance, proliferation [23], cell density [24], and hypoxia [25, 26]. There have been multiple attempts to validate the feasibility of FDG-based dose-painting to improve local tumor control by radiation therapy. In cases where the HVs are contained within the high FDG-uptake regions within the TV, the consequence of such a dose-painting strategy would be dose escalation to the HV in addition to other hyperglycolytic regions within the tumor. Our results indicate that the FDG threshold required in order to include the entire HV would necessitate inclusion of more than 90% of the hypermetabolic TV. This encompasses most of the FDG avid volume, and suggests that the hypoxic voxels are dispersed throughout the FDG-avid TV. In another word, our data suggest that it is not possible to inclusively escalate the dose to the HVs within the tumor by just boosting the dose to the hottest sub-regions within the tumor characterized with high FDG SUV. Even though this strategy might be successful for selected patients, it was not found to be universally applicable.
In a previous study in which a cohort of HNSCC, Nehmeh et al. showed that the tumor hypoxia can be spatially dynamic within the target volume, thus suggesting the presence of acute hypoxia [21]. In order to successfully escalate the radiotherapy dose to the hypoxic volume confirmation of its spatial stability within the tumor volume, i.e. chronic rather than acute hypoxia, is a pre-requisite. In another study, Lee et al. investigated the prognostic values of pre- and mid- treatment FMISO PET in a cohort HNSCC undergoing platinum-based chemo-radiation [27]. The authors showed excellent loco-regional control despite evidence of detectable hypoxia in the pre-treatment FMISO PET studies. They also showed that treatment outcome was independent of the hypoxic status of tumors at mid-treatment [27]. In contrast, Nicolay et al. showed a correlation between tumor hypoxia dynamics, as measured by FMISO PET, and, treatment response and outcome in HNSCC patients undergoing chemoradiation [28]. In another study, Widenmann et al. showed the potential of multi-parametric MRI as a surrogate to FMISO PET to longitudinally monitor tumor hypoxia in HNSCC patients undergoing chemoradiation [29]. This can have significant impact on patients management especially to the large availability of MRI, as well as due to its higher spatial resolution compared to PET.
Here, we also investigated the potential utility of FDG for predicting tumor hypoxia. Quantitative comparison between FDG and FMISO uptakes showed, on average, a moderate correlation (P < 0.01) between the two markers’ SUVavg for the primary tumors for both the MTVs, and the HVs (Fig. 2). This is in agreement with the results of Rajendran et al. in a study of 26 HNC primary tumors [13]. Discrepant results, however, were reported by Thorwarth et al., who did not find any correlation between FDG and FMISO uptake in 12 primary HNC tumors [14]. This lack of correlation may be due to the limited range of FDG SUVmax values (mean = 9.53; range 7.84 to 12.07) [14], which may require greater statistical power (i.e. many more lesions) in order to deduce a possible correlation between them. In contrast, the study by Rajendran et al. included patients with a wide range of FDG SUVmax (mean = 10.9; range 2.9 to 25.4) [2]. The latter is comparable to the range in our study (mean = 14.7; range 7.6 to 32.2). In contrast to our findings in primary tumors, we did not detect any correlation between FDG and FMISO SUV in lymph node metastases (Fig. 3). This was also noted by Gagel et al. [4]. The FMISO uptake appeared to be independent of the corresponding FDG uptake (regression slopes ~ 0) in LNs (Fig. 3), perhaps indicating that biologic processes contributing to glucose uptake (e.g., perfusion, hypoxia, proliferation) contribute to variable degree to the FDG signal in primary tumors versus nodal metastases. Similarity in the characteristics of the primary tumors and LNs microenvironments are vital if both are to be managed the same way, otherwise a more complicated lesion-based treatment approach would be necessary. Comparing the FMISO uptake between primary tumors and LNs on a patient-by-patient basis resulted in weak correlations between the corresponding SUVavgs (R = 0.18) when including the MTVs as well as for the SUVmax’s (R = 0.19 and R = 0.14 for the MTVs and HVs, respectively). Similar results have been reported for other noninvasive hypoxia markers including EF5 and fluoroazomycin arabinoside (FAZA) [20]. However, this is in contradiction with a report that showed a correlation between the oxygenation status of the primary tumor and that of the metastatic LNs using biopsy-based methods. A moderate correlation (R ~ 0.56) was, however, observed between the primary tumors and LNs SUVavgs for the HVs. Moreover, a moderate correlation yet statistically insignificant (R = 0.69; P = 0.96) has been shown between the HFs of the primary tumors and LNs (Fig. 4). Becker and co-workers reported similar results on the correlation between the oxygenation status of the primary tumor and that of the metastatic LNs in HNSCC using biopsy-based methods [11].
Both FDG and FMISO SUV’s are known to change as a function of time post-injection. The differences in uptake times post-injection, in both FDG (range 50 min to 180 min) and FMISO (patients cohort-1 range 114 min to 195 min; patients cohort-2 range 135 min to 181 min), may therefore be considered as a major source of uncertainty in this study. To our knowledge, correlation of intratumoral distributions of FDG (FMISO) between different time points post-injection is yet to be investigated. Till then, it would difficult to predict the effect of the ranges of times post-injection considered in this study on the correlations between the two radiotracers. Finally, qualitative comparison of FDG and FMISO intra-tumor distributions by means of the AVH’s showed the latter to be more homogeneously distributed in both primary and LNs (supplemental Fig. 5a, b). Using the area under the AVH as a measure of homogeneity, the differences between those of FDG and FMISO were shown to be statistically significant (paired t test, P = 0.0001 for the primary and P < 0.0001 for the LNs). Nevertheless, the primary tumors and LNs FMISO AVHs appeared to be comparable (paired t test, P = 0.58) (supplemental Fig. 5c). Finally, 7 out of the 20 subjects included in this study were HPV-positive. However, clinical data suggested there is no significant difference in the level, nor distribution of hypoxia in HPV-positive and HPV-negative tumors, as measured by a 15-gene hypoxia classifier [30] and FMISO PET [31].
One major limitation of this study is the small number of subjects included, which was a result of the complexity of the protocol. This could be the reason for weak to moderate correlations that were observed. Analysis of a larger cohort of subjects will be necessary before the findings in this study can be confirmed. Another limitation is the inaccuracy of image registration which can impact the accuracy of the correlation between the FDG and FMISO correlation, mainly in the voxel-wise analysis.
Conclusion
A moderate correlation was observed between FDG and FMISO distributions in the primary HNSCC tumors, but not for the LNs. There was a moderate correlation observed between the individual HFs of the primary tumors and their metastatic LNs. Our findings do not show a universal correlation between FDG and FMISO to exist for all tumors, and therefore FDG PET images cannot be used by themselves as a universal surrogate to identify or predict intra-tumor hypoxia. However, combining FDG PET data with contrast enhanced CT data, to provide supplementary spatial information on tissue perfusion, has been suggested as a path to improve the derivation of hypoxia information based on clinical standard of care scans [32].