A comprehensive literature review was undertaken and the details and focus of the resulting studies are illustrated in Figure 2. The majority (58%) were studies correlating [18F]FMISO PET either with tissue or blood hypoxic biomarkers, or other imaging biomarkers. Figure 3 maps out the studies geographically. The 40 studies included in this review were performed at just 13 institutions, across 8 countries, and 18 of the published studies were from Germany alone.
The studies have been summarised in evidence tables (see Tables 1-8), with some entered twice if they addressed 2 topics (for example repeat imaging and correlation with 18F-fluorodeoxyglucose (18F-FDG)). The results are discussed in detail below, in line with our ‘key questions’.
1) Correlation of [18F]FMISOPET with other biomarkers – is there a surrogate?
[18F]FMISO PET scans are both expensive and time consuming to conduct (as images are acquired 2-4 hours after tracer injection), therefore an easily accessible ‘surrogate’ hypoxia biomarker, which could help select patients who would benefit from [18F]FMISO imaging, would be desirable. Alternative biomarkers include hypoxia associated protein immunohistochemistry (IHC), hypoxic gene expression, oxygen electrode measurement (all tumour based) and serum osteopontin. Since all patients undergo a biopsy and blood tests at diagnosis, it would be ideal if one of these investigations could be applied to select patients for [18F]FMISO PET imaging. The studies correlating [18F]FMISO PET with other hypoxia biomarkers are summarised in Tables 1 and 2.
Studies correlating hypoxia protein immunohistochemistry with [18F]FMISO uptake.
Immunohistochemical markers of hypoxia include hypoxia-inducible factor 1-alpha (HIF1α), carbonic anhydrase 9 (CaIX) and glucose transporter 1 (Glut-1). These are endogenous proteins whose expression is upregulated in hypoxic conditions [19] and can be measured using IHC on tumour biopsy specimens. Table 1 outlines the six studies [6, 11, 20-23] which assessed the relationship between hypoxia on [18F]FMISO PET and IHC markers of hypoxia. Overall, the results are mixed, with three studies [6, 11, 20] reporting a positive correlation between hypoxia IHC markers and hypoxia on [18F]FMISO PET, and two studies [22, 23] concluding that there was no association. Nicolay et al [21] conducted the largest study, assessing the relationship with hypoxia IHC (both CaIX and HIF1α) and [18F]FMISO PET conducted at different time points (week 0, 2 and 5) in 49 patients undergoing chemoradiotherapy. There was no correlation between either HIF1a or CaIX expression and [18F]FMISO hypoxia in treatment-naïve patients. They did however find an association between hypoxia IHC and ‘adverse hypoxia dynamics’ on [18F]FMISO PET, i.e., delayed resolution of hypoxia on PET scans during radiotherapy treatment. A major issue with using tumour IHC is sampling bias. Hypoxia is typically heterogeneously distributed across a tumour, so a single biopsy sample may not be representative of the whole tumour.
It appears that there may be a relationship between HIF1α /CaIX/Glut-1 expression and [18F]FMISO imaging, but current evidence is insufficient to propose a proxy biomarker of [18F]FMISO uptake. This question could be answered more fully by using archival tumour samples from previous [18F]FMISO trials to assess correlation with HIF1α or CaIX. It should be noted, however, that the markers measure different aspects of hypoxia and may not correlate with each other. For example, HIF1α expression represents the transcriptional changes that occur in response to a chronically hypoxic tumour microenvironment whereas [18F]FMISO uptake is a direct indicator of intracellular hypoxia, both acute and chronic.
Studies correlating other hypoxia biomarkers with [18F]FMISO uptake.
The other hypoxia biomarkers which have been studied in relation to [18F]FMISO imaging are oxygen electrode measurements, gene signatures, and plasma hypoxia markers. The details of these studies are summarised in Table 2. Oxygen electrodes allow direct measurement of hypoxia by inserting small needles into tumours to measure the partial pressure of oxygen (pO2). Three studies [24-26] correlated pO2 readings with [18F]FMISO hypoxia (in a total of 58 patients) and all found a strong correlation. These results were useful in that they validated [18F]FMISO PET as a means of detecting hypoxia, but they do not provide a practical representative biomarker of [18F]FMISO hypoxia. Oxygen electrode measurement is an invasive procedure which requires directly accessible tumours, so it cannot be used in a routine clinical setting.
Hypoxic gene signatures [27-29] refer to a collection of genes whose expression is upregulated or downregulated in response to hypoxia. They can be measured from a tumour biopsy specimen and are thought to represent the hypoxic phenotype of the overall tumour. Signatures are able to prognosticate in HNC [30] and also predict the benefit of hypoxia modification therapy in HNC [31]. To date, there is only one published study [7] analysing the relationship between hypoxia gene expression and [18F]FMISO uptake in a cohort of 42 HNC patients treated with radiotherapy. Correlations were assessed at baseline and at different time points during radiotherapy. There was only a weak association between hypoxic gene signatures and [18F]FMISO uptake at baseline (r=0.20) which increased at weeks 1 (r=0.38) and 2 (r=0.43) during radiotherapy.
The final study [32] in Table 2 investigated the association between [18F]FMISO PET imaging and plasma hypoxia markers (osteopontin, vascular endothelial growth factor (VEGF), galectin-3 and circulating tumour growth factor (CTGF)). The most promising result was obtained with serum osteopontin, a protein whose plasma concentration has been shown to increase in conditions of tumour hypoxia [32]. There was a moderate correlation between osteopontin levels and the baseline hypoxic volume (r=0.579), and residual hypoxia on [18F]FMISO PET imaging (p<0.05) during treatment. Of note, osteopontin has been shown to inversely correlate with pO2 in HNC and also to prognosticate and predict benefit from hypoxia modification therapy in HNC [33]. Given that it is easily obtained by a blood test, it could potentially be an ideal ‘screening’ biomarker to select patients who would benefit from [18F]FMISO PET, but further studies are required to validate this concept.
Correlation of [18F]FMISO PET with other imaging modalities
Correlation with 18F-FDG PET
18F-FDG PET is a routine investigation for many newly diagnosed HNC patients and hence it is convenient to assess correlation with [18F]FMISO PET. 18F-FDG PET provides assessment of glycolysis in tissue, which is a process affected by hypoxia [34]. HIF1α (activated in areas of low oxygen) upregulates both glucose transporters (GLUTs) and glycolytic enzymes [35], and therefore it is conceivable that 18F-FDG PET could be a surrogate marker of hypoxia. Table 3 details the studies correlating 18F-FDG and [18F]FMISO PET. The majority of these studies showed a weak to moderate correlation between the two imaging modalities [6, 11, 24, 26, 34, 36-38], one showing no association between 18F-FDG and [18F]FMISO PET [39], and only one a strong association (r=0.81) [40]. Two other studies [41, 42] that demonstrated a strong relationship used ‘second order’ features on 18F-FDG PET; one looked at ‘total lesion glycolysis’ (SUVmean multiplied by metabolic tumour volume on 18F-FDG PET) and found a correlation coefficient of 0.85 with ‘total lesion hypoxia’ (SUVmean multiplied by hypoxic volume) on [18F]FMISO PET. The other study [41], used a radiomics signature from the CT, and found that this, in combination with 18F-FDG PET, improved the ability to predict for hypoxia on [18F]FMISO PET (with an area under the curve (AUC) of 0.83). In contrast, Kroenke et al. [37] used ‘texture analysis’ of the tumour on 18F-FDG PET and found that this did not improve the ability of 18F-FDG PET to predict [18F]FMISO uptake.
Given the many studies conducted, all with varied outcomes, it seems unlikely that 18F-FDG PET can be used to select patients for [18F]FMISO imaging. Furthermore, although some studies were able to demonstrate a correlation between degree of 18F-FDG and [18F]FMISO uptake, there were instances of low [18F]FMISO /high 18F-FDG uptake [34] and vice versa, showing that 18F-FDG and [18F]FMISO PET provide complementary and separate information from each other. From a biological perspective this is not surprising, given that one relates mainly to tissue glucose consumption and the other to tissue hypoxia.
Correlation with MRI
In recent years, attention has turned to multiparametric MRI (mpMRI) and its ability to provide information about tissue perfusion (dynamic contrast enhanced/DCE-MRI), cellularity (diffusion weighted imaging/DWI-MRI), and oxygenation (transverse relaxation time/T2*MRI).
These are all processes central to the development of tumour hypoxia. The six studies which have compared mpMRI with [18F]FMISO PET are summarised in Table 4. Two studies [43, 44] identified a relationship between DCE-MRI and [18F]FMISO uptake, with reduced Ktrans (a measure of perfusion) in the hypoxic volume. Data on ADC are conflicting, with both decreased [45] and increased [46, 47] values being reported in the hypoxic volume. Of these modalities, T2* MRI is the most direct marker of hypoxia, as it measures the concentration of deoxygenated haemoglobin. The study which compared T2* MRI to [18F]FMISO PET [40] did not find a correlation, which again can be explained by the fact that they measure different processes; blood oxygenation versus intracellular oxygenations. MRI has the benefit of higher resolution compared to PET and would be ideal to identify tumour hypoxia for radiotherapy planning, but currently we do not have sufficient evidence to propose it for this role.
2) Is [18F]FMISOimaging repeatable and reproducible?
Studies repeating [18F]FMISOPET at baseline.
Three studies [48-50] repeated [18F]FMISO PET scans within a short time frame (2-3 days) of each other (without interval treatment) to assess repeatability. They are described in Table 5. Okamoto et al [49] demonstrated that [18F]FMISO imaging is highly repeatable and that the maximum uptake location of [18F]FMISO varied by only 4 mm between two repeat scans. In contrast Lin [48] and Nehmeh et al [50] reported similar hypoxic volumes in ~ 50% of patients, but a significant variation in the location of tumour hypoxia in the other half of patients. The proposed explanation is that [18F]FMISO captures both acute and chronic hypoxia [50], and so stable uptake may be representative of chronic hypoxia only. Although patient numbers in these studies were small (7-14), the findings lend caution to the concept of dose escalating to a hypoxic volume generated from a single [18F]FMISO PET study in a patient. This is well displayed in the study by Lin et al [48](a dose planning study), which showed a decrease in the prescribed uniform dose to the hypoxic volume of up to 12 Gy, between two serials scans (which were separated by 3 days), due to instability of the hypoxic region.
Studies repeating [18F]FMISO PET during radiotherapy treatment.
Several studies interrogated hypoxia dynamics during radiotherapy treatment and are displayed in Table 6. All the studies showed that in the majority of patients, reoxygenation occurs and the degree of hypoxia, measured by [18F]FMISO PET, reduces through the course of treatment. Interestingly, although several studies comment on ‘residual’ hypoxia during treatment, only two formally reported on the geographical stability of the hypoxic volume [51, 52]. Bittner et al. [51]found a 72% overlap of the hypoxic volume from week 0 to week 2, suggesting that ‘residual hypoxia’ is an appropriate term. In contrast, Carles et al. [52]looked at spatial variation of hypoxia and found that only 24% of patients had geographically ‘stable’ hypoxia throughout their treatment, and that these patients had a better prognosis in terms of locoregional control. These findings are hugely important, as they undermine the concept of dose painting to the hypoxic volume seen on a single baseline [18F]FMISO PET and suggest that systemic hypoxia modification therapy, or dose escalation to the whole GTV, may be more appropriate to compensate for spatiotemporal changes in hypoxia.
3) Where do locoregional recurrences occur in relation to the initial hypoxic volume?
Despite the wealth of published studies on [18F]FMISO PET in HNC, only four were identified which correlated recurrence patterns to the initial hypoxic volume (see Table 7). From these, one study [53] concluded that recurrences arise from the original hypoxic subvolumes, with a median overlap of 42% between recurrence volume and the initial hypoxic volume. Two studies [9, 54] showed that a significant proportion of recurrences (33-50%) occur outside the pretreatment hypoxic volume. Nishikawa et al. [55]analysed pretreatment [18F]FMISO PET images from 21 patients with nasopharyngeal carcinoma (of whom nine recurred) to generate a risk model. They found that within the imaged tumour region, voxels with [18F]FMISO tumour to muscle ratio (TMR) > 2.42 predicted a recurrence rate of 30% within the same voxel. The AUC for this prediction model was only 0.59 however, and the authors concluded that the predictive value of pretreatment [18F]FMISO PET was insufficient for up-front dose escalation to the regions with high uptake, i.e. the hypoxic volume.
Overall, these findings suggest that although hypoxia is a known cause of radioresistance, there is not enough evidence to suggest that recurrences arise from the hypoxic regions identified on pretreatment imaging, especially when determined from a single scan. It should be noted, however, that the recurrence data stem from a total of only 70 patients, across 4 studies. Further knowledge is therefore required on disease recurrence and its relation to hypoxic volumes.
4) What have we learned so far from dose modification studies?
Given that the hypoxic uptake on [18F]FMISO PET carries important prognostic information [15], trials are underway to determine if radiotherapy can be dose de-escalated for patients with a good prognosis (absence or early resolution of hypoxia) and conversely escalated for patients with hypoxic tumours. So far 3 dose modification trials using [18F]FMISO PET as a biomarker have been published; see Table 8.
Dose de-escalation studies
The first [18F]FMISO de-escalation study was published in 2016 [23] on 33 patients with human papilloma virus (HPV)-positive oropharyngeal cancer. The radiotherapy dose to the metastatic lymph nodes was reduced by 10 Gy to 60 Gy in patients who had resolution of hypoxia at week one of radiotherapy. Ten patients had their radiotherapy dose de-escalated and remained recurrence free at two years. The second de-escalation study [56] was in a cohort of 19 HPV-positive oropharyngeal cancer patients who were treated with resection of the primary tumour and radiotherapy to the nodes followed four months later by a neck dissection. The radiotherapy dose was reduced to 30 Gy in 15 patients who had no hypoxia at either pre- or intra-treatment [18F]FMISO PET imaging. Eleven of the 15 patients had a pathological complete response. As these were pilot studies, neither had a comparative cohort in which patients received radiotherapy dose de-escalation despite hypoxia on PET. Given that the tumours were HPV-positive, it is possible their outcomes would have still been favourable, despite the observed hypoxia on PET.
A larger scale de-escalation study (clinical trial identifier NCT03323463; n=300) at Memorial Sloane Kettering is currently underway using [18F]FMISO PET to select patients to receive a de-escalated dose of radiation (30 Gy) if no hypoxia is observed on pre/intra-treatment imaging. In this study, all patients will receive two cycles of concomitant chemotherapy and surgical resection is no longer mandatory. However, this trial will not determine if it is the absence of hypoxia (on [18F]FMISO PET) that renders the patients suitable for treatment de-escalation as no randomisation to standard treatment vs. de-escalation is planned.
Dose escalation trial
So far one randomized phase II study has been published [57], which looked at dose escalation to the hypoxic volume alone on [18F]FMISO PET. Patients with a hypoxic volume pretreatment were randomised to receive standard chemoRT (70Gy in 35 fractions) or escalation of up to 10% with 77 Gy to the hypoxic volume only. The trial closed prematurely due to slow accrual (53 patients over 8 years). Thirty-nine patients had hypoxic tumours, of whom 19 received dose-escalation. The authors reported a non-significant improvement of 25% in local control for the dose escalation arm. Furthermore, of the patients treated with dose escalation, only a 2% mean elevation of radiotherapy dose was achieved, rather than the planned 10%. This trial highlights the difficulties of carrying out large-scale prospective imaging trials with [18F]FMISO. One of the reasons given for poor recruitment was scanner and tracer availability. In addition, the modest 2% dose escalation is much smaller than the 10% frequently quoted in planning studies and highlights the need to use real life patient data.