PET of Hypoxia: Current and Future Perspectives

ANALYSIS OF HYPOXIA PET IMAGES

Analysis of the pixel variation within normal tissues in these studies has led to an operational definition that activity concentrations greater than 1.3 or 1.4 times the concentration in blood reflect locoregional hypoxia. An early clinical study using ¹⁸F-fluoromisonidazole in a diverse range of tumor types found that, using this criterion to define hypoxia, 36 of 37 contained a measurable fractional hypoxic volume (2). An important caveat associated with these low tumor-to-blood ratios is that they are not entirely unexpected. In vitro experiments show that the uptake of ¹⁸F-fluoromisonidazole in cells under hypoxic (0.5% O₂) versus normoxic (21% O₂) conditions is typically about 6:1. Thus a subtumor region containing 100% hypoxic tissue could theoretically produce a signal 6-fold higher than a normoxic region. However, histologic analysis shows that the hypoxia exists as patches or ribbons dispersed through the disease. These staining patterns suggest that individual PET voxels consist of far less than 100% hypoxic elements—typically 5%–30%. Thus, ¹⁸F-fluoromisonidazole uptake within a hypoxic tumor voxel consisting of 20% hypoxia plus 80% normoxic tissue would not be expected to yield a ratio of 6:1 but only 2:1 (0.2 × 6 + 0.8 × 1 = 2.0). By comparison, the large uptake ratio between tumor and muscle seen with ¹⁸F-FDG is a consequence of every tumor cell, rather than a fraction, exhibiting facilitated uptake and entrapment. A visual representation is shown in Figure 1.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

(A) Histologic tumor section indicating hypoxia tracer pimonidazole (green) and vascular perfusion marker Hoechst 33342 (blue). (B) Same image with six 5-mm² areas superimposed, representing approximate resolution of a clinical PET scanner. (C) Single 5-mm² region at high magnification, containing approximately 20% pimonidazole-positive area. (D–F) Same pimonidazole distribution, but with a gaussian blur of 1 mm (E) and 5 mm (F) applied. This series demonstrates that PET voxel values are a function of both maximal uptake of tracer and hypoxic fraction of voxel. Averaging signal results in lowering overall tumor-to-blood ratios for hypoxia tracers. Increasing PET spatial resolution (E vs. F) would result in improved visualization of regions with higher tumor-to-blood ratios.

The low signal ratios between normoxic and hypoxic tumor locations are also a consequence of partial-volume averaging, giving rise to statistical uncertainties associated with image-based measurements. Nehmeh et al. (3) performed repeated baseline ¹⁸F-fluoromisonidazole PET/CT examinations in immobilized head and neck cancer patients for which accurate registration was possible. The 2 studies, performed 3 d apart, were designed to verify the constancy of hypoxia distribution over the duration required to generate a head and neck radiotherapy plan and begin treatment. Statistical analysis of the voxel variability of the ¹⁸F-fluoromisonidazole signal intensity at corresponding image locations between 2 PET baseline scans showed significant image variability. Whereas such variability can also be a consequence of spatial shifts in acute hypoxia, the poor count-to-noise ratio of late ¹⁸F-fluoromisonidazole hypoxia PET images is a concern for the clinical use of the spatial information within these images for radiotherapy planning purposes. This problem is not unique to ¹⁸F-fluoromisonidazole, however, and represents a significant challenge to any radiotracer that attempts to identify a tumor subregion or a specific subset of cells within the tumor.

DYNAMIC HYPOXIA PET

The limitations in defining hypoxia from late single-time-point ¹⁸F-fluoromisonidazole PET images might be overcome using a kinetic modeling approach in which the rate of irreversible tracer trapping derived from a dynamic imaging series is used as the quantitative measure for each tumor subregion. An example of such a dynamic dataset is shown in Figure 2. This approach was first implemented in clinical studies by Thorwarth et al. at Tübingen University (4) and demonstrated the importance of kinetic analysis to separate the PET signal component associated with hypoxia-specific radiotracer binding from signal associated with unbound freely diffusible tracer.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2.

Results of a dynamic PET scan as a function of time after injection with ¹⁸F-fluoromisonidazole. Single reconstructed PET slice is displayed through center of head and neck tumor. First 5 frames were of 1-min duration, followed by 5 frames of 5-min duration. At 30 min after injection, patient was removed from scanner and then reimaged at 90 min and 180 min after injection. Images were coregistered using low-dose CT scan (depicted in final image in series). This series shows evolution of ¹⁸F-fluoromisonidazole distribution within patient from initial blood pool to selective sequestration within hypoxic tumor subvolume.

One difficulty of performing voxelwise compartmental analysis is the statistical uncertainty associated with PET data from a single voxel. These uncertainties were studied by Wang et al. (5) using simulated PET datasets and the Philips Voxulus compartmental analysis toolkit. One thousand simulated dynamic scans of a phantom within inserted regions exhibiting hypoxic and necrotic tissue characteristics and noise characteristics representative of actual patient dynamic PET scans were generated for compartmental analysis. At a 15% image noise level, the ¹⁸F-misonidazole-trapping rate constant (k₃) exhibited an SD of 13.5%. This uncertainty is relatively large but can be reduced if the kinetic parameters are averaged over a multivoxel region of interest, albeit at the cost of decreased spatial resolution. Parametric maps of k₃, that is, images whose voxel intensity reflect a calculated irreversible trapping rate constant for that location, reflect a more reliable display of the intratumor variation of hypoxia than a PET image of the concentration of radiotracer at the time of image acquisition. This is especially important when considering interventions that might reduce blood flow, and consequently the delivery of the hypoxia imaging agent to the tumor, yet incur a transient increase in the amount of tumor hypoxia (6).

CURRENT CLINICAL HYPOXIA PET TRACERS

The 2′-nitroimidazole compounds exhibit a rate of uptake that is strictly dependent on the oxygen concentration, with the rate of accumulation increasing as partial pressure of oxygen (pO₂) falls below 10 mm Hg in most cases (7). This property has resulted in the synthesis of several fluorinated and iodinated agents that use the nitroimidazole basic structure. Of these, ¹⁸F-fluoroazomycinarabinofuranoside (¹⁸F-FAZA), and its iodinated counterpart (¹²³I/¹²⁴I-IAZA), are promising alternatives to ¹⁸F-fluoromisonidazole. Head-to-head comparisons between ¹⁸F-FAZA, ¹²⁴I-IAZA, and ¹⁸F-fluoromisonidazole in preclinical animal studies imaged at 3 h after injection demonstrated faster vascular clearance of ¹⁸F-FAZA, resulting in an increased tumor-to-blood ratio (5.19) relative to that for ¹⁸F-fluoromisonidazole (3.98). ¹²⁴I-IAZA exhibited a slightly poorer tumor-to-blood ratio at 3 h after injection but allowed (on account of the longer physical half-life of the iodine isotopes) later-time-point imaging (8). More recent clinical studies have successfully evaluated the feasibility of ¹⁸F-FAZA for imaging hypoxia in head and neck tumors, glioma, lymphoma, and lung (9).

A very different chemical class of hypoxia radiotracer that has been intensively investigated in both preclinical and clinical studies is the dithiosemicarbazone compounds. The potential of these agents for hypoxia imaging was first reported by Fujibayashi et al. from Eukui Medical School in Japan (10) and by Holland et al. from Washington University, St. Louis (11). The ability to rapidly label the precursor compound with a variety of copper isotopes of varying half-lives, including ⁶⁰Cu (t_1/2 = 23.7 min), ⁶¹Cu (t_1/2 = 3.35 h), ⁶²Cu (t_1/2 = 9.74 min), and ⁶⁴Cu (t_1/2 = 12.7 h), provides this marker with tremendous flexibility.

One potential difficulty with copper-diacetyl-bis(N4-methylthiosemicarbazone) (ATSM) is the universality and reliability of its use as a hypoxia imaging agent among different tumor types. In vitro experiments examining ⁶⁴Cu-ATSM uptake have shown rapid but quite variable absolute uptakes among different cell lines at the same pO₂ (12), suggesting that the observed high uptake in tumors may only partly be a direct consequence of hypoxia (13). Nevertheless, extremely high-contrast images of copper-ATSM have been obtained in a variety of tumor sites, suggestive of significant tumor hypoxia (11). ACRIN trial 6682, assessing the utility of copper-ATSM as a biomarker of treatment outcome in cervical carcinoma, is currently under way. Whether hot spots within PET images of copper-ASTM represent hypoxia is unknown, but they do provoke further investigation into the uptake and retention mechanism of this radiotracer. It is possible that that pO₂ is only one of several mechanisms affecting copper-ATSM distribution and that copper-ATSM uptake might better represent a general prognosticator of poor treatment response than of tumor hypoxia per se. However, in this context the absolute uptake of ¹⁸F-fluoromisonidazole is not purely pO₂-dependent but is also partially subject to changes in vascular delivery and cellular reductase expression (6,14).

HYPOXIA PET TRACERS: THE NEXT GENERATION

A wide range of fluorinated nitroimidazole compounds have previously been examined in an attempt to identify those with favorable whole-body clearance or more rapid pO₂-sensitive bioreduction. Early examples include metronidazole, etanidazole, EF1, EF3, and SR4454 (15). The primary determinant of the pharmacokinetics of each of these compounds is the octanol–water partition coefficient (logP). In general, the more lipophilic compounds (positive logP) display rapid tissue equilibration and higher first-pass tumor uptake rates but slower background clearance and higher liver activity. The hydrophilic compounds (negative logP) display the reverse characteristics: rapid renal clearance, lower liver uptake, and reduced production of metabolites, at the cost of lower absolute uptake and more heterogeneous tumor distribution. It is becoming increasingly apparent that a single hypoxia tracer is unlikely to be of optimal utility in every clinical situation and that particular disease characteristics will favor the use of a given hypoxia tracer–radioisotope combination (14). The newer generation of 2-nitroimidazole hypoxia tracers under clinical investigation seeks to exploit pharmacokinetic differences in ¹⁸F-fluoromisonidazole and will—it is hoped—find specific roles in which these characteristics can be maximally exploited.

CLINICAL APPLICATIONS OF HYPOXIA IMAGING

The ability to determine the degree and extent of tumor hypoxia is important both as a prognostic biomarker (23) and as a means of selection of patients for hypoxia-directed therapies. The intratumor spatial distribution of hypoxia can potentially serve as a target in radiotherapy planning (24) and also indicate the use of a growing number of adjunct therapies (25). Of these, the current crop of hypoxia-activated nitroaromatic prodrugs such as TH-302 and PR-104 seems most likely to make a significant clinical impact. Because the activation of these compounds shares a similar mechanism with the bioreduction of the 2′-nitroimidazoles, there appears a rationale for the incorporation of hypoxia PET into the evaluation and implementation of these drugs in the clinic.