Imaging tumour hypoxia with oxygen-enhanced MRI and BOLD MRI

Abstract

Hypoxia is known to be a poor prognostic indicator for nearly all solid tumours and also is predictive of treatment failure for radiotherapy, chemotherapy, surgery and targeted therapies. Imaging has potential to identify, spatially map and quantify tumour hypoxia prior to therapy, as well as track changes in hypoxia on treatment. At present no hypoxia imaging methods are available for routine clinical use. Research has largely focused on positron emission tomography (PET)-based techniques, but there is gathering evidence that MRI techniques may provide a practical and more readily translational alternative. In this review we focus on the potential for imaging hypoxia by measuring changes in longitudinal relaxation [R ₁; termed oxygen-enhanced MRI or tumour oxygenation level dependent (TOLD) MRI] and effective transverse relaxation [R ₂*; termed blood oxygenation level dependent (BOLD) MRI], induced by inhalation of either 100% oxygen or the radiosensitising hyperoxic gas carbogen. We explain the scientific principles behind oxygen-enhanced MRI and BOLD and discuss significant studies and their limitations. All imaging biomarkers require rigorous validation in order to translate into clinical use and the steps required to further develop oxygen-enhanced MRI and BOLD MRI into decision-making tools are discussed.

Introduction

Hypoxia occurs when the rate of oxygen delivery to tissue is inadequate to meet demand.¹ Disordered angiogenesis in tumours makes delivery inefficient, leading to hypoxia. This is a hallmark of cancer² and has been recognised for several decades as being a negative prognostic factor in most solid human cancers.^3–5 Furthermore, since the 1950s⁶ hypoxia has been known to cause radioresistance, which results in failure in radiotherapy treatment.⁷ More recently, tumour hypoxia has been implicated in the failure of chemotherapy regimens and numerous targeted therapies.⁸

In oncology there are three main approaches to circumventing these negative effects of tumour hypoxia. Modifiers such as carbogen gas (95%O₂:5%CO₂) and nicotinamide have been shown to alleviate tumour hypoxia in combination with existing (chemo)-radiotherapy,⁹ through improving blood flow and tumour oxygenation. The hypoxic subregions within tumours can be targeted selectively, for example with hypoxia-activated prodrugs^10–12 or drugs that modify oxygen consumption.^13,14 Finally, there is interest in modulating the dose distribution of delivered radiation based on heterogeneity within tumours,¹⁵ focusing on spatial differences in pathophysiological features such as glucose metabolism^16,17 as well as tissue hypoxia.^16,18

Current clinical practice does not identify or quantify hypoxia in tumours. When patients present with cancer, their disease distribution and volume is initially staged.¹⁹ Later, the change in these features following therapy is assessed by response monitoring.²⁰ However, these evaluations are based on anatomical and morphological staging and response systems that regard all abnormal tumour tissue as being equal in importance and relevance.²¹ Various imaging methods are available that can take radiology beyond this approach and incorporate measurements of tumour function.²² In this review, we discuss the various clinically available MRI methods that can map tumour hypoxia, with particular focus on oxygen-enhanced MRI [OE-MRI; also known as tumour oxygenation level dependent (TOLD) MRI and blood oxygenation dependent (BOLD) MRI] and we critique the biomarkers derived from these methods.

Requirements for a biomarker of hypoxia

A biomarker is a “defined characteristic that is measured as an indicator of normal biological processes, pathogenic processes or responses to an exposure or intervention, including therapeutic interventions”.^23,24 Biomarkers are used in cancer medicine to diagnose, predict and to stratify patients into different treatment groups, and also to monitor response to therapy.²⁵

For assessing hypoxia, biomarkers must be technically valid, shown to measure tumour biology and relate to clinical outcome.^26,27 Unfortunately at present, no biomarkers are used routinely to evaluate tumour hypoxia,²⁸ although several investigational approaches exist (Table 1). Oxygenation in tumours can be directly measured using needle electrodes, a technique that was invaluable in first proving the associations with hypoxia and treatment response using Eppendorf histography.^3,4,29,30 However, this technique was limited to accessible tumours and is not generally available.

Table 1.

Summary of methods available to study hypoxia in the clinic. *e.g. [¹⁸F]-fluoromisonidazole (FMISO), [¹⁸F]-fluoroazomycinarabinoside (FAZA) or [¹⁸F]-flortanidazole (HX4). **[⁶⁴Cu]-diacetyl-bis(N⁴-methylthiosemicarbazone)

Technique	Invasive to tumour	Contrast agent requirements	Measured entity	Spatial resolution	Further comments
In vivo
pO2 histography	Yes	None	[O2(s)] inferred from rate of arrival of O2 molecules at electrode	Sub mm scale	Very limited clinical availability
PET with [18F]-labelled nitroimidazoles*	No	i.v. injection of tracer	Retained [18F] or [64Cu] implying insufficient [O2] to reverse effect of tissue reductases	Few mm scale except for rodent MRI 0.2 to 0.5 mm for rodent MRI	Some validation; limited availability
PET with [64Cu] ATSM	No				Seldom used compared to [18F] PET
DCE-MRI	No	i.v. injection of gadolinium-based contrast agent	Blood flow and permeability		Indirectly associated with hypoxia
R 1 weighted OE-MRI	No	Hyperoxic inhalation	[O2(s)]		Emerging techniques requiring further validation
R 2* weighted BOLD MRI	No	optional hyperoxic inhalation	compartmentalised deoxyhaemoglobin
Ex vivo
Hypoxia RNA gene signatures	Yes	None	gene expression associated with hypoxia	µm scale	Requires tissue: provides important cross- validation for other hypoxia biomarkers
HIF 1α	Yes	None	HIF 1α transactivation
GLUT 1 and CA-IX	Yes	None	HIF 1α transactivation
Pimonidazole	Yes	Oral or i.v. administration	Retained nitroimidazole implying insufficient [O2] to reverse effect of tissue reductases
Circulating osteopontin	No	None	Chronic hypoxia	N/A

BOLD, blood oxygenation level dependent; DCE, dynamic contrast-enhanced ; OE, oxygen-enhanced; PET, positron emission tomography;RNA, ribonucleic acid.

Assessment of tissue-level hypoxia is common in pre-clinical experiments, but mapping the spatial distribution of hypoxic tumour tissue is also possible in humans. Strategies available use either exogenous administered compounds such as nitroimidazoles (e.g. pimonidazole) that bind to macromolecules in cells under hypoxic conditions,³¹ or use endogenous markers such as proteins CA-IX and GLUT1.³² Alternative approaches use gene signatures of hypoxia.³³ However, biopsies are challenging or impossible in some tumour types, provide only a limited subsampling of the tumour, and are difficult to interpret in the presence of temporally fluctuating hypoxia. Furthermore, repeat measurements are normally impractical.²⁸ In distinction, serological markers (e.g. osteopontin) have shown some utility³⁴ and can be performed on repeat sampling, but cannot distinguish levels of hypoxia in different tumours within the same patient.

Imaging is an attractive option since it can provide serial non-invasive sampling of whole tumour volumes. Imaging can both identify subregions within one tumour that vary in their hypoxic profiles, and simultaneously distinguish hypoxic and normoxic tumours from one another in the same patient.¹⁵ At present no MRI, positron emission tomography (PET) or other imaging biomarker has sufficient technical, biological and clinical validation to have been translated into routine clinical practice.³⁵ In the sections below, we review the current status of MRI biomarkers of hypoxia and discuss what further work is required to translate these techniques into clinical use.

MR contrast mechanisms: potential methods for imaging hypoxia

MRI is attractive as it offers several independent contrast mechanisms which interrogate different facets of hypoxia. In humans, MRI voxels are large (typically a few millimetres in each orthogonal dimension) in comparison with typically 0.1 mm distance between a hypoxic domain and its nearest capillary.³⁶ Although some subvoxel hypoxia heterogeneity can be expected, the resolution is more than adequate for radiotherapy planning. Registration of MRI with radiotherapy, however, is more difficult than for CT, single photon emission CT and PET, because of the different mechanisms of image formation.

Two substances of particular interest in imaging hypoxia are the dioxygen molecule, in solution, O_2(s), which has two unpaired electrons³⁷ and the deoxyhaemoglobin monomer, Hb, which has four unpaired electrons. Both molecules have an effect on longitudinal relaxation rate (R ₁), the reciprocal of the longitudinal relaxation time (T ₁). Independently, the heterogeneous intravoxel distribution of Hb affects the effective transverse relaxation time (T ₂*), through its effect on T ₂’ (Figure 1; details below). This effect on T ₂* (unlike the effect on T ₁) depends in a complex way on the spatial arrangement of the microvasculature and haematocrit, but in general higher concentrations of deoxyhaemoglobin in a voxel, are associated with short T ₂*.

Figure 1.

Schematic representation of the change in R ₁ and ΔR ₂* induced by hyperoxic gas.

The unpaired electrons endow both substances with modest longitudinal relaxivities (r ₁) at 1.5T of 0.17 s⁻¹.mM⁻¹ for O_2(s) and 0.008 s⁻¹.mM⁻¹ for the Hb monomer.^38,39 These relaxivities and concentrations are too small to allow hypoxic tumours to be identified from their T ₁ alone. However, switching the inhaled gas between air and 100% oxygen causes arterial hyperoxia. This in turn perturbs regional tumour concentrations of O_2(s) and Hb imparting a heterogeneous change in R ₁ (termed ΔR ₁) which is directly related to tumour hypoxia. This experimental design is referred to as OE-MRI or TOLD contrast.^40,41

In addition to their longitudinal relaxivities, O_2(s) and Hb make a paramagnetic contribution to tumour magnetic susceptibility which can be detected directly through susceptibility-weighted imaging.⁴² The paramagnetic susceptibility of Hb has an additional and important use as it induces extreme field gradients around erythrocytes, massively enhancing signal dephasing and increasing the effective transverse relaxation rate (R ₂*; reciprocal of T ₂*)–the so-called BOLD effect.⁴³ However, rather than measure native R ₂*, most studies exploit a similar experimental design to the above OE-MRI where the change in R ₂* (termed ΔR ₂*) is measured following perturbation with 100% oxygen. Unfortunately, unlike OE-MRI, there is no direct linear relationship between R ₂* or hyperoxia-induced ΔR ₂* and hypoxia.

An alternative experimental design uses carbogen gas as a hyperoxic challenge rather than 100% oxygen. This stems from the fact that many early BOLD studies were applied with the aim of measuring the effects of carbogen therapy. However, the interpretation of this design is more challenging because of the need to interpret the physiological effects of both O₂ and CO₂.⁴⁰ Studies in animal models and in humans have shown that the effects of the two gases differ between various normal tissues.^44,45 Furthermore, carbogen can be unpleasant to breathe. Because of these facts, and because most subsequent studies have not been designed to monitor the effect of carbogen as a therapy, many investigators have resorted to using 100% oxygen as a challenge instead of carbogen.

Other MRI biomarkers also provide indirect insight into hypoxia. Inadequate perfusion is a proximate cause of hypoxia and can be interrogated by arterial spin labelling or dynamic contrast-enhanced MRI (DCE-MRI).^46,47 Cell death and necrosis are a common consequence of hypoxia and can be reflected in elevated ADC or T ₂.²² Finally, a few investigational hypoxia-sensitive MRI probes have been used in pre-clinical studies, but have not progressed far in the clinic.^48,49

OE-MRI: R ₁ contrast MRI and hypoxia

In well-oxygenated voxels, inhaling excess O₂ increases R ₁ because supply is already adequate to meet the demand of local mitochondria, and excess oxygen remains dissolved in blood plasma and interstitial tissue fluid.⁵⁰ The measured change, ΔR ₁ induced by breathing oxygen is given by:

$${\displaystyle \mathrm{△}R1=R_{1(t)}-R_{1(0)}=\mathrm{△}[O_2].r_{1,0_2}+\mathrm{△}[Hb].r_{1,Hb}}$$

where R ₁₍₀₎ is R ₁ at baseline, R _1(t) is R ₁ at time t after the switch to O₂ inhalation, depending on the changes in concentration of the two relaxive substances Δ[O₂] and Δ[Hb] together with their respective relaxivity constants r _1,O2 and r _1,Hb. The second, Hb-dependent, term is small and in many settings can be neglected, leaving ΔR ₁ simply proportional to change in tissue oxygen concentration Δ[O₂]. The R ₁ values and concentrations are voxel-averages, R ₁ is assumed to be mono-exponential and measurements are assumed not to be confounded by inflowing blood.

Multiple studies have shown that OE-MRI produces measurable signal changes in normal tissues and is feasible on both clinical scanners^40,51,52 and at the high field strengths used for pre-clinical studies,⁴⁵ with signal changes of up to 20% reported (Figure 2). Although the relaxivity of O_2(s) declines at high field, this disadvantage is offset by the lower R ₁₍₀₎ and higher SNR at high field. The technique has also been used in several non-oncological settings over the last two decades in approximately 40 published studies of respiratory physiology and disease.^53–55

Figure 2.

OE-MRI in normal tissues: mean of the ΔR ₁ recorded in the spleens of 16 healthy volunteers. There is a positive ΔR ₁ induced following both 100% oxygen (O₂) and carbogen (CB), although the response seen with O₂ is slightly attenuated with CB gas. The ΔR ₁ returns to baseline once air breathing is resumed. Adapted from reference.⁴⁰ OE-MRI, oxygen-enhanced MRI.

Since OE-MRI can measure oxygen delivery, investigators in the 2000s began to explore its role in oncology. In most studies of pre-clinical xenograft and orthotopic tumour models^41,56–62 and in some human tumors,^63–65 positive values of oxygen-induced ΔR ₁ were reported, reflecting the average effect of O₂ inhalation in tumours. These studies covered a range of field strengths. R ₁ changes, and the concomitant signal intensity changes in T ₁- weighted MRI, are typically ≤5%, but with care the technique is feasible on both pre-clinical and clinical MRI platforms.

These studies of OE-MRI in oncology created interest in it being a method for use in studies of hypoxia. However, measuring positive ΔR ₁ in OE-MRI quantifies and maps oxygen delivery in tissues with fully saturated haemoglobin, but does not directly identify tissue hypoxia per se. This inspired a second look at OE-MRI signal changes in tumours. For the first time, analyses highlighted that some tumour subregions were refractory in R ₁ to oxygen challenge,^61,66 or even exhibited negative ΔR ₁. These regions had low haemoglobin oxygen saturation, so excess O₂ molecules were immediately bound to Hb and did not significantly alter tissue [O_2(s)].²⁸ At the same time this depletion of paramagnetic Hb may induce a barely-detectable decrease in R ₁. Decreasing R ₁ suggests a vascular volume fraction which is substantial (since only vasculature contains Hb), but nonetheless inadequate to supply the local mitochondria, perhaps because of chaotic architecture. Retrospective evaluation showed that this finding was also present in several other studies from other research groups.^41,56,61

Based on these data, we investigated whether the regions that are perfused but lack oxygen enhancement (with an OE-MRI and DCE-MRI biomarker termed perfused Oxy-R) could identify hypoxic subregions within tumours.³¹ This study showed for the first time the translational potential for OE-MRI,³⁶ since signals were accurate, precise, and sensitive to changes in tumour pO₂ (Figure 3A). The relationship of ΔR ₁ to change in pO₂ has also been shown by another independent research group.⁶⁷ Furthermore, perfused Oxy-R fraction quantified the hypoxic fraction in multiple models and detected dynamic changes in hypoxia induced by a vasomodulator (Figure 3B).

Figure 3.

OE-MRI validation in tumours: (A) Positive increase in R ₁ is induced following 100%oxygen (O₂) in a mouse model of renal carcinoma (786–0 R). OE-MRI changes are mirrored by change in tumour pO₂. (B) When OE-MRI signals are examined in perfused tissue, the oxygen refractory voxels (termed perfused Oxy-R; blue) are distinguished from oxygen-enhancing voxels (yellow); non-perfused voxels (grey) are excluded. The perfused Oxy-R fraction correlated significantly to the hypoxic volume measured on immunofluorescence.Adapted from reference.³¹ OE-MRI, oxygen-enhanced MRI.

The potential value of OE-MRI is now being investigated by several research groups. The method has been shown capable of distinguishing radiation necrosis from malignant high-grade glioma in mouse models⁶² (Figure 4). This indicates a possible diagnostic application. The potential use in radiotherapy prognosis has been suggested in a small study of rats with Dunning R3327-AT1 tumours treated with radiotherapy, where those tumors with greater oxygen-induced ΔR ₁ during therapy had greater growth delay.⁶⁸ In distinction, comparable results were not reported in mice bearing glioma and rhabdomyosarcoma xenografts where carbogen-induced challenges were performed.⁶⁹ Further pre-clinical studies will be required to help determine the best clinical trial design to qualify the role of OE-MRI parameters as prognostic biomarkers.

Figure 4.

OE-MRI application: The ΔR ₁ induced following carbogen inhalation distinguished tumour from radiation necrosis. Both pathologies had lower ΔR ₁ than healthy brain in the contralateral hemisphere. Adapted from reference.⁶² OE-MRI, oxygen-enhanced MRI.

Perhaps the most easily envisaged OE-MRI application may be in evaluating response to therapy. Although in their infancy, OE-MRI studies have shown considerable initial promise in this regard. A pre-clinical study using Calu6 and U87 xenografts has shown that the OE-MRI biomarker ‘perfused Oxy-R’ is sensitive to changes in hypoxia induced by hypoxia modifying targeted therapies.⁷⁰ Here, both the hypoxia-activated cytotoxic prodrug banoxantrone, and the oxygen consumption modifier atovaquone were shown to be active in the xenograft models, with reduction in the volume of tumor identified by OE-MRI, relative to control. Similarly, reduction in hypoxia has been shown with high-dose single fraction radiation as well as with fractionated chemoradiotherapy in the same xenograft models.^68,71

BOLD: R ₂* contrast MRI and hypoxia

The contribution of blood deoxyhaemoglobin to the R ₂* relaxation rate of each voxel is not simple

$${\displaystyle \mathrm{\Delta }{R}_2^{}=R_{2\left(t\right)}^{}-R_{2\left(0\right)}^{}=f\left(\mathrm{\Delta }\left[Hb\right]\right)}$$

Where f is an unknown non-linear function of the voxel dimensions and capillary geometry and Δ[Hb] is the change in tissue deoxyhaemoglobin concentration caused by hyperoxia.

Note that, unlike eq [1] for R ₁, eq [2] is not a voxel-average, and the relationship between the relaxation rate and the concentration of the relaxive substance cannot be simply described by a relaxivity. ΔR ₂* is determined by the vascular geometry, vascular volume and change in blood oxygenation, and may reflect the potential to enhance oxygen delivery to a tumour.

The oxygenation of haemoglobin depends on the arterial blood p_aO₂, but this is not in equilibrium with tissue pO₂ if viable mitochondria are present, because there must be an O₂ gradient from the vessels to the mitochondria. Measurements of tumour R ₂* therefore cannot provide an index of tumour oxygenation. The relationship of R ₂* weighted image response and tumour pO₂ has been investigated by invasive Eppendorf histography with carbogen breathing and showed a weak correlation.⁷² A stronger correlation of carbogen-induced decreases in R ₂* with tumour oxygen tension, measured by oxygen microelectrodes, has been observed in rat mammary carcinomas.⁷³ Carbogen-induced decreases in R ₂* of rat intracranial gliomas were shown to correlate with an increase in pO₂ measured by EPR oximetry.⁷⁴ Simultaneous measurements of tumour R ₂* and pO₂ have been achieved using an MRI-compatible fibre-optic pO₂ sensor.^75,76

Collectively these studies demonstrated that the R ₂* signal response to carbogen is temporally correlated with changes in tumour pO₂, but that there was no correlation between absolute R ₂* and pO₂. Associations of tumour R ₂* and oxygen-induced ΔR ₂* with hypoxia and improved tumour oxygenation, measured using immunohistochemical detection of reduced 2-nitroimidazole adducts, have been demonstrated in a range of pre-clinical tumour models^77–80 (Figure 5). Taken together, these data suggest that BOLD MRI can be used to assess changes in tumour oxygenation and provide good evidence that a hyperoxia-induced decrease in R ₂* is indicative of increased tumour oxygenation in vivo.⁸¹

Figure 5. .

BOLD imaging validation in tumours: Parametric R ₂* maps of a GH3 prolactinoma during inhalation of a) medical air and b) carbogen (CB). Intense (white) regions (relatively fast R ₂*) in the initial air breathing reflect the presence of paramagnetic Hb, whilst dark areas (relatively slow R ₂*) are consistent with the presence of oxyhaemoglobin. Inhalation of CB results in a clear decrease in R ₂*, indicating a decrease in Hb. Composite fluorescence images showing the distribution of reduced 2-nitroimidazole adduct formation from c) CCI-103F (red), administered during air breathing, and d) pimonidazole (green), administered during CB breathing, obtained from the same GH3 tumour, are also shown. Spatially, both CCI-103F and pimonidazole adduct formation are co-localised, but the extent of hypoxia staining is reduced following CB, providing histopathological validation of carbogen-induced ΔR ₂* as a non-invasive imaging biomarker of increased tumour oxygenation. BOLD, blood oxygenation level dependent.

As highlighted earlier, hyperoxia increases blood oxygenation, and the magnitude of the change in tumour R ₂* is dependent on blood volume, which is itself a determinant of the hypoxic fraction.^82,83 BOLD MRI has been shown to correctly predict the relative effects of radiosensitisers on tumour hypoxic fraction.⁸⁴ One pre-clinical study sought to test the hypothesis that the baseline tumour R ₂* and carbogen-induced ΔR ₂* measured prior to radiotherapy were prognostic for treatment outcome. Prior to irradiation, tumour R ₂* was quantified while the host breathed air and subsequently carbogen, and correlated with the subsequent tumour growth inhibition in response to ionizing radiation. Overall, tumours which exhibited a significantly faster baseline R ₂* and a significantly greater carbogen-induced ΔR ₂* were more responsive to radiotherapy (Figure 6).⁸⁵

Figure 6. .

BOLD imaging application: Pre-treatment R ₂* and carbogen (CB)-induced ΔR ₂* for individual GH3 prolactinomas (white triangles) and RIF-1 fibrosarcomas (black triangles), plotted against tumour volume at day 7 post-irradiation with 15 Gy as a percentage of the volume pre-treatment. GH3 prolactinomas displayed a fast baseline R ₂*, large CB-induced ΔR ₂* prior to radiotherapy and greater reduction in tumour volume post-irradiation. In contrast, RIF-1 fibrosarcomas displayed a slow baseline R ₂*, negligible ΔR ₂* response to CB and a smaller growth inhibition. The data suggest that quantitation of tumour R ₂* and CB-induced ΔR ₂* provide prognostic indicators of radiotherapeutic response. Adapted from reference.⁸⁵ BOLD, blood oxygenation level dependent.

Despite potential distortion artefacts arising in the vicinity of air and tissue interfaces in some anatomical sites, BOLD MRI has been implemented on standard clinical scanners and good reproducibility of human tumour baseline R ₂* maps demonstrated.⁸⁶ BOLD MRI has been used to quantify R ₂* and carbogen-induced changes in R ₂* of human head and neck cancers prior to radiotherapy.⁸⁷ In this study, all 11 patients studied showed a carbogen-induced tumour ΔR ₂* (statistically significant in 7) prior to ARCON therapy, and subsequently showed a low tumour recurrence rate after a considerable follow-up time. A clinical study on hypoxia in prostate cancer compared tumour R ₂* and relative blood volume (rBV, measured by DCE MRI) with tissue sections immunohistochemically stained for pimonidazole.⁸⁸ Fast R ₂* was correlated with pimonidazole staining and found to have a high sensitivity in depicting tumour hypoxia (88%), which was further enhanced by the addition of low rBV information (95%) without a change in specificity (36 and 29%, respectively), suggesting that the combination of native R ₂* with rBV were effective in mapping intraprostatic tumour hypoxia. In a follow-up study, 17 patients with prostate cancer were investigated using R ₂* measurements before and during a period of carbogen gas breathing.⁸⁹ 64% exhibited a reduction in tumour R ₂* during carbogen inhalation, with a significant mean reduction of 22%, suggesting the presence of tumour hypoxia in the native state which was improved by carbogen inhalation. In exploring the relationship between BOLD MRI and invasive oxygen electrode measurements in human prostate cancer, a significant positive correlation was found between tumour R ₂* and the fraction of tumour exhibiting pO₂ values < 5 mmHg, and a negative trend between R ₂* and pO₂.⁹⁰

That said, opposite observations of R ₂* values in relation to hypoxia have been observed in other tumour types. For example, in a pre-clinical study of mammary tumours, basal R ₂* was negatively correlated with pimonidazole staining.⁷⁷ The relationship of native R ₂* to tissue hypoxia appears to vary according to the underlying histology, and this relationship needs to be defined across a range of different tumour types.

Native R ₂* and cycling hypoxia

Since the paramagnetic susceptibility of Hb induces extreme field gradients around erythrocytes it enhances signal dephasing and increases R ₂*. Indeed, this effect is so pronounced that low tumour signal in T ₂* weighted MRI suggests hypoxia. This effect has been exploited in a small number of studies, where the native R ₂* has been measured in the absence of a challenge with hyperoxic gas.

Cyclical hypoxia arises from fluctuations in erythrocyte flux through the abnormal tumour vasculature.^91–93 Given the sensitivity of R ₂* to deoxygenated erythrocytes, continuous BOLD MRI measurements have been exploited pre-clinically to non-invasively image cyclical hypoxia through changes in the oxy/deoxyhaemoglobin ratio at high spatial and temporal resolution in both xenografts and patient tumours in vivo.^94–96 Frequencies in the range of 0.00027–0.001 Hz (corresponding to 15 to 60 min) were measured, comparable to the periodicity originally reported from classical pre-clinical invasive measurements of cyclical hypoxia.^97,98 Interestingly, in patients with head and neck squamous cell carcinoma, R ₂* fluctuations spatially correlated with parts of lymph nodes with low K ^trans values, typically in the vicinity of necrotic nodes. The R ₂* fluctuation fraction was higher in the non-responding patient group, suggesting that the presence of such fluctuations may be predictive of a worse outcome following treatment for HNSCC⁹⁶ (Figure 7).

Figure 7. .

Native R ₂* evaluation of cycling hypoxia: An example map of R ₂* perturbation is shown of a lymph node metastasis from a patient with squamous cell head and neck cancer (with zoomed in section). An example of the power spectra tested for non-random fluctuations is shown. Adapted from reference.⁹⁶

Comparison of R ₁ and R ₂* contrast MRI

Tumour subregions with low haemoglobin oxygen saturation are refractory to oxygen challenge on R ₁ OE-MRI. Low haemoglobin oxygen saturation implies very low blood [O_2(s)], and even lower tissue [O_2(s)] if there are functional mitochondria. Even if oxygen challenge substantially increases haemoglobin oxygen saturation, the absolute changes in [O_2(s)] are small, causing undetectable change in R ₁. It is logical to suspect that the R ₂* in these voxels should be high at baseline and also decrease on oxygen challenge because HbO₂, unlike Hb, causes no paramagnetic effect on magnetic susceptinility shift inside erythrocytes.

Several studies have examined the relationships between ΔR ₁, native R ₂* and ΔR ₂* in xenografts^56–59,61 or in patient tumours.^65,99 These generally describe complex and non-linear relationships, possibly reflecting spatial or temporal heterogeneity with subregions of luxurious and inadequate perfusion present simultaneously and sometimes seen (with temporal disconnect) in the same voxel. However, potential confounds exist in both measurements; for example ΔR ₁ can be confounded by inflow effects, ΔR ₂* can be confounded by presence of haemorrhage and both parameters can be confounded by oxygen-induced vasoconstriction and by vascular steal (Figure 8).

Figure 8. .

Schematic representation of the theoretical relationship between gas-induced ΔR ₁ and ΔR ₂*.

Since it is unclear from these studies as to whether ΔR ₁ and the R ₂*-based biomarkers measure the same underlying tumour pathophysiology, we performed a study in a xenograft model and in seven patients with renal cell carcinoma to explore the relationship between the two techniques.⁸⁰ Tumour-wise and voxel-wise ΔR ₁ and ΔR ₂* comparisons did not show correlative relationships in 786–0 R renal cancer xenografts. However, parcellation analysis revealed that perfused Oxy-R regions had faster native R ₂* (102.4 s^–1 vs 81.7 sec^–1) and greater negative ΔR ₂* (−22.9 s^–1 vs −5.4 s^–1) when compared with perfused Oxy-E and non-perfused tumour subregions. Similar findings were present in human renal cell carcinomas.

Future directions

Both R ₁ and R ₂* gas-challenge MRI techniques have potential to identify, spatially map and quantify tumour hypoxia. There is a significant body of evidence that biomarkers derived from the two techniques can reflect underlying low pO₂ and resultant tissue hypoxia (identified by immunohistochemistry) in tumour subregions. However, nearly all of this data are in rodent tumour models and further investigation is required to add to the handful of small studies performed to date, to confirm that these findings are replicated consistently in a range of human cancer types. A small number of studies have also explored technical validation – such as repeatability – in a single centre setting. While these data are promising, further data are required to confirm high-to-excellent reproducibility in a multicentre setting using scanners that cover a range of vendors, MRI field strengths and sequences.³⁵

However, the key questions for R ₁ and R ₂* gas-challenge MRI will be to demonstrate readily measurable and consistent value of their biomarkers in clinical applications, such as identifying tumour hypoxia, tracking change in hypoxia on therapy or spatially mapping hypoxia to guide differential dose radiotherapy (dose painting).¹⁰⁰ This is an area of research priority for the next decade, which if successful could result in the rapid translation of these techniques into clinical decision making.