Comparison of ¹H BOLD and ¹⁹F MRI to investigate tumor oxygenation

Abstract

¹⁹F MRI oximetry and ¹H blood oxygen level dependent (BOLD) MRI were used to investigate tumor oxygenation in rat breast 13762NF carcinomas and correlations between the techniques were examined. A range of tissue pO₂ values was found in the nine tumors while the anesthetized rats breathed air with individual tumor pO₂ ranging from a mean of 1 to 36 torr and hypoxic fraction HF₁₀ (<10 torr) ranging from 0 to 75% indicating a large intra- and inter-tumor heterogeneity. Breathing oxygen produced significant increase in tumor pO₂ (mean ∆pO₂ =50 torr) and decrease in HF₁₀ (p<0.01). ¹H BOLD MRI observed using a spin echo planar imaging (EPI) sequence revealed a heterogeneous response and significant increase in mean tumor signal intensity (∆SI =7%, p<0.01). R₂* measured by multi-gradient echo (MGRE) MRI decreased significantly in response to oxygen (mean ∆R₂* = −4 s⁻¹; p<0.05). A significant correlation was found between changes in mean tumor pO₂ and mean EPI BOLD ∆SI accompanying oxygen breathing (r² >0.7, p<0.001). Our results suggest that BOLD MRI provides information about tumor oxygenation and may be useful to predict pO₂ changes accompanying interventions. Significantly, the magnitude of the BOLD response appears to be predictive for residual tumor hypoxic fractions.

Introduction

Tumor oxygenation has been widely recognized as a potent factor influencing tumor response to various therapies, especially radiotherapy and hypoxia appears to promote tumor malignant progression and metastasis (1). Given the importance of tumor oxygenation, many measurement techniques have been developed (1,2). While each method has specific attributes, many are highly invasive or cannot be applied to longitudinal studies of oxygen dynamics.

BOLD (blood oxygen level dependent) MRI, extensively used in studying brain function, is increasingly being applied to assess blood oxygenation and vascular function in tumors non-invasively (3–12). The underlying rationale is that the paramagnetic deoxyhemoglobin creates microscopic field gradients, which enhance the transverse relaxation rate, R₂*, of water protons in blood and in the tissue adjacent to blood vessels. Decrease in deoxyhemoglobin concentration leads to a decreased R₂*, and thus, to an increased signal intensity in T₂*-weighted MRI (13,14). Gradient-recalled echo (GRE) or spin echo planar imaging (EPI) is sensitive to changes in R₂*. However, BOLD contrast is also influenced by other factors such as blood flow, blood volume and vascular architecture (3,15). Several recent studies have attempted to correlate BOLD MRI with tissue pO₂ measured by various techniques, notably, oxygen electrodes, oxygen sensitive fiber optic probes, ESR and ¹⁹F MRI (4,11,16,17). Some of these studies have indicated a strong quantitative correlation with tissue pO₂ (11), some found a qualitative relationship (16,17), while others suggested a lack of direct correlation, yet consistent temporal trends (4).

We have developed a method for measuring tumor oxygenation and dynamics based on ¹⁹F NMR EPI following direct intratumoral injection of the reporter molecule hexafluorobenzene (HFB): FREDOM (Fluorocarbon Relaxometry using Echo planar imaging for Dynamic Oxygen Mapping) (2). This technique provides quantitative pO₂ measurements at multiple specific locations simultaneously within a tumor, and reveals acute dynamic changes at individual locations with respect to interventions, such as hyperoxic gas breathing and vascular modifiers. The aim of this study was to compare ¹⁹F oximetry (FREDOM) with ¹H BOLD MRI in evaluating tumor oxygenation in response to hyperoxic gas (100% oxygen) challenge.

Materials and Methods

Tumor Model

Rat mammary carcinoma 13762NF (originally obtained from the Division of Cancer Treatment, NCI) was implanted syngeneically in a skin pedicle surgically created on the foreback of Fisher 344 adult female rats (~150 g, n = 9, Harlan), as described in detail previously (18). Tumors were allowed to grow and were investigated when tumor volume was 0.2 to 2.1 cm³ (mean volume = 1 cm³). Investigations were approved by the Institutional Animal Care and Use Committee.

MRI experiments

MRI was performed using a 4.7 T horizontal bore magnet with a Varian Unity Inova system. Each rat was given ketamine hydrochloride (120 µl; 100 mg/ml, Aveco, Fort Dodge, IA) as a relaxant (i.p.) and maintained under general anesthesia (air and 1% isoflurane (Baxter International Inc., Deerfield, IL)). The oxygen reporter molecule hexafluorobenzene (HFB, 50 µl, Lancaster, Gainesville, FL) was injected directly into the tumor along two or three tracks in a single central plane of the tumor, coronal to the rat’s body using a Hamilton syringe (Reno, NV) with a custom-made fine sharp needle (32G), as described in detail previously (2). A tunable (¹H/¹⁹F) volume RF coil was placed around the tumor-bearing pedicle. Each animal was placed on its side in the magnet with no change in position during the whole study. A thermal blanket was used to maintain body temperature. A single 2 mm slice coronal to the rat body containing the strongest fluorine signal was chosen for both ¹⁹F pO₂ and ¹H BOLD studies. ¹H and ¹⁹F MR images were acquired using a spin-echo sequence. Overlaying the ¹⁹F MR image on the corresponding ¹H image revealed the distribution of HFB.

¹H BOLD

Echo-planar imaging BOLD

A spin echo planar imaging sequence with pulse burst saturation recovery (PBSR) signal preparation was applied, as described previously (8). The initial saturation was designed to minimize in-flow effects. A series of 55 images including 5 baseline with air breathing (images 1–5) and 50 with oxygen breathing (images 6–55) was acquired on the 2 mm coronal section at 5 s intervals using MR parameters: τ = 500 ms (≡ TR), TE = 53.7 ms, field of view (FOV) = 40 × 40 mm, matrix = 32 × 32.

GRE R₂*

Multi-gradient echo (MGRE) images with 8 echoes were acquired on the same 2 mm slice during air breathing (baseline) and repeated immediately after the EPI BOLD sequence (~ 5 min after start of O₂ breathing) for 8 of 9 tumors. Acquisition parameters were: repetition time (TR) =195 ms, initial echo time (TE) = 7 ms, echo spacing = 6 ms, flip angle = 45°, FOV = 40 × 40 mm, matrix = 128 × 128, 2 averages, acquisition time = 6 min 40 s.

¹⁹F Tumor tissue oximetry - FREDOM

Following a re-equilibration period of air breathing (> 15 min), tumor oxygenation was estimated on the basis of ¹⁹F pulse burst saturation recovery (PBSR) EPI relaxometry of the HFB, as described previously (2). This approach provided pO₂ maps with 1.25 mm in plane resolution and ~3 µl voxel size (FOV = 40 × 40 mm, matrix = 32 × 32, 2 mm thick) in 6.5 minutes. The spin-lattice relaxation rate [R₁ (s⁻¹) = 1/T₁] was estimated on a voxel-by-voxel basis using a three-parameter monoexponential function, and pO₂ was estimated using the relationship pO₂ (torr) = (R₁ – 0.0835)/0.001876 (2). Seven consecutive pO₂ measurements including two baseline and five oxygen breathing were acquired on the same 2 mm section as used for ¹H BOLD studies.

Data Analysis

Data were processed using IDL 5.3/5.4 (Research Systems, Boulder, CO). Signal intensity (SI) in the EPI BOLD study was assessed on a voxel-by-voxel basis and averaged at every time point for the whole tumor section. The signal intensity change (ΔSI) of each tumor was normalized to the mean baseline value expressed as a percentage change using the equation:

$$(\mathrm{\Delta }SI)=(S{I}_M-S{I}_b)/S{I}_b·100\%,$$

where SI_M and SI_b refer to maximum mean SI and mean baseline SI, respectively.

R₂* maps were generated using all eight images with variable echo time by fitting an exponential model on a voxel-by-voxel basis. Mean R₂* of the whole section was determined for baseline air and oxygen intervention. ΔR₂* maps were obtained by subtracting R₂* oxygen map - R₂* baseline map.

For the FREDOM data, typically 40–100 voxels provided an R₁ fit, and potential pO₂ value. Since noise may give an apparent relaxation curve (R₁) fit, data were selected by applying thresholds of T₁ error < 2.5 s and ratio T_1error/T₁ < 50%. The two criteria are used since T₁ can have a very large range from 1.5 to 12 s and thus the absolute error is particularly important for long T₁s and the ratio for short T₁s. While these criteria appear quite lax, only those voxels, which provided consistently reliable data throughout the whole time course of seven measurements, were included for further analysis.

Statistical significance was assessed using an Analysis of Variance (ANOVA) on the basis of Fisher's Protected Least Significant Difference (PLSD; Statview, SAS Inst. Inc., Cary, NC) or paired Student's t-tests. ANOVA was applied for comparison of multiple repeat measurements and the PLSD examines the importance of individual measurements on the overall population. The assumption is that inhaled gas at various time points is the independent variable, while pO₂, R₂* and BOLD signal changes are dependant variables. Paired Student t-tests were used to compare individual pairs of data such as pO₂ in a specific tumor during air or oxygen breathing.

Results

Overlay of ¹⁹F on the corresponding ¹H MR image confirmed that HFB was distributed in both peripheral and central regions of a 2 mm thick section, located in the central plane of a representative tumor (Figure 1). The following EPI BOLD, GRE R₂* and ¹⁹F oximetry measurements all interrogated this thin slice.

Figure 1. Distribution of oxygen reporter probe.

A. ¹H T₁-weighted MR image of 2 mm slice through tumor showing signal voids corresponding to presence of HFB reporter molecule.

B. Overlay of ¹⁹F signal intensity on ¹H image showing distribution of oxygen reporter probe hexafluorobenzene (HFB) in both central and peripheral tumor regions (#2 in Table 1). Bar = 0.5 cm.

¹H BOLD

Echo-planar imaging

Normalized SI maps at different time points after switching to oxygen breathing showed heterogeneous response (Figure 2). Significant mean signal enhancement during oxygen inhalation was observed in all nine tumors with a mean ΔSI = 6.7 ± 1.4% (range from 1% to 12%; Table 1). However, individual voxel data showed some regions with negative response (Figure 2). The percentage of voxels with negative response ranged from 12% to 38% in the nine tumors. Baseline signal was quite stable, but there was a rapid response within about 25 s of switching the inhaled gas to oxygen. Some tumors showed a continual increase, which was usually biphasic and approached a stable plateau after about 3 mins. In other cases there was a transient maximum followed by a stable lower value, just marginally above baseline.

Figure 2. Variations in T₂* weighted images with oxygen challenge (BOLD).

A. Normalized spin echo planar images of two tumors acquired at several time points after switching to oxygen breathing. In response to oxygen breathing the two tumors (upper, #2 0.4 cm³; lower, #6 1.4 cm³) showed heterogeneous signal enhancement. Regions of decreased signal intensity were also observed (dark regions).

B. Variations of normalized signal intensity change (ΔSI) versus time for these tumors with respect to oxygen challenge. Both tumors showed an initial rapid response. For tumor #2 (○) this became biphasic reaching a plateau with 9% increase after about 200 s of oxygen breathing. The second tumor (#6, ■) rapidly reached a peak value of ΔSI = 1.4% after 35 s, then gradually decreased to a plateau at about 0.5% increase. Data points are mean values ± standard error.

Table 1.

Comparison of pO₂, BOLD and R₂* in individual tumors

Tumor	Tumor volume (cm3)	19F oximetry					BOLD EPI (%)	R2* (s−1)
#
		21% O2		100% O2		ΔpO2 (torr)		21% O2	100% O2	ΔR2*
		pO2 (torr)α	HF10 (%)	pO2 (torr)β	HF10 (%)
1	0.3	36 ± 1	0	172 ± 7**	0	136	12.2 ± 0.6	54 ± 1	51 ± 1*	−3
2	0.4	17 ± 2	29	76 ± 4**	0	59	8.7 ± 0.5	68 ± 1	65 ± 1**	−3
3	0.4	13 ± 2	31	62 ± 8**	4	49	7.6 ± 0.6	81 ± 1	77 ± 1**	−4
4	0.2	13 ± 1	29	81 ± 6**	4	68	12.4 ± 0.4	116 ± 2	102 ± 2**	−14
5	2.1	12 ± 1	40	33 ± 4*	0	21	2.3 ± 0.6	71 ± 1	73 ± 1	2
6	1.4	13 ± 1	38	26 ± 5**	27	13	1.6 ± 0.5	60 ± 1	56 ± 1**	−4
7	1.0	24 ± 5	25	72 ± 12**	0	48	4.6 ± 0.8	56 ± 1	51 ± 1**	−5
8	2.1	1 ± 2	75	16 ± 9*	38	16	2.4 ± 0.6	90 ± 1	89 ± 1	−1
9	0.8	35 ± 5	6	74 ± 9**	0	39	8.2 ± 0.6	NA	NA	NA
Mean	1.0 ± 0.2	18 ± 4	31 ± 7	68 ± 15*	8 ± 4*	50 ± 13	6.7 ± 1.4	74 ± 7	70 ± 6*	−4 ± 2

NA: not measured;

^*p < 0.05,

^**p < 0.001 from baseline (21% O₂).

^αMean pO₂ over all voxels in the two repeated baseline measurements.

^βmean highest pO₂ observed in all voxels during 5 oxygen breathing measurements.

GRE R₂*

Baseline maps revealed distinct heterogeneity with R₂* ranging from 6 to 450 s⁻¹ (T₂*: 2 – 166 ms, Figure 3). In response to oxygen challenge, a small, but significant decrease in R₂* was seen predominately in the tumor periphery (p < 0.001; Figure 3). There was no correlation between baseline R₂* and ΔR₂* on a voxel-by-voxel basis (r² = 0.1; Figure 3B). For the group of eight tumors, a significant decrease in mean R₂* (suggesting increased oxyhemoglobin level) was found (mean ΔR₂* = −4 ± 2, p < 0.05, Table 1). One tumor (no. 5; Table 1) showed contrary behavior with increased mean R₂* with oxygen breathing. A very close correlation was observed between the mean R₂* values during air versus oxygen breathing (Figure 3C). A general trend was observed when the mean BOLD response for individual tumors was compared with ΔR₂* (Figure 3D). In essence large ΔR₂* was associated with large ΔSI, and an increase in R₂* coincided with a small signal change. However, most tumors showed a ΔR₂* of −4 ms and this was associated with a large range in BOLD signal change.

Figure 3. Response of R₂* to oxygen breathing.

A. R₂* maps showed heterogeneous baseline R₂* values for tumor #2. Top left: R₂* while breathing air; Center R₂* after 5 min breathing oxygen; right difference map obtained by subtracting the oxygen map from the air map (mean ΔR₂* = −2.6 s⁻¹).

B. A voxel-voxel comparison (n = 2056) between baseline R₂* and ΔR₂* showed a poor correlation (r² = 0.1).

C. Comparison of mean tumor R₂* measured while rat breathed oxygen versus air showed a strong correlation (r²>0.96).

D. Comparison of mean signal change in T₂*-weighted image versus change in mean R₂* in tumors accompanying change in gas from air to oxygen breathing. No linear correlation was observed, but greater change in R₂* was accompanied by changes in signal intensity.

¹⁹F Tumor tissue oximetry – FREDOM

pO₂ maps showed a range of baseline pO₂ values and a heterogeneous response to oxygen breathing (Figure 4, Table 1). Oxygenation appeared clustered with higher pO₂ regions appearing close to the periphery when overlaid on the anatomical ¹H MR images of the corresponding tumor slices. Time course of pO₂ dynamics showed differential response in both rate and magnitude (Figure 4B). For the group of 9 tumors, baseline pO₂ varied from essentially hypoxic (0.3 torr) to well oxygenated (36 torr; Table 1). With respect to oxygen challenge, mean pO₂ increased significantly in all the tumors (ΔpO₂ = 50 torr; Table 1) and hypoxic fractions (< 10 torr) decreased significantly, from a mean tumor baseline 31% to 8% (p<0.05). In most tumors (7 of 9) the HF₁₀ was essentially eliminated (<5% residual), but in two tumors a substantial hypoxic fraction remained albeit considerably diminished compared with baseline. Histograms for the pooled individual voxels (n= 265) from the nine tumors in response to oxygen challenge revealed significant increase in pO₂ to a mean (75 ± 4 torr) and median (63 torr) (p < 0.001) (Figure 5A, B). Hypoxic fractions HF₅ (<5 torr) and HF₁₀ (<10 torr) decreased significantly from baseline values of 18% and 34% to 8% and 10%, respectively (p < 0.01). In common with previous observations in this tumor line baseline pO₂ tended to decrease with tumor volume (r²>0.52) and as a corollary HF₁₀ increased with tumor volume (r²>0.49) (19,20). A strong inverse correlation was found between tumor mean pO₂ and HF₁₀ (Figure 5C). The change in pO₂ (ΔpO₂) tended to increase for tumors with higher baseline pO₂ (r²>0.4), but the relationship was much stronger when comparing pO₂ during oxygen breathing versus baseline pO₂ (r²>0.6; Figure 5D).

Figure 4. Variation of pO₂ with oxygen challenge.

A. pO₂ maps obtained at successive times overlaid on T₁-weighted ¹H images of two tumors (#2 and #6). A range of pO₂ values was observed in both tumors under baseline conditions. In response to breathing oxygen, all the individual locations (34 voxels) in tumor # 2 (upper row) responded significantly and became well oxygenated. By contrast, some of initially hypoxic regions in tumor #6 (lower row) remained hypoxic, while others became well oxygenated.

B. Variation in mean pO₂ of each tumor. Tumor #2 (○; mean baseline = 17 torr) showed significantly increased pO₂ within 7 min of oxygen breathing, and continued to increase reaching 76 torr during the final measurement (49 min). Tumor #6 (■, mean baseline pO₂ = 13 torr) reached a peak value (26 torr) after 14 min, but the settled back to a lower level. * p < 0.05; ** p < 0.001.

Figure 5. Tumor oxygen tension distribution.

A. Pooled pO₂ values for individual regions (265) from the nine tumors showed a range of baseline pO₂ values from hypoxia (<5 torr) to 55 torr, with a mean (x) = 15 ± 1 torr and median (m) = 13.1 torr, while rats breathed air. Binning is based on ranges, e.g., 10 refers to 5 ≤ pO₂ <10 torr.

B. Oxygen breathing produced a significant increase in pO₂ with mean (75 ± 4 torr) and median (63 torr) (p < 0.001). Hypoxic fractions HF₅ (<5 torr) and HF₁₀ (<10 torr) decreased significantly from baseline values of 18% and 34% to 8% and 10%, respectively (p < 0.01).

C. Correlation between mean baseline pO₂ and hypoxic fraction showed inverse relationship (r² > 0.87).

D. Dependence of pO₂ achieved with oxygen breathing on baseline pO₂ (r² > 0.6)

Correlation of pO₂ with ¹H BOLD

For the group of nine tumors, a significant linear correlation was found between mean ΔpO₂ and ΔSI, detected by ¹⁹F oximetry and ¹H EPI BOLD (r² > 0.7, p < 0.001; Figure 6). However, there was a weak correlation between baseline pO₂ and ΔSI (r² = 0.3, data not shown). Likewise comparison of ΔR₂* with baseline pO₂, maximum pO₂ during oxygen breathing or ΔpO₂ indicated no correlations (r²<0.01). Every tumor exhibiting a large BOLD response (ΔSI>3%) had a negligible residual hypoxic fraction (HF₁₀<5%) during oxygen breathing. In two of three tumors showing a small BOLD response a large HF₁₀ remained (Figure 7).

Figure 6. Correlation of pO₂ and BOLD responses to oxygen challenge.

A significant linear correlation was found between mean increase in tissue pO₂ (ΔpO₂) and mean spin echo planar BOLD signal increase in the nine tumors with respect to oxygen intervention (r² > 0.7; p < 0.001).

Figure 7. Assessment of residual hypoxic fraction.

Comparison of the final hypoxic fraction (HF₁₀) with oxygen breathing as a function of BOLD signal response (ΔSI) suggests strong predictive value. For most tumors (6 of 9) a large BOLD response coincided with low residual HF₁₀. A small BOLD response indicated a large residual hypoxic fraction in 2 of 3 tumors.

Discussion

Oxygenation in 13762NF rat breast tumors was found to cover a considerable range with both intra- and inter-tumor heterogeneity, but a tendency towards lower pO₂ and greater hypoxia in larger tumors, as also reported previously by us (19,20). Likewise tumor response to modulation by hyperoxic gas breathing, which resulted in increased pO₂ and reduced residual hypoxic fraction was in line with previous observations (19–21). Here, we have undertaken consecutive ¹H MRI BOLD and ¹⁹F MRI oximetry investigations to examine potential correlations. A strong correlation was found between mean BOLD signal response (ΔSI) and change in mean pO₂ (Figure 6) supporting previous observations in various tumor types reported by others (4,11,17). More significantly, a small BOLD response usually indicates a substantial residual hypoxic fraction, while a large BOLD response to breathing oxygen indicates that the tumor is well oxygenated or becomes well oxygenated (Figure 7).

Quantitative oximetry (absolute pO₂ values) has been shown to predict for local recurrence and disease free survival in several human cancers (notably, cervical and head and neck (22,23)). To date, electrodes have provided the only quantitative clinical pO₂ measurements in tumors, and electrodes are not only invasive, but the Eppendorf Histograph cannot easily show changes in pO₂ with respect to interventions. Furthermore, the Histograph is no longer commercially available. ¹⁹F MRI based on perfluorocarbon reporters has been used to measure pO₂ in the human eye (24), but it remains generally restricted to pre-clinical animal studies (2) for two fundamental reasons: lack of access to clinical ¹⁹F MRI capabilities and lack of FDA approval for human use of PFCs. ESR can also measure pO₂ distributions in rodent tumors (16,25,26), but while it has been used to measure oxygenation in humans based on India ink tattoos, it is also handicapped by lack of clinical instrumentation.

BOLD contrast MRI is an attractive surrogate for clinical pO₂ measurements, since endogenous hemoglobin itself serves as the reporter molecule. A few studies of human tumor BOLD have been presented (5,27–29), but further validation relative to other techniques is of the utmost importance. Fundamental reports from Thulborn (30) and Wright (31) together with applications of fMRI in the brain lay a strong foundation for vascular oxygen measurements. Several groups (3,4,6,8,10–14,16,17,26,32) have explored BOLD responses in diverse tumor types with respect to varying oxygen concentrations and carbogen. There is considerable evidence that signal changes in T₂*-weighted images reflect changes in pO₂. However, some investigations have shown a lack of direct correlation, e.g., Baudelet and Gallez (4) reported that a 10% change in signal could correspond to a small (< 25 torr) or large (approaching 100 torr) change in pO₂ in syngeneic FSA mouse tumors. Importantly, both represent large changes in pO₂ by radiobiological standards. Elas et al. (16) showed correlation between EPRI based on vascular trityl spin probe and sequential BOLD measurements in FSA tumors in mice. As with our study their two measurements were sequential rather than concomitant and they had the added complexity of coregistering images from separate modalities.

Fan et al (17) previously compared ¹⁹F MR oximetry with BOLD response on a voxel-by-voxel basis in R3230AC rat breast tumors. In common with many experiments they infused perfluorocarbon emulsion, which progressively sequestered in the tumor tissue, while clearing from the vasculature. They reported that ¹⁹F and ¹H measurement agreed in 65% of pixels, viz. when ¹⁹F MR showed increased pO₂, then ¹H MR linewidth decreased reflecting less deoxyhemoglobin, as assessed using HiSS (High Spatial/Spectral resolution T₂* sensitive) measurements. Correlations were even stronger for subsets of pixels selected as showing no pO₂ change or a BOLD change. Overall they concluded that regions identified as hypoxic tended to show a small BOLD response to carbogen inhalation in the R3230AC rat breast tumors and ¹H MRI gave very few “false positives” (17). Our results are in accord with these previous observations. Tumors exhibiting a large BOLD response also showed a greater mean pO₂ response (Figure 6). Most significantly, tumors exhibiting a large BOLD response (here, defined as > 4%) showed essentially no residual hypoxic fraction (HF₁₀< 5%; Figure 7). We found no false positives and only one false negative. While breathing air 7 of 9 tumors had HF₁₀ > 20% (Table 1). In all but two cases this fell below 5% with oxygen breathing, essentiality eliminating the hypoxic fraction. These remaining two tumors showed a particularly small BOLD effect (ΔSI <2.5%). Meanwhile six of seven responsive tumors exhibited a large BOLD effect (>4%). We do note that the 13762NF tumors show relatively low hypoxic fractions (HF₁₀) compared with many reports for tumors implanted in rodents. However, the values are closely in line with measurements reported using the Eppendorf Histograph in breast tumors in patients (33)

We should note differences between Karczmar’s team’s approach and ours. The use of systemically delivered PFC emulsions as reporter molecules tends to provide signal from well perfused tumor regions only. Indeed. Fan et al. (17) predominantly detected ¹⁹F signal from the tumor periphery, as also noted by others (34). While our current approach allowed interrogation of central tumor regions, as well as periphery, it required direct injection of HFB into the tumor, which is invasive and samples limited regions only. Thus, it is particularly reassuring that the two approaches provide commensurate results.

BOLD changes reflect vascular oxygenation, whereas FREDOM measures tissue pO₂. As expected, the BOLD changes occurred much more rapidly (seconds, Figure 2) than the pO₂ changes (minutes, Figure 4). In the future it will be interesting to examine ¹H T₁-weighted tissue water response or so-called TOLD (Tissue Oxygen Level Dependant) response, as reported by Matsumoto et al. (35), since then both vascular and tissue changes can be assessed by ¹H MRI. Matsumoto et al. (35) examined response to hyperbaric oxygen breathing, but others have reported T₁ changes associated with hyperoxic gas in normal tissues (36) and tumors (3,10). The kinetic response of the mean BOLD signal was consistent with previous global near infrared observations in this tumor type (20,37). Changes in deoxyhemoglobin concentration generally followed a biphasic time course as also seen in the BOLD response (Figure 2B). This probably represents rapid arteriolar oxygenation followed by more sluggish response in the distant parts of the vascular tree. Meanwhile pO₂ response was more sluggish with continued increase over 30 mins (Figure 4B) for a responsive tumor. These observations are in line with previous observations based on oxygen sensitive fiber optic probes and polarographic electrodes in this tumor type (38).

BOLD MRI or susceptibility-weighted R₂* measurement based on the intrinsic paramagnetic properties of deoxyhemoglobin have been increasingly applied to assess tumor vasculature (7,9). An increase in BOLD SI or a decrease in R₂* may be related to decreased blood deoxyhemoglobin. However, BOLD SI change is also related to several other factors, e.g., changes in tumor blood flow, volume, hematocrit and the ability of red blood cells to traverse tumor capillaries (3). Here, we applied a presaturation sequence across the whole tumor to minimize flow effects. It has been shown that BOLD MRI is probably less sensitive to changes in tumor oxygenation in regions containing very sparse vasculature, and hence, little deoxyhemoglobin (9). There may be concern that poorly vascularized tumors cannot show a measurable BOLD effect, due to lack of deoxyhemoglobin (9,14). As a corollary, we propose that poorly vascularized tumors will also be hypoxic. Thus, a small or absent BOLD effect will be indicative of hypoxia. Indeed, Rodrigues et al. (12) showed that well vascularized GH3 prolactinomas tended to have a much higher R₂* than sparsely vascularized Radiation-induced fibrosarcomas (RIF-1). GH3 tumors showed a large ΔR₂* in response to carbogen breathing, whereas the RIF-1 tumors showed essentially no change. Breathing carbogen enhanced the response of GH3 tumors to a single high dose of radiation (15 Gy), whereas there was no effect on RIF-1 tumors. Our range of R₂* values (54 – 116 s⁻¹) is commensurate with previous reports for animal tumors at 4.7 T (12). In terms of potential clinical applications the ability to stratify patients based on oxic or oxygenatable tumors versus hypoxic (and resistant to modulation) could be significant.

BOLD contrast is related to changes in local deoxyhemoglobin concentration. However, baseline R₂* reflects not only blood deoxyhemoglobin level, vascular blood flow and volume, but also local tissue architecture, i.e., cell density, edema, necrosis. Thus, the baseline R₂* and change in R₂* (ΔR₂*) likely depends on tumor type. Recently, Robinson et al. (9) showed heterogeneous inter-tumoral R₂* among a variety of tumors, in which the GH3 prolactinoma had the highest mean R₂* = 89 s⁻¹ and RIF-1 had the lowest value of 58 s⁻¹. In response to carbogen breathing, a significant decrease in R₂* (−23 s⁻¹) was found in the GH3 prolactinoma, whereas the RIF-1 fibrosarcomas showed a little increase in R₂* (1 s⁻¹). In the current study, seven out of eight tumors had a modest decrease in R₂* (mean = −4 s⁻¹), while R₂* increased in one tumor (# 5) in response to oxygen breathing. As expected, the largest ΔSI in response to oxygen breathing coincided with a large decrease in R₂* and the one tumor with increased R₂* showed small ΔSI. However, most tumors had similar ΔR₂* (about 4 s⁻¹) yet a large range of ΔSIs (Figure 3D). Generally, there was no significant correlation between baseline ΔR₂* values and R₂* (Figure 3B, Table. 1). There was a strong correlation between mean R₂* measured during air or oxygen breathing (Figure 3C).

In the past, we had used a thick section (essentially projection) for ¹⁹F MR oximetry (FREDOM) studies (2). The 2 mm thick slice used here allowed more satisfactory spatial correlation between ¹⁹F oximetry and ¹H BOLD contrast. The success of FREDOM depends largely on the signal-to-noise ratio (SNR) of the acquired images. No obvious decrease in SNR due to reduced slice thickness was seen in the current study, which we attribute to greater emphasis on injecting the HFB in a narrow plane and the ability to image oblique angles in the upgraded Varian Inova system. On average, 29 voxels (range 10 to 43) provided reliable pO₂ readings which were traceable throughout the oxygen challenge in each tumor.

We had hoped to correlate individual voxels in both ¹⁹F and ¹H EPI (FREDOM vs. BOLD). However, the directly corresponding voxels did not allow meaningful correlation. Signal voids were observed, which had only 10 to 25% as much ¹H signal as surrounding tumor (Figure 1A). Overlaying ¹⁹F on ¹H images showed that the low ¹H signal intensity regions corresponded with ¹⁹F signal (Figure 1B). In addition the R₂* values were typically an order of magnitude smaller (14 to 37 s⁻¹, as opposed to 100 s⁻¹) with smaller and sometimes opposite changes in R₂* compared with surrounding voxels in response to breathing oxygen. We have previously observed such signal voids in ¹H images of tumors with HFB (19) and the new alternative ¹H MRI pO₂ reporter under development hexamethyldisiloxane (39). While ¹H and ¹⁹F voxels did not allow direct correlation, we believe that the judicious placement of reporter molecule in central and peripheral locations can provide a representation of the whole tumor, and thus, correlation with non-labeled regions is reasonable. Importantly, pO₂ values and dynamics observed using FREDOM are highly consistent with other oximetry methods in rat breast and prostate tumors, such as electrodes (38,40), fiber optic probes (37,41) and immuno histochemistry (42).

Oxygen breathing produced a significant increase in ¹⁹F measured pO₂, EPI BOLD SI and decrease in GRE R₂* on the same section of tumor. Data indicated a strong correlation between %ΔSI and ΔpO₂ (Figure 6). Although a significant increase in global mean SI was found in all tumors by BOLD in the current study, voxel-by-voxel data analysis showed oxygen breathing induced signal loss in many voxels for each tumor, averaged as 20% of the total voxels. Similar findings have been observed in other tumor types by others and us (6,8,10,14). Indeed, Fan et al. (17) previously remarked that correlates between mean BOLD and pO₂ were stronger when the whole tumor was considered, as opposed to individual regions. In a slightly different context Baudelet et al (43) reported a better correlation between mean tumor BOLD signal response and vascular k_ep than for individual voxels.

The ability to identify hypoxia could have far-reaching implications for radiotherapy. We have previously shown correlations between tumor pO₂ measured by ¹⁹F MRI (FREDOM) and response to single high dose irradiation in Dunning prostate R3327-HI and AT1 tumors (44,45). Furthermore, we could correctly assess the ability to alter pO₂ and modulate response to irradiation. Karczmar’s team also demonstrated the ability of BOLD (viz. HiSS) to predict the relative efficacy of tumor oxygenating agents (46). ¹H MRI BOLD assessment would be far more practical in the clinic and we believe our results together with the previous report from Fan et al. (17) provide strong impetus for translation.