Correlations of noninvasive BOLD and TOLD MRI with pO₂ and relevance to tumor radiation response

Abstract

Purpose

To examine the potential use of BOLD (Blood Oxygenation Level Dependent) and TOLD (Tissue Oxygenation Level Dependent) contrast MRI to assess tumor oxygenation and predict radiation response.

Methods

BOLD and TOLD MRI were performed on Dunning R3327-AT1 rat prostate tumors during hyperoxic gas breathing challenge at 4.7 T. Animals were divided into two groups. In Group 1 (n=9), subsequent ¹⁹F MRI based on spin lattice relaxation of hexafluorobenzene reporter molecule provided quantitative oximetry for comparison. For Group 2 rats (n=13) growth delay following a single dose of 30 Gy was compared with pre-irradiation BOLD and TOLD assessments.

Results

Oxygen (100%O₂) and carbogen (CB; 95%O₂/5%CO₂) challenge elicited similar BOLD, TOLD and pO₂ responses. Strong correlations were observed between BOLD or R₂* response and quantitative 19F pO₂ measurements. TOLD response showed a general trend with weaker correlation. Irradiation caused a significant tumor growth delay and tumors with larger changes in TOLD and R₁ values upon oxygen breathing exhibited significantly increased tumor growth delay.

Conclusion

These results provide further insight into the relationships between oxygen sensitive (BOLD/TOLD) MRI and tumor pO₂. Moreover, a larger increase in R₁ response to hyperoxic gas challenge coincided with greater tumor growth delay following irradiation.

Introduction

Hypoxia is increasingly regarded as an important factor and potential prognostic biomarker for tumor aggressiveness and response to therapy (1). The fundamental studies of Gray et al. demonstrated the influence of hypoxia on response to radiation (2). More recently, polarographic needle electrode measurements of pO₂ distributions and hypoxic fractions in patients confirmed poor outcome for hypoxic tumors in several disease sites including cervix (3,4), head and neck (5), and prostate (6). There have also been correlations based on PET of radionuclide labeled hypoxia markers or biopsies and histology, but to date there is no routine non-invasive approach available in the clinic to assess tumor oxygenation and help predict treatment outcome (1).

Oxygen-sensitive MRI is attractive, since it is non-invasive and avoids the need for an exogenous reporter agent. Increasingly, reports suggest the feasibility of BOLD (Blood Oxygen Level Dependent) and TOLD (Tissue Oxygen Level Dependent) contrast MRI in patients (7–14). BOLD effects have been evaluated in numerous animal models (15–25) and several studies have demonstrated a correlation with measurements of pO₂ (18–20,24–26). BOLD MRI has attracted increasing interest as a non-invasive indicator of hypoxia based on intrinsic effective transverse relaxation rate (R₂* =1/T₂*) or response to breathing hyperoxic gas (ΔSI or ΔT₂*). It provides an indication of tumor blood oxygenation, but is also sensitive to local hematocrit, vascular volume, pH, flow, and vessel density (22,23,27). Meanwhile, TOLD MRI is based on T₁-weighted contrast. The measured R₁ (=1/T₁) is sensitive to tissue oxygenation and is distinct from BOLD imaging (28–30). Hyperoxic gas breathing has also been evaluated in several disease sites using T₁W MRI (29). We hypothesized that both techniques provide complimentary information in assessing tumor oxygenation.

These reports prompted us to explore BOLD and TOLD measurements based on differences in signal intensity in T₁- and T₂*-weighted images, as well as relaxation rates (R₁ and R₂*) to predict radiation response. We chose 19F MRI relaxometry based on the reporter molecule hexafluorobenzene (HFB) as a correlative standard, since it provides quantitative assessment of partial oxygen pressure (pO₂) in vivo (31). The method is finding increasing use for assessing oxygen dynamics and is convenient for correlation with proton MRI (18,32–34). Here, we examined BOLD and TOLD MRI as surrogate biomarkers to assess tumor oxygenation, as correlated with quantitative pO₂ measurements. We also tested BOLD and TOLD MRI as a prognostic tool to predict radiation treatment outcome of a well-characterized Dunning R3327 rat prostate cancer: AT1.

Materials and methods

Animal model

This study was approved by the Institutional Animal Care and Use Committee. Twenty two male Copenhagen rats were implanted with syngeneic Dunning R3327-AT1 prostate tumors. The AT1 is an anaplastic tumor with a volume doubling time of 5.2 days (35) and TCD₅₀ (single radiation dose for 50% tumor control probability) of 75.7 Gy (36). Tumor tissue fragments were implanted subcutaneously in the thigh and tumor volume was calculated as V= (π/6)*a*b*c, where a, b, c, are the three orthogonal diameters measured using a caliper.

Tumor oximetry

MRI experiments were performed two to five weeks post tumor implantation depending on the desired tumor size. MRI used a horizontal bore 4.7-T magnet system (Varian, Palo Alto, CA) with a single-turn solenoid coil tunable to ¹H or ¹⁹F. Animals were anesthetized with isoflurane (1.5%) in air (1 L/min) delivered via a nose cone and kept warm (37 °C) using a circulating warm water blanket. T₂-weighted (T₂W) images with 0.3 mm in-plane resolution were acquired using a Fast Spin Echo (FSEMS) sequence: TR/TE_eff =2000/48 ms, Echo Train Length (ETL) =8, FOV=40×40 mm with 128×128 acquisition matrix, NEX=8, 1 mm slice thickness without gap, and 15 slices acquired transaxially to the leg.

BOLD images were acquired using a multi-echo gradient echo sequence (MGEMS): TR =150 ms, 10 echo time values ranging from 6 to 69 ms with echo spacing of 7 ms, flip angle (FA) = 20°, acquisition time 21 s. TOLD images were acquired using a gradient echo sequence (GEMS): TR/TE= 30/5 ms, FA=45°, acquisition time 3 s. The two sequences were interleaved in an automated fashion to obtain BOLD and TOLD (IBT) MR images using the same FOV, matrix size and imaging plane, as for the high resolution T₂W images to allow effective co-registration

Animals were divided into two groups. For Group 1 rats (n=9), IBT was acquired while animals breathed air (8 minutes baseline) followed by oxygen (100% O₂) for about 30 minutes. To compare the effect of hyperoxic gas challenge, this was followed by re-equilibration breathing air for approximately 10 minutes followed by carbogen (CB, 95% O₂, 5% CO₂) for about 30 minutes. In addition, quantitative pO₂ measurements were obtained using the FREDOM (Fluorocarbon Relaxometry using Echo planar imaging for Dynamic Oxygen Mapping) approach approximately 24 hours after IBT (31). Hexafluorobenzene (HFB, 50–100 μl) was injected directly into the tumor in a fan pattern to ensure distribution of HFB throughout a plane of the tumor approximately coinciding with the IBT study. Pulse burst saturation recovery echo planar imaging (EPI) was used to measure the spin-lattice relaxation rate, R₁, of HFB by arraying 14 delay times (τ), as described previously (31). Five pO₂ measurements were obtained for each gas: baseline air, O₂, return to air and finally CB. The FREDOM parameters were: TR=50 ms, TE=21 ms, τ ranges from 0.2 to 90 s, NEX =1 to 12 (depending on τ), FOV=40×40 mm with 32×32 acquisition matrix, slice thickness 10 mm, giving an acquisition time of 6½ minutes.

Rats in Group 2 (n=13) underwent only ¹H MRI using the same parameters as mentioned above. Based on results for Group 1, only 100% O₂ challenge was examined (i.e. no CB), and quantitative T₁ measurements were additionally acquired using a spin echo sequence (SEMS): TE= 20 ms, TR= 0.1, 0.2, 0.3, 0.5, 0.7, 0.9, 1.5, 2.5, 3.5 s, FOV=40×40 mm with 64×64 acquisition matrix, though this was zero-filled to 128×128 to allow voxel-wise comparison with IBT. Group 2 rats entered the radiation treatment protocol within 24 hrs of MRI as follows.

Radiation treatment

Radiation was delivered using a small animal X-ray irradiator (XRAD 225Cx, Precision X-Ray, Inc., North Branford, CT) operating at 225 kV and 13 mA, producing a dose rate of 3.5 Gy/min delivering a total dose of 30 Gy. An image guidance system utilizing a digital imaging panel for digital radiography, fluoroscopy, and cone-beam CT (CBCT) was used to ensure accurate localization to the tumor. Absolute dose calibration was performed in accordance with the recommendations of the AAPM TG-61 protocol (37).

Rats were divided into three sub-groups: Group 2a (n=2) were not irradiated (control), Group 2b (n=5) breathed air during irradiation and Group 2c (n=6) breathed 100% O₂ for about 15 minutes before irradiation and during irradiation. Tumor volume (V), measured with calipers, was normalized to the day of irradiation (V₀) and response was evaluated as the time to quadruple (T₄). One animal died before T₄ and the day of death was taken as T₄.

Data analysis

MRI data were processed using Matlab (MathWorks Inc., Natick, MA). Changes in signal intensity (%ΔSI) with respect to O₂ and CB challenge were calculated from the MGEMS and GEMS sequences. For BOLD images, a single TE=20 ms was selected. Signal intensity change was calculated as:

$$\mathrm{\Delta }SI=\frac{{SI}_t-{SI}_b}{{SI}_b}·100\%,$$ (1)

where SI_b is the mean baseline signal intensity during air breathing and SI_t is the mean signal intensity with 100% O₂ or CB inhalation excluding the first minute of transition to new gas.

T₂* maps were generated by fitting the multi-echo T₂*W image signal intensity (MGEMS) to TE, as a monoexponential function on a voxel-by-voxel basis:

$$SI=S_0·e^{-TE/T_2^{\ast }}+k,$$ (2)

where SI is signal intensity at each echo time (TE), S₀ represents the original magnetization and k is a constant. Initial estimates for S₀ and T₂* were calculated by first solving Equation (2) based on SI at the first two TE values and these preliminary estimates were used as initial values to fit Equation (2). The change of T₂* due to 100% O₂/CB challenge was then calculated as:

$$\mathrm{\Delta }{{\text{T}}_2}^{\ast }={{{\text{T}}_2}^{\ast }}_{\text{t}}-{{{\text{T}}_2}^{\ast }}_{\text{b}},$$ (3)

where T₂*_b represents mean baseline during air breathing, and T₂*_t is the mean value with 100% O₂/CB inhalation. Similarly, change in R₂* was calculated as (ΔR₂* =R₂*_t − R₂*_b).

Tumor pO₂ was estimated on a voxel-by-voxel basis using a monoexponential function, and pO₂ (torr) determined using the relationship (31):

$$p{O}_2=\frac{R_1-0.0835}{0.001876}.$$ (4)

Voxel-by-voxel measurements allowed calculation of HF₅ (fraction of voxels with pO₂<5 torr). Only voxels with consistently reliable fitting (coefficient of determination, R²>0.95) throughout the experiment were selected for further statistical analysis (the number of acceptable voxels ranged from 37 to 99).

Statistical Analysis

Mean baseline SI of the T₂*W and T₁W images was calculated for each ROI and compared with 100% O₂ and CB breathing. An average of all data points was used to determine the response to challenge excluding the first minute of transition to new gas. Significance of response to 100% O₂ and CB challenge was assessed using Student’s t-tests.

Results

Tumors showed distinct heterogeneity in terms of baseline T₂* and pO₂, while breathing air, with higher pO₂ near the periphery. All measurements were stable during air breathing (8 minutes) with mean values and standard deviations presented for each tumor in Table 1. Mean baseline pO₂ ranged from −0.5±0.6 torr to 17.9±1.3 torr for individual tumors with a mean of 5.8±6.5 torr (median 4.1 torr) for the small tumors (<1 cm³), while larger tumors were more hypoxic (−0.7±1.6 torr to 10.1±0.4 torr, mean=3.1±6.0 torr, median=0.1 torr; size >2.9 cm³, Table 1). Hypoxic fraction (HF₅) was more distinct and ranged from 40 to 80% for the small tumors and was over 95% for the larger tumors (Table 2). Spatial response of each parameter to gas challenge is presented for one tumor (#4; Fig. 1) and dynamic changes are compared for two representative tumors showing small and large responses respectively (Fig. 2).

Table 1.

Transverse relaxation times and pO₂ of Dunning prostate R3327-AT1 tumors

Rat #	Size (cm3)	T2* (air) ms	T2* (O2) ms	pO2 (air) torr	pO2 (O2) torr
1	0.7	30.6±0.9	32.6±1.8	3.2±1.0	17.3±1.3
2	0.6	28.7±0.5	31.6±1.0	17.9±1.3	46.9±6.0
3	0.6	20.5±0.2	21.0±0.2	7.1±0.5	17.1±2.5
4	0.5	21.1±0.1	23.5±0.2	5.0±1.0	40.1±2.6
5	0.5	24.7±0.8	25.9±0.6	2.0±3.1	9.1±0.3
6	0.7	23.8±0.4	24.1±0.3	−0.5±0.6	3.1±2.3
7	3.0	18.4±0.6	16.4±0.9	−0.7±1.6	−1.0±1.6†
8	2.9	27.1±2.0	26.2±1.8	10.1±0.4	12.6±1.1
9	8.8	39.8±1.5	44.6±2.0	0.1±0.3	−1.1±0.4
	Mean	26.1±6.5	27.3±8.2	4.9±6.1	16.3±16.9

T₂* and pO₂ values are means for each tumor and responses were significant (p<0.01) except ^† (NS).

Table 2.

Effect of oxygen and carbogen breathing on BOLD and TOLD MRI

		BOLD (%ΔSI)		ΔT2*(ms)		TOLD (%ΔSI)		ΔpO2 (torr)		HF5 (%)
Rat #	Size (cm3)	Oxygen	CB	Oxygen	CB	Oxygen	CB	Oxygen	CB	Oxygen	CB
1	0.7	2.7±1.2*	0.5±1.2	2.0±1.9*	3.4±2.65*	−1.0±2.4	−3.9±2.3*	14.1±7.4*	15.5±3.5*	79 to 62	67 to 56
2	0.6	8.2±2.9*	3.0±0.9*	2.9±1.3*	0.9±0.7*	−0.4±1.4	−2.8±2.0*	29.0±7.1*	13.8±3.8*	43 to 38	28 to 28
3	0.6	5.3±1.8*	4.3±1.19*	0.5±0.4*	0.1±0.4	0.5±1.1	−1.5±2.2*	10.0±3.6*	7.3±1.3*	53 to 41	50 to 48
4	0.5	18.7±5.3*	20.0±0.8*	2.4±0.6*	2.5±0.2*	7.3±2.2*	8.7±1.9*	35.2±2.7*	36.4±1.6*	66 to 29	45 to 31
5	0.5	4.3±1.4*	4.7±1.5*	1.2±0.7*	0.6±0.7*	1.2±1.0*	1.6±0.7*	7.1±1.3*	15.2±1.2*	43 to 17	27 to 14
6	0.7	1.2±2.1	3.8±1.4*	0.3±0.5	1.1±0.6*	0.2±0.8	2.1±1.3*	3.6±1.0*	6.8±3.6*	63 to 41	49 to 32
7	3.0	−1.7±0.7*	−1.4±0.8*	−1.6±0.9*	−2.1±1.3*	−1.1±0.8*	−3.1±1.2*	−0.3±1.5	−3.0±0.3	97 to 95	97 to 97
8	2.9	−1.4±1.1*	−1.5±1.3*	−0.9±2.4	−2.8±4.0*	−0.3±0.9	−1.9±1.3*	2.6±1.0*	4.8±0.5*	95 to 92	94 to 92
9	8.8	2.8±0.8*	1.7±0.7*	4.8±2.3*	6.9±2.8*	0.7±0.5*	−0.3±0.7	−1.2±0.4*	−0.6±0.5	96 to 96	96 to 87
Average		4.4±6.2	3.8±6.6	1.3±2.0	1.2±2.9	0.8±2.5	−0.1±3.9	11±13	11±12	71 to 57	61 to 54

Quantitative pO₂ measurements were obtained using FREDOM and HF₅ was calculated.

^*p<0.01

Figure 1. Oxygen sensitive MRI.

Semi-quantitative BOLD and TOLD, and quantitative T₂* and pO₂ response maps overlaid on high resolution T₂W image of a small Dunning prostate R3327-AT1 tumor (#4, 0.5 cm³) with respect to oxygen challenge, return to air and carbogen breathing. Response maps compare final image with each gas versus baseline. ¹⁹F oximetry was performed following direct intratumoral injection of hexafluorobenzene reporter the next day. Each map shows considerable intratumoral heterogeneity.

Figure 2. Changes in tumor oxygenation in response to oxygen breathing.

Mean changes in BOLD (red ○), TOLD (black □), %ΔT₂* (green △) for two tumors exhibiting very different response to A) 100% O₂ and B) CB challenge. Tumor #4 (open symbols) and # 6 (solid symbols). Corresponding mean changes in pO₂ to C) 100% O₂ and D) CB challenge. The hypoxic tumor showed little response by all measures, whereas the better oxygenated tumor was rapidly responsive.

Tumor response to 100% O₂/CB challenge was diverse, but the oxygen sensitive MRI parameters were highly consistent, as shown for representative tumors in Fig. 2. Tumor #4 showed large BOLD and TOLD responses (Fig. 2A and 2B) with corresponding large increase in pO₂ (Fig. 2C and 2D), whereas tumor #6 showed smaller response in each parameter. Rapid BOLD signal increase and T₂* enhancement were observed followed by slower TOLD response in tumor #4. By comparison baseline pO₂=5.0±1.0 torr (HF₅ =66%) increased by pO₂=35.2±2.7 torr (HF₅ =29%). Meanwhile, tumor #6 showed minimal BOLD and TOLD responses, lower baseline pO₂ and smaller changes with 100% O₂/CB breathing.

The mean change in SI (%ΔSI) for BOLD, TOLD, %ΔT₂*, and pO₂ for the group of nine AT1 tumors (Group 1) is presented in Fig. 3. Following the oxygen challenge, the gas was switched back to air and BOLD and TOLD measurements returned to stable baseline. The measurements were repeated while switching the gas to CB (Fig. 3B) and responses were very similar. FREDOM was performed on the second day to obtain corresponding pO₂ measurements (Table 1). Upon oxygen breathing the pO₂ increased to varying extents (mean ΔpO₂ = 11.1±13.0 torr, Table 2), followed by slow decrease to baseline during air breathing. CB showed similar pO₂ enhancement (Figs. 1, 3 and Table 2). Converting R₁ relaxation values to pO₂ led to some apparently negative pO₂ values. The most negative pO₂ in an individual voxel was −9.3 torr and the median negative value for all the tumors ranged from −1.5 to −8.5 torr giving an average negative hypoxic fraction of 22 (± 10)%. Such negative values may be interpreted as having a high likelihood of being hypoxic.

Figure 3. Oxygen dynamics assessed by MRI.

Mean BOLD (T₂*W %ΔSI, blue ◆), TOLD (T₁W %ΔSI, yellow ▲), and %ΔT₂* (green ●) signal response to the hyperoxic gas challenge, together with pO₂ (red) for Group 1 tumors (n=9). Baseline measurements were quite stable during air breathing and changed to various extents in response to A) oxygen (O₂), or B) carbogen (CB) breathing. Arrows indicate the transition to each gas. Proton MRI measurements were observed in an interleaved manner and corresponding ¹⁹F oximetry was performed following injection of hexafluorobenzene reporter the next day.

Strong correlations were observed between ΔpO₂ and BOLD (R²>0.82, p<0.001) for eight animals during CB and O₂ breathing (Fig. 4A). One tumor (#9 ■) was excluded from the analysis, because the HFB was primarily injected into the center of this large tumor (8.8 cm³) and therefore did not reflect the whole tumor. Somewhat weaker relationships were found between ΔpO₂ and TOLD (R²>0.47, p<0.004, Fig. 4B) and ΔpO₂ and ΔR₂* (R2>0.58, p<0.001, though R2>0.75 for O₂ challenge alone, Fig. 4C). ΔpO₂ tended to increase with baseline pO₂ (R²>0.47, and for O₂ challenge alone R²>0.61, Fig. 4D). We note that tumor #4 showed exceptionally large responses of each parameter. Lest this one tumor bias the relationships, we also tested correlations in the absence of tumor #4. Now the correlations between ΔpO₂ and BOLD weakened somewhat (R²>0.59), ΔpO₂ and R₂* was essentially unchanged (R2>0.5), ΔpO₂ and TOLD was non-existent (R²<0.01), but pO₂ during hyperoxic gas breathing and pO₂ actually increased (R²>0.72).

Figure 4. Correlation between BOLD, TOLD, and pO₂ for 9 AT1 tumors.

Graphs show correlations between ΔpO₂ and (A) BOLD (B) TOLD and C) ΔR₂* in response to oxygen (●) and carbogen (○) challenges (Group 1 tumors). The HFB injection in one tumor (■) was in the center and the data were excluded from the correlations, as being unrepresentative of the whole tumor. D) Correlation between mean pO₂ during oxygen (●) or carbogen (○) breathing and baseline air-breathing pO₂.

Mean responses for BOLD, TOLD, ΔT₂* and ΔpO₂ to the two gas challenges were very similar (Table 2). Moreover, strong correlations were observed between the responses to the two gases for individual tumors (Fig. 5). Regional responses and local effects may ultimately be more important than mean values across a whole tumor and therefore the relative response at individual locations (voxels) were compared and are presented for tumor #4 (Fig. 6). Substantial concordance was observed for the two gases and notably pO₂ was very similar including the crucial hypoxic fractions (Fig. 6c).

Figure 5. Comparison of response to oxygen and carbogen for 9 AT1 tumors.

A) Correlation between mean BOLD %ΔSI (◆), Δ%SI TOLD (○), and %ΔT₂* (▲) responses for each tumor in Group 1 to oxygen and carbogen challenge assessed non-invasively using ¹H MRI. B) Correlation between oxygen and carbogen in modulating tumor pO₂ assessed using ¹⁹F MRI relaxometry following direct intratumoral injection of HFB reporter. C) Correlation between residual hypoxic fraction (pO₂ < 5 torr) in individual tumors, while breathing oxygen or carbogen.

Figure 6. Comparison of response to oxygen and carbogen for individual sites in a representative tumor.

Correlations observed between response to oxygen and carbogen challenge for individual voxels in a representative tumor (#4) based on A) T₂W signal response (BOLD), B) T₁W signal response (TOLD), and C) pO₂ confirming similarity of the effects of both gases.

For Group 2 (13 AT1 tumors) IBT studies were performed with respect to O₂ only and quantitative T₁ measurements were added. Two tumors served as unirradiated controls (Group 2a, mean V₀=0.4±0.1 cm3). The other tumors received a 30 Gy irradiation dose, while rats breathed air (n=5; V₀=0.5±0.3 cm3) or 100% O₂ (n=6; V₀=0.5±0.2 cm3). Growth was measured up to four times the initial tumor size or 80 days (Fig. 7). One rat died before the tumor reached T₄ and this date was used as T₄. Untreated controls had the fastest growth rate with a T₄ of 8±1 days (Fig. 8A). The irradiated tumors showed significantly slower growth rates with T₄ =39±11 days (Group 2b) and 50±18 days (Group 2c). A correlation was found between T₄ for the animals breathing 100% O₂ during radiation and pre-irradiation TOLD (%ΔSI, Fig. 8B; R²> 0.79; p<0.05) or R₁ (Fig. 8C; R²> 0.57) response to oxygen challenge. Significantly, Group 2c tumors could also be stratified by magnitude of response, where large responders (ΔSI > 1.5% and ΔR₁ (R_1(air) − R_1(O2) > 0.005 s⁻¹) showed significantly larger T₄ ~70 days vs. 35 days (p< 0.05). No such correlation was seen for semi-quantitative ΔSI or quantitative R₂* BOLD responses.

Figure 7. Influence of radiation on Dunning prostate R3327-AT1 tumor growth.

Growth curves for the 13 individual AT1 tumors: sham-irradiated tumors (Group 2a, n=2, yellow filled triangles), and single dose 30 Gy, while rats breathed air (Group 2b, n=5, filled circles) or oxygen (Group 2c, n=6, red filled squares).

Figure 8. Tumor growth delay.

A) Effect of air and oxygen breathing on radiation response assessed by time for tumor to quadruple is size (T₄); * p<0.01. Correlation between (T₄) and B) TOLD or C) ΔR₁ for those tumors irradiated during oxygen breathing (n=6).

Discussion

This study showed distinct trends between BOLD, TOLD and absolute pO₂ responses to hyperoxic gas breathing challenge. Quantitative BOLD and TOLD measurements were readily implemented and gave consistent results. However, TOLD response was more closely related to tumor growth delay following single high dose irradiation.

The smaller tumors (volume <1 cm³) were significantly better oxygenated than the larger ones (Tables 1 and 2) in accord with previous studies regarding AT1 tumors (38–40), although individual tumors showed a wide range of hypoxic fractions. Generally, the smaller tumors showed larger pO₂ changes upon 100% O₂ or CB breathing compared to larger tumors. Likewise, smaller tumors tended to exhibit larger changes in BOLD (6.3±6.3%), TOLD (1±3.7%), and T₂* (1.5±1.1 ms) compared to larger tumors BOLD (−0.3±2.0%), TOLD (−0.9±1.3%), and T₂* (0.7±4.1 ms), although these differences did not reach statistical significance for the groups. Tumors found to have low baseline pO₂ tended to exhibit smaller pO₂ response to 100% O₂ or CB challenge (Table 1, Figs. 2, 4) and as a corollary those with greater baseline hypoxic fraction retained a large hypoxic fraction (Table 2). Each of the oxygen sensitive parameters behaved consistently, whereby tumors exhibiting larger BOLD, TOLD and T₂* response to 100% O₂ and CB challenge showed high baseline pO₂ when examined by ¹⁹F MRI the following day (Figs. 2, 4).

Negative values were observed in FREDOM measurements on occasion accounting for 8.2% to 38.8% of total pixels. Although negative pO₂ values are physically impossible, we believe they arise from noise generating uncertainties in the curve fitting. Provided that errors and uncertainties are considered, such negative values are consistent with high probability of actual values close to zero (i.e. hypoxic). Errors in the original calibration could also be the cause of apparently negative pO_2. We do note slightly different calibrations reported for hexafluorobenzene at 4.7 T and 37 °C with the anoxic intercept reported to be as low as 0.074 s⁻¹ (31,41). During many years of investigations we have found the calibration used here to provide highly consistent results, although we do note that the reported calibration curve and phantom showed a range of relaxation rates for 0% O₂ and thus, there is clearly some uncertainty (42). We also note that the Eppendorf Histograph, sometimes considered as the “gold standard” for in vivo oximetry often indicated some negative pO₂ values (43). In this study, we correlated ΔpO₂ with BOLD and TOLD parameters. Therefore, the effect of errors in actual pO₂ measurement, although valid, is minimized.

Strong correlations were found between both quantitative and semi-quantitative BOLD responses and ΔpO₂ measurements (Figs. 4A, C) in line with previous reports for rat mammary tumors 13762NF (18) and R3230AC (44). Indeed, general correlations between BOLD response (variously assessed as change in signal intensity, change in local linewidth or quantitative T₂*) and change in pO₂ have been reported in diverse tumors based on various oximetry techniques (e.g., ESR (19,25), fiber optic probes (45), electrodes (26) and ¹⁹F MRI (18,24)). The relationship is often far from linear, presumably reflecting the O₂-hemoglobin binding characteristics (22). Some reports indicate that T₂* alone may be indicative of hypoxia in tumors (46,47), though vascular extent can exert a strong influence (23,48). As such, adding TOLD measurements ought to help identify hypoxia irrespective of vasculature, since T₁ is reported to be directly sensitive to pO₂ (30).

Combined BOLD and TOLD was recently reported in rabbit tumors (49) as well as normal tissues and tumors in patients (29). While the BOLD response is generally larger (often approaching 2–3 fold, c.f., Figs. 2A, 3 and Table 2) and hence easier to identify, TOLD at a minimum provides validation for changes in oxygenation. More significantly, TOLD response was found here to be related to radiation response, whereas no such stratification or correlation was seen with BOLD. T₁ measurements typically require longer acquisition times and application of sequential pulse sequences, as presented here, may significantly increase study times, making it impractical for the clinic. In this regard innovative simultaneous measurements are being developed (50,51). In terms of potential clinical applications the signal to noise of individual images would be expected to be lower at 1.5 or 3 T in a large bore magnet, but this may be compensated by larger voxels. Relaxivity of oxygen in blood and tissue is reported to show complex effects on R₁ and R₂*, but effective studies of R₁ and R₂* response to oxygen breathing have been reported recently in human tumors (51).

Correlation between oxygen sensitive parameters is interesting, but ultimately the ability to predict therapeutic response is vital. As such the observation (Fig. 8) that tumors exhibiting significantly greater TOLD response (ΔR₁ and ΔSI) had significantly longer tumor growth delay is promising. Specifically, those tumors showing small TOLD response to 100% O₂ challenge (<1.5%ΔSI; <0.005 s⁻¹ change in R₁) showed no benefit compared with tumors in air-breathing rats. Binary stratification may be valuable in terms of defining the utility of O₂-breathing during irradiation. It is also interesting to note the direct linear relationship between the time to quadruple in size (T₄) and %ΔSI or ΔR₁ (Fig. 8B, C). A similar result was reported in orthotopic C6 gliomas in rats subject to single high dose irradiation (52).

Others reported that, GH3 prolactinomas, which are highly vascularized, showed a significant BOLD response with respect to CB breathing challenge and experienced increased tumor growth delay in response to single dose radiation (17). There appeared to be a trend between magnitude of response (ΔR₂*) and tumor volume 7 days after irradiation, although this was not explicitly stated. Meanwhile, RIF-1 tumors growing in mice showed minimal BOLD response and no benefit from CB, but a two-fold range in tumor volume 7 days following irradiation (17). We found no correlation between BOLD response and tumor growth delay for the AT1 tumors. Thus the relative utility of BOLD and TOLD may depend on tumor type.

FREDOM was used in this study to estimate the tumor pO₂ and was correlated with BOLD and TOLD MRI. Relatively high pO₂ values (>100 torr) were observed upon 100% O₂/CB breathing in the tumor periphery (Fig. 1), which has also been noticed in previous studies using the AT1 tumor type (53). Thus, mean change in BOLD and pO₂ measurements may have been dominated by extreme responses to 100% O₂ or CB breathing. However, T₁ oxygen dependence is more directly related to tissue pO₂ and the TOLD responses tend to be considerably smaller than BOLD (typically, about 1/3) and thus, are less susceptible to large local perturbations. Thus, the TOLD response appeared less well correlated with ¹⁹F MRI pO₂, but is more reflective of the whole volume and hence radiation response. Therefore, a large BOLD response may be relevant, but for relatively hypoxic tumors, such as the AT1 here, where most tumors exhibit a small response, then TOLD appears more relevant. Assessing these values with TOLD MRI provided better prediction to treatment response.

There remains controversy regarding the use of 100% O₂ versus CB to modify tumor hypoxia with the goal of enhancing radiation response. Some studies have reported differential response to 100% O₂ and CB (23,27,29). Here, as in our previous studies with various tumors (1,42,54,55), we found close correlations between the effects of the two gases (Figs. 3, 5 and 6). Most notably, the residual hypoxic fraction was essentially identical (Figs. 5C, 6C). Therefore, we used 100% O₂ breathing with respect to irradiation since this will be more readily acceptable to patients. Indeed, our recent investigations regarding the feasibility of BOLD in humans have used oxygen breathing challenge (13,14).

Most patients still receive conventionally fractionated radiation therapy (CFRT), but hypofractionated regimens are gaining popularity due to the efficiency of fewer patient visits, opportunity for better treatment planning and development of enhanced targeting. Furthermore, in specific tumors SBRT (stereotactic body radiation therapy) has been found to provide significantly enhanced outcome (56). It is expected that hypoxia may influence each dose with little opportunity for the progressive reoxygenation thought to accompany CFRT (57). As such SBRT and adjuvant administration of hypoxic cell selective cytotoxins could be considered (58).

As noted by others, BOLD contrast imaging is sensitive to changes in vascular volume and flow as well as the inter conversion of oxy and deoxyhemoglobin (22,23). Additional challenges are provided by tissue interfaces and motion. In pre-clinical studies of subcutaneous tumors as used here, the tissues may be readily immobilized. In terms of semi-quantitative BOLD based on changes in signal intensity alone, tumor motion may prevent effective image co-registration for meaningful subtraction: indeed, we faced such a problem with bladder filling when examining cervical cancer in human patients (13). This was largely overcome by acquiring sagittal images, as opposed to transaxial. Likewise breast cancer imaging was more successful in the sagittal orientation since breathing artifacts were minimized. (14). Quantitative determination of R₂* maps largely overcomes the spatial correlation issues, since histograms of population values may be compared for tumor tissue, even if voxel overlap is imperfect. Perhaps the most challenging situation is for imaging lung tumors in patients. Here, artifacts can arise from blood flow, and respiratory and cardiac motion. Breath-hold imaging has been attempted to immobilize the lung tumor during imaging, but, variation in diaphragm/lung position during different breath-holds can be encountered. In addition, most lung cancer patients cannot tolerate long breath-hold attempts. Respiratory-gating has been used to acquire images during expiration throughout the MRI exam to ensure the same anatomy was imaged, whereas sacrificing temporal resolution in the dynamic BOLD study (59). TOLD imaging tends to be slower and therefore even more susceptible to motion. In some cases, EPI sequences have proven successful as recently demonstrated in human mediastinal organs (kidney, spleen) (50), but in lung EPI is subject to severe distortions. As shown here, both R₁ and R₂* responses are related to pO₂. We have recently shown that combined measurement of R₁, R₂ and blood volume fraction can provide estimates of pO₂ without the need for exogenous reporter molecule (60). Currently, such measurements are relatively slow requiring 14 minutes per pO₂ map and it remains to be seen whether they a provide superior prediction of radiation response.

There have been many attempts to assess and modulate tumor hypoxia prior to radiation. Approaches, as simple as inhalation of hyperoxic gases, have been successful in preclinical models (38,39), but translation to the clinic has shown marginal efficacy (1). It is suggested that the general lack of success has been inability to identify those tumors (patients) that would benefit from modified therapy. The correlations between BOLD, TOLD and pO₂ measurements reported here provide further impetus for the use of non-invasive ¹H MRI with respect to a hyperoxic gas challenge as surrogate biomarker for tumor oxygenation emphasizing potential value for clinical applications. Notably, the ability to stratify tumors based on hypoxia is particularly timely, since there are increasing opportunities to tailor therapy to tumor characteristics, potentially enhancing success through personalized medicine.