SEARCH

SEARCH BY CITATION

Keywords:

  • correlation;
  • BOLD;
  • pO2;
  • Tmath image;
  • tumor

Abstract

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

Blood oxygen level-dependent (BOLD) contrast-based functional MRI (fMRI) has been reported as a method to assess the evolution of tumor oxygenation after hyperoxic treatments, because of its sensitivity to changes in blood flow and deoxyhemoglobin content. However a number of questions remain: 1) In view of tumor heterogeneity, how good is the correlation between the MR parameters in gradient-echo imaging (signal intensity (SI) or effective transverse relaxation time (Tmath image)) and local tumor oxygen partial pressure (pO2)? 2) Is the magnitude of the change in SI or Tmath image a quantitative marker for variation in pO2? 3) Is initial Tmath image a good marker for initial pO2? To address these questions, murine tumors were imaged during respiratory challenges at 4.7 Tesla, using fiber-optic microprobes to simultaneously acquire tumor pO2 and erythrocyte flux. The BOLD signal response (SI and Tmath image) was temporally correlated with changes in pO2. However, the magnitude of the signal bore no absolute relation to pO2 across tumors, i.e., a given change in SI corresponded to a 25 mmHg pO2 change in one tumor, but to a 100 mmHg change in another. The initial Tmath image value did not reliably predict tumor oxygenation at the beginning of the experiment. In conclusion, the major advantages of the technique include noninvasiveness, high spatial resolution, and real-time detection of pO2 fluctuations. Information afforded by the BOLD imaging technique is qualitative in nature and may be combined with other techniques capable of providing an absolute measure of pO2. Magn Reson Med 48:980–986, 2002. © 2002 Wiley-Liss, Inc.

Oxygen is a key environmental factor in the development and growth of tumors, and their response to treatment. Because solid tumor hypoxia is associated with resistance to radiotherapy (1, 2), treatments aimed at acutely increasing the amount of oxygen delivered to tumors during radiotherapy have been developed, e.g., inhaling high-oxygen gas. Functional MRI (fMRI) studies based on blood oxygen level-dependent (BOLD) contrast (3) are of potential value in assessing the oxygenation status of tumors, particularly with regard to interventions designed to alter tumor oxygenation. Several groups have monitored the tumor response to high-oxygen-content gases by gradient-echo (GRE) MRI (4–8). Significant increases in intensity of Tmath image-weighted GRE images have been observed during hyperoxia using 100% O2 or carbogen (95% O2/5% CO2) breathing. The factors underlying the BOLD contrast have been the subject of intensive research, especially in neuroscience (9–12). In tumors, the BOLD effects observed in 2D GRE images have been attributed to both changes in blood flow and changes in blood oxygenation; the technique is also referred to as flow and oxygenation-dependent (FLOOD) imaging (13).

Since the key factor determining tumor response to radiotherapy is thought to be local pO2, rather than blood flow or hemoglobin saturation, there is a critical need to know how the BOLD signal correlates with tumor pO2. In attempts to understand this relationship, a good correlation was found between the averaged increase in tumor Tmath image and the averaged increase in tumor pO2 as measured by oxygen microelectrodes positioned subsequent to removal of the animal from the magnet (14). The simultaneous acquisition of tumor GRE images and pO2 was achieved in another study (15) using an MR-compatible fiber-optic oxygen sensor based on fluorescence quenching of a ruthenium dye. Here (15) carbogen breathing caused an increase in both pO2 and signal intensity (SI) averaged over the whole tumor.

In order to definitively assess the value of the method, however, possible temporal and spatial heterogeneity within tumors must be taken into account. Furthermore, it would be of value to know whether variation in SI or Tmath image quantitatively reflects the tissue pO2. The aim of the present study was to simultaneously measure the MR parameters and the tissue pO2 in the same tissue volume. We developed an experimental set-up (comparable to that described by Maxwell et al. (15)) to allow the simultaneous MRI of tumors in which fiber-optic oxygen/flow sensors were implanted. Respiratory gases were used to modulate tumor pO2 (a range of oxygen concentrations and carbogen). In addition to GRE imaging, multi-GRE images were also acquired to calculate the absolute effective transverse relaxation rate (Rmath image) and thereby avoid the confounding effects of inflow and oxygenation (16). Because of the frequently heterogeneous distribution of vascular response throughout the tumor (17), data obtained in the immediate vicinity of the fiber-optic probe were compared with those obtained in the whole tumor.

METHODS

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

MRI

MRI was performed with a 4.7 Tesla (200 MHz, 1H) bore system (40-cm inner diameter; Bruker Biospec, Ettlingen, Germany). A surface coil 2 cm in diameter was used for radiofrequency (RF) transmission and reception. A preliminary multislice T2-weighted fast spin-echo sequence (TR = 3 s, effective TE = 63 ms) was used to determine the position of the fiber-optic sensor tip in the tumor, and a single 1.3-mm slice was selected for the imaging. GRE images were acquired in 25.6 s, using 64 phase and frequency-encode steps over a 3-cm FOV. The in-plane spatial resolution was 470 μm, and was chosen to be compatible with the volume of tissue sampled by the OxyLite™/OxyFlo™ probes. Acquisition parameters were: TR = 200 ms, flip angle = ∼45°, two averages, kHz receiver bandwidth = 12.5. The TE in the GRE images was 20 ms. For the multi-GRE sequence, a six-echo sequence was used, with TE = 6.1 ms and echo spacing = 6.02 ms.

In Vitro Studies

We first established the dependence of blood Rmath image on deoxyhemoglobin content in our MR system operating at 4.7 Tesla. Nine aliquots of human blood from a single donor were gassed with an N2/O2 mixture to achieve equilibration at defined oxygen tensions. After equilibration, oxygen saturation levels and hemoglobin contents were measured using a OSM3 radiometer, and pO2 values were measured using a CIBA Corning 288 blood gas analyzer. Capped 10-ml syringe samples were imaged for Rmath image determination using a multi-GRE sequence.

In Vivo Studies

Syngeneic FSa fibrosarcoma tumors (18) were inoculated into the thighs of male C3H/HeOuJIco mice and were studied at a mean diameter of 8 mm (14 days after implantation). The mice were anesthetized with 1% isoflurane delivered by a calibrated vaporizer. A combined fiber-optic probe (∼500 μm outer diameter, OxyLite™/OxyFlo™ instruments; Oxford Optronix Ltd., Oxford, UK) was introduced into the tumor, providing real-time and simultaneous measurement of tumor pO2 (by luminescence quenching) and microvascular erythrocyte flow (by laser-Doppler flowmetry) (19). Data were collected continuously at a sampling frequency of 20 Hz and further averaged to give a sampling frequency synchronized to that of the MR images. Laser-Doppler signals were recorded in blood perfusion units (BPUs), which is a relative units scale, and tissue pO2 in mmHg. A thermocouple sensor, implanted at a different location inside the tumor to prevent susceptibility artifact, was also used to monitor tissue temperature and to automatically compensate for the pO2 measurement. Warm air was flushed into the magnet to maintain the body temperature at around 37°C.

Continuous imaging was performed during the respiratory challenge. A 5-min baseline period was first acquired while the mice breathed air. Then the gas regime was sequentially changed at 5-min intervals: air - [50% O2] air - [100% O2] - 21% O2 - [carbogen] - air. The protocol was first applied during GRE acquisition and was repeated 10 min later on the same mouse during the multi-GRE acquisition.

Image Data Analysis

For each time course, the susceptibility effects (Tmath image) were estimated on a voxel-by-voxel basis using a six-parameter monoexponential function. The goodness of fit was checked and a threshold value for the determination coefficient (r2 > 0.6) was chosen to select only voxels showing a sufficiently monoexponential pattern of decay.

In the text, we will refer to “local” SI or Tmath image when signals were measured in a 3 × 3 square around the probe (the signal from the central voxel where the probe was located was not considered because of the possible susceptibility artifact). We will refer to SI and Tmath image as “averaged over the whole tumor” when the signals were measured in a region of interest (ROI) encompassing the whole tumor.

RESULTS

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

Blood Rmath image Dependence on O2 Saturation

No statistical difference in the hemoglobin (Hb) content was found in the different blood samples (mean ± SD: 15.5 ± 0.1 g/100 ml), so we could directly relate Rmath image to the fraction of oxygenated hemoglobin in blood (Y). The hemoglobin saturation ranged from 5.3% to 96.3%. A positive correlation between Rmath image and [1−Y] was observed (Fig. 1a), and the least-squares fit of the data indicated that both the linear and the quadratic terms were significant. In Fig. 1b, the experimental O2-Hb dissociation curve is shown. Blood pO2 ranged from 6.5 to 192 mmHg. Figure 1c shows the relationship between Rmath image and blood pO2. The relationship between Rmath image and pO2 is approximately linear in the range of 0–30 mm Hg. Because the binding coefficients of mouse and human blood O2-Hb are very similar, the overall interpretations of BOLD contrast in human blood and mouse blood are comparable.

thumbnail image

Figure 1. Blood Rmath image dependence on oxygen saturation at 4.7 Tesla. a: Relationship between Rmath image and 1-Y. Y: Fraction of oxygenated hemoglobin in blood. –: Linear curve fit; —: Quadratic curve fit. b: Experimental oxygen-hemoglobin dissociation curve. —: Sigmoidal curve fit. c: Relationship between Rmath image and pO2. —: Sigmoidal curve fit.

Download figure to PowerPoint

In Vivo Results During High-Oxygen Gas Breathing

The initial local tumor pO2 during air-breathing was 6.3 ± 4.6 mmHg (mean ± SD, N = 5). The stability of pO2 and the MR parameters during the baseline acquisition was estimated: the average standard deviation (SD) was 0.6 mmHg for pO2, 2.3% for SI, and 0.69 ms for Tmath image.

Administration of high-oxygen-content gas produced an increase in tumor oxygenation. The mean values (±SD) were: 32.2 ± 22.5 mmHg, 41.4 ± 35.9 mmHg, and 45.2 ± 37.3 mmHg for 50% O2, 100% O2, and carbogen, respectively. The pO2 values observed in 100% O2-breathing mice were significantly higher than in 50% O2-breathing mice (paired t-test, P < 0.001). Mean pO2 values obtained with carbogen were not significantly different from those seen with 100% O2. Similar results were observed by Hunjan et al. (20). The response to carbogen gas breathing produced an additive increase in pO2 in only one-fifth of the tumors. In Fig. 2, we show the typical time courses of pO2, SI, and Tmath image observed in two tumors presenting a distinct response to the respiratory challenge. The left-hand tumor showed a higher baseline pO2 (10 mmHg) and a far more dramatic response (40 or 100 mmHg), whereas the right-hand tumor (baseline 2 mmHg) did not respond to oxygen alone. By contrast, the right-hand tumor did respond to carbogen. In all five tumors, we found a good correlation between the changes in SI and the changes in tumor pO2. The same correlation was observed between Tmath image and tumor pO2. Changes in local SI or local Tmath image were greater than those averaged over the whole tumor. However, the local values were noisier, as may be expected given the limited number of voxels represented. Voxel-by-voxel analysis showed spatial heterogeneity in the tumor response to the respiratory challenges. In Fig. 3, we present computed difference maps of SI and Tmath image superimposed on a T2-weighted anatomical image after 50% O2, 100% O2, and carbogen breathing compared to the image recorded when the mice were breathing air. A color scale reflecting the intensity of the change was used when a statistically significant difference (P < 0.01) was observed relative to the control (air breathing). The results are expressed as % change in SI, and as absolute change in Tmath image values. Although most voxels present a significant increase in SI and Tmath image due to breathing of high-O2-content gas, some regions showed no response and some isolated voxels showed an opposite change. In the typical tumor shown in Fig. 3, the number of voxels that present a significant response to carbogen breathing was higher than it was under 100% oxygen breathing. The response around the probe was comparable when breathing 100% oxygen and carbogen, which is consistent with the results obtained using the OxyLite™. The measurements obtained near the probe tip did not necessarily reflect the global response of the tumor.

thumbnail image

Figure 2. Time course of pO2, SI, and Tmath image during respiratory challenge (air-50% O2-air-100% O2-air-carbogen-air) for two tumors presenting distinct responsiveness. A: pO2 measured by a fiber-optic microprobe during the GRE sequence. B: % changes in SI, locally (−) and averaged over the whole tumor (−+). C: pO2 measured by a fiber-optic microprobe during the multi-GRE sequence. D: Changes in Tmath image, locally (−) and averaged over the whole tumor (−+). Note the variations in (A and C) pO2 during the respiratory challenge, and the real-time response of the MR parameters (B) SI and (D) Tmath image. Note also the greater response when the MR parameters were measured locally compared to measurements averaged over the whole tumor.

Download figure to PowerPoint

thumbnail image

Figure 3. Mapping of the response in SI and Tmath image during respiratory challenge. Computed difference maps superimposed on a T2-weighted anatomical image after 50% O2, 100% O2, and carbogen breathing compared to air breathing. a:% change in SI. b: ΔTmath image. A color scale reflecting the intensity of the change was used when a statistically significant (P < 0.01) difference with the control (air breathing) was observed. Note the heterogeneity of response inside the tumor (gray shaded). The location of the fiber-optic probe tip that transects the imaging plane is indicated by the arrow and corresponds to the dark voxel.

Download figure to PowerPoint

Since SI in the Tmath image-weighted images correlated with pO2, we wanted to know whether the variation of SI was quantitatively related to the variation of pO2. In Fig. 4, we show typical results illustrating this relationship, as established for two different tumors. It is clear that while SI does vary in line with pO2, the relationship is not proportional, as a 10% change in SI reflected a shift of anywhere from <25 mmHg to ∼100 mmHg pO2.

thumbnail image

Figure 4. Two typical results (mice 2 and 4) showing the relationship between the local changes in SI and the corresponding pO2 recorded during the experiment. The mean initial pO2 is at 0%ΔSI. Note the change of slope at high pO2. A 10% change in SI can reflect a variation of <25 mmHg (□) or ∼100 mmHg pO2 (▪).

Download figure to PowerPoint

Is Tmath image a better marker for pO2? As Tmath image increased in parallel with pO2 for each tumor, it was interesting to evaluate the correlation of Tmath image with pO2. The variation in localRmath image (= 1/T*2) is shown in Fig. 5 as a function of pO2. In addition to the existing inverse relationship, it becomes clear that, from one mouse to another, similar pO2 values were not associated with comparable local Rmath image.

thumbnail image

Figure 5. Variation in local Rmath image as a function of variations in pO2. Each symbol represents a different tumor. Note that the sensitivity of Rmath image to changes in pO2 is variable from one tumor to another. Note also that a given value of Rmath image is not predictive of the pO2 value.

Download figure to PowerPoint

There was an attempt to fit the data in a four-parameter sigmoidal curve to estimate the relationship between Rmath image (or SI) and pO2. For some tumors, the parameters were not significant and/or illogical. This may be explained by the fact that the recorded pO2 did not always reach values above 30 mmHg during respiratory challenge. For these tumors, the sigmoid model was not appropriate, as no plateau phase was observed. For these reasons, we made a distinction between changes in SI or Rmath image occurring in the range of 0–30 mmHg and those occurring at a pO2 higher than 30 mmHg. Linear least-squares analysis was performed for each case. The relationship between Rmath image and pO2 recorded in vitro (Fig. 1c) justifies the approximation. Slopes of the double linear fit are shown in Table 1. The range of pO2 observed during each study is given. The slopes are rather different (by a factor of 2) between individual mice. Finally, we looked at the value of initial Tmath image as an indicator of the responsiveness to carbogen. In Table 2, we present the relative frequency of voxels in individual tumors that showed (or did not show) a statistically significant change in Tmath image after carbogen breathing compared to air breathing. In four out of five tumors, we found that responsive regions were associated with higher initial Tmath image values than regions that did not respond to carbogen. However, the absolute value of Tmath image is not a good predictive marker, as a given value of Tmath image can correspond to a region that does or does not respond to carbogen according to the tumor in question.

Table 1. Analysis of the Relationship Between %ΔSI or ΔRmath image and pO2, Measured in the Vicinity of the Probe*
MouseSlope: % ΔSI/10 mmHgSlope: ΔRmath image/10 mmHgb
pO2 rangea[0–30 mmHg]>30 mmHgpO2 rangea[0–30 mmHg]>30 mmHg
  • Two or one least squares linear fit were performed whether the recorded pO2 reached values above 30 mmHg or not. The magnitude of the change is reflected by the slope.

  • a

    In mmHg (pO2 range reflects the minimum and the maximum value of pO2 recorded during the duration of the experiment (30min).

  • b

    In s−1. Estimated parameter ± standard error.

1[1–22]4.76 ± 0.75[3–24]−3.76 ± 0.29
2[5–100]5.81 ± 0.680.61 ± 0.09[12–100]−1.06 ± 0.26−0.24 ± 0.06
3[12–74]2.89 ± 1.85 (NS)4.92 ± 1.40[6–55]−1.42 ± 0.30−0.25 ± 0.53 (NS)
4[2–18]7.82 ± 0.59[0–9]−1.93 ± 0.38
5[2–13]9.29 ± 1.24[2–12]−2.13 ± 0.26
Table 2. Analysis of Initial Tmath image as an Indicator of the Response to Carbogen*
Mouse% Activated voxelsa (total number of voxels)Initial Tmath image values (ms)
Activated voxelsNonresponding voxels
  • Comparison of initial Tmath image values between activated and nonresponding voxels inside the tumor.

  • a

    % Tumor voxels with significant change compared to air breathing (t-test, P < 0.05). Mean ± SD.

  • ***

    P < 0.001

  • *

    P < 0.05, t-test.

112% (298)33.4 ± 12.821.3 ± 10.9***
236% (102)43.8 ± 24.221.5 ± 9.8***
316% (143)24.0 ± 15.715.5 ± 7.0***
410% (524)34.9 ± 13.335.9 ± 19.2 NS
515% (410)23.6 ± 11.119.6 ± 13.5 *

DISCUSSION

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

Because of its sensitivity to changes in blood flow and deoxyHb content, BOLD fMRI was previously reported as a very sensitive method to assess the variations of tumor oxygenation after hyperoxic treatments. This was first demonstrated in the pioneering studies of the groups of Karczmar and Griffiths (4–6), and later applied by many groups in oncology (7, 8, 21). Correlation between the evolution of pO2 and the evolution of SI (in GRE imaging) or Tmath image has so far been established using data averaged over whole tumors, leaving several questions concerning the validity of the method unanswered: 1) Given tumor tissue heterogeneity, how good is the correlation between SI or Tmath image and pO2 when measuring these parameters from micro-regions within tumors? 2) Is the magnitude of the variation in SI or Tmath image a quantitative marker for pO2? 3) Is the initial Tmath image a good marker for the initial pO2? The purpose of this study was to address these questions and experimentally assess the value of Tmath image-weighted GRE imaging using deoxyhemoglobin as an endogenous marker for tumor pO2. Simultaneous direct acquisition of tumor pO2 has allowed us to relate tissue oxygenation directly to change in image intensity or Rmath image.

In response to respiratory challenges, tumor Tmath image and SI increased in Tmath image-weighted GRE images (Fig. 2), as reported by the previously cited groups. These changes were positively time-correlated with tumor pO2, indicating that the method correctly reflects the evolution of tumor oxygenation. This positive correlation was observed when MR parameters recorded over the whole tumor were used, as shown by Maxwell et al. (15). In the study carried out for this paper, MR parameters were measured in voxels that possess a volume comparable to that sampled by the OxyLite™ microprobe (Fig. 2). This latter observation validates the use of Tmath image-weighted sequences to map the evolution of the pO2 within tumors, and is of particular value because it permits the observation of heterogeneity within the tumor during respiratory challenges.

Can the method be used to quantitatively assess the evolution of pO2? What is the relationship between SI and pO2? In vivo, the observed increase in SI was more pronounced when passing from 5 to 15 mmHg than when passing from 40 mmHg to 50 mmHg. A plateau was reached at high pO2 (typically above 30 mmHg). At least four parameters may be at the origin of this complex relationship between the BOLD response and tumor pO2: 1) the curvilinear dependence of blood Rmath image on blood deoxyHb content, 2) the hemoglobin-O2 dissociation curve, 3) the “inflow” effect, and 4) the blood volume fraction.

To shed some light on this issue, it is interesting to look at the BOLD effect that was assayed in vitro as part of this study. At 4.7 Tesla, we found a quadratic dependence of blood Rmath image on the deoxyHb fraction (Fig. 1). A similar relationship was previously reported between both blood R2 (22, 23) and Rmath image (24), and oxygen saturation. In normal tissues (pO2 above 15 mmHg), the quadratic term is not relevant and the relationship is often assumed to be linear. However, very low HbO2 saturations may be typical in tumors compared to normal tissue (25, 26). The quadratic term is more relevant in studies of tumor oxygenation, which indicates that the technique should be more sensitive in deoxygenated regions. Moreover, in this study, the BOLD contrast was correlated directly to tumor pO2 and not to blood oxygen saturation. The decreased sensitivity that was observed above 30 mmHg tumor pO2, could probably be explained by the fact that at this tumor pO2, most of the microvessel hemoglobin already exists in oxygenated form. A further increase in blood pO2 would not further increase hemoglobin saturation (no further change in SI or Rmath image would be observed), while it would continue to improve tumor oxygenation. Overall, this means that small changes in pO2 can be better detected in tissue with low pO2, e.g., tumors or renal medulla (27).

Another source of nonlinearity in the relationship between SI and pO2 is the influence of the so-called “inflow effect.” We tried to evaluate the influence of this parameter using blood flow data (OxyFlo™) and the SI at TE = 0 (Io) (16). No significant changes were observed in our model in response to the respiratory challenges. However, this does not exclude an inflow effect, as subtle changes in blood flow may go undetected by these methods. Is Tmath image a better marker for pO2 because this parameter is independent of the inflow effect? Tmath image values and tumor pO2 were positively time-correlated. However, no correlation was found between the Tmath image values and the initial tumor pO2 level, indicating that Tmath image is a bad predictor of tumor oxygenation. It should be emphasized that the Tmath image value for a given voxel is not only determined by microscopic field heterogeneity due to the presence of deoxyHb, but also by other factors such as the macroscopic field homogeneity and the T2 relaxation process. Without any baseline pO2 estimation, and because of the curvilinear relationship, the variation of Tmath image could not be quantitatively related to the response in pO2. Finally, a comparison of responses between different locations within the same tumor, or using different tumors, should be made with caution: in addition to the question of magnetic field homogeneity, deoxyHb content can be influenced by the local blood volume fraction, which may vary within tumors (and necessarily between tumors) and thus may influence the BOLD sensitivity (28–30).

In conclusion, GRE sequences and Tmath image measurements offer the capability to monitor oxygenation changes in tumors. The major advantages of the technique include its noninvasive nature, high spatial resolution, and real-time detection of changes in oxygenation. This method should be considered qualitative because it does not provide an absolute measure of pO2, and because calibration of the BOLD effect to tissue oxygenation is not straightforward. Individually, SI and Tmath image changes are positively correlated with tumor pO2, but the magnitude of change could not easily be compared between voxels of the same tumor or necessarily between tumors. In order to evaluate the basal level of pO2 in the tumor, BOLD imaging should be combined with additional techniques that can provide absolute quantification of pO2. Among MR techniques these include 19F NMRI (20), electron paramagnetic resonance (EPR) spectroscopy (31, 32), EPR imaging (33), and Overhauser imaging (34).

REFERENCES

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES