Examining correlations of oxygen sensitive MRI (BOLD/TOLD) with [¹⁸F]FMISO PET in rat prostate tumors

Abstract

Hypoxia is regarded as a potential prognostic biomarker for tumor aggressiveness, progression, and response to therapy. The radiotracer ¹⁸F-fluoromisonidazole ([¹⁸F]FMISO) has been used with positron emission tomography (PET) to reveal tumor hypoxia. Meanwhile, blood oxygen level dependent (BOLD) MRI and tissue oxygen level dependent (TOLD) MRI offer insight into oxygenation based on endogenous signals without the need for radiolabels. Here, we compared BOLD and TOLD MRI with [¹⁸F]FMISO uptake using Dunning prostate R3327-AT1 tumor bearing rats. BOLD and TOLD MRI were acquired with respect to an oxygen gas breathing challenge. The following day, dynamic PET was performed up to 90 minutes following IV injection of [¹⁸F]FMISO. Tumors showed distinct heterogeneity based on each technique. Correlations were observed between magnitude of mean BOLD or TOLD MRI signal responses to oxygen-breathing challenge and initial distribution of [¹⁸F]FMISO. Correlations were observed for whole tumor as well on a regional basis with stronger correlations in the well perfused tumor periphery indicating the strong influence of perfused vasculature. After 90 minutes most correlations with signal intensity became quite weak, but correlations were observed between hypoxic fraction based on FMISO and fractions of tumor showing BOLD or TOLD response in a subset of tumors. This emphasizes the importance of considering regional heterogeneity and responsive fractions, as opposed to simple magnitudes of responses. Although the data represent a small cohort of tumors they present direct correlations between oxygen sensitive MRI and PET hypoxia reporter agents in the same tumors, indicating the potential utility of further investigations.

Introduction

Hypoxia is known to be associated with tumor aggressiveness, metastatic spread and resistance to various treatments [1,2]. As such there has been much interest in developing non-invasive imaging approaches to assessing tumor oxygenation and hypoxia [2-5]. ¹⁸F-fluoromisonidazole ([¹⁸F]FMISO) has been developed as an imaging radiotracer for tumor hypoxia using positron emission tomography (PET) [3,6]. Meanwhile, blood oxygen level dependent (BOLD) MRI and tissue oxygen level dependent (TOLD) MRI assess vascular oxygenation and tissue oxygenation through the paramagnetic endogenous contrast caused by deoxyhemoglobin and the oxygen molecule, respectively [7-9]. Measurement of BOLD and TOLD with an oxygen breathing intervention can provide information on oxygen delivery [8]. In this study we have explored correlation between these approaches. Specifically, we compared [¹⁸F]FMISO uptake with BOLD and TOLD MRI in the well characterized Dunning prostate R3327-AT1 tumor growing subcutaneously in rats [10,11].

Many solid tumors exhibit oxygen deficiency (hypoxia) as a result of rapid growth and inefficient vasculature arising from disorganized angiogenesis. Tumor hypoxia is increasingly regarded as an important factor for malignant progression and response to therapy [12,13]. The direct measurement of tumor hypoxia historically required invasive procedures such as the insertion of polarographic oxygen-sensitive electrodes, which has been considered as the gold standard for measuring the hypoxic fraction in tumors [2]. Noninvasive imaging offers several advantages, including the feasibility of longitudinal measurements on the same subject, the generation of complete three-dimensional maps of tumor hypoxia, and the potential application to image-guided therapy. Quantitative dynamic oximetry has also been achieved using the direct injection of specific oxygen sensitive reporter molecules such as perfluorocarbons, siloxanes and chars permitting oximetry based on ¹⁹F or ¹H MRI or ESR [14-18]. However, these methods are invasive and are therefore restricted to easily accessible tumors. In addition, they are limited to a relatively small number of measurements. An alternative approach is the use of nitroimidazoles to visualize hypoxia itself [3].

Extensive studies have reported radiotracers that selectively accumulate in hypoxic tumors and detection using PET or SPECT. The most extensively developed agent for PET is ¹⁸F-fluoromisonidazole ([¹⁸F]FMISO) [3,6]. [¹⁸F]FMISO is a lipophilic compound that enters cells by diffusion. Under normoxic conditions, there is equilibrium of intra-cellular and extra-cellular [¹⁸F]FMISO, while under hypoxic conditions, a sequence of single electron reductions takes place and it binds to intracellular macromolecules. Several groups reported that there is retention of [¹⁸F]FMISO based on tissue oxygenation as assessed by pO₂-polarography [19]. Hence, the accumulation of [¹⁸F]FMISO in tumors is related to hypoxia. FMISO has been applied in multiple clinical trials, but other agents with modified properties, notably lipophilicity have been developed [3,6,20].

MRI provides functional and anatomical information on tumor vasculature and physiology at high spatial resolution. Recently, oxygen sensitive MRI has attracted increasing interest as a non-invasive indicator [2,5,21,22]. Blood oxygen level dependent (BOLD) MRI is sensitive to vascular oxygenation since deoxyhemoglobin is paramagnetic causing local susceptibility gradients influencing the effective transverse relaxation rate (R₂*) [23-26]. Some studies examine baseline R₂* as an indicator of hypoxia [27-29], but we favor examining the response ΔR₂* to an intervention such as an oxygen breathing gas challenge [8]. Tissue oxygen level dependent (TOLD) MRI is based on T₁-weighted images or R₁ (=1/T₁) relaxation rate measurements. R₁ is directly influenced by paramagnetic molecular O₂ and is therefore sensitive to the local concentration and consequently partial pressure (pO₂) [30,31]. Increasingly studies examine combined BOLD and TOLD measurements to provide additional insights into tissue oxygenation [7-9,32-34]. The aim of our study was to compare uptake of [¹⁸F]FMISO with BOLD and TOLD MRI to establish potential correlations relevant to assessing hypoxia. Previous studies in Heidelberg have examined [¹⁸F]FMISO uptake in the Dunning prostate R3327-AT1 tumor [35], while extensive studies in Dallas have explored oxygen sensitive MRI [8,36]. Others have interrogated correlations between oxygen sensitive MRI and nitroimidazoles based on immunohistochemistry [22,37]. One brief study examined uptake of FMISO in comparison with BOLD MRI in rat osteosarcomas [38] and others have compared BOLD MRI and hypoxia based on PET, but in separate cohorts of animals and separate publications [39,40]. We have now undertaken sequential combined investigations using BOLD, TOLD and FMISO based PET on individual tumors and believe this is the first study of this type.

Materials and methods

Tumor model

This study was approved by the Institutional Animal Care and Use Committee. Six adult male Copenhagen rats (bred in-house at UT Southwestern; originally from Charles River) were implanted subcutaneously with Dunning R3327-AT1 tumor. AT1 is an anaplastic, low metastatic, comparatively fast growing tumor with tumor volume doubling time of 5.2 days [10]. Tumors were originally obtained from Dr. J. T. Isaacs (Johns Hopkins, Baltimore, MD) [41]. Tumors were allowed to grow (1.4 to 2.6 cm³) and then investigated using MRI and PET.

MRI data acquisition

MRI was performed using an Agilent (Varian) 4.7 T MRI scanner (Palo Alto, CA) and a 35 mm single turn solenoid volume coil placed around the tumor and thigh, as described previously [42]. Briefly, the rats were maintained under general gaseous anesthesia (1.5% isoflurane in air (1 L/min); Baxter International, Deerfield, IL) and kept warm (37°C) using a circulating warm water blanket. Shimming was performed on the tissue water resonance. High resolution T₂-weighted anatomical images were acquired transaxially to the tumor and thigh using a fast spin echo sequence (TR=2000 ms, effective TE=48 ms, Echo Train Length =8, 128×128 data matrix, 40×40 mm FOV, 1 mm slice thickness without gap). Interleaved BOLD and TOLD sequences (IBT) with baseline (air) followed by an oxygen gas breathing challenge were performed. BOLD data were acquired using a multi-echo gradient echo sequence with: TR=150 ms, 10 echoes TE range =6 to 69 ms with echo spacing of 7 ms, flip angle =20°, and acquisition time 21 s for 128×128 data matrix, 1 mm slice thickness and 40×40 mm FOV. TOLD was obtained using a gradient echo sequence with TR=30 ms, TE=5 ms, flip angle =45°, acquisition time 3 s, 128×128 data matrix, 1 mm slice thickness and 40×40 mm FOV to match the BOLD acquisition.

PET imaging

The next day PET was performed on a Siemens Inveon Multimodality PET/CT scanner (Siemens Medical Solution Inc., Knoxville, TN, USA). Anesthesia was induced with 3% isoflurane in air at room temperature and maintained under 2% isoflurane anesthesia. The CT images were acquired at 80 kV and 500 µA with a focal spot of 58 µm. The total rotation of the gantry was 360° with 360 rotation steps obtained at an exposure time of approximately 125 ms/frame. Each rat was injected by tail vein with [¹⁸F]FMISO (570-600 µCi in 200 µl saline, prepared in our laboratory, according to standard published procedures [43]). Immediately after injection, a dynamic PET scan was initiated and continued over a period of 90 min. PET data were reconstructed into 18 frames, where each frame represents the average value of the respective 5 min interval and PET dynamic time activity curve was generated. PET images were reconstructed using Fourier and Ordered Subsets Expectation Maximization 3D (OSEM3D) signogram and reconstruction algorithms, respectively. Reconstructed CT and PET images were fused and analyzed using Inveon Research Workplace (IRW) software.

Histology

Pimonidazole hydrochloride (Hypoxyprobe-1; NPI, Burlington, MA) was injected into the tail vein at a dose of 60 mg/kg. Sixty minutes later the rat was sacrificed and tissues prepared for routine immunohistochemical detection of pimonidazole hypoxia marker as described previously [44].

Data analysis

For PET quantification, volumes of interest (ROIs) were placed in specific areas. An ellipsoid shaped ROI was used to select the whole tumor. The tumor center was defined with the diameters reduced to 70% in all three dimensions (1/3 of the tumor volume) and the outer layer was defined as tumor periphery (2/3 of the tumor volume). PET data: for presentation a slice was selected by visual inspection of CT images from the 3D data to match the MRI orientation. An ROI was also placed on the contralateral muscle. The resulting quantitative data are expressed as percent of the injected dose per gram of tissue (%ID/g) and as the ratio of uptake of [¹⁸F]FMISO (%ID/g) in the region of interest (ROI) vs. muscle. Voxel by voxel analysis provided estimation of hypoxic fraction based on the standard accepted threshold of 1.2 times of %ID/g of muscle [3].

For MRI, ROIs were manually drawn for central and peripheral tumor regions. The changes in signal Intensity (%ΔSI) with respect to oxygen challenge were calculated as (Equation 1):

SI_b is the mean baseline signal intensity during air breathing and SI_t is the mean signal intensity with oxygen inhalation based on T₁- or T₂*-weighted (echo time =20 ms) imaging. R₂* (1/T₂*) maps were generated by fitting the multi-echo gradient echo images to TE, as a single exponential function on a voxel-by-voxel basis. Changes of T₂* and R₂* due to oxygen challenge were calculated on a voxel by voxel basis as (Equations 2, 3):

For each ROI, baseline temporal standard deviation in T₂* and T₁w SI was used to determine the responsive voxels as measured by each method, respectively. A voxel was labeled responsive if T₂*w or T₁w ΔSI was larger than n times the corresponding baseline standard deviation (n=1.5, 2.0, and 2.5 tested separately). The response fraction was defined as the fraction of responsive voxels over all voxels. Data were processed using Matlab (Math Works Inc., Natick, MA). Pearson correlation coefficients (r²) were calculated to compare measurements. Statistical analyses were performed using Microsoft Excel 2007 (Microsoft, Seattle, WA) and considered significant when P<0.05. Local and regional parameter and response were compared using linear regression.

Results

MRI showed distinct heterogeneity in the Dunning prostate R3327-AT1 tumors (Figures 1 and 2). Global and regional measurements as well as responses to oxygen breathing challenge are shown in Table 1A, 1B, 1C and 1D. As expected strong correlations were observed between semi quantitative mean BOLD and TOLD response (r²>0.93, P<0.01) as well as between BOLD and changes in T₂* (ΔT₂*; r²>0.90) or R₂* (ΔR₂*; r²>0.95). Both T₂* and R₂* are presented since mean R₂* ≠ (mean T₂*)^-1. Tumor oxygenation is often reported as hypoxic fraction rather than mean signal response and fractions of tumor responding to oxygen breathing intervention are shown in Table 2. The apparent enhancing fraction depends on the threshold chosen and three different thresholds are presented, based on changes of 1, 1.5, and 2 times baseline (temporal) standard deviation on a voxel by voxel basis. BOLD responsive enhancing fractions were correlated irrespective of threshold above 1.5 SD: specifically, r²>0.95. Meanwhile, the threshold appears more critical for TOLD with r²>0.85 for 1.5 vs. 2 SD.

Figure 1.

Comparison of PET and Oxygen-sensitive MRI in a hypoxic Dunning prostate R3327-AT1 tumor. (A) [¹⁸F]FMISO dynamic activity for two tumor regions (central and peripheral). High [¹⁸F]FMISO activity was observed in the peripheral region at early time points, but decreased rapidly. The activity in the tumor center increased gradually over 90 mins. Both regions reached the “hypoxia threshold” suggested to be 120% of muscle. (B) Transaxial image slice selected from 3D PET data set to match MRI in (D). Ring enhancement was observed around the tumor periphery indicating perfusion related [¹⁸F]FMISO activity at 3 mins. post injection. Heat maps correspond to local radioactivity. (C) Accumulation of [¹⁸F]FMISO at 90 mins shows hypoxia in the tumor center. (D) This tumor was highly responsive to oxygen-breathing challenge as seen in the corresponding oxygen-sensitive MRI. Heat maps of the tumor are overlaid on anatomical image where the thigh muscle and bone marrow are clearly seen. Scale bar 1 cm.

Figure 2.

PET and Oxygen-hyphen sensitive MRI in a Dunning prostate R3327-AT1 tumor showing low [¹⁸F]FMISO uptake. A. [¹⁸F]FMISO activity in the peripheral region decreased initially and gradually reached the hypoxia threshold. The activity in the tumor center also increased gradually but never reached the hypoxia threshold. B. Ring enhancement showed perfusion related [¹⁸F]FMISO activity at 3 mins. C. Accumulation of [¹⁸F]FMISO at 90 mins indicated a ring of hypoxia in the peripheral region but no hypoxia in the tumor center. D. This tumor was relatively unresponsive to oxygen breathing challenge in terms of oxygen-sensitive MRI.

Table 1A.

MR parameter responses to oxygen breathing challenge and uptake of FMISO

Tumor	BOLD (%Δ SI)	TOLD (%Δ SI)	T2* Air ms	T2* O2 ms	ΔT2* ms	R2* Air s-1	R2* O2 s-1	ΔR2* s-1	F-MISO %ID	T/M 3 mins	T/M 90 mins
1	-0.87	0.36	27.79	27.22	-0.57	47.71	48.54	0.83	0.34	1.07	1.19
3	5.67	3.55	26.82	28.67	1.85	61.76	58.30	-3.46	0.35	1.90	1.16
4	5.12	2.47	18.10	19.05	0.95	76.29	72.08	-4.21	0.40	1.71	1.45
5	-2.21	-1.44	31.05	30.60	-0.45	41.03	42.50	1.47	0.39	1.26	1.25
6	4.04	2.97	31.71	33.05	1.34	38.24	36.28	-1.96	0.38	1.98	1.34
Mean	2.35	1.58	27.09	27.72	0.62	53.01	51.54	-1.47	0.37	1.58	1.32
SD	3.63	2.07	5.44	5.32	1.08	15.88	14.06	2.53	0.03	0.40	0.15

Table 1B.

Data based on mean of tumor periphery (outer 40%) across single image slice

Tumor	BOLD (%Δ SI)	TOLD (%Δ SI)	T2* Air ms	T2* O2 ms	ΔT2* ms	R2* Air s-1	R2* O2 s-1	ΔR2* s-1	FMISO %ID	T/M 3 mins	T/M 90 mins
1	0.27	0.58	28.41	28.2	-0.21	48.82	48.6	-0.22	0.31	1.24	1.08
3	5.35	4.22	21.24	22.41	1.17	81.37	77.70	-3.66	0.37	2.11	1.20
4	2.71	1.99	16.81	17.12	0.31	87.57	84.61	-2.96	0.34	1.89	1.23
5	-0.89	-1.22	26.72	26.96	0.24	45.95	45.94	-0.01	0.39	1.44	1.27
6	5.02	3.54	26.24	27.66	1.42	45.49	42.72	-2.76	0.34	2.06	1.21
Mean	2.49	1.82	23.88	24.47	0.59	61.84	59.91	-1.92	0.35	1.75	1.20
SD	2.78	2.21	4.77	4.71	0.68	20.81	19.65	1.68	0.03	0.39	0.07

Table 1C.

Data based on mean of tumor center (central 60%) across single image slice

Tumor	BOLD (%Δ SI)	TOLD (%Δ SI)	T2* Air ms	T2* O2 ms	ΔT2* ms	R2* Air s-1	R2* O2 s-1	ΔR2* s-1	F-MISO %ID	T/M 3 mins	T/M 90 mins
1	-3.21	-0.14	26.56	25.28	-1.28	45.52	48.41	2.89	0.42	0.69	1.47
3	nd	nd	35.23	38.13	2.9	32.60	29.63	nd	0.31	1.09	1.02
4	8.29	3.45	19.67	21.65	1.97	60.62	54.15	-6.47	0.61	1.10	2.23
5	-3.8	-1.74	37.71	36.19	-1.52	33.59	37.19	3.61	0.36	0.46	1.18
6	2.65	2.02	40.69	41.91	1.22	26.55	25.86	-0.69	0.53	1.53	1.91
Mean	0.98	0.90	31.97	32.63	0.66	43.08	39.05	-0.17	0.45	0.97	1.56
SD	5.68	2.30	8.66	8.71	1.97	13.08	12.06	4.60	0.12	0.41	0.50

Table 1D.

Data based on mean of muscle across single image slice

Muscle	BOLD (% SI)	TOLD (% SI)	T2* Air ms	T2* O2 ms	ΔT2* ms	R2* Air s-1	R2* O2 s-1	ΔR2* s-1	FMISO %ID
1	5.07	1.57	17.01	17.98	0.98	60.71	57.2	-3.5	0.29
3	3.15	2.14	15.27	15.8	0.53	66.87	64.77	-2.1	0.30
4	1.63	-0.52	16.61	16.87	0.26	61.56	60.75	-0.81	0.27
5	-0.34	-2.88	16.95	16.83	-0.12	59.78	60.26	0.48	0.31
6	2.21	3.2	10.99	10.93	-0.06	105.22	104.53	-0.7	0.28
Mean	2.34	0.70	15.37	15.68	0.32	70.83	69.50	-1.33	0.29
SD	1.99	2.42	2.55	2.77	0.45	19.42	19.77	1.52	0.02

Table 2.

Fraction of tumor slice enhancing with respect to gas breathing challenge

Tumor	TOLD	TOLD	TOLD	BOLD	BOLD	BOLD	PET
	Threshold 1SD (%)	Threshold 1.5SD (%)	Threshold 2SD (%)	Threshold 1SD (%)	Threshold 1.5SD (%)	Threshold 2SD (%)	Hypoxic fraction (%)
1	26	12	6	40	25	17	49
2	14	4	2	37	23	14
3	20	9	4	50	33	20	39
4	19	7	2	55	40	27	61
5	23	9	3	29	17	9	49
6	20	7	2	39	18	9	53

PET also showed distinct heterogeneity in each tumor with particular differences between central and peripheral regions, which evolved over 90 mins (Figures 1 and 2). The PET system failed during acquisition of data for Tumor #2, so that there were only 5 successful data sets. At early times there was relatively intense signal in the tumor periphery consistent with high perfusion and this declined over the first 15 mins to settle at about 1.2 (tumor to muscle: T/M). Meanwhile, the central regions increased continuously consistent with central hypoxia (Tumor #4; Figure 1). Tumors #5 (Figure 2) and #1 each showed very different behavior with much lower initial central tumor signal, which remained less than muscle even after 90 mins. Images of PET signal intensity at 3, 13 and 90 mins are presented for a central tumor slice to match the MRI TOLD, BOLD, and R₂* measurements for all five tumors in Figure S1.

At early time points (e.g., 3 mins) correlations were found between relative [¹⁸F]FMISO activity and mean TOLD ΔSI(%), BOLD ΔSI (%), and ΔR₂* on a global basis (r²>0.65) and for tumor periphery (all r²>0.75), while weaker correlations were found for central regions (r² between 0.37 and 0.62) (Figure 3). At 90 mins post infusion peripheral tumor regions showed a highly consistent value T/M=1.2±0.07 and lack of correlation with MRI (Figure 3E-H). In central tumor regions comparison of [¹⁸F]FMISO activity and MRI showed correlative trends, however removing the obvious outlier (Tumor #3) yielded strong correlations between BOLD, TOLD, ΔR₂* and PET (all r²>0.9).

Figure 3.

Correlations between MRI measurements and [¹⁸F]FMISO activity at two time points suggested the contribution from both perfusion and hypoxia conditions. Top panel: correlation of [¹⁸F]FMISO activity at 3 mins with MRI measurements (TOLD ΔSI(%), BOLD ΔSI(%), R₂*(air) and R₂*) of two tumor regions (center and periphery: Peri). Bottom panel: correlation of [¹⁸F]FMISO activity at 90 mins with MRI measurements.

MRI responsive enhancing fractions of the tumors based on TOLD and BOLD were calculated based on various thresholds (Table 2). Estimates of responsive fractions were found to be strongly influenced by the threshold, as presented for thresholds defined as changes greater than 1, 1.5 and 2 times the baseline standard deviation. As expected, the greater the threshold, the smaller the determined responsive fraction. While the absolute fractions change with threshold, the relative fractions were quite consistent with responsive TOLD fractions being correlated (r²>0.85). Meanwhile, the BOLD response was less sensitive to threshold with correlations r²>0.95 for 1, 1.5 or 2 SD. No obvious correlations were observed between BOLD and TOLD responsive fractions. Comparison of oxygen-sensitive MRI response fractions with PET hypoxic fraction indicated that both TOLD and BOLD showed significant correlations when 4 out of 5 tumors were considered, excluding the outlier (Figure 4A, 4B). In each case a larger hypoxic fraction associated with smaller oxygen responsive fraction: the correlation was much stronger for TOLD than BOLD.

Figure 4.

Responsive fractions compared with hypoxic fraction. Correlation of hypoxic fraction determined by ¹⁸FMISO PET in tumor slice with (A) T₂*-weighted fraction of slice enhancing by more than 2SD with oxygen breathing challenge (r²>0.65) and (B) T₁-weighted fraction of tumor image slice enhancing by more than 2SD with oxygen breathing challenge (r²>0.98). In each case correlation is based on 4 out of 5 tumors (red) with blue outlier neglected (Tumor #1 for TOLD and Tumor #4 for BOLD).

Discussion

We successfully achieved oxygen-sensitive MRI and hypoxia-sensitive PET in a group of Dunning prostate R3327-AT1 tumors growing subcutaneously in rats. Each approach revealed heterogeneity in the tumors. Strong correlations between MRI and early distribution of [¹⁸F]FMISO indicated that each method is strongly influenced by tumor perfusion. At later times correlations became more subtle, but strong inverse correlations were observed between hypoxic fraction determined by PET and enhancing fractions observed by oxygen sensitive MRI with respect to an oxygen gas breathing challenge.

BOLD and TOLD measurements matched our previous observations in AT1 tumors: baseline R₂* during air breathing was similar [36], as was the response to oxygen breathing [8,42]. We also found that mean BOLD and TOLD measurements were correlated and BOLD magnitude was typically about double TOLD [36]. Both MRI and PET indicated tumor heterogeneity, notably well perfused tumor periphery, which is typical of the Dunning prostate R3327-AT1 tumor type [45-48]. Figure 3 (3 mins) indicates peripheral perfusion based on PET and the responsive regions for MRI. At 90 minutes the periphery shows relatively low [¹⁸F]FMISO essentially at the threshold of 1.2 and was unrelated to MRI. Meanwhile, the signal at the tumor center correlated closely with MRI for 4 out of 5 tumors. Three of 5 tumors showed T/M>1.2 indicating hypoxia. We found that higher [¹⁸F]FMISO corresponded with larger BOLD and TOLD response. This suggests that hypoxia is present (hence T/M>1.2), but that tumor oxygenation could be modified in response to oxygen breathing. Indeed, we have previously observed baseline hypoxia in this tumor type [8,46,49], which could be modified by oxygen breathing, as indicated by BOLD and TOLD responses.

The correlations of the magnitude of BOLD and TOLD responses with FMISO uptake are interesting, but spatial extent of uptake may be more meaningful, specifically, relating to the classic hypoxic fraction (Figure 4). Here, both BOLD and TOLD showed significant intuitive correlations: in 4 of 5 tumors a large enhancing area defined by MRI correlated with a smaller hypoxic fraction determined from FMISO PET. In each case an outlier was observed (Figure 4). Separate tumors were outliers in the BOLD and TOLD observations and in each case the fractional MR enhancement was larger than expected for the extent of FMISO uptake or conversely FMISO uptake was larger than expected. Discrepancies may arise from various factors. It is important to note that the PET data were acquired from 3D data sets thereby reflecting the whole tumor, whereas MRI was limited to a single slice through a central region. For the outliers other tumor regions may have been exceptionally hypoxic.

Use of an exogenous reporter may be subject to issues of delivery; specifically severe hypoxia is likely caused by lack of perfusion in which case the [¹⁸F]FMISO may not reach the target tissue effectively. It is noticeable that every tumor showed extensive peripheral [¹⁸F]FMISO at early time points (3 minutes in Figures 1 and 2 and 13 minutes in Figure S1), but in some cases central perfusion was exceptionally low (Figure 2). At 90 minutes 4 of 5 tumors showed signal throughout the tumor, albeit with highly variable concentration. However, Tumor #5 (Figures 2 and S1) showed minimal uptake in the tumor center. This may reflect lack of hypoxia, but appears to be consistent with the low perfusion. Indeed, BOLD and TOLD response was particular low implying lack of vasculature and hypoxia. This suggests a need to account for perfusion, as widely recognized in the need to consider both early and late time points [35,50]. Likewise, O’Connor et al. have shown that consideration of the uptake of pimonidazole (a nitroimidazole widely used for immunohistological detection of hypoxia) correlates better with TOLD measurements in some tumors, if poorly perfused regions are identified by dynamic contrast enhanced MRI and excluded from consideration [22]. Here, we infused pimonidazole in some rats and staining emphasized predominant central hypoxia, though we have insufficient data to seek correlations (Figure S2). Exogenous reporter agents are subject to delivery to cells and uptake. Indeed, it is reported that uptake of [¹⁸F]FMISO was limited in tumors following vascular disruption based on DMXAA, thereby reflecting reduced perfusion as opposed to increased hypoxia [51].

We chose [¹⁸F]FMISO as the PET reporter for hypoxia, since it has been most widely used in both pre-clinical and human investigations [3,52]. However, we recognize ongoing efforts to develop enhanced agents and several reports comparing the relative efficacy to detect hypoxia in tissues [53]. FMISO has an octanol-water partition coefficient around 0.4 leading to extensive initial uptake in normal tissues, and hence relatively low contrast [3]. ¹⁸F-2-(2-nitro-1H-imidazol-1-yl)-N-(3,3,3-trifluoropropyl)-acetamide ([¹⁸F]EF3) is more lipophilic and might be expected to partition better into cells, but was reported to not be superior to FMISO [54]. ¹⁸F-2-(2-nitro-1H-imidazol-1-yl)-N-(2,2,3,3,3-pentafluoropropyl)-acetamide ([¹⁸F]EF5) is still more lipophilic and has been tested in clinical trials [55,56], but its preparation is more complex. A head to head comparison of several fluorinated nitroimidazoles indicated more rapid clearance for the relatively hydrophilic agents ¹⁸F-2-(4-((2-nitro-1H-imidazol-1-yl)methyl)-1H-1,2,3-triazol-1-yl)propan-1-ol ([¹⁸F]HX4) and ¹⁸F-fluoroazomycin arabinoside ([¹⁸F]FAZA) vs. [¹⁸F]FMISO in rhabdomyosarcomas, suggesting potential benefit for determining hypoxia with higher contrast at earlier times [57]. Chelated metal ions promise facile preparation of radiolabeled reporter agents [58,59]. Copper-diacetylbis(N4-methylthiosemicarbazone) (Cu-ATSM) has been evaluated extensively, but cell line dependent discrepancies in biodistribution have been reported [60]. A study comparing several agents showed higher uptake of ⁶⁴Cu-diacetyl-bis(N4-methylsemicarbazone) [⁶⁴Cu]Cu-ATSM]) in an SQ20b mouse tumor, but a lack of correlation with the immunohistochemistry hypoxia markers pimonidazole or endogenous CAIX [61]. Meanwhile, the three fluorinated agents [¹⁸F]FMISO], [¹⁸F]HX4, and [¹⁸F]FAZA showed similar tumor distributions, which were consistent with immunohistochemistry. There is also choice in nitroimidazole for immunohistological validation with pimonidazole and CCI-103F being readily available and widely used including pulse chase investigations to allow assessment of changes in hypoxia [37,42]. We selected pimonidazole for the current study, since it is much easier to use than the exceedingly hydrophobic CCI-103F, which must be administered in peanut oil IP [42].

It is of course important to note that FMISO uptake is indicative of hypoxia, whereas oxygen sensitive MRI is likely more closely related with the ability to manipulate tumor oxygenation. Our past studies suggest that tumors which show a very small BOLD or TOLD response are hypoxic and cannot be modulated by the oxygen breathing challenge [8,62]. Nonetheless the correlations observed here do suggest that oxygen-sensitive MRI is related to tumor hypoxia. The correlations observed between BOLD, TOLD and [¹⁸F]FMISO uptake provide further impetus for the use of non-invasive oxygen sensitive MRI to evaluate tumor oxygenation.

Disclosure of conflict of interest

None.