Introduction

Tumor microenvironment is a major determinant of tumor growth and development [1]. Aspects of the microenvironment, particularly pO2, modulate transformation and tumor growth as the immediate source of the conditions to which malignant cells respond [2]. Tumor microenvironments determine delivery and modification of therapeutic agents to cellular targets, often frustrating beneficial effects from agents showing promise in cellular experiments [3]. Microenvironmental lack of oxygen, hypoxia, determines radiation curability of tumors [49]. Changes in these microenvironments reflect response to therapy and may allow early prediction of ultimate therapeutic effect [10]. However, many microenvironment measurements are invasive opening questions as to their relevance to “real” native animal environments [1113].

Vascular endothelial growth factor A (VEGF) is a crucial protein generated in response to hypoxia and is the principal stimulus for tumor blood vessel formation, angiogenesis [14, 15]. VEGF-stimulated angiogenesis insures the survival and growth of human solid tumors beyond the short distance over which oxygen may diffuse before it is consumed [16]. However, the quantitative relationship between pO2 and VEGF production in native environments remain substantially unexplored. Electron paramagnetic resonance (EPR) images (EPRI) give spatial distributions of quantitative tumor oxygenation (EPROI) in native, “real” unperturbed conditions using injected non-toxic small molecule pO2 reporters of the environment [1721]. The excitation frequencies at which images are obtained are those of newer human magnetic resonance imaging (MRI), with magnetic fields that are less than 10 millitesla, 1/658 that of MRI. Spectroscopic imaging allows direct interrogation of the highly specific and sensitive EPR spectral response of the reporter molecules to various aspects of tissue and tumor microenvironment such as pO2 [22]. But a major difficulty in correlating pO2 with cell and tissue hypoxic response has been the rapid variation of pO2 with location. Comparisons of wildly fluctuating pO2 values with unlocated biopsies blur any real correlation with the VEGF hypoxia protein and prevent understanding of this varying microenvironment. In principle, local oxygen measurements with a polarographic electrode could be stereotactically registered with a subsequent stereotactically guided biopsy to try to relate small regions with high and low pO2 with hypoxia protein production. However, we are unaware of such registered measurements. Moreover, the electrode might perturb the environment sufficiently to create significant artifact.

Key to probing this relationship between the extent of hypoxia and the amount of VEGF production is co-location of the biopsy with the location of the pO2 values in the oxygen image. EPR oxygen images have spatial resolution of approximately 1 mm with a pO2 resolution of approximately 1–3 mmHg (torr) [23]. We previously demonstrated that rapid spatial variation of tumor pO2 seen in EPROI with gradients of 50 Torr/mm correlated with Oxylite™ measurements and gradients co-localized with the image pO2 voxels. To do this, we designed and built a sterotactic delivery system to bring the Oxylite™ tip to the same location in the tumor as the image voxel [19]. For the work described here, we modified this device to accept an 11-gauge bone biopsy needle. This allowed us to obtain biopsies whose contents could be registered or co-located with the proper EPR oxygen image voxels. Using this technique, we quantified for the first time in a native state, the in vivo relationship between EPROI oxygen concentrations and the VEGF hypoxia protein response of cells in a tumor. This technology opens the study of the relationship between in vivo EPROI and a variety of hypoxia proteins. A biopsy localized in a microenvironment image, e.g., a pO2 image, is a precondition to fully understanding the nuanced, quantified interplay between all aspects of the tumor microenvironment and its effect on tumor growth in situ in animals. These pO2 images will be eventually obtained in humans to guide individual human tumor therapy. It is crucial to use this stereotactic biopsy technique to understand the local response of animal tumors to oxygen specifically where it is occurring.

Materials and Methods

Short Description of the Basic Stereotactic Location Method

The basic observation predicating this method development is that oxygen physiology varies rapidly with position as shown below in EPR oxygen images. Thus physiologic correlates need to be obtained at the same position as the voxels in the EPROI. We previously dealt with a similar problem in validating the EPROI. A stereotactic system was developed to launch an Oxylite™ fiberoptic oxymetric probe into animal tissue in the imager so that the oxygen sensitive tip of the Oxylite™ could be located in the image and registered with the corresponding image voxels. This validated the pO2 values measured in each EPROI voxel [19]. For the measurements described here, biopsies, from which the quantitative levels of a hypoxia response protein were derived, needed to be located in the EPROI. The Oxylite™ was replaced with a stainless steel 11-gauge bone biopsy needle with an inner diameter of 1.8 mm. Biopsies were obtained from tumor volumes that contained ~200 image voxels. The substantially larger diameter of the needle stiffened it relative to the 200 μm glass fiber diameter of the Oxylite™ eliminating concern about the needle bending creating location uncertainty. Moreover, the larger size of the biopsy reduced sensitivity to slight variations (order, 0.5 mm) in relative location of biopsy and image co-location. By recording the stereotactic coordinates of the needle tip when it touched a known location in the image coordinate system, the locations of the biopsy and those of the image could be related or registered. This provided a means for imaging the animal tissue and tumor pO2 in a native condition which is violated only at biopsy, which was obtained in a matter of minutes. Thus, pO2 data was obtained with minimal perturbation of the animal, avoiding artifact and distortion of the data.

Spin Probe

EPRI uses a specific spin probe injected into the animal specimen to report specific aspects of the tissue environment, in this case and for this probe, oxygen [24]. The spin probe was OX063 trityl radical (methyl-ris[8-carboxy-2,2,6,6-tetrakis[(2-hydroxyethyl]benzo[1,2-d:4,5-d′]bis[1,3]dithiol-4-yl]-, trisodium salt, MW 1427), a kind gift from GE Healthcare, London, GB. A single inhomogeneously broadened spectral line with a peak-to peak line width of 16.5 μT and a spin packet linewidth of 1.6 μT in hypoxic solutions is further broadened in the presence of oxygen by 0.054 μT/Torr. Other free paramagnetic species, self-broadening, viscosity broadening, and power saturation can broaden the line as well but these effects are small and are accounted for via an oxygen calibration [18]. A concentration of 0.25 g/mL of OX063 was achieved by dissolving the intensely green powder in distilled water. Any residual particles were cleared using a 40 μm filter. We injected a 0.2 mL of this solution (1.4 mmol/kg body weight) intravenously into a mouse weighing 25 g. This corresponds to ~1/5 of the LD50 for the spin probe [25]. The spin probe is a triacid salt. The solution was, therefore, slightly hypertonic and was well tolerated. The tumor half-life of the spin probe of ~30 min was much longer than that in other tissues.

Animal Preparation

Seven tumors were used for the comparison of EPRI with VEGF concentrations. FSa cells were a kind gift from Kathy Mason, M.D. Anderson Hospital, Houston, TX. Cells were injected intramuscularly into the mid distal hind legs of 6–8 week old C3H/HeN/Hsd (Harlan Sprague Dawley, Indianapolis, IN) female mice. Tumors were grown to 9 to 11 mm diameter in 10 days. Animals were both immobilized and anesthetized for both biopsy and EPROI measurements, minimizing animal movement during the procedures. Gas anesthesia with 1-2% isofluorane mixed with medical grade air was adjusted to regularize breathing and prevent animal motion during the imaging and the biopsy procedure. Respiration frequency and depth were monitored continuously. A 24 G Angiocath was used to cannulate the mouse tail. The leg/tumor was immobilized with soft elastic vinyl polysiloxane dental mold material (GC America Inc, Alsip, IL) as shown in Fig. 1. A Physitemp (Clifton, NJ) digital needle probe thermometer was used to measure the skin temperature. The skin temperature was maintained at 33.5°C with opposed heating lamps. The trityl spin probe accumulates rapidly in the mouse bladder and creates a changing source of confounding signal and consequent artifact. Using a double lumen urethral catheter and Harvard 22 syringe pump (Harvard Apparatus, Holliston, MA), we flushed the bladder at 15 mL/h to eliminate this [26].

Fig. 1.
figure 1

Experimental setup. FSa tumor in the resonator, immobilized with vinyl polysiloxane (blue). Stereotactic device was used to register the location of the 11-gauge bone biopsy needle in the tumor volume.

The beginning of imaging to the end of the biopsy procedure was approximately 2 h. The complete experiment, including the animal preparation (approximately 45 min), could be up to 3 h long. We carefully maintained a steady level of anesthesia and monitored the breathing rate and temperature during the experiment to ensure that the oxygenation level of the tissues was similar for EPRI and biopsy.

All animal experiments were performed according to the US Public Health Service “Policy on Humane Care and Use of Laboratory Animals” and the protocols were approved by the University of Chicago Institutional Animal Care and Use Committee (ACUP No 69681). The University of Chicago Animal Resources Center is an AALAC approved animal care facility.

EPR Imaging

4D EPR images were obtained using a spectroscopic imager with an air-core magnet and Anderson type gradient coils [27] operating at 250 MHz and 9 mT as described previously [28]. The Zeeman field was modulated at 4.98 KHz with over-modulation. The peak-to-peak linewidth of the OX063 trityl is 16.0 μT. We used an accurate line shape simulation that allows operation with high modulation amplitude (17 μT) to increase signal-to-noise ratio (SNR) in images without sacrifice in the determination of the intrinsic or spin packet line width [29, 30]. The quality factor (Q) of the resonator was approximately 280 without the tumor and 200 with the tumor installed. The power delivered to the resonator was 25 μW and well below the saturation level (200 μW).

EPROIs were acquired and reconstructed using a 4D tomographic spectroscopic imaging technique. This involves the acquisition of spectra with fixed gradients, referred to as projections. Each projection involved the acquisition of 256 field points scaled to uniformly populate the field interval determined by the gradient as has been described previously [23]. The projections were subjected to a Gaussian filter whose width was four points and then subsampled to 64 points. A field of view of 3 cm was used. This determined our voxel size to be (3√2 cm/64) or 0.66 mm in the spatial direction. This voxel size determination has been discussed extensively elsewhere [18, 28].

Projections were obtained with 14 gradient amplitudes and a subset of ten polar and ten azimuthal direction angles chosen to sample the solid angle uniformly [31]. An image required approximately 45 min. The low SNR at higher gradient projections were partially compensated for by signal averaging of multiple field sweeps, up to ten averages for the highest gradient. Each sweep was obtained with 256 field points with acquisition time of 3 ms/point, 3 ms time constant, and 12-dB/octave filter using a SR830 lock-in amplifier (Stanford Research Systems, Sunnyvale, CA). The spatial field of view (FOV) and the spectral FOV were 3 cm and 0.1 mT. Spectroscopic image reconstruction used filtered back-projection as described previously [18, 28, 32]. All image voxels with spin probe amplitude less than or equal to 15% of the maximum spectral amplitude in the image were discarded. This 15% threshold was used to avoid fitting voxels with no signal or very poor signal. Small areas within several images had maximum intensities that did not survive this threshold. The oxygen values for these areas were discarded. The oxygen maps reconstructed and fitted from the images have pO2 resolution of ~3 Torr and a measured spatial resolution of ~1 mm. Oxygen partial pressure resolution is limited by the accuracy of the spin probe spectral linewidth measurement. The actual source of the uncertainty is a complicated function of signal noise, sampling frequency and the various forms of data filtering used in the reconstruction. In homogenous phantom imaging experiments we have found that linewidth scatter is 0.16 μT, which is equivalent to 3 Torr via the oxygen vs linewidth calibration which is discussed in ref. [18]. Each image has a voxel size of (0.66 mm)3. The basic spectral response to oxygen is the broadening of the line in response to the collision with local molecular oxygen. [33] Thus, to first order the technique is insensitive to the amount of spin probe present, which is proportional to the amplitude or height of the signal. As have others, minimal self-broadening corrections were used for low (less than 1 mM concentrations) of spin probe [30].

Selecting the Location of the Biopsy Needle in the Oxygen Image

Immediately after the EPR imaging, the pO2 image of the leg with the tumor was reconstructed, and pO2 values from all the voxels were printed out in a series of tracks, with a precise location of the track in the tumor. Biopsy position was picked, essentially on the fly, as one that has predominantly low values (≤20 Torr) at one end, and high values (≥20 Torr) at the other end of the track along the y-axis, the axial direction of the resonator. Fig. 2 shows an example of the voxels in the EPROI chosen for analysis corresponding to the biopsy in the tumor. Fig. 2 represents the voxels from a sagital and a coronal view of a typical EPROI. The voxels shown inside and including the bright red square were selected for analysis. The figure shows the central track corresponding to the center of the needle plus and minus two voxels transverse to the track. This track size was chosen based on the inner diameter of the needle (1.8 mm) and the resolution of the measurement (~1 mm). Given the voxel size of 0.67 mm, this is best approximated by a transverse square of 5 × 5 voxels.

Fig. 2.
figure 2

Location of voxels in axial and sagittal slices from 3D EPR oxygen image, corresponding to the location of a biopsy, which is outlined in red. The pO2 for each voxel is indicated in torr (mmHg) by the color bar to the right of the image.

Biopsy Technique

While the animal was still in the EPR imager and under anesthesia, the sterotactic positioning device for location of the biopsy needle was rapidly and easily installed (Fig. 1). The chosen position of the biopsy was marked on the skin of the tumor. Then the skin was cut with small surgical scissors. Biopsies were obtained with an 11-gauge (1.8 mm inner diameter) bone biopsy needle (Becton Dickinson, Franklin Lakes, NJ). The biopsy needle was inserted into the puncture wound and then fully through the tumor to approximately 10-12 mm depth. The needle was rotated carefully to isolate the biopsy from the tumor tissue and then withdrawn slowly, with a small suction applied using a syringe. Depending on the tumor size, the length of the biopsy was between 7 and 12 mm. These 1.8-mm diameter tubes of tissue passed through both well oxygenated and hypoxic regions. Based on the information from EPROI, biopsy obtained was then divided longitudinally into hypoxic and oxic samples. We obtained a total of 17 samples from seven tumors imaged.

VEGF Determination in Tumor Biopsy

Biopsies were weighed, homogenized, and transferred into a lysis buffer containing 50 mM Tris pH 7.4, 3 mM EDTA, 20 mM β-glycerophosphate, and proteases inhibitors cocktail (chemicals were from either Fisher Scientific, Pittsburgh, PA, USA, or Sigma Aldrich, St. Louis, MO, USA). After three freeze–thaw cycles to break the cell membranes, the homogenates were centrifuged for 5 min at 5,000×g (Eppendorf centrifuge 5,415°C), 4°C and supernatants were collected. The total protein concentrations were determined using Bradford assay (Bio–Rad Laboratories, Hercules, CA, USA).

VEGF protein concentration was determined using the ELISA Quantikine Mouse VEGF Immunoassay (R & D systems, Cat. no. MMV00, Minneapolis, MI, USA). No significant cross-reactivity tested for several other factors, including other members in the VEGF family, was observed. The intra-assay precision is 4.3–8.2%, average percent recovery is 91% in mouse tissue homogenates, and the technique is sensitive to less than 3.0 pg/mL. The amount of the VEGF in the biopsy was expressed as pg of VEGF protein per mg of the total protein in the biopsy.

Statistical Analysis

The biopsy volume, projected into the EPROI as shown in Fig. 2, selected an average of 100 voxels. We accepted voxels surrounding the inner volume of the biopsy needle in order to account for the cubic sampling and the 1 mm resolution of the EPROI. This number allowed us to calculate descriptive statistics from the projected volume in the EPROI including median pO2, mean pO2, fraction of voxels less than 10, 6, and 3 Torr. The fraction of voxels less than 10, 6, and 3 Torr is referred to, respectively, as HF10, HF6, and HF3. Errors in VEGF concentration were determined from three separate VEGF concentration sub-samples from each sample. Uncertainties in the hypoxic fraction were calculated by computing the uncertainty in a fraction propagated from uncertainties in the numerator and denominator. These uncertainties were estimated making the assumption for each number a Poisson distribution with standard error equal to the square root of the numbers of voxels in each group, e.g., HF10 error was calculated as:

$$ {\text{ $ \Delta $ HF}}_{{10}} = {\text{HF}}_{{{\text{10}}}} *{\sqrt {\frac{{{\text{ $ \Delta $ NHv}}^{{\text{2}}} }} {{{\text{NHv}}^{{\text{2}}} }} + \frac{{\Delta {\text{Nv}}^{{\text{2}}} }} {{{\text{Nv}}^{{\text{2}}} }},} } $$

where NHv is the number of hypoxic voxels, and Nv is the total number of voxels in the biopsy.

The Pearson product moment coefficient of the correlation between oxygen statistics from EPROI and VEGF mass per mass of sample protein R, and p, the significance of each correlation were calculated using STATA software.

Results

EPR Oxygen Images Show Steep Gradients in Oxygen Tension in the Tumor Over Distances of 1 mm

Fig. 1 shows an FSa fibrosarcoma in the leg of a C3H mouse immobilized in the resonator/sample holder of the EPR oxygen imager. The 11 gauge biopsy needle is fixed in the stereotactic needle holder which is attached to the imager, locating the needle tip in the image coordinate system of the imager. The biopsy needle is seen extending into the tumor. Fig. 2 shows the oxygen image obtained from the FSa fibrosarcoma. The colorbar in Fig. 2 correlates color with oxygen partial pressures in each (0.66 mm)3 voxel. The image shows axial and sagittal planar pO2 slices of the tumor. Oxygen gradients of 40 Torr/mm are seen in the image. These gradients are typical of most tumor EPROI. The red box represents the boundaries of the biopsy voxels contained in the needle shown in Fig. 1. The last 5.3 mm of the needle provide the material from which the VEGF was sampled in this particular measurement, shown in the red box in Fig. 2a, b. From the pO2 voxels within and including the box, EPROI statistical summaries of the roughly 200 pO2 voxels were obtained including mean pO2, median pO2, and fraction of the voxels with pO2 less than 3, 6, and 10 Torr. These fractions of the total number of voxels with pO2 less than a certain value are EPROI hypoxic fractions, referred to as HF3, HF6, and HF10.

All pO2 Statistical Summaries from the pO2 Image of the Biopsies Correlate Significantly with VEGF Protein Content in the Biopsy

The first column of Table 1 shows the Pearson product moment correlations (R value) for each of the biopsy pO2 voxel statistical summaries and the second column shows their significance. All of the correlations have p ≤ 0.05. The p values of the hypoxic fractions are approximately a factor of two smaller than those of the median and mean pO2 indicating that the significance of the hypoxic fractions are twice that of the mean and median pO2. R values rather than R 2 are used because it is a measure of the part of the linear regression line due to true variation and not due to random variation. Fig. 3 shows a plot of the VEGF concentrations in each biopsy vs the HF10. The linear regression is also shown, corresponding to the stated R of 0.63. As can be seen in Fig. 2, within a biopsy there are steep gradients in pO2. This weakens the correlation between mean or median pO2 and favors the hypoxic voxel fractions. This is consistent with the smaller p values and improved significance of the hypoxic fraction statistics relative to overall voxel pO2 average and median statistics.

Table 1 Correlations and their significance between VEGF concentrations and various hypoxic fractions and the mean and median oxygen pO2 in the biopsy specimen determined from the voxels of the biopsy track cast into the EPR oxygen image with different margins
Fig. 3.
figure 3

Correlation between hypoxia and VEGF content. Fraction of voxels in the biopsy with pO2 less than 10 Torr, HF10 from the EPR oxygen image vs VEGF concentration in picograms per microgram of total protein. For method of error calculation, see “Material and Methods”, “Statistical Analysis”.

Discussion

In prior work, in vitro measurements of human MX-1 breast cancer cells showed a quantitative relationship between VEGF levels and pO2 in the atmosphere in which cells were grown [12]. Ambient pO2 levels between 3% and 0.1% led to several fold increase in the amount of VEGF mRNA, as well as the increased level of the protein itself, with half-maximum levels at 1.5–3%, i.e., 10 and 20 Torr of oxygen [11]. Inspiratory hypoxia (at 10% O2), a qualitative measure of tumor hypoxia, increased the VEGF levels in MX-1 tumors grown in nude mice [11]. Using a cranial window chamber, Fukumura et al. [13] orthotopically grew human gliomas in SCID mice. Quantitative pO2 images locally correlated with VEGF expression as seen using a coupled green fluorescent protein. These studies showed the spatial registration of oxygen and VEGF induction as well as quantified and localized pO2 correlation with VEGF expression. This was accomplished at the cost of an invasive and artificial experimental paradigm, and limiting sensitivity to superficial structures near the window.

The data presented here represent an insight into the quantitative relationship between pO2 and the production of VEGF with the pO2 information obtained under native in vivo conditions. Prior work investigating the parameter that optimally predicts of the effect of pO2 on the outcome of tumor control with radiation choose a 10 Torr measure of in vivo tissue hypoxia. This includes the pioneering work of Kolstad in cervical malignancies [34] and Gatenby [35] looking at radiation control of lymph nodes and that of Hockel et al. in uterine cancer and Brizel et al. in soft sarcoma [8, 36]. Other authors found HF2.5 to be a prognostic factor in head and neck cancers [37]. In vitro measurements show the pO2 for half the oxygen enhancement is consistently closer to 3 Torr. Our own laboratory has shown that the optimal EPROI pO2 predictor of tumor control by radiation, in addition to radiation dose is the HF10. These EPROI were obtained in native conditions as those described here. They also show a high significance of the correlation with the HF3 but to a slightly smaller degree in correlation magnitude and significance than the HF10 [38]. It is to be expected that different pO2 measurement techniques measure different aspects of local oxygen and will highlight somewhat different pO2 values [39]. A review by Peter Vaupel, et al., also demonstrates the strong correlation between tumor cure and hypoxic fractions at threshold values of from 2.5 to 10 Torr [40].

Since normoxic VEGF protein half-life is 15–20 min and the half-life of VEGF mRNA is 30 min [41], while hypoxic VEGF mRNA half-life increases to 6–8 h, VEGF protein levels appear to be more sensitive to chronic hypoxia [4143] rather than transient hypoxia lasting only a few minutes [4447]. The tissue level of VEGF is an equilibrium between VEGF production and secretion and conveyance away through the circulation, a chronic condition. EPR oxygen images are obtained in approximately 45 min. It is therefore also more sensitive to and diagnostic of chronic hypoxia. Thus, VEGF concentration and EPROI hypoxia are responding to the same type of microenvironmental hypoxia.

Conclusions

EPR imaging of pO2 and registered localized interrogation of the hypoxic signaling response is a new paradigm for biologic measurement. EPR oxygen images provide information that is spatially localized. The spatial differentiation allows spatial patterns to be discerned, characterizing the individual subject [38]. On the other hand, registration of the oxygen images with localized biopsy, shown for the first time in this work, allows multiple experiments to be obtained from the same location in an animal tumor to determine how the oxygen tension is related, in vivo, with the quantified production of a protein synthesis response to quantified low oxygen tensions. As images improve in quality, simple linear correlation between pO2 hypoxic fraction and VEGF concentration will give way to probing of pO2 thresholds for VEGF signaling or other, non-linear behavior that is more likely to characterize individual response both to hypoxia and to therapy. Because EPROI show individualized patterns of hypoxia, they will allow individualized delivery of cancer therapy or individualized risk assessment for myocardial infarction, stroke, or peripheral vascular disease. For all of these diseases, response to therapy can be individualized using EPROI.