Monitoring Tumor Hypoxia Using 18F-FMISO PET and Pharmacokinetics Modeling after Photodynamic Therapy

Photodynamic therapy (PDT) is an efficacious treatment for some types of cancers. However, PDT-induced tumor hypoxia as a result of oxygen consumption and vascular damage can reduce the efficacy of this therapy. Measuring and monitoring intrinsic and PDT-induced tumor hypoxia in vivo during PDT is of high interest for prognostic and treatment evaluation. In the present study, static and dynamic 18F-FMISO PET were performed with mice bearing either U87MG or MDA-MB-435 tumor xenografts immediately before and after PDT at different time points. Significant difference in tumor hypoxia in response to PDT over time was found between the U87MG and MDA-MB-435 tumors in both static and dynamic PET. Dynamic PET with pharmacokinetics modeling further monitored the kinetics of 18F-FMISO retention to hypoxic sites after treatment. The Ki and k3 parametric analysis provided information on tumor hypoxia by distinction of the specific tracer retention in hypoxic sites from its non-specific distribution in tumor. Dynamic 18F-FMISO PET with pharmacokinetics modeling, complementary to static PET analysis, provides a potential imaging tool for more detailed and more accurate quantification of tumor hypoxia during PDT.


Cell lines and animal models. The MDA-MB-435 and U87MG cell lines were obtained from American
Type Culture Collection (ATCC, Manassas, VA). The MDA-MB-435 cells were cultured in Leibovitz's L-15 medium supplemented with 10% FBS (GIBCO, Grand Island, NY) at 37 °C and 5% CO 2 whereas the U87MG cells were cultured and passaged in Eagle's minimal essential medium supplemented with 10% FBS at 37 °C and 5% CO 2 .
Athymic nude mice were obtained from Harlan laboratories (Frederick, MD, USA). All experiments with live animals were conducted under protocols approved by the National Institutes of Health Clinical Center Animal Care and Use Committee (NIH CC/ACUC). The methods were carried out in accordance with the approved guidelines. The MDA-MB-435 and U87MG tumor models were generated by subcutaneous injection of 5 × 10 6 cells in l00 μ l of PBS into the right shoulder of nude mice. The mice were used for imaging and photodynamic therapy when the tumor volume reached around 100 mm 3 (10-15 days for U87MG and 15-20 days for MDA-MB-435).
PDT treatment. When the tumor size reached ~100 mm 3 , mice received an intravenous injection of 2-[1-hexyloxyethyl]-2-devinyl pyropheophorbide-alpha (HPPH) 18 (0.47 μ mol/kg). At 24 h post-injection, mice were anesthetized by inhalation of isoflurane (1% in 1 L/min oxygen). Each mouse was exposed to laser irradiation at a wavelength of 671 nm (Laserglow Technologies, ON, Canada) with a fluence of 75 mW/cm 2 for 30 min 19 . Laser energy was measured by a LPE-18 laser energy meter (Coherent Portland, OR, USA). During PDT treatment process, mice were positioned in a specially-designed mouse holder, and were anesthetized by inhalation of isoflurane (1% in 1 L/min oxygen).
Static PET imaging. Static PET imaging was performed with an Inveon ® PET scanner (Siemens, USA) to measure the 18 F-FMISO tumor uptake before and after the PDT treatment. Approximately 3.8 MBq of 18 F-FMISO was injected intravenously to tumor-bearing mice. Ten-minute static PET scan was performed at 2 h post-injection. The mice were anesthetized by inhalation of isoflurane (1% in 1 L/min oxygen) during the 10-min scan. The static PET acquisition was performed before, 1 h and 24 h after the PDT treatment. PET images were reconstructed using 3D ordered-subsets expectation maximum followed by maximum a posteriori algorithm with a smoothing parameter of 0.1 (OSEM-3D-MAP). The tumor region of interest (ROI) was defined by applying a threshold on the reconstructed PET images to eliminate tumor necrotic region. The mean radioactivity of the 18 F-FMISO tumor uptake was calculated with decay correction from the entire tumor region of interest (ROI), and was compared before and after PDT treatment.
Dynamic PET imaging. Two-hour dynamic PET imaging was performed with an Inveon ® PET scanner (Siemens, USA) before and 1 h, 6 h, 24 h after PDT treatment. Because the circulation time of 18 F-FMISO was relatively long compared to other tracers with active transport mechanism (e.g. FDG) 15,20 , the acquisition time for the dynamic imaging was set to 2 h. For each dynamic scan, approximately 4 MBq 18 F-FMISO was injected intravenously into the mice at the start of PET acquisition. Mice were anesthetized by inhalation of isoflurane (1% Scientific RepoRts | 6:31551 | DOI: 10.1038/srep31551 in 1 L/min oxygen) during the 2 h scan. Same group of mice were used for 1 h and 24 h post PDT, while a separate group of mice were used for the 6 h post PDT time point in order to avoid the residue of radioactivity remained from the 1 h post treatment group. Dynamic PET images were reconstructed into 52 time frames using (OSEM-3D-MAP). Frame rates were: 10 × 30 s, 10 × 60 s, 10 × 120 s, 10 × 150 s, 12 × 300 s. Pharmacokinetic modeling. The pharmacokinetic modeling 15,21,22 was conducted to analyze the time series data from the dynamic PET imaging. To extract the tumor TAC from the PET data, the last frame of the dynamic PET image was used to define the tumor ROI. A threshold was applied to eliminate the necrotic region. The defined ROI was then applied to all time frames. Each point of the tumor TAC curve was given by the mean tracer tumor uptake within the defined ROI of each time frame. The TAC curve generated from the left ventricle ROI, which was defined with first minute frame, was used as the plasma input function.
The irreversible two-tissue compartment model 15,20,21,23 was chosen as the mathematical model in the present study to analyze the 18 F-FMISO's transportation and internalization rate from the plasma to the tumor region. The transport of 18 F-FMISO from vessel to hypoxia sites is considered to be purely diffusive 15 , and based on the diffusion equation 24 , the tracer's concentration within the compartment is derived as: , C a (t) are the 18 F-FMISO activity concentration (Bq/mL) as a function of time in the compartment corresponding to plasma, non-specific diffusive tumor area, and specific accumulative area, respectively. The plasma input function Cp(t) was the TAC curve generated from left ventricle ROI. K 1 (mL/g per min) is the transport rate constant of tracer from plasma to tissue, k 2 and k 3 are kinetic transportation rate constants (in min −1 ) between diffusion and accumulation compartments. The activity concentration was calculated by solving the above differential equations. The signal intensity measured in a given ROI on PET images was a weighted sum of C p (t), C d (t), and C a (t), expressed as: where w p is the fractional blood volume in tumor, w d and w a are the relative contribution of diffusion and accumulation compartments to the rest of tumor volume (1− w p ). Given the measured TAC of tumor ROI and blood input function, the K 1 , k 2 and k 3 values were calculated by fitting the measured TAC with the analytical ROI function C ROI (t) (equation (2)) using non-linear least-square regression. The goodness of the fitting was evaluated using the residual analysis 25 , and the error (between the data value and estimated function value) was limited to be less than 5%. The compartmental modeling was conducted using a home-made Matlab object-oriented toolbox. Input parameters include plasma input TAC data, tumor ROI TAC data, and initial guess of K 1 , k 2 , k 3 , w p , and w d values. According to literature 15,21,23 and preliminary fitting test, the K 1 , k 2 , and k 3 were initialized to 0.01; w p and w d were initialized to 0.1. The k 3 values fitted from the PET data were compared before and after treatment. K i (mL/g/min) determined in Gjedde-Patlak plot in non-compartmental graphic analysis 23,26,27 , was also calculated and compared for both tumor models at different time points.
K i represents the influx rate constant of the irreversibly trapping tracer from plasma to the tumor as an entire system. Thus it takes into account both specific and nonspecific distribution of 18 F-FMISO inside tumor. Parametric mapping 23 of the influx rate K i and tracer's specific internalization rate k 3 was generated using the Siemens Inveon Research Workstation. K i mapping was applied for the whole body area using Patlak graphical analysis, with a reference tissue selected in the muscle area. The k 3 map was calculated locally in the tumor area based on the irreversible compartment model presented above. Higher intensity at certain voxel of the parametric map image means higher kinetic rate value at this position. The parametric maps were also overlapped with the last frame of dynamic PET image. The tumor volume was approximated based on the 18 F-FMISO tumor uptake of the last frame of dynamic PET. The percentage of overlapping of K i and k 3 map volume comparing to the approximated entire tumor volume was calculated and compared at different time points post PDT.
Immunohistochemistry. Immunohistochemistry analysis was performed to observe ex-vivo tissue hypoxia with HydroxyprobeTM kit (Hydroxyprobe Inc., Burlington, MA). Mice bearing U87MG tumors and MDA-MB-435 tumors were injected with 60 mg/kg of pimonidazole before and 2 h after PDT treatment. After 1 h to 90 min of incubation of pimonidazole, mice were euthanized and tumors were harvested. Tissues were fixed in a 4% formaldehyde solution at room temperature overnight. Tissue sections were prepared and stained following the procedure suggested in the manufacturer's instructions 28 . The staining slices were then visualized with ×10 magnification using Olympus BX41 microscope. All microscopic images were adjusted with the same white balance, thus tumor hypoxia was reflected directly by the pimonidazole staining color intensity (in brown color). The images were recorded with the field of view avoiding the edge of stained tumor sections. The hypoxic fraction (% of positively stained area) was quantified using ImageJ, in threshold recorded images to highlight the Scientific RepoRts | 6:31551 | DOI: 10.1038/srep31551 area positively stained by pimonidazole (in dark brown color). The threshold intensity depends on the staining intensity of each image and the results were averaged. Statistical analysis. Quantitative 18 F-FMISO uptake values were expressed in %ID/g, and the kinetic parameters k 3 and K i were in min −1 and mL/g/min respectively. These values were presented in form of mean ± SD. The differences between two tumor models at different time points were evaluated using Student's t test. P value less than 0.05 was considered to be statistically significant. The Pearson correlation coefficient r was calculated to evaluate the correlation between parameters K i and k3, where r = 1 was total positive correlation and r = 0 was no correlation.

Results
The quantitative analysis of static PET performed for both MDA-MB-435 and U87MG tumor models was presented in Fig. 1. According to Rajendran et al. 29 , tissue-to-blood activity ratio (T/B) higher than 1.2 indicates hypoxia in the target tissue. The baseline tumor uptake of U87MG was 0.75 ± 0.2%ID/g (n = 4) at 2 h post injection, while the heart uptake of at 2 h post-injection was at the same magnitude as the tumor uptake regardless of the inter-subject difference (mean 0.82 ± 0.1%ID/g). The T/B ratio of U87MG tumor was approximately 0.9, indicating intrinsically non-significant hypoxia in U87MG tumors. The baseline tumor uptake of MDA-MB-435 tumor was 1.92 ± 0.35%ID/g (n = 4), while much higher contrast was observed compared to the heart uptake (mean heart uptake 0.65 ± 0.3%ID/g). The T/B ratio was approximately 3, indicating intrinsic hypoxia in MDA-MB-435 tumors. At 1 h post PDT, there was a dramatic increase in 18 F-FMISO uptake of the MDA-MB-435 tumor (increased to 6.27 ± 0.69%ID/g, p = 0.0015) comparing to its baseline. However, at 24 h post treatment, the MDA-MB-435 tumor uptake decreased to 3.88 ± 0.34%ID/g. While the U87MG tumor uptake increased to 3.41 ± 0.15%ID/g at 1 h, and further increased to 6.8 ± 0.3%ID/g at 24 h post treatment. The overall increase of 18 F-FMISO tumor uptake from baseline before treatment to that at 24 h post PDT treatment was 700% (p < 0.001) in U87MG tumor and 200% (p < 0.001) in MDA-MB-435 tumor. A clear difference in 18 F-FMISO tumor uptake was shown on the PET images for U87MG tumor model before and 24 h after PDT, while for MDA-MB-435 tumor model this difference was less visible on PET images.
In order to monitor the kinetics of 18 F-FMISO retention to hypoxic sites after treatment, dynamic PET scan with kinetic modeling was performed before and after PDT. The mean value of 18 F-FMISO kinetic rates (K 1 , k 2 , k 3 , K i ) were analyzed for the whole tumor ROI and was compared before and after treatment ( Fig. 2A, Table 1). In U87MG tumor model, 18 F-FMISO tumor uptake intensity shown on the dynamic PET frame was nearly the same as the background intensity (< 0.9%ID/g) before PDT. The reversible transportation of tracer between plasma and tumor interstitial might still exist under normoxic condition, however, the 18 F-FMISO retention to hypoxic sites was negligible. The definition of the tumor ROI based on PET image and TAC curve extraction were not feasible. Therefore compartmental analysis was not applicable for the U87MG model before PDT treatment. To simplify The dynamic data of the two tumor models were further analyzed at voxel level with K i and k 3 parametric maps and were compared at different time points post PDT treatment. The whole body K i mapping was shown in Fig. 3. As K i took into account both non-specific and specific distribution, K i mapping was thus similar to the uptake intensity PET image. However, the K i parametric map revealed higher (p < 0.001) tumor to tissue contrast ratio It is interesting to compare the parametric maps of K i and k 3 around the tumor region with the tumor volume measured from PET image. For the U87MG tumor model, the K i mapping area overlapped with approximately 95% of the entire tumor region at 1 h post PDT treatment while k 3 mapping (Fig. 4) was shown only in a part of the tumor region near the tumor boundary area (approximately 21.5 ± 4% of the tumor volume, n = 4). At 24 h post PDT, the K i mapping area overlap decreased to 80 ± 8%, however, the approximate percentage of the k 3 mapping area increased to 68.5 ± 10% and was mostly located in the center of the tumor region. For the MDA-MB-435 tumor model, the K i mapping overlapped with approximately 100% of the entire tumor region at 1 h post PDT treatment, which decreased to 75 ± 5% at 24 h post PDT. The k 3 mapping covered approximately 34.8 ± 5% of tumor volume at 1 h post PDT. However, this percentage decreased to approximately 19.5 ± 5% at 24 h post PDT. The tumor hypoxia development in both MDA-MB-435 and U87MG tumors before and after PDT was also visualized qualitatively with pimonidazole staining (Fig. 5). The hypoxic fraction of MDA-MB-435 tumor model was 30 ± 9% vs. 10 ± 5% for U87MG tumor (n = 3) before PDT, indicating higher intrinsic tumor hypoxia in MDA-MB-435 tumor. At 2 h post PDT, the hypoxic fraction increased significantly in both tumor models (55 ± 12% for MDA-MB-435, p < 0.005, vs. 35 ± 15% for U87MG, p < 0.001). It is notable that intense brown spots were observed in the MDA-MB-435 tumor after PDT, suggesting that tumor hypoxia was increased with high heterogeneity; while in the U87MG tumor, more homogeneous enhancement was observed.

Discussion
The responsiveness of tumor to phototherapy has been monitored previously by neovasculature imaging [30][31][32] . The oxygen consumption and tumor hypoxia during PDT has also been characterized in vitro [33][34][35] . Moore et al. assessed tumor hypoxia change in response to PDT in vivo using scintigraphic imaging 9 . There have also been studies of tumor hypoxia by static PET imaging 8,13 . However, hypoxia measurement with static PET alone may not be able to fully reflect tumor hypoxia condition 14 . In the present study, dynamic PET using hypoxic specific tracer 18 F-FMISO combined with pharmacokinetics modeling was performed in addition to static PET.
The change of hypoxia status under PDT in two tumor xenograft models (U87MG and MDA-MB-435) was investigated in this work. The reason that these two models were chosen for comparison was that MDA-MB-435 is intrinsically hypoxic while U87MG is initially low to non-hypoxic before treatment 36 . It would be interesting to know whether different behaviors of hypoxia change would occur after PDT. Our results suggested that although U87MG tumor was initially low to non-hypoxic, therapy-induced hypoxia developed continuously subsequent to PDT (Fig. 1). On the contrary, the MDA-MB-435 tumor with intrinsic hypoxia responded quickly to PDT and produced acute tissue hypoxia at early time points post PDT as mean 18 F-FMISO tumor uptake value increased dramatically at 1 h post PDT but then decreased at 24 h post PDT. Therefore, different patterns of tumor hypoxia variation during PDT could be observed with static PET imaging. However, the details related to these variations, e.g. whether the static tumor uptake was due to temporary increase/decrease of blood flow remained unclear. Dynamic PET with pharmacokinetics was then performed to further investigate the detailed information related to tumor hypoxia during PDT.
The k 3 refers to the internalization rate of the tracer from non-specific diffusion to specific accumulation. Higher k 3 value means higher specific 18 F-FMISO internalization to hypoxic region. Although the mean tumor uptake in U87MG kept increasing (Fig. 1B), the mean k 3 value remained consistent from 1 h to 24 h post PDT ( Fig. 2A). The k 3 parametric mapping (Fig. 4) of U87MG tumor further indicated that the percentage of 18 F-FMISO specific internalization increased and became dominant at 24 h post PDT as tumor hypoxia developed gradually over the entire tumor region. The pimonidazole staining (Fig. 5) also showed that the tumor hypoxia developed post PDT and distributed relatively homogeneously over the staining region. This continuous development of hypoxia condition of U87MG was possibly due to its high tumor vascularity 36 that allowed to maintain perfusion for a prolonged period of time post PDT, thus delayed the development of hypoxia due to vascular circulation shutdown. The high vessel density in U87MG model was also observed by high K 1 and k 2 values (Table 1), indicating a high blood supply and efflux post PDT. The influx rate constant K i refers to tracer's total irreversible trapping from plasma to the tumor as an entire system and it takes into account both non-specific distribution through vasculature and specific internalization of the given tracer. The strong correlation between K i and k 3 (Fig. 2B) further indicated that the 18 F-FMISO specific trapping to hypoxic region was positively correlated with blood supply. Therefore the tumor uptake of 18 F-FMISO and its variation in response to PDT observed in static PET were mainly related to the blood supply through tumor vasculature for U87MG tumor.
However, for MDA-MD-435 tumor model, similar to the variation of 18 F-FMISO tumor uptake observed from static PET (Fig. 1B), the mean k 3 value increased dramatically at 1 h post PDT and decreased at later time points, which corroborated with the pimonidazole result. This might be reflective of rapid induction of apoptosis and necrosis by acute tissue hypoxia 37 at early time following PDT treatment; consequently less effective specific 18 F-FMISO internalization was present at late time post PDT in MDA-MB-435 tumor. The dramatic decrease of K i at 24 h (Table 1) also showed the shutdown of tracer influx through blood flow to tumor region. The relatively low K 1 and k 2 value (Table 1) as well as the relatively independent variation between K i and k 3 for MDA-MB-435 (Fig. 2C) indicated that 18 F-FMISO's specific trapping to hypoxic sites was less dependent on tumor vasculature when compared to U87MG. Therefore the variation in 18 F-FMISO tumor uptake in response to PDT observed in static PET was not only due to the blood supply through tumor vasculature but also due to tumor intrinsic hypoxic environment for MDA-MB-435 tumor.
The parametric mapping addressed the distribution of the tracer kinetics related to tumor hypoxia at voxel level. Since the influx rate K i referred to the kinetics rate of total tracer "trapping" into the tumor, it was expected that the whole body K i map would show a similar pattern as the 18 F-FMISO uptake image. However, K i map revealed higher tumor to tissue contrast when compared to the dynamic PET image frame. The acquisition time of the last frame of 2 h dynamic PET image was 300 seconds versus 10 minutes for static PET image at 2 h post PDT. The shorter acquisition time resulted in less radioactive incidents acquired to generate one image, thus the contrast in dynamic PET scan was worse than the 10 minutes static scan. However K i parametric mapping helped to overcome this issue by improving image contrast essentially around tumor region. It could also be beneficial for other cases where static PET images show low tumor to tissue contrast. Moreover, even though the static PET image was able to provide higher image contrast, the information presented was based on a specific time point (e.g. 10 min at 2 h post PDT) where the uptake could be contributed by a temporary radiobiological event e.g. temporal increase of blood flow 14 . While K i and k 3 maps were generated based on information of a much larger time window (the entire 2 hour post injection), there would be little to no temporal bias affecting the analysis.
The ability to map specific tracer retention in hypoxic sites with k 3 parametric map and differentiate from its non-specific distribution is one of the key points that dynamic PET analysis can offer while static PET images do not. Bartlett et al. 38 reported that parametric mapping especially k 3 parametric map showed better correlation with the direct pO 2 measurement than the static PET image voxel intensity. Therefore, dynamic PET with parametric mapping could potentially offer a more accurate and more detailed way than the uptake based static PET image to assess tumor hypoxia and address the heterogeneity 29 of the tumor hypoxia during therapy at voxel level in vivo.
Previously, Casciari et al. claimed that the bioreduction product of FMISO was diffusible and was able to leave cells 12 . Bruehlmeier et al. also suggested the use of two-tissue reversible compartment model 39 to assess tumor hypoxia with 18 F-FMISO PET. The internalization of 18 F-FMISO to hypoxic sites was considered to be reversible; the transportation rate from internalization compartment back to diffusive compartment (k 4 , in min −1 ) was taken into account. However, based on the PET data used in the present study, two-tissue irreversible model gave better fit than the reversible model. Table 2 showed the goodness fit evaluated by residual analysis of both irreversible and reversible compartment models applied to the dynamic PET data of MDA-MB-435 model. The fitting errors of reversible model were all greater than 5% at different time points; the k 4 fit value was in the order of 10 −4 to 10 −5 min −1 . This shows that the reduced or non-reduced 18 F-FMISO that is capable of exiting from hypoxic sites is minor compared to the tracer that is constantly trapped. Considering the exchange of 18 F-FMISO between the diffusive and accumulative compartments as a dynamic equilibrium process, the k 3 in the irreversible model implemented in the present study referred to the absolute trapping rate of tracer that irreversibly stayed inside cells.
Our studies clearly indicated the need of combining both static image analysis and dynamic PET followed by parametric mapping to monitor tumor hypoxia before and after treatment. There are, however, a few concerns of using dynamic PET in practice. Dynamic PET requires relatively long acquisition time (1~2 h) with the study subject remaining anesthetized during the entire scan. There is also technical complexity from data acquisition to kinetics analysis; while static PET scan is advantageous for its simplicity and convenience. Therefore, although the approach of dynamic PET with pharmacokinetics modeling can assess tumor hypoxia more accurately following PDT than static PET, it will unlikely replace static PET in routine practice.

Conclusion
In this study, 18 F-FMISO PET and pharmacokinetics modeling were used to monitor tumor hypoxia before and after PDT. A clear difference in tumor hypoxia in response to PDT was observed between two tumor types: U87MG and MDA-MB-435 xenografts. In particular, dynamic PET with kinetics analysis provided information on tumor hypoxia in distinguishing the specific tracer retention to the hypoxic tumor region and the non-specific distribution in the tumor vasculature. Voxel level analysis of parametric map images allowed us to assess the heterogeneity of tumor hypoxia during PDT treatment. 18 F-FMISO PET imaging with pharmacokinetics modeling offers a potential imaging tool to fully characterize tumor hypoxia during treatment.  Table 2. Goodness fit (ɛ) of both irreversible and reversible two-tissue compartment models applied to dynamic PET data of MDA-MB-435 tumor model.