[¹⁸F]-FLT Positron Emission Tomography Can Be Used to Image the Response of Sensitive Tumors to PI3-Kinase Inhibition with the Novel Agent GDC-0941

The phosphatidyl inositide (PtdIns) signaling network of enzymes and lipid messengers is a key nexus for the control of almost all cellular processes and is often deregulated in disease (1). The formation of PtdIns (3, 4, 5) P₃ from PtdIns (4, 5) P₂ by class IA phosphoinositide 3-kinase (PI3K) enzymes gives rise to signaling resulting in cell proliferation, motility, invasion, and metastases; activation of this enzyme by mutation (or deregulation of upstream effectors such as HER2 or the negative regulator PTEN) is found in a wide range of cancers (2). Drug discovery efforts have aimed at inhibiting this enzyme, and GDC-0941 is a potent and selective inhibitor of class I PI3K (3, 4) that is currently undergoing clinical trials in a range of malignancies. A current problem for drug discovery pipelines is the lack of biomarkers for use in such trials to allow effective biologic doses to be determined and to allow better patient management by the early assessment of therapeutic response. Noninvasive imaging with positron emission tomography (PET) offers the possibility of assessing whole-tumor response; assessment of thymidine incorporation via the salvage pathway with 3′deoxy-3′-[¹⁸F]fluorothymidine ([¹⁸F]-FLT) has been shown to predictive of response in a range of preclinical models and a variety of drugs (5).

In this study, we assess whether changes in uptake of [¹⁸F]-FLT are predictive of response to PI3K inhibition by GDC-0941 or genetic methods in subcutaneous and orthotopic models of cancer with activation of PI3K signaling (via deletion of PTEN or mutation of PIK3CA), as a guide to the use of such imaging agents in clinical trials. A subsidiary aim was to see whether imaging could be used to assess longitudinal drug response in an orthotopic model.

Subcutaneous and orthotopic tumors: HCT116 (PIK3CA H1047R/KRAS G13D/PTEN wild-type) and U87 (PIK3CA wild-type/KRAS wild-type/PTEN-negative) cells were implanted subcutaneously (0.1 mL of a 5 × 10⁷/mL stock in PBS) in female nude mice (CBA nu/nu, aged 8–12 weeks). Cell lines were obtained from the American Type Culture Collection (ATCC) and were not authenticated. Tumor volumes were measured at least 3 times weekly with calipers, and the volume calculated as (tumor length × tumor width²)/2. For the orthotopic model, HCT116 cells (5 × 10⁶/50 μL in serum-free RPMI) were injected aseptically into the spleen of anesthetized CBA nude mice through an incision made in the skin underneath the left ribcage into the inferior tip of the spleen using a 27-gauge needle. The skin was closed using a surgical staple and mice were administered buprenorphine (0.1 mg/kg/sc). Dosing for therapies commenced in the evening after the baseline PET scan, when subcutaneous tumors were sized at approximately 200 mm³. GDC-0941 was prepared in 0.5% hydroxypropyl methyl cellulose and administered by oral gavage (0.1 mL/10 g). Dosing was adapted from published data (3). Mouse welfare was monitored at least daily. In all treatment groups, pimonidazole [60 mg/kg intraperitoneally (i.p.); Chemicon International Inc.] was given 2 hours before sacrifice.

Inducible dominant-negative PI3K p85 subunit (Δp85α) tumors: HT29 (PIK3CA P449T/KRAS wild-type/PTEN wild-type) cells (obtained from the ATCC and not authenticated) expressing inducible Δp85α or parental controls were grown as xenografts by sc injection of 0.1 mL of 5 × 10⁷/mL stock in serum-free medium into either side of the mid-dorsal region of the back of 6- to 8-week-old female SCID/Bg mice as previously described (6). Mice were housed in individually vented caging systems on a 12-hour light:dark environment and maintained at uniform temperature and humidity. Tumor size was measured 3 times a week with calipers, and the volume calculated as above. When tumors reached about 300 mm³, mice were imaged (see below) and then gavaged with doxycycline [0.2 mL orally, 10 mg/mL; daily]. Imaging was repeated 72 hours later. On sacrifice, tumors were excised and bisected. One half was snap-frozen in liquid nitrogen and used to prepare lysates and the other half was fixed in 10% formalin for immunohistochemistry. All procedures were carried out in accordance with the Scientific Procedures Act 1986 and National Cancer Research Institute Guidelines 2010 (7) by approved protocols following institutional guidelines (Home Office Project Licenses 40-3212 and 40-3306 held by Professors K.J. Williams and C. Dive).

Animals underwent dynamic baseline scanning when xenograft size had reached about 200 to 300 mm³ for subcutaneous tumors or at 20 days postimplant for intrasplenic HCT116 tumors Animals were anesthetized with 1% to 2% isoflurane, the tail vein was catheterized and they were placed in the animal bed (Minerve Small Animal Environment System Bioscan) and transferred to an Inveon preclinical PET scanner (Siemens). Animals bearing the dual flank PI3K knockout/control tumors were similarly prepared and scanned using a Quad-HIDAC system (Oxford Positron Systems). At the start of the acquisition, mice were injected with either 10 MBq of [¹⁸F]-FLT (synthesized in-house using published protocols; ref. 8) or [¹⁸F]-FDG (Erigal) intravenously via the tail vein. List mode data were collected for 60 ([¹⁸F]-FDG) or 105 to 120 minutes, in accordance with maximal acquisition times advised (7). Anesthesia was maintained during image acquisition via a nose cone with respiration and temperature monitored throughout using appropriate systems. After imaging, HCT116 or U87 bearing animals were recovered in a warmed chamber, randomized into treatment groups, and re-scanned at 18 hours and 186 hours after vehicle or therapy. Animals bearing the dual flank PI3K knockout/control tumors were recovered, administered doxycycline (see above), and imaged 3 days later. Immediately after the last scan, tumors were excised, weighed, and half fixed in 10% formalin, half snap-frozen in liquid nitrogen for Western blot analysis.

Inveon acquired data: before reconstruction, the list mode data were histogrammed with a span of 3 and maximum ring differences of 79 into 3-dimensional (3D) sinograms with 21 to 33 time frames (5 × 60 seconds, 5 × 120 seconds, 5 × 300 seconds, and 6–18 × 300 seconds) for image reconstruction. Images were reconstructed using the 3D-OSEM/MAP algorithm (4 OSEM3D iterations and no MAP iterations, with a requested resolution of 1.5 mm). Regions of interest (ROI) were drawn manually over tumor and either heart contents ([¹⁸F]-FLT) or a section of liver ([¹⁸F]-FDG) using Inveon Research Workplace software and further normalization was conducted using the injected dose (from the dose calibrator) and animal weight to give a standardized uptake value (SUV). SUV_max was calculated from the maximum voxel value within the ROI and SUV_mean as the average over all voxels, normalized uptake values were calculated by dividing SUV_max or SUV_mean from the tumor by that from the heart contents.

For Quad-HIDAC data: images were reconstructed for 5-minute timeframes using OPL-EM (9). Absolute calibration of the images was achieved by reference to an ¹⁸F source imaged in the field of view in each scan. ROI analysis was undertaken manually using in-house developed software running under IDL6.0 (Research Systems Inc.). Further normalization was conducted using the injected dose (from the dose calibrator) and animal weight to give a SUV; SUV_mean was calculated as the overall average of all voxels, which was applied to all reconstructed timeframes to give a time–activity curve.

Mice were anesthetized (2% isoflurane) and connected to a heartbeat monitor for the duration of the experiment. Skin temperature measurements obtained directly before and after the 20-minute procedure indicated that the body temperature remained constant (36ºC–38ºC). T1-weighted (TR/TE = 700/8.6 ms) and T2-weighted (TR/TE = 1130/15 ms) sequences based on RARE (4 echoes, 4 averages, acquisition matrix = 256 × 192, reconstructed at 256 × 256, with 15 × 1 mm² slices; ref. 10) were used to obtain abdominal images for each mouse using a Magnex 7-Tesla, horizontal-bore magnet (Agilent Technologies) interfaced to a Bruker Biospec Avance III console (Bruker Biospin Ltd) with a transmit/receive 2.5 cm surface coil.

Protocols were used as previously described (11). Primary antibodies used were total-AKT phospho-AKT, GSK-3β, phospho-GSK-3β, phospho-ERK (Cell Signaling Technology); HIF-1α (BD Transduction Laboratories), CA-IX (kindly provided by Dr. Jaromir Pastorek, Bratislava, Slovakia); β-actin (Sigma) and TK1 (ab15580, Abcam). Protein expression was quantified by densitometry relative to the loading control protein β-actin, using ImageJ.

Frozen tumors were sectioned in 12-μm slices and thaw mounted on Indium Tin Oxide slides, before a α-cyano-4-hydroxycinnamic acid matrix 20 mg/mL [50:50 acetonitrile:water, 0.2% trifluoroacetic acid (TFA)] was applied with a thin layer chromatography (TLC) sprayer at 1,000 nL/s, total application per section 120 μL. Slides were analyzed on a Kratos AXIMA MALDI Time of Flight (TOF)-TOF CFR+ imaging mass spectrometer (Shimadzu) in reflectron mode with laser power set at 63 and a 400 to 800 mass range. Spot size was 100 μm, 100 laser shots were collected per spot with the laser rastering every 25 shots.

Tumor volume was monitored over the 8 days of therapy in treated and untreated animals. There was a significant reduction in tumor growth between treated and untreated U87 tumors (69.2 ± 9.0 vs. −1.0 ± 3.8 mm³/d; P < 0.01; Fig. 1A). HCT116 showed a variable response, though overall there was no difference in growth between treated and untreated HCT116 tumors (46.2 ± 11.9 vs. 37.3 ± 8.4 mm³/d; Fig. 1B). No significant change in mouse body weight was observed with drug treatment.

Downregulation of phospho-AKT (pAKT) and phospho-GSK3β (pGSK3β) levels have been reported as pharmacodynamic (PD) biomarkers of GDC-0941 action (2); therefore, these were assessed in U87 and HCT116 tumors treated with vehicle/GDC-0941. pAKT and pGSK-3β levels were downregulated in all treated U87 tumors (Fig. 2B) as evidenced by reduced pAKT/total AKT and pGSK3β/total pGSK3β ratios (0.28 ± 0.03 vs. 0.12 ± 0.03 and 0.66 ± 0.11 vs. 0.14 ± 0.08, respectively; P < 0.01; Fig. 2C). This correlated with a reduction in tumor growth as expected in this model. However, PD biomarker response was much more variable in HCT116 tumors and did not correlate with tumor growth inhibition. As we have previously observed in PTEN-deleted tumors (12) GDC-0941 also downregulated the hypoxic mediators hypoxia-inducible factor 1 (HIF-1) and carbonic anhydrase 9 (CA-IX) in sensitive U87 tumors (Supplementary Fig. S1), as well as upregulating the apoptotic signaling as evidenced by an increase of cleaved PARP (Supplementary Fig. S2).

To assess whether the reduction in tumor growth in sensitive tumors could be monitored by imaging, uptake of [¹⁸F]-FLT was measured at baseline and after 1 and 8 days of therapy in both models. For U87 tumors, there was a significant reduction in [¹⁸F]-FLT uptake (NUV_max) at 18 hours posttherapy (2.17 ± 0.38 vs. 1.59 ± 0.29, P < 0.01), which was not seen in the control group (2.17 ± 0.39 vs. 2.02 ± 0.09; Fig. 3A and B); TK1 expression levels relative to actin were also decreased (1.00 ± 0.28 vs. 0.27 ± 0.16, P < 0.01; Fig. 3E). For HCT116 tumors, there was no significant difference in NUV_max at baseline and 18 hours in the treated (2.36 ± 0.78 vs. 2.38 ± 0.49) or untreated (2.30 ± 0.81 vs. 2.78 ± 0.34) tumors (Fig. 3C and D); similarly TK1 expression levels relative to actin were unchanged (0.80 ± 0.11 vs. 0.95 ± 0.18; Fig. 3F). Correlation of the change in NUVmax between baseline and 18 hours with tumor growth inhibition over the treatment period in all U87 and HCT116 tumors studied was significant (P < 0.05; Fig. 3G). There was no significant difference in NUV_max between treatment groups at baseline. In a separate experiment, we assessed [¹⁸F]-FDG uptake in response to therapy and found significant decreases in uptake for treated U87, but not HCT116, tumors (Supplementary Fig. S3A–S3D); this HCT116 cohort was dominated by drug-responsive tumors, however, with a significant reduction in growth (53.8 ± 4.3 vs. 19.2 ± 3.9 mm³/d; Supplementary Fig. S3F) that did not result in a significant decrease in [¹⁸F]-FDG uptake.

To further substantiate the link between PI3K inhibition and the reduction of [¹⁸F]-FLT uptake, PI3K signaling was inhibited using a HT29 colorectal cancer cell model that inducibly expresses a dominant-negative isoform of the p85α regulatory subunit of PI3K (Δp85α); inducible inhibition of PI3K by this method has been shown to reduce xenograft growth via cell-cycle arrest (Fig. 4A; ref. 6). Here, we show that uptake of [¹⁸F]-FLT into tumor was significantly reduced after 72-hour induction of Δp85α tumors with doxycycline (SUV_max: 4.15 ± 0.42 vs. 2.24 ± 0.30) compared with parental controls (4.56 ± 0.36 vs. 4.01 ± 0.44, P < 0.05; Fig. 4B and C), correlating with a reduction in tumor growth.

The responsiveness of the HCT116 cell line to GDC-0941 was also assessed in an orthotopic liver metastases model. The presence of liver metastases after intrasplenic inoculation of HCT116 cells was assessed by T2-weighted MRI 14 days after primary inoculation, and baseline uptake of [¹⁸F]-FLT was established at day 21. GDC-0941 was then given twice daily at 50 mg/kg for 8 days with further [¹⁸F]-FLT scans at 18 hours and MRI at the end of therapy. Uptake of [¹⁸F]-FLT was significantly reduced at 18 hours (NUV_max of 2.76 ± 0.16 vs. 1.99 ± 0.13, P < 0.05; Fig. 5A and B). Consistent with the predictive nature of [¹⁸F]-FLT in subcutaneous models, a reduction in the burden of HCT116 liver metastases in mice treated with GDC-0941 for 8 days was revealed by MRI (Fig. 5C) and upon assessment of liver to body weight ratios between treated and untreated animals (10.9 ± 1.1% control vs. 5.9 ± 0.5% for treated animals, P < 0.01; Fig. 5D).

Levels of GDC-0941 were measured in HCT116 tumors with MALDI mass spectrometry. Drug, detected as the (M+H)⁺ ion (514.64), was distributed homogenously through the liver, with less penetrating metastatic tumors (Fig. 6A and B, tumors arrowed). The amount and distribution of GDC-0941 in liver metastases was comparable to that of xenograft tumors (Fig. 6C).

Targeting the PI3K pathway has become a major focus in current drug development programs in oncology due to the prevalence of mutations that activate signaling via this network in cancer. GDC-0941, a dual-class PI3K inhibitor whose structure is designed to bind to the ATP pocket of the catalytic p110 subunit (4, 13), is currently in phase I and II clinical trial and exerts antiproliferative effects on a range of cell lines (3, 4, 14, 15). This has been shown to correlate with a reduction in the number of cells in S-phase as well as the induction of apoptosis (14). Initial phase I results indicate that inhibition of this target by GDC-0941 and a variety of other targeted agents may not have broad efficacy across a range of malignancies; as targeted therapeutics have shown high efficacy rates in specific patient subpopulations (12), this suggests patient stratification before treatment is an important unmet clinical need. GDC-0941 has shown activity in cell lines harboring mutations in the PI3K pathway components PTEN, PIK3CA, and HER2 (2, 16), and genetic response signatures consisting of mutations to PIK3CA, HER2, dual PIK3CA/HER2, PIK3CA/PTEN, or HER2/PTEN have been shown to confer sensitivity to GDC-0941 in breast cancer cell lines (14). These have yet to be clinically validated, however, and this process is likely to be confounded by the significant heterogeneity found in tumors (17). This is also likely to confound existing biomarkers of drug target inhibition for agents targeting this pathway, which in any case may be less than robust (18), perhaps due to the complexity and redundancy of cancer signal transduction networks (19).

Molecular imaging offers a way of assessing the biologic status of whole tumors. [¹⁸F]-FLT has been developed as a generic marker of proliferation, and uptake has shown prognostic significance in some cancers (20, 21). As uptake depends on both ENT1 transporter levels and the activity of the thymidine salvage pathway enzyme TK1, it may not reflect proliferation in cells that rely on de novo synthesis and so may not be pathognomonic for DNA synthesis (22, 23). Identifying response in tumors that have high [¹⁸F]-FLT uptake in clinical trials, however, may allow the determination of biologically effective doses, as well as helping patients to avoid lengthy cycles of ineffective treatment; as phase I results for GDC-0941 have shown at best stable disease [and there is evidence from trials of targeted agents in lung cancer (12) and gastrointestinal stromal tumor (24) that overall survival benefit of treatment may result from stable disease as opposed to tumor regression], global measures of changes in proliferative indices are likely to result in better patient management. Links between response and [¹⁸F]-FLT uptake have been shown in a number of preclinical studies of classical and targeted chemotherapy (18, 23, 25–36). Clinically, early response to therapy determined with [¹⁸F]-FLT PET imaging has been shown to predict survival in some malignancies (37); however, in others, significant decreases in uptake have not correlated with overall survival (reviewed in ref. 4); thus, [¹⁸F]-FLT may still have use as a negative indicator, that is, survival is not likely in tumors that do not show such a response.

The current study shows that GDC-0941 inhibits tumor growth in PTEN-negative U87 glioma subcutaneous xenografts as previously reported (2, 3) and that this effect can be measured using [¹⁸F]-FLT PET acutely after therapy and before tumor growth inhibition becomes apparent. The dose used (50 mg/kg/q2d) translates to 2 doses of about 245 mg in humans (38), which is comparable with the maximum tolerated dose (MTD) of 450 mg every day established in phase I trials (39), with dose-limiting toxicity (DLT) of grade III maculopapular rash. The timing of acute scans was chosen as inhibition of growth is likely to be rapid in xenograft models compared with human tumors; other studies have seen response as early as 6 hours after therapy (40). Thus dose and time point are likely to be reflective of the clinical situation. [¹⁸F]-FLT uptake at 186 hours was reduced to a similar degree for U87 tumors in both treated and control groups compared with baseline but this was not significant (data not shown). Lessening of reduction in uptake due to GDC-0941 at day 8 may be the precursor to eventual drug resistance, and although previous studies show inhibition of growth for up to 19 days at the dose used here (2), there is clear upregulation of MAPK/ERK signaling at day 8 in sensitive cells that may provide a mechanism for this (Supplementary Fig. S4) and is consistent with previous studies on the increased efficacy of dual PI3K/MAPK inhibition (41). The reduction in the control group may be due to a Gompertzian decrease in proliferation due to increased size.

Acute reduction also correlated well with cell-cycle arrest as measured by TK1 expression and downregulation of pAKT by greater than the 30% suggested for tumor stasis in a PK/PD model (42). Target inhibition and efficacy is not seen in the PIK3CA-mutant/moderate HER2-expressing HCT116 model but neither is there reduction in the number of cycling cells as measured by TK1 expression or tracer uptake (both tumors chosen for the study have high uptake of [¹⁸F]-FLT and most likely rely on the salvage pathway to obtain thymidine for DNA synthesis). Although HCT116 cells are sensitive to drug in vitro, tumors grown in vivo have been reported to be more resistant to this drug (42), and although growth arrest after GDC-0941 tumors has been shown at higher doses that may be difficult to achieve clinically (43), the lack of correlation between pathway and growth inhibition in the HCT116 subcutaneous model suggests that factors other than drug delivery, such as heterogeneity in the tumor microenvironment, may be at work. Interestingly, when [¹⁸F]-FDG uptake in response to therapy was studied in a further cohort, HCT116 tumors were more responsive to therapy (Supplementary Fig. S3). This was not reflected in tracer uptake changes, however, which may indicate that [¹⁸F]-FDG could be a less sensitive biomarker for PI3K inhibition.

To validate the correlation between change in tracer uptake and drug response, we used a genetic model of PI3K inhibition where inducible expression of a dominant negative regulatory subunit has been shown to result in growth inhibition rather than cell death in the PIK3CA-mutant colorectal cancer HT29 model in vivo. This showed that at 72 hours, when tumors are growth arrested, there is a sharp decrease in the uptake of [¹⁸F]-FLT. Although this is as expected, the correlation of imaging endpoints to the activity of a single pathway component may not provide proof of pathognomonicity, as, for example, isogenic HCT116 cells where PTEN was deleted showed a decrease rather than an increase in the uptake of [¹⁸F]-FLT (44). Changes in PI3K signaling can result in cell death as well as growth arrest, and GDC-0941 caused apoptosis in U87 tumors as previously described (ref. 2; Supplementary Fig. S3); that this was not seen after downregulation of PI3K signaling in the HT29 model further shows the different responses to pathway alterations in different cellular milieu. We have shown that [¹⁸F]FLT uptake is reduced in a genetic model of apoptosis by an unknown mechanism (data not shown) so it is possible that the reduction in uptake may be due to these apoptotic, as well as antiproliferative, effects.

Interestingly, HCT116 tumors were sensitive to the same dose of drug in an orthotopic metastases model. Although the liver is more highly perfused than tumor, liver metastases are predominantly hypovascular (45), and there was clearly less drug in liver metastases than the surrounding tissue, consistent with drug levels being similar in both environments. It is likely that the orthotopic tumor environment more closely mimics the clinical situation (46), suggesting that differences in response may be due to orthotopic microenvironmental signaling or selection pressures on HCT116 metastases. Consistent with hypovascularity, liver metastases have been shown to be highly hypoxic (47, 48), and a link between reduction in PI3K and hypoxic signaling by GDC-0941 has been shown in thyroid cancer models (16). HIF-1, CA-IX, and blood vessel density (Supplementary Fig. S1 and data not shown) were also reduced in sensitive U87 tumors after 8 days of GDC-0941 therapy, so it is possible that downregulation of hypoxic response by GDC-0941 has more impact in the metastatic environment resulting in the reduction in growth. [Other studies have shown that drug antivascular effects do not affect the uptake of either [¹⁸F]-FDG or [¹⁸F]-FLT uptake, however (ref. 49), suggesting such effects do not confound the use of these imaging biomarkers].

In conclusion, our study shows that [¹⁸F]-FLT is a strong candidate for monitoring the efficacy of GDC-0941 in vivo.

K.J. Williams has a commercial research grant and is a consultant/advisory board member of GE Healthcare. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Cawthorne, I. Wilson, K.J. Williams

Development of methodology: C. Cawthorne, C.J. Morrow, M. Babur, G. Brown

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Cawthorne, N. Burrows, R.B. Gieling, D. Forster, J. Gregory, A. Smigova, M. Babur, K. Simpson, C. Hodgkinson, K.J. Williams

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Cawthorne, N. Burrows, D. Forster, K.J. Williams

Writing, review, and/or revision of the manuscript: C. Cawthorne, N. Burrows, C.J. Morrow, D. Hiscock, I. Wilson, K.J. Williams

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Cawthorne, D. Forster, M. Babur, C. Hodgkinson

Study supervision: C. Cawthorne, C. Dive

Synthesized radiotracer used in data acquisition: M. Radigois

Supervised the MALDI imaging component of the study: A. McMahon

The authors thank Katrina Copeland, Jonathan Bevan, and Neil O'Hara for assisting with quality control of radioisotope production; Karen Davies for technical assistance during the MRI scanning; and Peter Julyan for use of his PET analysis software. The authors also thank Shimadzu for the loan of the Kratos Axima MALDI TOF-TOF imaging mass spectrometer.

This study was financially supported by grants from GE healthcare, and also EU FP7 Metoxia Grant agreement no. 222741 (K.J. Williams) and Cancer Research UK C147/A12328 (C. Dive).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.