Background
Biomarkers are crucial to the development of new drugs and optimization of the existing options, by facilitating selection of the population to treat, confirming proof-of-concept and acting as early markers of tumour-response. The latter can be provided in the clinic by non-invasive functional imaging, for example positron emission tomography (PET) measurements of 2′-deoxy-2′-[18 F]fluoro-glucose (FDG) and 3′-deoxy-3′-[18 F]fluorothymidine (FLT), dynamic contrast-enhanced magnetic resonance imaging for vascular parameters and diffusion-weighted imaging for apoptosis [
1,
2]. However, they are not always easy to implement, and furthermore may be inappropriate for the mechanism-of-action (MoA) of a particular drug and cannot always detect true responses to the treatment [
3‐
6]. We have recently described a rapid, robust MRI-method, which detects the response of solid tumours to drugs with different MoA in several different experimental models [
7]. The method quantifies the spin–lattice relaxation of protons (T
1) in tumours both rapidly and accurately using an IR-TrueFISP method. Across several models, the fractional change in tumour T
1 (ΔT
1) correlated with the percentage of cells positive for the antigen Ki67 (a marker of cycling cells), but not with other markers such as apoptosis, necrosis or blood volume, all of which showed no consistent change with drug-treatment [
7]. Recently, a preclinical study in a neuroblastoma mouse model treated with three different drugs showed a consistent decrease in T
1[
8], and a clinical study reported a small decrease in T
1 in patients with colorectal cancer metastasis undergoing bevacizumab therapy [
9].
To investigate further what ΔT
1 reflects about the tumour biology, we have compared ΔT
1 with magnetic resonance spectroscopy (MRS) markers of proliferation and apoptosis
in vivo[
10], as well as histology and immunohistochemistry
ex vivo following treatment with the allosteric mTOR inhibitor, everolimus (Afinitor) in two different murine tumour models, RIF-1 and B16/BL6. Everolimus was selected for these studies because although the drug has significant clinical activity in several different types of cancer, there is currently no confirmed molecular marker that can stratify the patient population [
11]. Using the RIF-1 model, we demonstrate that ΔT
1 is a highly sensitive and specific predictor of response to everolimus and also the microtubule stabilizer patupilone. Collectively, these data further suggest that incorporation of T
1 measurements in clinical trials should be an important aid to drug development and optimization of existing drugs.
Methods
Tumour Models
All animal experiments were carried-out strictly according to the local Swiss animal welfare regulations. The protocol was approved by the local veterinary authorities (Kantonales Veterinäramt Basel-Stadt, permit number 1974). C3H/He female mice (20–25 g) and C57/BL6 mice (20 g) were obtained from Charles River (France) and were acclimatized to local conditions for at least one week prior to experiments. Three experiments were performed in the RIF-1 fibrosarcoma model in C3H/He mice, one experiment was performed in the B16/BL6 melanoma model in C57/BL6 mice. All animal experiments were performed under isoflurane anesthesia, and every effort was made to minimize suffering.
Tumour volume (TVol) and animal body-weight (BW) measurements were made at least twice per week including just before treatment (baseline) and the endpoint. TVol was determined using calipers to measure three orthogonal dimensions and applying the formula: l*h*w*π/6.
Murine RIF-1 fibrosarcoma
Freshly cultured RIF-1 tumour cells were injected subcutaneously (5 × 10
6 in 50 μL phosphate-buffered saline) in the upper thigh of anesthetized C3H/He mice, as previously described [
12]. After 2 weeks, tumours were of sufficient size (at least 200 mm
3) for the studies and were divided into two equal groups and treated daily with compound or vehicle. Experiment 1: treatment with everolimus (n = 7) compared to vehicle (n = 7), experiment 2: treatment with everolimus (n = 8) compared to vehicle (n = 8), experiment 3, previously published in [
7]: three different doses of patupilone (each group n = 8) compared to vehicle (n = 8).
Murine B16/BL6 melanoma
Freshly cultured B16/BL6 tumour cells expressing the enzyme luciferase were injected intra-dermally (5 x 10
4 in 1 μl) into the dorsal pinna of both ears of anesthetized C57/BL6 mice as previously described [
12,
13]. These black melanoma cells rapidly metastasize from the primary ear tumour to the regional lymph-nodes, in particular the neck. After 2 weeks, mice were divided into two equal groups (n = 10) and treated daily with everolimus (10 mg/kg p.o.) or vehicle for 6 days (experiment 4). MRI was performed on the metastasis in the cervical lymph nodes on day 5. In one mouse in the vehicle group there was no measurable lymph node metastasis.
Compounds/drugs and their application
All compounds utilized in this study were obtained from the Novartis chemical department. The compounds and their respective vehicles were prepared each day just prior to administration to animals and the administration volume individually adjusted based upon animal body weight. Everolimus (RAD001) was obtained as a microemulsion and was freshly diluted in a vehicle of 5% glucose and administered by oral gavage (p.o.) to mice daily in a volume of 10 ml/kg at 10 mg/kg. Patupilone (epothilone B, EPO906) was dissolved in polyethylene glycol-300 (PEG-300) and then diluted with physiological saline (0.9% w/v NaCl) to obtain a mixture of 30% (v/v) PEG-300 and 70% (v/v) 0.9% saline. Treatment with vehicle (PEG-300/saline) or patupilone (3, 5 or 6 mg/kg) was once weekly using an i.v. bolus of 2–3 sec in the tail vein.
Experimental design
Mice were divided into different treatment groups so that each group had the same mean TVol, and magnetic resonance (MR) measurements were made before treatment (baseline) i.e. day 0 and at the endpoint. For everolimus, the endpoint was day 5, and for patupilone it was day 7. T
1 was measured in all four experiments. MRS was performed in experiment 1 only. Bioluminescence was measured
ex vivo in experiment 4 (see below). At the end of everolimus-experiment 1, animals were sacrificed by CO
2 inhalation, the tumours ablated and prepared for histology and immunohistochemistry (IHC) as previously described [
7].
Magnetic Resonance in vivo
Animals were anaesthetised using 1.5% isoflurane (Abbott, Cham, Switzerland) in a 1:1 mixture of O
2/N
2 and placed on an electrically warmed pad for canulation of one lateral tail-vein as previously described [
7]. MRI experiments were performed on a Bruker DBX 47/30 or Avance 2 spectrometer (Bruker Biospin, Ettlingen, Germany) at 4.7 T equipped with a self-shielded 12 cm bore gradient system.
Quantitative T1imaging
The spin–lattice relaxation of protons (T
1) was measured with an inversion recovery (IR) TrueFISP (true fast imaging with steady state precession sequence, [
14]) imaging sequence as previously described [
7]. The basic sequence was a series of 16 TrueFISP images acquired at a time interval, TI, following a global 180° inversion pulse (TI = 210 ms to 5960 ms in 324 ms increments). Each TrueFISP image (one slice containing the central part of the tumour) was acquired with a flip angle α of 30°, a matrix size of 128 × 96, a field-of-view of 30 × 22.5 mm
2, a slice thickness of 2 mm, an echo time (TE) of 1.7 ms, and a repetition time (TR) of 3.4 ms. Pixelwise T
1 calculation was done using the method described in [
15]. A region of interest (ROI) comprising the entire tumor was drawn manually on the resulting T
1 map and the mean T
1 of the central tumour slice was calculated in this ROI. MR image analysis was performed off-line with in-house written software based on IDL 6.0 programming environment (Research Systems Inc., Boulder, CO, USA).
1H-MR spectroscopy
Localized shimming with FASTMAP method was performed on a 2.5 mm3 voxel to obtain line widths of <20 Hz. Point resolved spectroscopy (PRESS) experiments at the same voxel position (voxel size = 8 mm3, TE = 20 msec, TR = 1500 msec, SW = 4000 Hz, TD = 2048, with external volume suppression) were performed. One spectrum was acquired with water suppression (400 averages) and one spectrum without water suppression (1 average). The total time for MRS was 10–12 min for each mouse. The water signal (one peak) of the non-suppressed spectrum was used as an internal reference for relative quantification of metabolites using the metabolite to H2O ratio (Cho/H2O for choline, etc.). Peaks in the water-suppressed spectrum were identified by their chemical shifts, so total choline (Cho) was at 3.2 ppm, CH3-lipids at 0.9 ppm, CH2-lipids at 1.3 ppm, creatine at 3.0 ppm, and polyunsaturated fatty-acids (PUFA) at 5.3 ppm and very weakly at 2.8 ppm; however the latter peak was not used for any calculations.
Histology and immunohistochemistry
A tumour slice of 3–4 mm thickness was cut from the largest circumference of the tumour, immersion-fixed in 4% (w/v) phosphate-buffered formalin (pH 7.4; J.T. Baker, Medite, Service AG, Dietikon, Switzerland) at 4°C for 24 hours and processed into paraffin as previously described [
4]. IHC was performed on paraffin sections of 3 μm using the following antibodies for detection of (i) cleaved Caspase-3 (rabbit polyclonal antibody #9661, Cell Signaling, Danvers, MA, USA) (ii) Ki67 (rat monoclonal antibody, clone TEC3, #M7249, DAKO, Glostrup, Denmark) and (iii) CD31 (rabbit polyclonal antibody, #E11114, Spring Biosciences, Pleasanton, CA, USA).
Image acquisition and analysis of histological slices
For quantification, the entire section was scanned using the MiraxScan system (Carl Zeiss AG, Jena, Germany). The absolute size of viable, necrotic and complete tissue areas was measured on the full scans using MiraxViewer software (Carl Zeiss AG, Jena, Germany). Quantification of Ki67 positive and negative nuclei in the complete viable areas was performed in a fully automated manner with TissueMap software at Definiens AG, Munich, Germany. Results were summarized as the total area, percentage-viable and percentage-necrotic area, total number of nuclei and the cell density (number of nuclei per mm2) in both the viable and total (including therefore necrotic) areas. Cleaved caspase-3 positive particles were quantified as positive pixels per total pixels in a semi-automated fashion with the AnalySIS® FIVE software (OlympusSIS, Münster, Germany) on six images (346.7 x 260 μm2 each) per section excluding necrotic areas and border zones of necrotic areas. CD31 stained slides were scanned with the Aperio ScanScopeXT slide scanner (Aperio, Vista, CA, USA) and vessels were quantified with the Aperio ImageScope software, using the Microvessel Analysis v1 Algorithm.
Bioluminescence
Because of the black pigmentation of the C57/BL6 mice, bioluminescence (BioL) could not adequately be measured in vivo and therefore was determined ex vivo as follows. After 6 days, the cervical lymph-nodes were removed and weighed and then iced. Individual lymph-nodes were homogenized at 4˚C with 1 mL cold phosphate-buffered saline (without Ca2+ and Mg2+), rinsed in the same buffer, and 200 μl triplicates placed in a 96-well plate with 50 μl D-luciferin (1 mg/mL). BioL was measured at an emission wavelength of 560 nm using the imaging chamber of the IVIS™ system (Caliper Life Sciences Inc, Hopkinton, MA, USA) for 1 min at room temperature.
Data analysis
Results are presented as mean ± SEM except where stated and all available data are shown. The T/C ratio is commonly used to quantify tumour growth inhibition, where T and C represent the means of the relative tumour volumes (tumour volume divided by its initial volume) of the treatment and control mice, respectively [
16]. Longitudinal changes, such as in tumour volume (ΔTVol) or in T
1 (ΔT
1), were expressed as change between endpoint and baseline divided by value at baseline (fractional change in %). The T/C ratio was calculated for all parameters. For parameters measured at one time point only (such as histological read-outs), T/C was calculated as ratio of means of treatment and control mice, respectively. Differences between groups were analysed using a 2-tailed t-test. For the
in vivo biomarker analyses which involved longitudinal analyses in the same animals, differences were analysed by a) 2-way repeated measures ANOVA and b) t-test at the endpoint; the latter method is therefore associated with the respective T/C. The different dose groups in experiment 3 were tested with 1-way ANOVA vs. control group. Quantification of the linear-relationship between the various parameters measured
in vivo and
ex vivo were analysed by Pearson’s correlation to provide the correlation coefficient (r) and the significance (p). Application of the non-parametric Spearman’s correlation did not affect the results except in one case (see Results). To facilitate comparison of PUFA levels which were not always detectable, a 2-sided Fisher’s exact test was also used. For all tests, the level of significance was set at p < 0.05 (two-tailed) where *p < 0.05, **p < 0.01, ***p < 0.001 versus vehicle.
To determine the sensitivity and specificity of a change in the imaging marker T
1 as a marker of tumour response to treatment, receiver-operator curves (ROC) were generated [
17,
18] using Graphpad Prism (GraphPad Software, La Jolla, CA, USA) considering mice treated with the drugs everolimus (experiment 1 and 2) or patupilone (experiment 3). Briefly, responders (R) to drug-treatment were defined as showing no change or regression, in TVol, defined as ΔTVol ≤ 10%, while all others were considered non-responders (NR). Each of these tumours was then classified as R or NR by the ΔT
1 at different discrete cut-offs to generate at each ΔT
1 value a table of positive and negative predictions for determination of specificity and sensitivity at each value. The plot of 1-specificity versus sensitivity generated the ROC curve and the area under this curve (AUC) was quantified by the trapezoidal method.
Discussion
We have previously shown that a small but highly significant decrease in the mean spin–lattice relaxation of protons (T
1) of experimental tumours induced by various different types of chemotherapy is strongly correlated with the change in tumour volume and also the immunohistochemical proliferation marker Ki67 [
7]. Furthermore, in the RIF-1 model the antimetabolite 5FU also decreased levels of the proliferation marker choline and this too was correlated with the change in T
1 (ΔT
1). The data presented here on RIF-1 and B16/BL6 tumours confirm these observations for the allosteric mTOR inhibitor everolimus, providing further evidence that ΔT
1 reflects the number of remaining proliferating tumour cells following successful chemotherapy. The greater the decrease in T
1, the lower the percentage of proliferating cells after therapy. In the previous report, a sample area (10%) of an
ex vivo tumour slice was examined histologically, and thus, true cell density and also therefore the overall extracellular space could not be assessed either. In two models, using the cytotoxic patupilone on murine RIF-1 and rat mammary BN472, there were trends for cell-density to decrease by approx. 10% but neither reached significance [
7]. In this report, we have made a detailed histological study of the effect of everolimus on RIF-1 tumours grown s.c. in murine C3H mice.
RIF-1 cells are sensitive to everolimus with an IC50
in vitro of 2.6 ± 1.6 nM (insensitive cells have an IC50 > 1 μM, see references [
19,
20]), but this is still not as sensitive as the endothelial cells which have IC50 <1 nM, which likely explains the fact that everolimus has anti-tumour cell as well as anti-angiogenic properties [
19]. Daily treatment of mice bearing RIF-1 tumours caused tumour stasis, and consistent with this, histology at the endpoint of 5 days showed a decrease in total tumour area and a proportional decrease in the viable area of approx. 35% compared to vehicle (both not significant, p < 0.1). However, the total number of cells showed a similar trend for a decrease in proportion (p < 0.1) and thus the overall cell density in the viable areas was unchanged. Since necrosis was also not affected by everolimus (non-significant increase of 20%), this analysis showed that cell density and the extracellular space were unaffected. Many previous experiments
in vitro and using human tumour xenografts
in vivo have shown that T
1 is sensitive to a) the amount of water in the extracellular space (but not intracellular) and b) the amount of protein in the water [
21‐
25]. It is well recognised that inhibition of mTOR (the molecular target of everolimus) causes a decrease in cell size [
26], because cell cycle progression is blocked at G1 thus inhibiting protein synthesis and cell growth. Consequently, cell density might not change, but the extracellular space could increase. Unfortunately we could not measure the average cell size because defining where one cell ends and another begins is difficult and there was no automatic programme for such an approach. But in any case, an increase in extracellular space would lead to an increase rather than a decrease in T
1[
21‐
25], suggesting that if cell size changes occurred they could not explain the T
1 decrease that we have always observed following successful chemotherapy with many different agents [
7]. This suggests to us, that tumour cell and vascular destruction leads to the release of proteins and also paramagnetic ions into the extracellular space which causes the decrease in T
1; an effect which has been shown
in vitro[
22]. However, everolimus did not cause a decrease in the blood vessel density, as has been seen in several other tumour models [
20,
27‐
29], although this does not rule out an effect on the functional vasculature (previously measured as low in RIF-1 tumours [
30]) and/or that early vascular changes had normalised by day 5. Also in this model, there was no clear evidence of increased tumour cell kill since caspase-3 levels were unaffected, although there did appear to be a strong trend for an increase in the PUFAs of everolimus-treated tumours which has been associated with apoptosis in other experimental models [
10].
Immunohistochemistry (IHC) showed that approx. half of vehicle-treated RIF-1 tumour cells were positive for the nuclear antigen Ki67. Ki67 is considered to be a proliferation marker since it is expressed in all cycling cells (G1, S and G2M) but not therefore in cells in G0, and is a convenient IHC tool in the clinic for assessing tumour growth and response [
31,
32]. Given that the effect of mTOR inhibition is to block G1, it was not surprising that everolimus caused a marked decrease in the %Ki67
+ cells whether expressed as total number or density and there was a proportional increase in the cells negative for Ki67. Everolimus also decreased levels of total choline (Cho) in RIF-1 tumours, which is another marker of viable and proliferating cells, in this case reflecting membrane turnover. Cho tends to be higher in tumour than normal tissue [
33] and successful chemotherapy has also been shown to decrease in Cho in both experimental models [
7,
8,
10] and the clinic [
34,
35]. In the RIF-1 tumours, these proliferation markers correlated significantly with each other as well as with the ΔT
1, supporting the notion that ΔT
1 is a surrogate of the remaining number of proliferating cells in a tumour after therapy even though our histological analysis suggests that it cannot be measuring cell number or density directly. Support for this hypothesis came from the B16/BL6 model treated with everolimus where again the decrease in bioluminescence, which measures viable tumour cell number, was correlated to the ΔT
1.
It is worth repeating that we have found that six different types of chemotherapy including anti-metabolites, inhibitors of mTOR, microtubules, VEGF-R, PI3K and HSP90 [
7] [and unpublished] in several different tumour-types implanted in both mouse and rat hosts, all showed a decrease in T
1 in response to successful chemotherapy i.e. characterized by a significant change in TVol in comparison to vehicle-treated tumours. Furthermore, where a tumour was resistant to that particular type of chemotherapy (paclitaxel and patupilone), there was no change in T
1[
7]. This suggests that ΔT
1 is a generic marker of tumour response, because, as discussed above, it reflects overall tumour destruction. But, what is the level of sensitivity and specificity i.e. how useful could such a method be in the clinic? To answer this question, we used receiver-operating-characteristic curves (ROC) to analyse two different models in which mice bearing RIF-1 tumours were treated with either a single dose of everolimus or three different doses of patupilone. In both cases, the ROCs had a large area-under-the-curve (AUC) of 0.84 and 0.97 which is considered of very good to outstanding predictive ability [
17]. Consider for comparison, IHC markers of the mTOR pathway to predict everolimus activity
in vitro using cell lines in which ROC-AUCs of 0.86-0.88 were determined [
20]; also excellent predictive activity but no better or even lower than that we have shown here. Indeed, with our data, if a cut-off of an 8% decrease were selected (i.e. ΔT
1 = −8%), then the sensitivity to both drugs would be perfect in this model at 1.0 i.e. providing a negative predictive value of 100%. In other words, after two MRI-scans one could completely eliminate from the study any tumours with a T
1 decrease smaller than 8% since these should not benefit from further treatment.
Competing interests
All the authors are or were (CW) employees of Novartis Pharma AG, Basel, Switzerland. They declare no competing interests.
Authors’ contributions
CW carried out the MRI and MRS studies, performed data and statistical analysis, and drafted the manuscript. PA carried out the MRI studies and critically revised the manuscript. MS performed and analysed histological and IHC studies. VR performed IHC studies and prepared figures with IHC data. SF took care of the animal model, the animal treatment, and carried out bioluminescence assessments. PM took care of the study design and coordination, performed data and statistical analysis, and drafted the manuscript. All authors read and approved the final manuscript.