Introduction
Breast cancer development and growth is strongly influenced by the crosstalk of tumour cells with the surrounding extracellular matrix/stroma [
1‐
3]. The stroma can make up a significant proportion of a breast carcinoma [
4], and differs from normal stroma, bearing closer resemblance to granulation tissue and wound healing, with a high number of fibroblasts, deposition of type I collagen and fibrin, and the infiltration of inflammatory cells [
5]. The presence of a fibrotic focus, a central scar-like area within a carcinoma that represents a focus of exaggerated reactive tumour stromal formation, was first proposed as an indicator of increased tumour aggressiveness in invasive ductal breast cancer by Hasabe et al. [
6], and has since been linked to early disease relapse, lymph node and osteolytic bone metastasis, and reduced long-term survival [
7‐
9]. Hypoxia has also been associated with the formation of fibrotic foci [
5].
Advanced MRI techniques provide a means of defining non-invasive quantitative biomarkers to inform on biologically relevant structure-function relationships in tumours, thereby enabling an understanding of their behaviour and heterogeneous distribution [
10]. Imaging biomarkers for assessing tumour pathophysiology require evaluation before being routinely deployed in clinical trials; in particular, imaging-pathology correlation, and thus whether the imaging biomarker reflects underlying pathology is important to establish, but can often only meaningfully be studied in animal models [
11].
The newly generated tumour stroma shows similarities with granulation tissue and subsequent scar formation in wound healing and differs from the normal stroma by an increased number of fibroblasts, enhanced capillary density, deposition of type I collagen and fibrin and the presence of inflammatory cells.
Several MRI biomarkers have the potential to detect breast cancer fibrosis. The fibrous nature of collagen may increase the non-monoexponential contribution to the diffusion-weighted MRI (DWI) signal, arising from the propensity of water molecules to diffuse along the fibres, combined with reduced diffusivity from encountering more barriers to random diffusion, compared to surrounding tissue [
12‐
14]. Increased macromolecular collagen fibre content may also yield a greater destruction of signal arising from MT MRI from off-resonance saturation [
15], and the short-lived signal of collagen (T
2*~500μs) may be detectable with ultrashort-echo time (UTE) sequences [
16]. Dynamic contrast-enhanced (DCE) MRI remains a standard technique used in breast cancer MRI protocols and may be suitable for fibrosis detection in some tissues [
17], but the use of contrast adds complexity to clinical studies and can be contraindicated in certain patients.
This study aims to determine the ability of multi-parametric MRI incorporating several endogenous contrast mechanisms, such as DWI, MT-MRI and UTE-MRI, performed on a clinical imaging platform, to detect and quantify fibrosis in a chemically-induced rat model of mammary carcinoma previously shown to produce heterogeneous tumours with a range of fibrosis severity [
18].
Materials and methods
Animal procedures
This study was performed in accordance with the local ethics review panel, United Kingdom National Cancer Research Institute guidelines for animal welfare in cancer research, and the ARRIVE (Animal Research: Reporting In Vivo Experiments) guidelines [
19,
20]. Female Sprague-Dawley rats (200–250 g, n=18; Charles River, Margate, UK) were injected with 37.5 mg.kg
−1 of refrigerated
N-methyl-
N-nitrosourea (MNU, Sigma-Aldrich, Poole, UK) intraperitoneally, resulting in tumours that spontaneously developed at various sites associated within the mammary fat-pad [
18]. Tumour formation was detected by palpation and growth was monitored by calliper measurement; animals were imaged when tumours reached approximately 3cm
3 (using ellipsoid volume formula, (π/6)×L×W×D).
Animals were anaesthetised using a 4 ml.kg-1 intraperitoneal injection of fentanyl citrate (0.315 mg.ml-1) plus fluanisone (10 mg.ml-1 (Hypnorm; Janssen Pharmaceutical Ltd., High Wycombe, UK)), midazolam (5 mg.ml-1 (Hypnovel; Roche)), and water (1:1:2). Prior to imaging, an intraperitoneal injection of 60 mg.kg-1 pimonidazole (Hypoxyprobe, Burlington, VT, USA) in phosphate- buffered saline was given, in preparation for subsequent histological staining for hypoxia.
Magnetic resonance imaging
MR imaging was performed on a MAGNETOM Avanto 1.5T clinical scanner (Siemens Healthcare, Erlangen, Germany), to validate clinical sequences and support methodological transfer. For MRI, the animal was secured supine using an insulating vacuum beanbag to both retain body heat and prevent excessive movement. The animal was placed with the tumour centred on top of a small-loop temporomandibular joint (TMJ) coil, itself centred within the multi-element head receiver coil [
21]. Elements of the head coil array were used in parallel with the small-loop coil during all acquisitions. Scans were performed in the coronal plane, with full tumour coverage. Morphological T
2-weighted fast spin-echo images were obtained for anatomical localisation. Diffusion-weighted MRI (DWI), ultrashort-echo time (UTE) MRI and magnetisation transfer (MT) data were acquired centred on the lesion.
UTE data were acquired with a prototype three-dimensional (3D) multiple gradient echo (mGRE) sequence with 1.1 mm isotropic resolution; the first echo acquired was on the free induction decay (FID) immediately following the read pulse, followed by four regular gradient echoes. This acquisition was repeated in order to acquire four ultrashort-echo times (70–560 μs). DWI was based on a clinical patient protocol (nine b-values, 0–800 mm
-2s; see Table
1) acquired in free-breathing using a fat-suppressed two-dimensional (2D) single-shot prototype EPI sequence. MT data were acquired as a series of matched 3D GRE acquisitions, with 1.0-mm isotropic voxels, and two flip angles with/without an MT pulse set at 1.5 kHz offset. Detailed sequence parameters are given in Table
1, and were adapted from clinical imaging sequences; the total acquisition time was approximately 1 hour.
Table 1
MR imaging parameters for anatomical imaging (T2w), diffusion-weighted imaging (DWI), ultrashort-echo time imaging (UTE) and magnetisation transfer imaging (MT). Total protocol time: approx. 1 h
Sequence type | TSE | 2D EPI | 3D mGRE | 3D GRE |
Slices | 24 | 18 | 96 | 30 |
FOV (mm) | 120×72 | 150×105 | 103×103 | 128×96 |
Slice thickness (mm) | 1 | 1.5 | 1.07 | 1 |
Matrix Size | 256×152 | 102×72 | 96×96 | 128×96 |
TR (ms) | 800 | 2100 | 42 | 15 |
TE(ms) | 9.6 | 60.8 | 7.16, 11.64, 16.12, 20.60 | 2.52 |
UTE (ms) | - | - | 0.07, 0.14, 0.28, 0.56 | - |
NSA | 1 | 18 | 1 | 8 |
iPAT | GRAPPA 2 | GRAPPA 2 | - | GRAPPA 2 |
Fat sat. | No | Yes | Yes | No |
b-values (mm-2s) | - | 0, 20, 40, 60, 80, 100, 200, 400, 800 | - | - |
Variable flip angles (°) | - | - | - | 4°, 24° |
MT pulse | - | - | - | without/with (1.5 kHz) |
Time (min:s) | 4:54 | 15:58 | 4 × 5:36 | 4 × 2:28 |
MR image analysis
MRI analysis was performed using proprietary software (ADEPT, The Institute of Cancer Research, London, UK). All MR images were reviewed, and regions of interest (ROIs) were independently drawn by two observers, MR scientists (NPJ and DJC) with 5 and 32 years’ experience, respectively, in conducting preclinical MR studies. Repeatability of ROI delineation was assessed using the Sørenson-Dice similarity coefficient. Each ROI was drawn around the tumour on the imaging slice that macroscopically matched the histological section stained, MR parameters calculated on a voxel-by-voxel basis, and reported as the average value for repeated ROI median values per slice analysed together with calculation of repeat-measures coefficient of variation (CoV).
For DWI analysis, the perfusion-insensitive apparent diffusion coefficient (ADC) was estimated using images for b=200 mm
-2s and above [
22], with a single-exponential model (Eq.
1). All b-values were used for intravoxel incoherent motion (IVIM) fitting using a bi-exponential model (Eq.
2) to simultaneously derive estimates of pseudodiffusion fraction (f), pseudodiffusion coefficient (D*) and tissue diffusivity (D). The compound parameter fD* was also calculated. Initial estimate values for IVIM fitting were found using the segmented approach [
22], by estimating D using a monoexponential fit of images with b=200 mm
-2s and above (as per ADC) and f from the observed S
0 relative to the intercept of this curve at b=0 mm
-2s:
$$ {S}_b-{S}_o.\exp \left(-b. ADC\right) $$
(1)
$$ {S}_b={S}_o.\left[f.\exp \left(-b.{D}^{\ast}\right)+\left(1-f\right).\exp \left(-b.D\right)\right] $$
(2)
where the observed signal intensity at a given b-value is denoted S
b, and S
0 is the corresponding signal at b=0mm
-2s (equal to the total available signal S
total modulated by the apparent T
2 and the acquisition echo time, S
0=S
total.exp(-TE/T
2app) [
23].
For UTE imaging, T
2*
short was calculated using the first (ultrashort, < 1ms; see Table
1 for values) echo from successive imaging acquisitions, and the conventional T
2*
long using the remaining (i.e. not ultrashort) echoes from all acquisitions, using separate mono-exponential models (Eq.
3); the ratio of the calculated signal arising, analogous to f in the IVIM DWI model, from each of the two relaxation constants was also calculated. All DWI and UTE fitting was performed using a Markov Chain Monte Carlo (MCMC) Bayesian statistical approach [
24] as a robust least-squares estimator, with no data filtering.
$$ {S}_{TE}={S}_o.\exp \left(-\frac{TE}{T_2^{\ast }}\right) $$
(3)
MT acquisition images were used for calculation of magnetisation transfer ratio (MTR) (Eq.
4) [
25,
26], longitudinal relaxation constants in the presence/absence of the MT pulse using the variable flip angle (VFA) method [
27] (T
1 and T
1s, respectively), and B
1-independent MT saturation (δ) and apparent MT rate (k
a) (Eqs.
5 and 6) [
26,
28]:
$$ MTR=\left({S}_{ref}-{S}_{MT}\right)/{S}_{ref} $$
(4)
$$ {k}_a= MTR/{T}_{1s} $$
(5)
$$ \delta =\left({R}_{1 app} TR+{\alpha_{nom}}^2/2\right)\left({S}_{ref}-{S}_{MT}\right)/{S}_{MT} $$
(6)
where S
ref and S
MT are signal amplitudes from identical sequences acquired with and without the MT pulse, TR is the acquisition repetition time, R
1app is the spin-lattice relaxation rate (or T
1
-1), α
nom is the nominal acquisition flip angle in radians, and the small flip angle approximation is used [
26].
Histological staining and analysis
Following MR imaging, animals were killed by cervical dislocation, and the tumour excised and fixed in 10 % formalin. Tumours were then cut through the centre, and embedded in paraffin blocks, with orientation matched to the geometry of the imaging slices to facilitate subsequent image correlation.
Tumour sections (5 μm) were stained with haematoxylin and eosin (H&E), to allow assessment of necrosis and tumour grade, and picrosirius red, to assess collagen I/III deposition (fibrosis). Immunohistochemistry visualised using DAB was performed using FITC-conjugated mouse monoclonal antibodies against pimonidazole adducts, followed by rabbit anti-FITC antibodies, for the detection of hypoxic regions, or rabbit monoclonal antibodies against CD31 (EP3095; Millipore, Watford, UK) to assess vascular endothelial cells as a proxy for perfusion. Whole tumour images were acquired using a motorised scanning stage (Prior Scientific Instruments, Cambridge, UK) attached to a BX51 microscope (Olympus Medical, Southend-on-Sea, UK) driven by CellP (Soft Imaging System, Munster, Germany). Snapshots at ×200 magnification were also acquired from CD31-stained sections.
Tumour grade and degree of necrosis (semi-quantitative assessment) were evaluated by an expert pathologist (AN). Percentage area of each tumour section displaying pimonidazole adduct or picrosirius red positivity was measured using pixel counts from a customised routine operating on a Lab colour-space separation into stain and non-stain classes (Mathworks, Natick, MA, USA) of a digital image, and visually confirmed for accuracy. Microvessel density was assessed by counting CD31-positive vessels from six random fields (×200) distributed across the section and the number converted to vessels/mm2.
Discussion
The presence of a histologically-confirmed fibrotic focus has been shown to be a predictor of increased tumour aggressiveness, relapse, metastasis and poor long-term survival in breast cancer [
6‐
9]. Fibrotic foci are also associated with tumour hypoxia, an independent indicator of poor treatment response and prognosis [
5,
18,
31‐
33]. The ability to detect fibrosis within mammary carcinomas non-invasively would be of great value in helping guide personalised treatment. The validation of appropriate MRI techniques with potential to inform on fibrosis using preclinical models with matched histology can directly guide development of imaging studies in the clinical setting.
In this study, a range of endogenous MR imaging contrasts were measured in chemically-induced mammary carcinomas arising in rats injected with MNU; the tumours were highly heterogeneous and presented with a range of fibrosis levels as previously observed in this model and typical of the clinical setting [
18,
34]. The imaging performed in the study used exclusively clinical hardware, conferring greater translational relevance to the study, and the scanning was performed within a clinical timeframe using standard and prototype (UTE and DWI) sequences developed by the manufacturer for use on the clinical platform. It has previously been shown that this platform is suitable for preclinical work of this nature [
21,
35], and can return functional MR parameters with good measurement repeatability across several imaging biomarkers. Repeated analysis by independent observers showed excellent repeatability of ROI positioning and all derived MR parameters except the pseudo-diffusion parameters from the IVIM diffusion model.
The results from the MT measurements were striking in their significance, with the presence of increased collagen leading to significant reductions in T
1 measurements, as well as increased k
a and δ. After correcting for multiple comparisons, the correlations of these remained significant (p<0.0125). The MT ratio parameter, MTR, was correlated to picrosirius red stain fraction but fell short of significance. The similar parameter δ, less dependent on the influence of B
1 [
26], showed a stronger correlation and indicated that B
1 effects should be accounted for when analysing MT data. The fibrous macromolecule collagen has a much shorter spin-lattice relaxation time T
1 compared to normal tissue, and through magnetisation transfer to water protons reduces the apparent T
1 of an imaging voxel dependent on the partial volume of collagen. The presence of the MT pulse saturates the collagen protons, and with transfer to the interacting water molecules, an additional and greater reduction occurs, giving much lower T
1s. The apparent MT rate constant for the destruction of the water signal by the MT saturation, k
a, is an empirical rather than a true rate constant [
28], but does relate to the amount of collagen present, giving the observed correlation. Combining MT parameters using a PLSR analysis demonstrated a prediction error similar to that given from cross-validation using each parameter alone, indicating that different MT parameters provide statistically similar information on how collagen affects the tumour microenvironment. These results indicate that the MT measurement as performed was sensitive to the presence and proportion of collagen in the tumour, and can provide a non-invasive assessment of collagen content.
In diffusion-weighted imaging, the presence of collagen fibres will modify the diffusion characteristics of water molecules, providing additional barriers to free diffusion. In this study, the ADC and D values were negatively correlated with the picrosirius red staining, although with
p-values short of significance (p=0.0274 and 0.0253, respectively), suggesting that the measurement of true diffusion is affected by the presence of fibrosis, in line with observations in hepatic fibrosis [
36,
37]. These parameters were also found to contribute in the latent variables of the PLSR analysis, alongside MT parameters, although this model did not outperform the best individual MT parameters. The fibrous nature of collagen may also introduce heterogeneity to the diffusion hindrance, manifesting as a non-Gaussian diffusion component captured as a significant positive correlation of collagen presence with the pseudo-diffusion parameter fD*. The data for the pseudo-diffusion volume fraction f, often considered related to perfusion, showed no correlation with the endothelial marker CD31, which is likely reflective of the inherent difficulty in reliably fitting IVIM data, but also the complexity of tumour perfusion [
30,
38]. In contrast, the non-significant correlation of f with necrosis (r=-0.607, p=0.0165) may suggest that f does not solely capture vascular fraction [
23] and may be related to the degree of non-Gaussian diffusion introduced by the presence of collagen fibres [
14]. The high CoV values associated with the pseudo-diffusion parameters, however, indicate that caution is required in interpreting these results.
In this study, the use of ultrashort echoes in order to visualise collagen did not give rise to a significant correlation. The conventional measurement of T2*long, using echo times longer than the relaxation time of collagen, showed a correlation to picrosirius red, suggesting that the overall voxel T2* is sensitive to the presence of fibrosis, and decreases with increasing collagen content.
The design of this study includes several limitations, which are nonetheless linked to its strengths. The use of clinical scanner hardware and imaging sequences means that while the scanner was not optimised for small animal studies, the techniques used were shown to be immediately translatable to clinical work. The mammary carcinoma model used in this work yielded tumours that varied considerably in presentation, growth rate, and composition; this reflects the clinical presentation of breast cancer and supports the potential of these results for clinical translation.
We have demonstrated the use of a multi-contrast MRI protocol to investigate the properties of chemically-induced mammary carcinoma in a preclinical setting, and have shown the potential of a clinical MT sequence to detect the presence of fibrosis non-invasively. Results from MT parameters outperformed those from multiple-b-value DWI and UTE imaging in detecting and quantifying intratumoral collagen, potentially providing information of biological relevance to support clinical assessment. Given that the presence of fibrosis is known to be a prognostic factor in mammary carcinoma, and may be induced following radiation therapy [
39,
40], the results of this study support the inclusion of MT protocols in clinical breast MRI examinations.