Hypoxia and perfusion in breast cancer: simultaneous assessment using PET/MR imaging

Objectives Hypoxia is associated with poor prognosis and treatment resistance in breast cancer. However, the temporally variant nature of hypoxia can complicate interpretation of imaging findings. We explored the relationship between hypoxia and vascular function in breast tumours through combined 18F-fluoromisonidazole (18 F-FMISO) PET/MRI, with simultaneous assessment circumventing the effect of temporal variation in hypoxia and perfusion. Methods Women with histologically confirmed, primary breast cancer underwent a simultaneous 18F-FMISO-PET/MR examination. Tumour hypoxia was assessed using influx rate constant Ki and hypoxic fractions (%HF), while parameters of vascular function (Ktrans, kep, ve, vp) and cellularity (ADC) were derived from dynamic contrast-enhanced (DCE) and diffusion-weighted (DW)-MRI, respectively. Additional correlates included histological subtype, grade and size. Relationships between imaging variables were assessed using Pearson correlation (r). Results Twenty-nine women with 32 lesions were assessed. Hypoxic fractions > 1% were observed in 6/32 (19%) cancers, while 18/32 (56%) tumours showed a %HF of zero. The presence of hypoxia in lesions was independent of histological subtype or grade. Mean tumour Ktrans correlated negatively with Ki (r = − 0.38, p = 0.04) and %HF (r = − 0.33, p = 0.04), though parametric maps exhibited intratumoural heterogeneity with hypoxic regions colocalising with both hypo- and hyperperfused areas. No correlation was observed between ADC and DCE-MRI or PET parameters. %HF correlated positively with lesion size (r = 0.63, p = 0.001). Conclusion Hypoxia measured by 18F-FMISO-PET correlated negatively with Ktrans from DCE-MRI, supporting the hypothesis of perfusion-driven hypoxia in breast cancer. Intratumoural hypoxia-perfusion relationships were heterogeneous, suggesting that combined assessment may be needed for disease characterisation, which could be achieved using simultaneous multimodality imaging. Key Points • At the tumour level, hypoxia measured by 18F-FMISO-PET was negatively correlated with perfusion measured by DCE-MRI, which supports the hypothesis of perfusion-driven hypoxia in breast cancer. • No associations were observed between 18F-FMISO-PET parameters and tumour histology or grade, but tumour hypoxic fractions increased with lesion size. • Intratumoural hypoxia-perfusion relationships were heterogeneous, suggesting that the combined hypoxia-perfusion status of tumours may need to be considered for disease characterisation, which can be achieved via simultaneous multimodality imaging as reported here. Electronic supplementary material The online version of this article (10.1007/s00330-020-07067-2) contains supplementary material, which is available to authorized users.


Introduction
Hypoxia is a common characteristic of the tumour microenvironment and arises due to avid metabolism and poor perfusion as a result of the structurally and functionally aberrant microcirculation found in tumours [1]. In breast cancer, the presence of hypoxia has been confirmed with pO 2 histography and occurs irrespective of histological type, molecular subtype, grade or patient characteristics [2,3]. In vitro studies have shown that hypoxia promotes a dedifferentiated phenotype in ductal carcinoma in situ [4] and downregulates the expression and function of oestrogen receptor-α (ERα) [5]. Several clinical and preclinical studies in breast cancer have demonstrated that overexpression of hypoxia-related proteins is associated with an aggressive phenotype, poor prognosis and resistance to treatment [6][7][8].
Although tumour hypoxia can be broadly categorised as diffusion or perfusion limited, it is generally accepted that the tumour microenvironment is a highly dynamic entity, exhibiting temporally varying perfusion patterns and heterogeneous oxygen-tension gradients [9]. Experimental evidence suggests that oxygen levels continually fluctuate owing to transient changes in perfusion [10]. These changing perfusion and oxygenation levels induce a variety of gene expression profiles resulting in a unique micromilieu that is pivotal for tumour growth and metastatic dissemination [11]. Given the temporal variation in oxygenation and perfusion within tumours, sequential multimodal imaging investigations may not always be effective in assessing the association between these parameters, as similar tumour status cannot be guaranteed between imaging sessions. Simultaneous assessment of the hypoxia and perfusion in tumours can mitigate confounders associated with the dynamic character of these processes, and thus allow additional pathophysiological characterisation of breast cancer.
Imaging methods, including positron emission tomography (PET) and magnetic resonance imaging (MRI), have been used for the non-invasive assessment of the tumour microenvironment. Dynamic contrast-enhanced (DCE) MRI has shown utility in characterising tumour perfusion and vascular permeability in clinical studies [12], while diffusion-weighted imaging (DWI) can provide surrogate measures of tumour cellular density [13]. PET with 18 F-labelled nitroimidazoles can provide specific measures of intracellular hypoxia [14]. In breast cancer, 1 8 Ffluoromisonidazole ( 18 F-FMISO) has been used for the evaluation of response to anti-angiogenic and HER2-targetted treatment [15,16] and shown potential utility as a predictor of response to primary endocrine therapy [17,18]. Additionally, high 18 F-FMISO uptake at baseline has been associated with shorter disease-free survival [18] and disease-specific death [19].
Despite the intrinsic link between tumour hypoxia and perfusion, multimodal imaging approaches to characterise this aspect of cancer pathophysiology have been limited in the clinical setting [16,[19][20][21][22][23][24]. To effectively assess relationships between temporally varying microenvironment parameters, combined PET/MR imaging presents an attractive option as it permits examination of the tumour under the same physiologic conditions, while also conferring methodological advantages in the spatial registration of data from the two modalities.
The primary objective of this study was to examine the association between hypoxia and vascular function in patients with treatment-naïve breast cancer using simultaneous 18 F-FMISO-PET/MRI. To our knowledge, this is the first such study in breast cancer.

Study participants
Women aged > 18 years with histologically confirmed primary breast cancer and a tumour diameter > 10 mm on mammography and/or ultrasound were eligible for the study (February 2017 to November 2018). Pregnancy, lactation, previous surgery or radiotherapy for cancer or benign breast disease, inadequate renal function and contraindications to MRI were exclusion criteria for the study. The research was approved by a National Research Ethics Committee (14/EE/0145). All study participants provided written informed consent before PET/ MRI examination.

PET/MRI acquisition
Participants underwent a 60-min simultaneous PET/MR scan of the breasts in the prone position on a SIGNA PET/MR scanner (GE Healthcare), using a 16-channel bilateral breast array (RAPID Biomedical) 120 min (median [range], 120.2 [119.8-127.5] min) after injection of 306 ± 14 MBq 18 F-FMISO. The uptake period post injection (p.i.) was used to enhance hypoxic-to-normoxic tissue contrast and allow the free 18 F-FMISO concentrations in tissue and blood to reach equilibrium [25,26], a requirement for influx rate constant (K i ) determination by Patlak analysis [27].
MRI: The MRI protocol involved a 2-point Dixon sequence for PET attenuation correction, T 1 -and T 2 -weighted images, DWI, and a DCE series. Sequences were also acquired to measure B 1 + transmission-field non-uniformity, using a Bloch-Siegert method, and baseline T 1 (T 10 ) as required for the pharmacokinetic analysis of DCE-MRI data [32]. DCE-MRI acquisition involved five pre-contrast images, followed by 43 phases after intravenous bolus injection of 0.1 mmol/kg of Gadovist (Bayer Healthcare). MRI sequence details are given in Supplemental Table 1.

Image analysis
Tumour regions were manually delineated in OsiriX, version 8.0.2 (Pixmeo SARL), by three radiologists in consensus (1, 3 and > 20 years of experience in breast MRI). Regions were drawn on the peak-enhancing volumes of the DCE-MRI series on all contiguous axial sections encompassing the invasive part of the tumour and including multifocal/multicentric disease (Supplemental Methods III). Synchronous bilateral cancers were regarded as independent lesions [33].
DCE-MRI: Pharmacokinetic analysis of the DCE-MRI series was performed in MIStar, version 3.2.63 (Apollo Medical Imaging), using the extended Tofts' model [34] to calculate contrast influx rate constant, K trans ; efflux-rate constant, k ep ; extravascular-extracellular volume fraction, v e ; and plasma volume fraction, v p (Supplemental Methods III). DWI: Calculation of apparent diffusion coefficient (ADC) maps was performed in OsiriX, using b values of 0 and 900 s/ mm 2 . Mean lesion ADC was calculated by manually outlining whole tumour regions on the b = 900 s/mm 2 image (Supplemental Methods III; [35] ).
PET: Image frames from 150 to 180 min p.i. were averaged, rigidly registered to the peak-enhancing phase of the DCE-MRI series and subsequently employed for the determination of 18 F-FMISO uptake as mean and maximum standardised uptake values normalised by body weight (SUV mean , SUV max ) and maximum tumour-to-plasma (T max / P) and tumour-to-muscle (T max /M) ratios within the regions defined on the DCE-MRI. The influx rate of 18 F-FMISO into the trapped (hypoxic) tissue compartment (K i ) was determined by Patlak-plot analysis, utilising all frames in the registered 18 F-FMISO series and the scaled population-based AIF. Hypoxic fractions (%HF) in tumour regions were calculated as the percentage of voxels with K i values > 2 × standard deviations (SD) of the mean K i of normoxic muscle (Supplemental Methods III).

Histology
Histopathological information including tumour histological subtype, grade, oestrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor-2 (HER2) status was obtained from core biopsies or surgical tumour specimens. Cancers with positive ER or PR expression were classified as hormone-receptor (HR) positive.

Statistics
Statistical analysis was performed in IBM SPSS Statistics for MacOS, v25.0 (IBM Corp.) or Matlab 2016b. Continuous data were assessed for normality using the Anderson-Darling test. Correlations between continuous variables were assessed using the Pearson correlation coefficient (r). t tests were used for comparison between means of two groups, and ANOVA when more than two groups were compared. Where data were not normally distributed, or normality could not be assessed, Mann-Whitney U and Mood's median or Kruskal-Wallis H tests were employed for comparisons between two or more groups, respectively. p values < 0.05 were considered statistically significant.

Results
A total of 32 women were enrolled into the study. Two participants withdrew before the PET/MR examination. PET/MRI data and DCE-MRI data from two participants were excluded owing to inadequate acquisition of DCE-MRI and poor pharmacokinetic-model fitting respectively. In total, data from 29 participants with 32 biopsy-confirmed primary breast cancers Two thirds of the lesions (21/32; 66%) were invasive ductal cancers (IDC). The majority of cancers (29/32; 91%) were either grade 2 or 3. HR-positive expression was noted for 31/32 (97%) lesions, with 24/32 (77%) cancers being HER2-negative. Tumour characteristics are summarised in Table 1. Additional clinical information is provided in Supplemental Table 2.

Relationship between 18 F-FMISO-PET and DCE-MRI parameters
Scatter plots indicating the relationships between DCE-MRI parameters and K i or %HF are illustrated in Fig. 1. An inverse relationship was observed between mean lesion K i and K trans , v e and v p ( Fig. 1(a-d); Supplemental Fig. 2 [36]), which was statistically significant for K i vs. K trans (r = − 0.38, p = 0.04), but not for K i vs. v e (r = − 0.30, p = 0.10) or v p (r = − 0.28, p = 0.12). Associations between %HF and DCE-MRI parameters followed similar trends, also indicating a decrease in hypoxia with increasing K trans , v e and v p (Fig. 1(e-h)). Statistically significant correlations were observed between %HF and both K trans (r = − 0.33, p = 0.04) and v e (r = − 0.38, p = 0.03). No correlation was observed between k ep and either K i (r = 0.08, p = 0.65) or %HF (r = 0.02, p = 0.90). Figure 2 presents axial slices through K i and K trans parametric maps of four tumours of different histological subtypes, indicating heterogeneous spatial relationships between hypoxia and perfusion; other DCE-MRI parametric images are given in Supplemental Fig. 3.

F-FMISO-PET and DCE-MRI parameters vs. tumour histology and grade
Hypoxic fractions > 1% were observed in 6/32 (19%) cancers with an additional 8/32 (25%) lesions displaying hypoxic fractions greater than zero but less than 1%; the remaining 18/32 (56%) tumours had a %HF of zero. Dot plots of %HF vs. tumour histological subtype and grade are presented in Fig. 3. K i , %HF and 18 F-FMISO uptake parameters showed no significant difference between different histological subtypes or grades (Tables 2 and 3). Similarly, no significant differences were observed between histological groups or grades for the DCE-  Table 4 presents correlations between imaging indices and tumour size as measured by longest diameter on MRI or pathological size. No or weak negative correlations were observed between tumour size and DCE-MRI parameters. Conversely, 18 F-FMISO-PET parameters correlated positively with size; , v e fractional volume of extravascular-extracellular space, v p plasma fractional volume, ADC apparent diffusion coefficient (mm 2 /s), K i 18 F-FMISO influx rate (mL/cm 3 /min), %HF percentage hypoxic fraction, SUV standardised uptake value (g/mL), T max /M maximum tumour-to-muscle ratio, T max /P maximum tumour-to-plasma ratio a Mood's median test b Mann-Whitney U test for malignancies of type IDC and ILC only (mixed and IMC lesions were not included in the comparison) c One-way analysis of variance (ANOVA)

Discussion
This study explored the relationship between tumour hypoxia and vascular function in breast cancer using combined 18 F-FMISO-PET/MRI. Hypoxic fractions and K i measured on 18 F-FMISO-PET showed inverse relationships with the DCE-MRI perfusion parameter K trans , consistent with the generally accepted view that tumour hypoxia is a consequence of inadequate oxygen supply to the tumour [1]. Previous clinical studies in cervical and head-and-neck carcinomas have demonstrated significant negative correlations between contrast enhancement or pharmacokinetic parameters from DCE-MRI and polarographic pO 2 measurements or pimonidazole immunohistochemistry [37,38]. These findings are consistent with our results in breast cancer.
However, PET and DCE-MRI parametric images exhibited largely heterogeneous intratumoural patterns with hypoxic islands on K i maps often colocalising with areas of increased K trans . This spatially discrepant relationship between hypoxia , v e fractional volume of extravascular-extracellular space, v p plasma fractional volume, ADC apparent diffusion coefficient (mm 2 /s), K i 18 F-FMISO influx rate (mL/cm 3 /min), %HF percentage hypoxic fraction (%), SUV standardised uptake value (g/mL), T max /M maximum tumour-to-muscle ratio, T max /P maximum tumour-to-plasma ratio * p < 0.05; ** p < 0.01 a Kruskal-Wallis H b Mann-Whitney U test for grade 1 and 2 cancers only (grade 1 lesions were not included in the comparison) c One-way analysis of variance (ANOVA) d Significant difference between grade 2 and 3 cancers (p = 0.01). Pairwise multiple comparison analysis utilised the Dwass-Steel-Critchlow-Fligner method  K trans contrast influx rate (mL/g/min), k ep contrast efflux rate (min −1 ), v e fractional volume of extravascular-extracellular space, v p plasma fractional volume, ADC apparent diffusion coefficient (mm 2 /s), K i 18 F-FMISO influx rate (mL/cm 3 /min), %HF percentage hypoxic fraction, SUV standardised uptake value (g/mL), T max /M maximum tumour-to-muscle ratio, T max /P maximum tumour-to-plasma ratio * p < 0.05; ** p < 0.01 a Pathological size as measured on tumour specimens from patients undergoing primary surgery (n = 21) and perfusion has been previously documented, with the coexistence of hypoxic and hyperperfused tumour sub-volumes [39]. Various biological mechanisms, including hypoxiainduced angiogenesis, interstitial fluid pressure, a fluctuating haemodynamic response, increased oxygen diffusion distances from the microvasculature and the presence of longitudinal oxygen gradients across tumour vessels, have all been proposed to explain the occurrence of hypoxia in highly perfused areas [40,41]. Thus, although the general trend of our results would support the widely accepted view that hypoxia develops in hypoperfused breast tumours, the diverse relationships observed in individual tumour sub-volumes indicate heterogeneity in hypoxia-perfusion patterns and reflect the variety of pathophysiological mechanisms occurring in cancers.
The weak relationship between PET hypoxia parameters with k ep suggests that the degree of tumour hypoxia is more strongly influenced by vascular flow rather than vessel permeability. Li et al [42] have previously suggested that k ep is a much more sensitive measure of vessel permeability than K trans , as the latter represents a combined measure of blood flow, vessel permeability and capillary-surface area. Our findings broadly agree with previous research in cervical and head-and-neck carcinomas, which illustrated weaker correlations between hypoxia and permeability-surface-area product than between hypoxia and blood flow [37,43]. The relationship between K trans and regional hypoxia observed in our study suggests this is due to fluctuations in tumour vascular flow rather than capillary permeability.
No or weak positive correlations were found between static 18 F-FMISO parameters (SUV mean , SUV max , T max /M, T max /P) and DCE-MRI metrics. In contrast, in human head-and-neck cancer, where hypoxia is often marked, 18 F-FMISO SUV measurements were negatively correlated with both K trans and k ep [20]. A plausible explanation for this disparity is the  IDC, invasive ductal carcinoma; ILC, invasive lobular carcinoma; IMC, invasive mucinous carcinoma; Mixed, carcinoma of mixed ductal and lobular type higher level of hypoxia typically encountered in head-andneck cancer, which will lead to uptake values being more dominated by hypoxia-specific 18 F-FMISO trapping rather than non-specific tracer accumulation. Due to the higher contribution of non-specific 18 F-FMISO accumulation at low hypoxia levels [44], the use of uptake values in cancers without marked hypoxia may not accurately reveal relationships between hypoxia and perfusion.
No significant correlation was observed between PET hypoxia parameters and tumour grade or subtype. Our sample size of non-IDC cases was small for evaluating the impact of histology on tumour hypoxic status, but the presence of non-zero hypoxic fractions was observed in all histological subtypes studied. Hypoxic fractions and higher K i were noted in both grade 2 and 3 tumours, and less so in grade 1 cancers. These findings are concordant with previously reported small differences in hypoxia between low-and high-grade breast malignancy [2].
Correlations between DCE-MRI functional parameters and pathological size or MR tumour diameter yielded moderate negative relationships and conversely positive associations between 18 F-FMISO-PET hypoxia parameters and size. The size-related hypoxia changes could be ascribed to diffusionlimited hypoxia, concomitant perfusion decreases or increased interstitial fluid pressure [45].
ADC has been shown to inversely correlate with cellular density [46], and therefore, a reduction in ADC should theoretically be accompanied by an increase in tumour hypoxia. Our findings indicated no association between ADC and PET hypoxia parameters. This result could be explained by the molecular subtype of lesions in our sample, which predominantly consisted of ER-positive/HER2-negative cancers. Due to lower blood flow, ER-positive or HER2-negative lesions exhibit lower ADC values than ER-negative or HER2positive cancers [47,48]. As ADC is affected not only by tissue cellularity but several pathophysiologic processes including blood flow, membrane permeability and the geometric architecture of the interstitial space [49,50], it is likely that the lack of association between the PET hypoxia parameters and ADC is a consequence of the combined effect of cellularity, perfusion and microvessel structure on ADC. This assertion is further supported by the weak correlations between DCE-MRI indices and ADC observed in this study. It should be noted, however, that inconsistent correlations between ADC and DCE-MRI parameters have been reported in tumours, including breast cancer [51][52][53].
We calculated hypoxic fractions based on a specific parameter for hypoxia namely influx rate constant K i . Despite the higher variability associated with kinetic parameter estimates, our choice was based on two considerations. First, several authors have reported lack of correlation between 18 F-FMISO uptake ratios and pO 2 measurements casting doubt on the accuracy of thresholds derived from static PET imaging for hypoxic quantification [54,55]. Kinetic parameters, including K i , have provided superior correlations with physiological measures of hypoxia from pO 2 histography and immunohistochemistry [54,55]. Second, these thresholds have mostly been defined on measurements from head-and-neck cancers and are not necessarily applicable to other tumour types, including breast cancer.
The main limitations of our study are the small sample size and that the majority of cancers were HR-positive ductal (Left-to-right) DCE-MRI image at peak enhancement; K trans map representing tumour perfusion for the lesion ROI overlaid on the peak-enhancing DCE-MRI image; K i map representing tumour hypoxia for the lesion ROI overlaid on the DCE-MRI image at peak enhancement; ADC map. K trans , contrast influx rate (mL/g/min); K i , 18 F-FMISO influx rate (mL/cm 3 /min); ADC, apparent diffusion coefficient (mm 2 /s) carcinomas. Though our findings cannot be generalised to the full spectrum of histological/molecular subtypes encountered in breast cancer, our study indicates the presence of hypoxia in all histological subtypes studied independent of nuclear grade. While the majority of lesions (56%) examined were found to be non-hypoxic, it should be noted that breast tumours are generally less hypoxic than cancers of the head and neck, cervix or lung and show greater variability in hypoxia among molecular subtypes, with basal-like subtypes being the most hypoxic [56].
Our demonstration of in vivo simultaneous measurement of perfusion and hypoxia is clinically important for three reasons. First, previous reports have indicated that tumours with a high hypoxia-perfusion ratio (i.e. hypoxia due to low perfusion) have a poorer prognosis and suboptimal treatment response [57,58]. In breast cancer, studies have described differences in the response to perfusion-related hypoxic exposure between molecular subtypes [59,60], emphasising the need for combined hypoxia-perfusion measurements to provide more accurate prognostic information or tailor treatment. Second, preoperative radiotherapy or radiochemotherapy regimes in early or locally advanced breast cancer have reported beneficial clinical outcomes [61,62]. Hypoxia and hypoperfusion are known to reduce the effectiveness of radiotherapy and chemotherapy, and the hypoxia-perfusion status of tumours at baseline could allow optimisation of these regimens. Third, tumour hypoxia can occur independently of hypoperfusion as evidenced in the oncology literature [39,40,57,58] and our findings. As such, the data presented here can be viewed as providing further indication of the benefit of noninvasive multimodal assessment of the tumour microenvironment for disease characterisation.
In conclusion, we found a negative relationship between tumour hypoxia, measured by 18 F-FMISO-PET, and markers of perfusion and vascular function from DCE-MRI, endorsing the hypothesis of perfusion-driven hypoxia in breast cancer. No associations were observed between 18 F-FMISO-PET parameters and tumour histology or grade, but hypoxic fractions increased with lesion size. The intratumoural heterogeneity observed in hypoxia and perfusion images is consistent with the known complex relationship between perfusion and the hypoxic tumour micromilieu. The combined hypoxiaperfusion status of tumours may need to be considered in determining treatment efficacy or informing therapy selection in breast cancer, which could be achieved using simultaneous multimodality imaging as reported here.
Informed consent Written informed consent was obtained from all subjects (patients) in this study.
Ethical approval Institutional Review Board approval was obtained. All procedures performed involving human participants were in accordance with the ethical standards of a National Research Ethics Committee (NRES Committee East of England -Cambridge Central; 14/EE/0145) and the Administration of Radioactive Substances Advisory Committee (ARSAC), UK.

Methodology
• prospective • observational • performed at one institution Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.