Hyperoxic BOLD-MRI-Based Characterization of Breast Cancer Molecular Subtypes Is Independent of the Supplied Amount of Oxygen: A Preclinical Study

Hyperoxic BOLD-MRI targeting tumor hypoxia may provide imaging biomarkers that represent breast cancer molecular subtypes without the use of injected contrast agents. However, the diagnostic performance of hyperoxic BOLD-MRI using different levels of oxygen remains unclear. We hypothesized that molecular subtype characterization with hyperoxic BOLD-MRI is feasible independently of the amount of oxygen. Twenty-three nude mice that were inoculated into the flank with luminal A (n = 9), Her2+ (n = 5), and triple-negative (n = 9) human breast cancer cells were imaged using a 9.4 T Bruker BioSpin system. During BOLD-MRI, anesthesia was supplemented with four different levels of oxygen (normoxic: 21%; hyperoxic: 41%, 71%, 100%). The change in the spin–spin relaxation rate in relation to the normoxic state, ΔR2*, dependent on the amount of erythrocyte-bound oxygen, was calculated using in-house MATLAB code. ΔR2* was significantly different between luminal A and Her2+ as well as between luminal A and triple-negative breast cancer, reflective of the less aggressive luminal A breast cancer’s ability to better deliver oxygen-rich hemoglobin to its tissue. Differences in ΔR2* between subtypes were independent of the amount of oxygen, with robust distinction already achieved with 41% oxygen. In conclusion, hyperoxic BOLD-MRI may be used as a biomarker for luminal A breast cancer identification without the use of exogenous contrast agents.


Introduction
Hypoxia, i.e., the lack of oxygen within tissue, has been identified as one of the most pervasive tumor microenvironment factors in cancer progression [1][2][3][4].Hypoxia induces aggressive and treatment-resistant tumor phenotypes, leading to rapid tumor progression and poor prognosis [5].In breast cancer, hypoxia has been shown to induce hallmarks of cancer including angiogenesis ("angiogenic switch") and the deregulation of cell metabolism ("metabolic switch") [6][7][8].Specifically, the response to hypoxia on a cellular level is mediated by the expression of so-called hypoxia-inducible factors (e.g., HIF1-α) [1,2,4].As a result, under persisting hypoxic conditions, HIF1-α accumulates within cells and eventually translocates into the nucleus to induce the expression of vascular endothelial growth factors (VEGFs) [9,10].The presence of VEGFs promotes the activation of blood-vessel inducers, switching on angiogenesis.The distribution of these key adaptive hallmarks of cancer progression is not uniform within a given primary tumor, resulting in intra-tumoral heterogeneity.Of note, breast cancer is a disease with remarkable intra-tumor heterogeneity where multiple cell populations co-exist within the primary tumor [11][12][13].Currently, however, treatment decisions for breast cancer are based on histological, immunohistochemical, and molecular profiling by invasive tissue sampling of small tumor subregions, which are not necessarily representative of the entire tumor.The treatment plans that are based on invasive tissue sampling from breast biopsies that are inherently prone to site selection bias may not be representative of the tumor in its entirety, which can be an important contributing factor for disease recurrence in up to 30% of breast cancer patients [14].To improve treatment outcome, deciphering hypoxia-induced breast cancer heterogeneity has to be accomplished prior to patient treatment.Considering this, novel imaging biomarkers targeting the entire tumor, including imaging biomarkers targeting hypoxia, are urgently needed to facilitate appropriate diagnosis and guide precision treatment.Apart from facilitating a non-invasive assessment of tumor physiology, imaging biomarkers will enable a holistic view on tumor heterogeneity induced by the hypoxic tumor microenvironment.
Magnetic resonance imaging (MRI) offers insights into the oxygen levels in the blood stream via the blood-oxygen-level-dependent (BOLD) effect.The BOLD effect is based on the differential magnetic properties of deoxy-and oxyhemoglobin, therefore exploiting blood as a ubiquitous intrinsic contrast agent [15][16][17].Oxygenized hemoglobin, with its lack of unpaired electrons, is weakly diamagnetic.Conversely, in the case of deoxygenized hemoglobin, unbound binding sites of iron atoms bound to hemoglobin increase local susceptibility gradients, which induces magnetic field distortions and results in alterations in the effective spin-spin relaxation rate that can be exploited as contrast in MR imaging.The resulting so-called BOLD effect has been a key mechanism in functional brain MRI [18,19] and has proven to be a promising imaging biomarker of hypoxia [15,20].In cancer imaging, the BOLD effect is mostly harnessed by measuring the effective spin-spin relaxation rate R 2 for the assessment of tumor oxygen provision may already be achieved at lower oxygen levels.Therefore, further refinement of hyperoxic gas challenges is necessary to pave the way for BOLD-MRI as a contrast-agent-free alternative in standard clinical practice.
We hypothesized that the molecular subtype characterization of breast cancer can be achieved with hyperoxic BOLD-MRI independently of the amount of oxygen.To this end, in this preclinical study, we investigated hyperoxic BOLD-MRI using different hyperoxic gas challenges of varying oxygen concentrations of 41%, 71%, and 100% O 2 in characterizing xenograft tumors of different breast cancer molecular subtypes.

Cell Culture and Mouse Model
This preclinical study was approved by the Austrian Federal Ministry of Education, Science and Research (project number BMFWF-66.009/0284-WF/V/3b/2017)and all study procedures were conducted in accordance with the European Community's Council Directive of 22 September 2010 (2010/63/EU).
Luminal A, Her2 + , and triple-negative breast cancer molecular subtypes are considered less aggressive, moderately aggressive, and highly aggressive breast cancer, respectively, based on their treatment prognosis in the clinical context.Therefore, tumor cells of luminal A (MCF-7), Her2 + (SKBR-3), and triple-negative (MDA-MB-231) breast cancer molecular subtypes were cultivated in RPMI (MCF-7) and DMEM (SKBR-3 and MDA-MB-231) cell culture medium, supplemented with 10% fetal bovine serum and 2% penicillin/streptomycin antibiotics.All tumor cells were kept in standardized conditions at 37 • C in a humidified incubator in an atmosphere containing 5% CO 2 during cultivation.
Subsequently, four-to six-week-old female athymic Balb/c nude mice (n = 28) from Charles River Laboratories (Wilmington, MA, USA) were inoculated subcutaneously into the right flank with 2 × 10 7 luminal A (n = 10), Her2 + (n = 8), or triple-negative (n = 10) tumor cells suspended in 100 µL of serum-free growth medium using a 27 G injection needle.The tumor growth of luminal A-derived tumor xenografts was supported by subcutaneous implantation of half an estrogen pellet (0.72 mg/day, 60-day release, Innovative Research of America, Sarasota, FL, USA) in the neck area.The tumors were allowed to grow for around 1-2 weeks to reach sizes suitable for MR examination with a maximal diameter of 10 mm.Of note, the mice were anesthetized using 3.0% isoflurane (Zoetis Österreich GmbH, Vienna, Austria) mixed with breathing air at a flow rate of 2.0 L/min for induction for all experimental and imaging (detailed below) procedures, and their breathing rate maintained at approximately 60 breaths/min using 1.5-2.0% of isoflurane in the anesthetic gas.

MRI
MRIs were performed in all mice using a 9.4 T Bruker Biospec 94/30USR system (Bruker, Ettlingen, Germany) equipped with a BGA 12S HP gradient system combined with a 40 mm 1 H transmit-receive volume coil (Bruker T13162V3).The system was operated using the Paravision 7 software suite (Bruker, Ettlingen, Germany).During imaging, the body temperature of the mice was kept at 37 • C using a water heating pad.The vital functions of all mice were monitored using an SA II monitoring system (SA Instruments, Stony Brook, NY, USA) with respiratory gating to limit movement artifacts.At first, a series of T 1 -weighted images were acquired as anatomical reference using a 2D RARE sequence with variable repetition times for T 1 mapping (VTR: 1472 ms, 2.000 ms, 3.000 ms, 4.000 ms, 5.000 ms, 7.000 ms, 8.000 ms, 9.000 ms; TE: 21.0 ms, matrix size: 128 × 128; number of slices: 10; slice thickness: 1 mm; spatial resolution: 0.278 × 0.222 mm; acquisition time: 16 min).Following scout image acquisitions, T 2 * mapping was repeated three times to acquire baseline (i.e., normoxic 21% O 2 in the anesthetic gas) BOLD measurements.Then, the O 2 fraction in the anesthetic gas was set to hyperoxic gas challenge conditions of 41%, 71%, and 100% O 2 using an air-oxygen blender (Sensor Medics Corporation, Yorba Lina, CA, USA).The oxygen levels were chosen to homogenously span the range between normoxic 21% and a maximum of 100% oxygen.For each BOLD measurement, a T 2 * multigradient echo sequence (MGE) was used (TR: 850 ms, TE: 14 echoes from 7-34.47 ms with an interval of 2.11 ms, matrix size: 128 × 128, number of slices: 10; slice thickness: 1 mm, spatial resolution: 0.278 × 0.222 mm, acquisition time for 3 cycles: 4 min).In between each hyperoxic measurement cycle consisting of three T 2 * map acquisitions, a five-minute waiting period was included into the scanning protocol to allow enough accumulation time for oxygen in the blood.At the end of imaging, mice were euthanized by cervical dislocation.

Image Postprocessing T 2
* and T 1 maps were generated using the image sequence analysis tool in the Paravision 7 software suite.Then, using in-house developed MATLAB code (version 2018a), T 2 * relaxivity (R 2 * ) was calculated as the inverse of T 2 * .Subsequently, R 2 * parameter maps averaged over the three baseline image acquisition cycles were generated.The difference between the averaged baseline R 2 * parameter map and challenged image acquisitions at each cycle was calculated using Equation (1): where R 2 * (challenge) corresponds to challenged R 2 * parameter maps per oxygen level, while R 2 * avg (baseline) relates to the average baseline R 2 * parameter map.
Following the above calculations, to discriminate between hyper-oxidation responsive and non-responsive voxels, all ∆R 2 * parameter maps were filtered according to Equation ( 2), as presented in [25]: where SD(∆R 2 Meanwhile, independently of the oxygen concentration, significant differences between luminal A and Her2 + , as well as luminal A and triple-negative breast cancer were observed.At 100% O 2 , these differences became most pronounced.No significant difference between Her2 + and triple-negative breast cancer was observed at any oxygen concentration.The characterization of breast cancer molecular subtypes based on ∆R 2 * using different hyperoxic gas challenges is illustrated in Figure 1.
Diagnostics 2023, 13, x FOR PEER REVIEW 5 of 10 Meanwhile, independently of the oxygen concentration, significant differences between luminal A and Her2 + , as well as luminal A and triple-negative breast cancer were observed.At 100% O2, these differences became most pronounced.No significant difference between Her2 + and triple-negative breast cancer was observed at any oxygen concentration.The characterization of breast cancer molecular subtypes based on ∆ 2 * using different hyperoxic gas challenges is illustrated in Figure 1.

Discussion
The results of our preclinical study show that there is a significant difference between luminal A and Her2 + , as well as luminal A and triple-negative BC in their response to hyperoxic BOLD-MRI independently of the amount of oxygen.While the significant difference is most pronounced at 100% O2., it is already achievable at 41% O2.
Breast cancer is known for its high level of intratumoral heterogeneity, where varying selective pressures of the hypoxic tumor microenvironment induce divergent physiological phenotypes between tumor subregions [12,13].Currently, treatment plans are based on invasive tissue sampling, which is inherently prone to site selection bias and can rarely capture the full extent of intratumoral heterogeneity, contributing to treatment failure and patients' deaths [14].It is therefore crucial to establish non-invasive imaging biomarkers that identify molecular subtypes of breast cancer and enable a comprehensive depiction of hypoxic tumor physiology.Recently, for example, Parkins et al. [26] performed non-invasive PET/MRI in a murine model of triple-negative breast cancer.The authors identified PET-based biomarkers that directly link hypoxia to the heterogeneous

Discussion
The results of our preclinical study show that there is a significant difference between luminal A and Her2 + , as well as luminal A and triple-negative BC in their response to hyperoxic BOLD-MRI independently of the amount of oxygen.While the significant difference is most pronounced at 100% O 2 ., it is already achievable at 41% O 2 .
Breast cancer is known for its high level of intratumoral heterogeneity, where varying selective pressures of the hypoxic tumor microenvironment induce divergent physiological phenotypes between tumor subregions [12,13].Currently, treatment plans are based on invasive tissue sampling, which is inherently prone to site selection bias and can rarely capture the full extent of intratumoral heterogeneity, contributing to treatment failure and patients' deaths [14].It is therefore crucial to establish non-invasive imaging biomarkers that identify molecular subtypes of breast cancer and enable a comprehensive depiction of hypoxic tumor physiology.Recently, for example, Parkins et al. [26] performed non-invasive PET/MRI in a murine model of triple-negative breast cancer.The authors identified PET-based biomarkers that directly link hypoxia to the heterogeneous alternation of PD-L1 expression, highlighting the potential of imaging biomarkers to characterize hypoxia-induced physiological heterogeneity and support treatment planning in breast cancer.MRI also provides other non-invasive contrasting techniques for the assessment of intratumoral hypoxia via the BOLD-effect [15,27,28].The BOLD effect is dependent on the fraction of oxygenized hemoglobin in the tumor microvasculature, making it a useful diagnostic tool for the identification of oxygen-deprived tumor subregions.BOLD-MRI using hyperoxic gas challenges is performed to introduce diagnostic variation of intratumoral deoxyhemoglobin concentration, thereby allowing the identification of structurally intact tumor regions.Until now, measurements of the BOLD effect have mostly been performed using hyperoxic gas challenges of 100% O 2 or carbogen (95% O 2 and 5% CO 2 ) [15,22,28].However, several studies reported that hyperoxic gas challenges, especially with carbogen, were not well tolerated by patients [22,28].Moreover, the oxygen equilibrium curve suggests that the amount of oxygenized hemoglobin barely changes beyond pO 2 = 70 mmHg, which corresponds to normoxic conditions of 21% O 2 in the breathing air [24].We therefore hypothesized that molecular subtype characterization of breast cancer with BOLD-MRI using hyperoxic gas challenges of less than 100% O 2 would lead to similar results as when 100% O 2 is used.Of note, the ability to use less than 100% O 2 would greatly improve the technique's applicability in standard clinical practice.
Indeed, our results show that an increase to 100% O 2 had no beneficial effect on the characterization of breast cancer.Instead, 71% O 2 or even 41% O 2 were sufficient to detect measurable decreases in the transverse relaxation rate (∆R 2 * ).Across all ∆R 2 * measurements, statistically significant differences between less aggressive luminal A and moderately aggressive Her2 + , as well as between luminal A and highly aggressive triplenegative breast cancer were observed.On the other hand, no significant difference in ∆R 2 * measurements between Her2 + and triple-negative breast cancer was observed.Further, under all oxygen levels used in this study, ∆R 2 * was the lowest in luminal A breast cancer.As for the lack of distinction between Her2 + and triple-negative breast cancer, such lack has been observed previously in studies using dynamic-contrast-enhanced MRI [29][30][31].As for the low ∆R 2 * measurements in luminal A breast cancer, this can be explained by the fact that the magnitude of BOLD signal decrease following a hyperoxic gas challenge depends on the vascular maturity in the tissue, i.e., the ability of vessels to transport oxygen-enriched blood [32,33], and clinical studies using contrast-enhanced MRI have shown that luminal A tumors develop the most functional blood vessels in comparison to Her2 + and triple-negative tumors [34].
To our knowledge, this is the first preclinical study to successfully discriminate luminal A from other breast cancer subtypes using hyperoxic BOLD-MRI.Previously, Virani et al. conducted a preclinical study including a hyperoxic gas challenge and found correlations between BOLD parameters and tumor hypoxia in mouse glioblastoma models [35].In another preclinical study, McPhail and Robinson [36] found correlations between carbogeninduced ∆R 2 * and the extent of hypoxia in chemically induced rat breast tumors.Similarly, Yang et al. [25] found associations between histologically-proven hypoxic tumor subregions and BOLD-MRI signal following a 100% O 2 challenge in rat breast tumors.Others have used BOLD-MRI measurements to predict treatment response in rat models of breast cancer [37] and prostate cancer [38,39].
Studies in humans have found correlations between R 2 * measured at 21% O 2 and tumor hypoxia in patients with invasive ductal carcinoma [40,41].Using a hyperoxic gas challenge, Rakow-Penner et al. [28] found that in a preliminary study BOLD-MRI allows for the discrimination of benign and malignant breast lesions from healthy breast tissue.However, so far, attempts to differentiate benign from malignant breast lesions have not been successfully performed [42,43].The study by Liu et al. [40] showed that there was no correlation of tumor R 2 * at 21% O 2 and breast cancer receptor status, highlighting the complexity of breast cancer characterization based on BOLD parameters.Indeed, the lack of association between BOLD signal and receptor status in previous studies may be explained by the low field strengths used in clinical routine.Because of the field dependency of R 2 * [44,45], changes in R 2 * following a hyperoxic gas challenge would be less pronounced at 1.5 Tesla or 3 Tesla compared to the 9.4 Tesla system used in our study.The increased signal-to-noise ratio at high field strengths allow higher resolution scans to be obtained, which leads to less disrupting partial volume effects.This provides more detailed information on tumor physiology, especially in highly heterogenous cancers such as breast cancer [11,13].
Our results add to the characterization of molecular subtypes of breast cancer with hyperoxic BOLD-MRI and without the use of intravenous Gadolinium-based contrast agents, which have been under debate over their potential side-effects and toxicity [46].Furthermore, the results of our study suggest that the adverse effects during carbogen breathing during BOLD-MRI in the clinical setting can be reduced by setting the oxygen level to 41%.This may significantly improve the translatability of hyperoxic gas challenges while reducing the adverse effects of oxygen toxicity during prolonged oxygen breathing and promote the introduction of ∆R 2 * as a promising contrast agent-free imaging biomarker of breast cancer.
Our study has some practical limitations, which include its preclinical nature using a limited number of luminal A, Her2 + , and triple-negative tumor xenografts only.We did not quantify the fraction of responsive voxels for each BC subtype, which may prove to be useful to the characterization of BCs based on hyperoxic BOLD-MRI.Future translational studies should test whether molecular subtype characterization of breast cancer with BOLD-MRI using hyperoxic gas challenges of less than 100% O 2 is feasible and useful in humans.In addition, hyperoxic gas challenges using lower levels of oxygen may also be of value to differentiate benign lesions (such as fibroadenomas) from malignant breast lesions, which was not within the scope of the study.

Conclusions
Hyperoxic BOLD-MRI enables a distinction between luminal A and Her2 + and between luminal A and triple-negative breast cancer independently of oxygen level in a hyperoxic gas challenge.This molecular subtype characterization was already achieved with 41% O 2 .Overall, we conclude that hyperoxic BOLD-MRI without the use of intravenous contrast agents is promising to provide imaging biomarkers for breast cancer characterization and the assessment of intratumoral heterogeneity.Future studies could focus on the verification of our results in a clinical setting via the implementation of hyperoxic BOLD-MRI into multiparametric MRI protocols.

Figure 2
Figure 2 shows exemplary ∆ 2 * maps masked onto a corresponding T1-weighted image for a triple-negative breast cancer xenograft under different hyperoxic gas challenges.The figure shows that from 41%, 71%, to 100% O2, the steady decrease in ∆ 2 *

Figure 2 shows exemplary ∆R 2 *
Figure 2 shows exemplary ∆R 2 * maps masked onto a corresponding T 1 -weighted image for a triple-negative breast cancer xenograft under different hyperoxic gas challenges.The figure shows that from 41%, 71%, to 100% O 2 , the steady decrease in ∆R 2 * is clearly visible, especially in the dorsal half of the lesion.