Tissue Hypoxia and Alterations in Microvascular Architecture Predict Glioblastoma Recurrence in Humans

Glioblastoma (GBM) represents the most frequent and malign primary brain tumor in adults that is characterized by a tumor bulk with rapid cellular proliferation and active neovascularization surrounded by the tumor border zone with diffuse infiltration into the adjacent brain tissue. This aggressive and invasive behavior of GBM cells limits complete surgical resection and challenges adjuvant therapies by irradiation means. Inadequate control of these infiltrative GBM cells is a major cause of treatment failure and tumor relapse, which occurs inevitable typically 6–7 months after initial diagnosis (1) in close proximity (1–2 cm) to the primary tumor bulk in the initial infiltration zone (2). Remarkably, fundamental knowledge of the (spatial-temporal) pathophysiology of GBM recurrence development is slowly emerging (3), however, new concepts for early and reliable detection of recurrent GBM are urgently needed.

GBM cells infiltrate the surrounding neural tissue through two principal pathways, namely, the perivascular space or the brain parenchyma, especially along white matter tracts (4). The extracellular spaces in the parenchyma, however, are narrow and more tortuous than the perivascular space. Experiments in animal models have shown that the vast majority (>85%) of GBM cells migrate along preexisting blood vessels (5, 6), a process known as vascular cooption (7), and disrupt the astrocyte–vasculature interaction (8). Subsequently, tumor cells organize themselves into cuffs around microvessels accompanied by upregulation of angiopoietin-2 (ANG-2) in the coopted vessels (7). This is associated with vascular regression leading to a decrease in tissue oxygen tension (hypoxia), hypoxia-induced VEGF expression (9), and eventually to the development of new vessels, a process known as neovascularization (9). Thereby, this infiltrative tumor niche either becomes part of the expanding tumor bulk or develops into a new tumor mass (10).

These biological processes of GBM progression are most likely relevant for the development of recurrences, and combined assessment of microvascular architecture, oxygen metabolism, and neovascularization activity might elucidate pathophysiologic mechanisms involved in therapy resistance and tumor recurrence. Given these facts, these parameters might represent clinically relevant novel imaging biomarkers for the early detection of GBM recurrence. Most of the available neuroimaging techniques, however, do not allow a combined investigation of these different processes, requiring multiple examinations with different modalities, which in turn are not well suited for in vivo investigations in humans due to their invasiveness (oxygen electrodes or biopsies), or their limited availability and high costs (PET). This hinders the implementation of such outcome measures for clinical routine use. To counterbalance these limitations, we previously presented a novel multiparametric MRI approach scoping to obtain quantitative information about microvascular architecture and perfusion, neovascularization activity, and oxygen metabolism in patients with glioma (11–14).

In this study, we hypothesized that our MRI approach offers the capability to detect pathophysiologic changes in the early developmental stage of GBM recurrence in humans. Hence, the aims of our study were (i) to investigate the pathophysiologic changes preceding GBM recurrence, (ii) to assess the interdependence of the physiologic parameters prior to radiological recurrence, and (iii) to determine pathophysiologic differences in the development of locally and distantly recurrent GBM.

The institutional review board approved this retrospective study. Written informed consent in accordance with the ethical standards of the Helsinki Declaration of 1975 and its later amendments was obtained from all enrolled patients. A prospectively populated institutional database was searched for patients with GBM (World Health Organization, WHO grade IV) who were treated according to standard of care, that is, maximal safe and radical resection, radiotherapy, and concomitant and adjuvant chemotherapy with temozolomide (15), and received magnetic resonance (MR) examinations with our MRI study protocol between July 2015 and April 2020. Inclusion criteria were as follows: (i) aged ≥ 18 years, (ii) pathologically confirmed GBM on the basis of the WHO histologic grading system, (iii) no previous diagnosis of GBM recurrence, (iv) no additional anti-GBM treatment, but the standard of care (i.e., no antiangiogenic therapy etc.), (v) two post-standard-therapeutic follow-up MRI scans without signs for radiological recurrence were performed in a period of maximal 1 year prior to radiological recurrence, and (vi) recurrence was determined by at least two board-certified radiologists in consensus on the basis of the updated Response Assessment in Neuro-Oncology criteria (16, 17) with clear radiological features of recurrence. Radiological recurrence was defined as endpoint of the study period for each patient.

Follow-up MRI examinations were performed every 3–6 months or on an unscheduled basis in case of clinical signs of tumor recurrence. MRI data acquisition was carried out on a 3-Tesla Whole-body Scanner (Trio, Siemens) equipped with the standard 12-channel head coil. The conventional MRI (cMRI) protocol for diagnosis of brain tumors in clinical routine included, among others, the following sequences: (i) an axial fluid-attenuated inversion recovery (FLAIR) sequence, (ii) an axial diffusion-weighted imaging sequence, (iii) pre- and post-contrast enhanced (CE) high-resolution three-dimensional T1-weighted (T1w) magnetization-prepared rapid acquisition with gradient echo sequences, and (iv) a gradient echo (GE) dynamic susceptibility contrast (DSC) perfusion MRI sequence with 60 dynamic measurements during administration of 0.1 mmol/kg bodyweight gadoterate-meglumine (Dotarem, Guerbet) at a rate of 4 mL/second using a MR-compatible injector (Spectris, Medrad). A 20-mL bolus of saline was injected subsequently at the same rate. The parameters of the cMRI sequences are summarized in Supplementary Table S1.

For MR-based examination of microvascular architecture and neovascularization activity using the vascular architecture mapping (VAM) approach, we used a DSC perfusion MRI sequence obtained with a spin echo (SE) echo-planar imaging read out (SE-DSC) using the same parameters and contrast agent injection protocol as described for the routine GE-DSC perfusion MRI (Supplementary Table S1). Our strategies to minimize patient motion and differences in time to first-pass peak, which may significantly affect the data evaluation, were described previously (13, 14).

For MR-based investigation of tissue oxygen metabolism using the quantitative blood oxygen level–depended (qBOLD) approach, we performed the following sequences: (i) a multi-echo GE sequence and (ii) a multi-echo SE sequence for mapping of the transverse relaxation rates R₂* ( = 1/T₂*) and R₂ ( = 1/T₂), respectively. All experimental sequences for VAM and qBOLD used the same geometric parameters (voxel size, number of slices, etc.) and identical slice position as for the routine GE-DSC perfusion sequence (Supplementary Table S1). The additional acquisition time (TA) for the VAM sequences (SE-DSC perfusion: TA, 2 minutes) and qBOLD (R₂* and R₂-mapping: TA, 1.5 and 3.5 minutes) was 7 minutes.

Processing of cMRI, VAM, and qBOLD data, as well as calculation of MRI biomarkers were performed with custom-made MATLAB (MathWorks) software. Details of the MRI data processing pipeline were published previously (11–14) and are described in Supplementary Fig. S1A. The procedure resulted in the MRI biomarker maps for microstructural density (ADC), macrovascular (CBV) and microvascular perfusion (μCBV), microvascular architecture, including microvessel density (MVD) and vessel size index (VSI), neovascularization activity represented by the microvessel type indicator (MTI), and oxygen metabolism, including oxygen extraction fraction (OEF), cerebral metabolic rate of oxygen (CMRO₂), and tissue oxygen tension (PO₂). All nine biomarker maps are summarized at the bottom of Supplementary Fig. S1A.

A retrospective quantitative analysis of the MRI biomarker data was conducted to obtain the time courses of the dynamic changes in the nine physiologic parameters that occurred prior to recurrence of GBM after standard therapy. This was executed in four steps, starting first with definition of regions of interest (ROI) on the CE T1w MRI with clear radiological features of GBM recurrence, that is, the endpoint of the study period for an individual patient. Second, these ROIs were transferred to the previous follow-up MRI data without signs of GBM recurrence (first and second follow-up). In a third step, MRI biomarker maps were coregistered to the respective CE T1w MRIs, the ROIs were copied, and the mean MRI biomarker values were computed. Finally, the MRI biomarker values were plotted against the time difference between the date of the follow-up MRI and the date of radiological recurrence, that is, the time before radiological recurrence (TBR; in days), which resulted in the time courses of the MRI biomarkers. Details of the procedure are described in Supplementary Fig. S1B.

Local GBM recurrence was defined as tumor progression within the wall of the resection cavity or within 20 mm from the margin, and distant GBM recurrence at more than 20 mm from the margin of the resection cavity (18).

Statistical analyses were performed using R (version 3.6.3, R Foundation) and SPSS (version 21, IBM). For the scatter plots of MRI biomarker values versus TBR nonparametric, regressions were performed using locally estimated scatterplot smoothing (LOESS) functions to obtain smooth curves. The LOESS trend lines were graphically analyzed for dynamic physiologic changes that precede GBM recurrence. A two-sample Kolmogorov–Smirnov test was used to search for significant differences in the biomarker distribution between local and distant recurrences. Furthermore, the MRI biomarker values were quarterly pooled for further statistical analysis. Significance of differences in quarterly pooled MRI biomarker values between locally and distantly recurrent GBM was calculated by using a Mann–Whitney U test. Significance of differences in MRI biomarkers between the four quarters to recurrence and to radiological GBM recurrence within the two subgroups (local and distant recurrence) was determined using the one-way ANOVA method. The Tukey test was used as post hoc procedure to be consistent with the assumption that homogeneity of variance was met and for correction for multiple comparisons. Homogeneity of variance was verified using the Levene test. When the assumption of homogeneity of variances was violated, Welch ANOVA in combination with the Games–Howell post hoc test was used. P values less than 0.05 were considered to indicate significance.

The institutional database contained almost 1,200 MR examinations using the study MRI protocol in 400 patients with brain tumors. From these, 106 patients suffered from brain tumors other than glioma (e.g., brain metastasis or cerebral lymphoma) and 103 patients had a low-grade (WHO grade II) or anaplastic glioma (WHO grade III). Furthermore, 135 patients with GBM were excluded because these patients had none or only one post-standard-therapeutic follow-up MRI scan without signs for radiological recurrence.

This resulted in a total of 56 patients (37 males; 60.8 ± 11.5 years; 29–80 years) who fulfilled the study inclusion criteria. Forty-nine patients (87.5%) showed a local recurrence of the GBM, and from these, 11 patients (19.6%) developed a distant recurrence too, that is, a combined tumor recurrence pattern. Seven patients (12.5%) only had a distantly recurrent GBM. Thus, in total, 18 patients suffered from a distant recurrence of the tumor. The mean time between first diagnosis and recurrence was 14.5 ± 6.6 months (5.6–31 months). Five patients (8.9%) initially had a GBM with a mutation of the IDH1 gene. A gross-total resection of the initial GBM was achieved in 26 patients (46.5%), a near-total resection in 25 patients (44.6%), and a subtotal resection in 5 patients (8.9%). Treatment of the recurrent GBM included a repeat craniotomy in 27 patients (48.2%, pathology confirmed GBM recurrence in all cases), a temozolomide rechallenge in 12 patients (21.5%), repeat radiotherapy or repeat combined radiochemotherapy in 6 patients (10.7%), second-line monotherapy with the antiangiogenic drug, bevacizumab, in 6 patients (10.7%), and palliative care without further treatment of the tumor in 5 patients (8.9%). Supplementary Fig. S2 shows an example for a repeat craniotomy of a recurrent GBM.

Calculation of MRI biomarker maps for perfusion (CBV and μCBV), microvascular architecture (MVD and VSI), neovascularization activity (MTI), and oxygen metabolism (OEF, CMRO₂, and PO₂) was successfully performed for all 168 MRI examinations of all 56 patients. These MRI biomarker maps, but not cMRI, provided indications for early changes in tissue physiology which occurred prior to recurrence of GBM in humans. Figure 1 shows an illustrative case depicting two follow-up MRI examinations (Fig. 1A and B) without signs for recurrence in cMRI (CE T1w, FLAIR, and CBV), but changes (red arrows) in μCBV and microvascular architecture (MVD), as well as increasing hypoxia (decreasing PO₂) and neovascularization activity (MTI) in the second follow-up (95 days prior to radiological recurrence; Fig. 1B). At radiological recurrence (Fig. 1C), massive contrast enhancement and peritumoral edema were recognized accompanied by increased perfusion (CBV and μCBV), microvascular architecture (MVD), and neovascularization activity (MTI), as well as a recovery of the tissue oxygen tension (increased PO₂) to normal levels (and even higher).

The most interesting aspect of another representative case (Fig. 2) was the very regional hypoxia (low PO₂) that was detected 176 days before radiological recurrence (enlarged image section in the rightmost image of Fig. 2B). This small hypoxic region might have been the starting point that initiated upregulation of neovascularization activity leading to increased perfusion and microvascular architecture, as well as recovery of the tissue oxygen tension, massive contrast enhancement, and peritumoral edema at radiological recurrence almost half a year later (Fig. 2C). A further representative case with distant GBM recurrence and less pronounced pathophysiologic changes is presented in Supplementary Fig. S3.

The MRI biomarker values obtained from 147 MRI examinations of the 49 patients who developed local recurrence of the GBM were plotted against the time before radiological recurrence (TBR in days) of the tumor. The same procedure was carried out separately with the MRI biomarker data from 54 MRI examinations in 18 patients with distant recurrence of the GBM. The resulting time courses provided information about the dynamic changes in perfusion, microstructural density, microvascular architecture, neovascularization activity, and oxygen metabolism that occurred before and during the development of local and distant recurrence of GBM, respectively (Fig. 3).

Both macrovascular (CBV) and microvascular perfusion (μCBV) revealed similar time courses for both local and distant GBM recurrence (Fig. 3A and B). The values for CBV and μCBV showed a constant course until they increased from 120 days to radiological recurrence. It is noteworthy that the increase was a little stronger for μCBV, that is, microvascular perfusion (Fig. 3B) increased faster compared with macrovascular perfusion (Fig. 3A). The density of microvessels (MVD), however, revealed a continuous decrease during the first two quarters that was intensified at about 190 days and reached a clear minimum at about 120 days prior to radiological recurrence. This was followed by a strong increase in MVD (Fig. 3D). VSI, the second parameter for microvascular architecture (Fig. 3E), however, showed a fairly opposite course in the 1-year period to recurrence, which was in accordance with previous findings (11): microvasculature either shows high density of vessels with small lumen or low density of large-lumen vessels. Notwithstanding, 40 days before radiological recurrence, a divergence from this principal tendency was observed, namely a simultaneous increase of both the mean vessel density and diameter. In other words, these findings indicated that early neovascularization activity (120–40 days to radiological recurrence) was more dominated by formation of smaller vessels which were transformed into larger-lumen vessels during later phases of tumor vasculature development. Compared with that, the time course of MTI was found to be rather simple and demonstrated an upregulation of neovascularization activity at about 120 days prior to radiological GBM recurrence (Fig. 3F).

The time courses of the MRI biomarkers for oxygen metabolism revealed marked differences between local and distant recurrence. For tissue that developed in the further course into local recurrence of a GBM, OEF (Fig. 3G) showed an inverse course compared with MVD over the whole 1-year period: a rather strong increase over the first two quarters which was even intensified at about 190 days and reached its maximum at 120 days prior to radiological recurrence. This was followed by a strong decrease because of the development of a tumor vasculature enabling adequate blood supply and eventually hyperperfusion. The metabolic rate of oxygen (CMRO₂, data in cyan in Fig. 3H), however, was rather constant over the first two quarters, started to increase at about 190 days to recurrence, and reached the maximum at about 60 days, which was 60 days later than the extrema of the other parameters. This means that CMRO₂ showed a time lag in the course compared with OEF and MVD. Consequently, tissue oxygen tension (PO₂, data in cyan in Fig. 3I), which is influenced by both OEF and CMRO₂, showed a decrease (i.e., increasing hypoxia) from the second quarter that was even intensified at a TBR of about 190 days and reached its minimum value (i.e., maximum hypoxia) at 90 days prior to radiological recurrence.

This clearly demonstrated a close correlation between the dynamics of perfusion (especially μCBV), microvessel architecture (especially MVD), and oxygen metabolism in the 1-year period prior to local recurrence. The time of the OEF maximum and the MVD and μCBV minima was coincidently at 120 days. Remarkably, however, the maximum for CMRO₂ (TBR of 60 days) and consequently the minimum for PO₂ (TBR of 90 days) occurred closer to the time of radiological recurrence, which was indicative of a metabolic switch from oxidative phosphorylation to aerobic glycolysis, also known as the Warburg effect, which is associated with a switch from an infiltrative to a proliferative phenotype (19). The time courses of the ADC values (Fig. 3C, data in cyan) were supportive for this explanation, because these revealed a decrease of microstructural density (increasing ADC values) until 120 days prior to recurrence, followed by an increase in microstructural density (decreasing ADC) probably due to tumor cell proliferation. The decrease in microstructural density in tissue that developed in the further course into local GBM recurrence, however, might be associated with post-radiotherapeutic effects (20). It is noteworthy that oxygen metabolism in brain tissue that developed in the further course into distant GBM recurrence (data in red in Fig. 3G–I) was rather constant over the first two quarters and the changes were weaker and occurred later compared with those for local recurrences. This might be explained by additional local effects of radiotherapy, which was previously associated with post-therapeutic tissue hypoxia (21).

Models for the physiologic changes that occur during development of local and distant GBM recurrence (Fig. 4A and B), respectively, were obtained via combination of the comprehensive data from the different time courses (Fig. 3). The diagrams provide insight into the interdependencies between vessel density/perfusion (as a combined parameter), tissue oxygen tension (i.e., hypoxia), neovascularization activity, and microstructural density, which preceded radiological recurrence of GBM in humans. The scenario might be as follows: during the first half of the 1-year period to local GBM recurrence, post-radiotherapeutic effects cause a continuous decrease in local vessel density/perfusion, oxygen tension, and microstructural density. Approximately 190 days prior to radiological recurrence, incipient cooption of GBM cells to microvessels induces vessel regression, which is associated with an additional decrease in vessel density/perfusion and oxygen tension (i.e., intensified hypoxia; Fig. 4A). After about 70 days of intensified decrease (at TBR of 120 days), the minimum of vessel density/perfusion is reached at 50% of the initial level because neovascularization activity is upregulated and starts to promote a strong continuous increase of vessel density/perfusion until recurrence of the GBM is radiologically detected. In addition, at TBR of 120 days, microstructural density also begins to increase slowly, but continuously, due to tumor cell proliferation. Tissue oxygen tension, however, continues to decrease over a further 30-day period after initiation of neovascularization and reaches its minimum 90 days prior to radiological recurrence. At radiological detection of GBM recurrence, vessel density/perfusion reaches on average 170% and tissue oxygen tension 125% of the initial level, at the beginning of the 1-year period. The model for the distantly recurrent GBM is very similar. The crucial turning points, however, occur 10 days later, possibly due to statistical fluctuations, and the minima were higher (Fig. 4B).

Statistical analysis of the quarterly pooled MRI biomarker data revealed that the most significant differences between the development of local and distant GBM recurrences (marked by a “†” in Supplementary Fig. S4), respectively, existed for MVD (for second and third quarters) and oxygen metabolism: OEF and CMRO₂ for the third and fourth quarters and PO₂ for the third and fourth quarters, as well as at radiological recurrence. In other words, the dynamics of changes in microvascular density and oxygen metabolism (including hypoxia) were significantly stronger for local recurrence compared with distant recurrence.

On the other hand, locally recurrent GBM per se showed more significant differences between the quarters (marked by a “*” in Supplementary Fig. S4) compared with distantly recurrent GBM. During local GBM recurrence, CBV, μCBV, MVD, and CMRO₂ increased significantly from the third to the fourth quarter and from the fourth quarter to radiological recurrence. Only oxygen metabolism, however, revealed significant early changes half a year prior to local radiological recurrence: OEF and CMRO₂ increased significantly and PO₂ decreased (i.e., tissue hypoxia increased) significantly from the second to the third quarter (Supplementary Fig. S4).

Recurrence is an inevitable event in the course of disease of most patients with GBM and poses a major diagnostic and therapeutic challenge as recurrent GBMs are commonly detected at an advanced stage, which limits the choice of treatment options. Gross-total resection of GBM recurrence is associated with significantly longer overall survival compared with no surgery or subtotal resection (22) and should, therefore, be performed whenever possible in patients with recurrent GBM. However, early and reliable detection of GBM recurrence is essential to enable the most radical repeat resection of the tumor.

In this study, we applied an extended MRI protocol, which is fully compatible with the requirements of clinical routine diagnostics (11, 13), to search for early alterations in macro- and microvascular perfusion, microvascular architecture, neovascularization activity, oxygen metabolism (including tissue hypoxia), and microstructural density that occurred during development of recurrence in patients with GBM. These physiologic MRI biomarkers were introduced and used in previous studies. The novelty of this study, however, is the noninvasive investigation of the dynamic pathophysiologic changes that precede the recurrence of GBM over a period of 1 year using these biomarkers. Our main findings were threefold: (i) our MRI approach enabled detection of early pathophysiologic changes during development of GBM recurrence, (ii) we developed a model describing the dynamic changes and the interdependence of the physiologic parameters that precede GBM recurrence, and (iii) differences in the pathophysiologic development of locally and distantly recurrent GBM were dominated by post-radiotherapeutic effects.

The earliest pathophysiologic alteration, which was detected with our MRI approach and might be associated with the development of GBM recurrence, occurred 190 days prior to radiological detection of recurrence. We observed a decrease in microvascular density with accompanied increases in oxygen extraction from vasculature and tissue hypoxia. This might be related to vascular cooption of GBM cell and formation of cuffs around microvessel (7). It is well-known that coopted vessels express ANG-2, which in absence of VEGF promotes endothelial cell apoptosis and vessel regression (9), that could have been macroscopically detected as decreasing MVD with our MRI approach. Zagzag and colleagues (23) identified, in a temporal study on a murine glioma model, vascular cooption as an initial step of a cascade of events. By using transmission electron microscopy (TEM), they observed vascular cooption 1 week after glioma cell (GL261) implantation, followed by endothelial cell apoptosis and vascular regression after 3 weeks. Involution of coopted vessels resulted in tumor hypoxia, upregulation of proangiogenic factors (hypoxia-induced VEGF expression), and a shift toward an angiogenic phenotype after 4 weeks. In our study the effects of the angiogenic switch (increasing MVD, CBV, μCBV, and MTI) were macroscopically observed 70 days after the assumed onset of vascular cooption and 120 days prior to detection of recurrence in cMRI. The significant difference in the time to the angiogenic switch (3 vs. 10 weeks) may be caused by the different study designs: microscopic investigation with TEM of implanted glioma cells in mouse brains versus macroscopic investigation with MRI of recurrent GBM in patients. Furthermore, this could also be heavily influenced by differences in treatment which patients versus these mice received.

A continuous increase in tissue hypoxia (decrease in PO₂) preceding the angiogenic switch was observed in our study too. Interestingly, hypoxia continued to increase over another 30 days after the angiogenic switch, which might be required to maintain and intensify the neovascularization activity. Otherwise, that is, without continued hypoxia after the angiogenic switch, the neovascular progression might be totally or in part be suppressed. This was a very interesting and unexpected finding of our study, because it could open up the possibility of developing new combined monitoring–treatment options tailored for recurrent GBM. Taking these study observations into account, a potential strategy could be as follows: follow-up examinations, including our physiologic MRI approach, are used to search for increasing MVD and μCBV, and MTI values suspect for incipient neovascularization activity combined with decreasing PO₂ (increasing hypoxia). If this pattern is detected, one could initiate antiangiogenic (24) or antihypoxic therapy (25, 26) or a combined treatment targeting both angiogenesis and hypoxia (27). Of course, repeat gross-total resection or minimal invasively stereotactic laser ablation (28) for recurrent GBM may be alternative treatment options too.

The usefulness of cMRI and PET for GBM recurrence prediction has been already investigated in previously published in-human studies. Chang and colleagues (29) studied postoperative cMRIs of 26 patients with native GBM immediately after undergoing gross-total tumor resection and before chemoradiotherapy. Retrospective voxel-wise histogram analysis revealed that voxels of future GBM recurrence within the peritumoral region showed small, but statistically significant signal loss on ADC and anatomic FLAIR images months before the development of abnormal enhancement occurs. Furthermore, they calculated probability maps of tumor recurrence based on multiparametric logistic regression model using these signal changes. Metz and colleagues (30) even evaluated preoperative maps of fractional anisotropy and found significant differences between the areas of recurrence-free edema and areas with later GBM recurrence. However, the usefulness of these approaches for follow-up monitoring after chemoradiotherapy remains largely unknown. Furthermore, a prospective follow-up study after standard therapy concluded that conventional perfusion (macrovascular CBV) and diffusion MRI (ADC) do not reliably predict the GBM relapse sites (31), which is in line with our results. Two further studies (32, 33) compared postoperative cMRI and ¹⁸F-fluoroethyl-tyrosine PET data for prediction of GBM recurrence and revealed the superiority of metabolic imaging for that purpose. We chose a similar, but MRI-based approach in our study by consideration of oxygen metabolism, neovascularization activity, and microvascular architecture for early GBM recurrence detection.

There are several limitations in our study. Our VAM approach was limited by the requirement for additional DSC MRI perfusion sequences (SE-DSC) and a separate contrast agent injection. However, this ensures that the GE-DSC perfusion sequence, which is essential for clinical routine diagnosis, was kept unchanged regarding spatial and temporal resolution. Therefore, our approach allows the acquisition of VAM data with high signal-to-noise ratio, high spatial resolution, and coverage of the whole brain, which is mandatory for in clinical routine. These features are especially important for detection of small or multicentric lesions. The commonly used combined simultaneous GE-SE-DSC perfusion sequences (34–36), however, do not meet these requirements and necessitate performing the routine perfusion separately, requiring an additional contrast agent injection. Furthermore, the combined simultaneous GE-SE-DSC perfusion sequence was performed with a double dose or more (34–36). This resulted in the application of up to a triple dose. The model for assessment of oxygen metabolism also has several limitations. It assumes that the system is in the static dephasing regime (37), which is associated with the fact that OEF is predominantly weighted to the medium sized and larger vessels of the venous vascular network. Furthermore, the multiparametric qBOLD approach (38) provides an average blood oxygenation within the entire vasculature, which is different from the venous blood oxygenation (deoxygenated CBV) derived from the original qBOLD approach (39), and ignores the intravascular component. In addition, accumulation of hemosiderin and/or proteins, as well as non-BOLD susceptibility effects (R2*; due to uncorrectable macroscopic background gradients or white matter fiber orientation), and contrast agent leakage could bias the OEF estimation (40–42). Therefore, it is important to point out that the multiparametric qBOLD approach provides only an estimation of the oxygen metabolism with model-inherent limitations. The number of patients with local (N = 49) and especially distant recurrence (N = 18) was relatively small, which was related to our rather strict inclusion criteria. We performed a retrospective analysis of our prospectively obtained physiologic MRI data. Prospective studies evaluating the clinical usefulness of our pathophysiologic model of GBM recurrence deserve further attention. It is also important to note that a clear distinction between therapy-related changes in oxygen metabolism and microvascular architecture was not possible with our imaging approach. The causes of the physiologic changes prior to GBM recurrence, as described in this study, as well as the models that were developed, are hence speculative. Therefore, histologic and biological validation of our findings is required by correlation with results from IHC or invasive methods using animal models, as well as from clinical PET data of the same patients.

Conclusively, we were able to describe time courses of the pathophysiologic tumor characteristics that occur prior to the recurrence of GBM in patients after standard therapy. These findings might be helpful to implement both new MR-based imaging modalities for routine follow-up monitoring of patients with GBM after standard therapy and novel therapy options which, if combined, could enable earlier and specially tailored treatment of recurrent GBM. However, further preclinical, as well as prospective in-human studies are required for validation, evaluation, and application of our findings.

A. Stadlbauer reports grants from German Research Foundation during the conduct of the study. K. Podar reports personal fees from Celgene, Janssen, and Amgen (consulting and honoraria) and Eutropics, Celgene, and Janssen (advisory board) outside the submitted work. No disclosures were reported by the other authors.

A. Stadlbauer: Conceptualization, data curation, formal analysis, funding acquisition, investigation, visualization, methodology, writing-original draft, project administration. T.M. Kinfe: Validation, writing-original draft, writing-review and editing. I. Eyüpoglu: Writing-review and editing. M. Zimmermann: Software, methodology, writing-review and editing. M. Kitzwögerer: Data curation, formal analysis, project administration, writing-review and editing. K. Podar: Conceptualization, validation, writing-review and editing. M. Buchfelder: Resources, supervision, writing-review and editing. G. Heinz: Resources, data curation, supervision, writing-review and editing. S. Oberndorfer: Conceptualization, resources. F. Marhold: Resources, data curation, supervision, investigation, writing-original draft, project administration, writing-review and editing.

The authors acknowledge support by Open Access Publishing Fund of Karl Landsteiner University of Health Sciences, Krems, Austria.

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.