Vessel calibre—a potential MRI biomarker of tumour response in clinical trials

Abstract

Our understanding of the importance of blood vessels and angiogenesis in cancer has increased considerably over the past decades, and the assessment of tumour vessel calibre and structure has become increasingly important for in vivo monitoring of therapeutic response. The preferred method for in vivo imaging of most solid cancers is MRI, and the concept of vessel-calibre MRI has evolved since its initial inception in the early 1990s. Almost a quarter of a century later, unlike traditional contrast-enhanced MRI techniques, vessel-calibre MRI remains widely inaccessible to the general clinical community. The narrow availability of the technique is, in part, attributable to limited awareness and a lack of imaging standardization. Thus, the role of vessel-calibre MRI in early phase clinical trials remains to be determined. By contrast, regulatory approvals of antiangiogenic agents that are not directly cytotoxic have created an urgent need for clinical trials incorporating advanced imaging analyses, going beyond traditional assessments of tumour volume. To this end, we review the field of vessel-calibre MRI and summarize the emerging evidence supporting the use of this technique to monitor response to anticancer therapy. We also discuss the potential use of this biomarker assessment in clinical imaging trials and highlight relevant avenues for future research.

Introduction

An abnormal vasculature characterized by disorganized, permeable and dilated vessels is a hallmark of cancer tissue.^1,2 Unlike the ordered microcirculatory arrangement of blood vessels found in normal tissue, the heterogeneous tumour vascular bed comprises a chaotic network of mixed vessel types, from small-calibre, capillary-like vessels to oversized and permeable post-capillary venule-like vessels, and vascular malformations. The formation of vessels with a wide spectrum of calibres—that is, the cross-sectional width of the vessel—in cancer tissue is attributed to compression of the vessel walls by tumour and host cells, as well as vessel dilatation owing to imbalance between proangiogenic and antiangiogenic factors.^3–7 The formation and development of abnormal tumour vessels have pivotal roles in the progression of cancers towards metastatic phenotypes and in determining the success of anticancer therapy.^5,8 Whereas contrast-enhanced MRI is generally considered the gold standard clinical approach for in vivo radiographic imaging assessment of the structural and haemodynamic status of solid tumours,⁹ vessel-calibre MRI has emerged as a potentially useful approach for clinical trial response monitoring only over the past decade and is not widely established. With the advent of antivascular and antiangiogenic therapies that seem to be cytostatic rather than predominantly cytotoxic and, therefore, do not simply result in decreased tumour sizes, traditional assessment of radiographic response and progression-free survival according to MRI-based tumour-volume criteria can no longer be considered an adequate biomarker of therapeutic response.^9–15

Herein, we provide a comprehensive and critical review of vessel-calibre assessment by MRI. We present a historical overview, ranging from the early development of this approach to current clinical trials, and explain the basic concepts underlying the technique. We summarize the proangiogenic and antiangiogenic influences that determine the structure of tumour vasculature, and explain how blood vessels can be imaged in vivo using MRI protocols. In addition, we discuss potential reasons for the lack of uptake of this methodology, and provide recommendations for imaging that we believe will enable realization of the full potential of vessel calibre as a biomarker in clinical trials. Finally, we highlight key advancements and future directions relating to vessel-calibre MRI.

Vessel calibres in solid cancers

The processes of tumour vascularization and growth, angiogenic pathways and targeted therapies are among the most-studied facets of cancer,^5,16–19 and a detailed description of these aspects is outside the scope of this Review. Nevertheless, we provide a brief overview of the relevant mechanisms that directly affect tumour-vessel calibres in the following sections.

The vasculature of solid cancers

For a tumour to grow beyond a few millimeters in diameter, the formation and remodelling of new blood vessels is essential.¹⁶ Growth of the tumour vasculature is thought to occur in a variety of ways that includes: use of the pre-existing host vasculature (vessel co-option); growth of new vessels as branches from existing vessels (sprouting angiogenesis); splitting of existing blood vessels to form new branches (intussusception); de novo formation of blood vessels independent of existing vessels, either comprising endothelial cells (vasculogenesis) or tumour cells in the absence of endothelial cells and fibroblasts (vascular mimicry); and transdifferentiation of tumour cells into endothelial cells.⁵ Initially, many tumours grow by vessel co-option; however, at some point an angiogenic switch occurs that leads to overproduction of proangiogenic factors, such as vascular endothelial growth factor (VEGF), fibroblast growth factors (FGFs), angiopoietin-2 (ANG-2) and chemokines, which in turn leads to new vessel formation and vessel maturation (Figure 1).^1,5,17 The new vessels are predominantly immature and abnormal (overdilated, hyperpermeable, tortuous and disrupted), resulting in variable perfusion;^2,7 some tumour tissues receive too much blood, whereas others do not receive enough and are, therefore, starved of oxygen and nutrients.^2,7 In parallel, solid stress imparted by proliferating cancer cells, stromal cells and the extracellular matrix can lead to compression of vessels, reducing blood flow (Figure 1).^3,4,20,21 Consequently, the tumour vascular bed is spatially and temporally heterogeneous with regard to multiple vessel parameters, not least vessel calibre, which creates an irregular microenvironment that adversely affects drug delivery and introduces variation in cellular responses to therapeutic agents, resulting in decreased efficacy of cancer therapy.⁸ Indeed, this microenvironment is characterized by hypoxia (low oxygen concentration),^22,23 low pH, increased solid stress^4,20 and high interstitial fluid pressure²⁴—factors that can all contribute to tumour progression and resistance to various treatments (such as radiotherapy, chemotherapy and immunotherapy).^8,21

Figure 1.

Vessel calibres in solid cancers. Conceptual illustration showing key factors for vessel-calibre growth and remodelling within a tumour, including vessel dilatation after stimulation by proangiogenic factors such as VEGF, FGFs, ANG-2 and chemokines,⁵ and compression of vessels as a result of solid stress induced by tumour growth.⁴ In normal tissues, red vessels indicate oxygen-rich feeding arteries and arterioles, blue vessels are veins carrying deoxygenated blood and the intermediate regions indicate a transient capillary stage. In cancerous tissues, slow blood flow is indicated by reduced colour intensity. Abbreviations: ANG-2, angiopoietin-2; FGFs, fibroblast growth factors; VEGF, vascular endothelial growth factor.

Cancer therapies that affect vessel calibres

Treatments for cancer not only affect cancer cells, but also have broader effects on the tumour microenvironment, including the vasculature. Following radiation treatment, oxygen-deprived tumour tissue might become reoxygenated owing to radiation-induced death of surrounding radiosensitive and oxygen-rich cell populations that were previously obstructing nearby tumour blood vessels.⁵ This alteration in the microvascular environment could potentially lead to decompression of tumour vessels, increased vessel calibres and increased perfusion of tumour tissues.²³ In addition, some chemotherapeutic drugs might kill proliferating endothelial cells and, therefore, act as antiangiogenic agents,¹⁷ with the possible consequence of modifying blood-vessel calibres. Clearly, distinguishing treatment-related changes in tumour-vessel calibres from the alterations induced by tumour growth is important for patient management and clinical trial design.

Blocking angiogenesis was initially proposed as a strategy to starve tumours of blood flow and halt the delivery of nutrients required for cell survival, growth and proliferation.²⁵ However, the therapeutic potential of antiangiogenic agents also seems to be attributable to mechanisms other than simple destruction and ‘pruning’ of the tumour vasculature;⁸ Jain and colleagues^1,26,27 have proposed that antiangiogenic agents can transiently normalize the tumour vasculature, converting the heterogeneous, abnormally dilated and hyperpermeable vessels to a more-efficient state that results in improved blood perfusion and decreased vessel diameters. This effect might create a window of opportunity during which various concomitantly administered therapies (radiation, chemotherapy or immunotherapy, for example) are likely to be most effective.^27,28 Additionally, antiangiogenic therapy could potentially disrupt the vascular cancer-stem-cell niche.¹⁷ Probably, a range of mechanisms are in fact operative at different time periods during antiangiogenic therapy; thus, the proposed diverse mechanisms of action are unlikely to be mutually exclusive.^17,29

Challenges to antiangiogenic therapy

Heterogeneity within an individual tumour, as well as among and across cancer types, suggests that each tumour responds differently to therapies depending on the biology of the component cancer-cell populations and their microenvironment.⁶ As is true for cancer-cell-directed therapies, some tumours are intrinsically resistant and others develop resistance to antiangiogenic therapies.^5,18 Tumours can develop resistance by upregulating distinct compensatory proangiogenic pathways that can circumvent therapy, or might continue to grow because of the seeming inability of the antiangiogenic agent to prune established blood vessels. Both of these possibilities highlight the need for direct measurements of vascular response—or lack thereof.

Furthermore, studies of a number of anti-VEGF agents have highlighted the importance of dosing regimens tailored to individual tumours and time-dependent drug scheduling.²⁸ Whereas low doses of anti-VEGF agents might induce vascular normalization and improve the delivery of adjuvant therapies, high doses might result in an excessive decrease in the number of functional tumour blood vessel pathways and, therefore, compromise drug delivery.⁸ For example, in patients with non-small-cell lung cancer, tumour perfusion and the net influx rate of docetaxel, a taxane targeting the microtubular network, were reduced within 5 h after administration of a 15 mg/kg dose of the anti-VEGF antibody bevacizumab.³⁰ A lower dose of bevacizumab might have had the opposite effect according to the concept of vascular normalization, although this possibility remains to be proven in patients. In two mouse models of breast cancer,³¹ low doses (5 mg/kg) of an anti-VEGFR2 antibody, DC101, increased perfusion of nanoparticles, whereas a higher dose (10 mg/kg) seemed to hinder tumour perfusion. Similarly, the size of the concomitantly administered therapeutic agent also seemed to be important in this study; the lower dose of DC101 was shown to reduce vessel diameters and improved the permeability of the tumour to 12 nm particles, but not 60 nm or 125 nm particles.³¹ Furthermore, the therapeutic efficacy of albumin-bound paclitaxel (nab-paclitaxel), which is around 10 nm in size, was increased by DC101 treatment, whereas the outcome of therapy with doxil, which is approximately 100 nm in size, was not affected by this anti-VEGF agent; doxil and nab-paclitaxel had similar efficacy in the absence of DC101 therapy. These results suggest that selected doses of DC101 can induce a vascular normalization effect that increases tumour perfusion, but only to smaller molecules, possibly by creating a more homogeneous vasculature, but with smaller vessel—or ‘pore’—diameters. In addition, a preclinical study of bevacizumab in mouse xenograft models of human ovarian and oesophageal cancers demonstrated decreased tumour uptake of trastuzumab and bevacizumab antibodies, and human immunoglobulin G, as well as reduced ex vivo mean vessel densities.³² Thus, the dose of bevacizumab used in this study might have resulted in ‘pruning’ of too many blood vessels and caused a decrease in the size of vascular pores, owing to increased pericyte coverage, to below that of an antibody.³¹ Together, these findings provide additional evidence of the gaps in our knowledge relating to the vascular responses to anticancer therapies that necessitate the development of assessments of vascular outcomes.

The dose-dependence of antiangiogenic agents has implications not only for conventional chemotherapeutics and low-molecular-weight targeted therapies, but also for the delivery of nanotherapeutics. Tumour blood vessels are more permeable to large molecules than many normal vessels, which provides an intriguing mechanism for selective delivery of macromolecules to tumours.³³ Moreover, large molecules are retained within tumours because of poor clearance by impaired lymphatic systems. These concepts are collectively known as enhanced permeability and retention (EPR).^34–37 Successful outcomes of EPR-based nanotherapeutics will probably—similarly to conventional therapies—be governed by the tumour type as well as the dose and timing of concurrent antiangiogenic therapies.³⁵ Furthermore, because the extracellular matrix can also retard the penetration of nanotherapeutics in tumours,³⁸ improved delivery of nanoparticles might require normalization of both tumour vessel and the extracellular matrix.⁸ Thus, vessel-calibre MRI might be equally important in optimizing and monitoring such approaches.

Traditional MRI in cancer therapy

Conventional MRI in monitoring response

Current approach

Use of conventional MRI for monitoring response to therapy in solid tumours has been reviewed extensively.^11,12,14,15 In short, most aggressive pathological lesions are easily identified using T₁-weighted MRI protocols after administration of a standard gadolinium-based contrast agent, on the basis of the increased vascularity and permeability of tumour vessels. T₁-weighted images are sensitized to the longitudinal magnetic-relaxation time of protons in water molecules (T₁), and paramagnetic contrast agents, such as gadolinium, increase the MRI signal intensity in the blood by shortening the T₁-relaxation times. These images are usually complemented by T₂-weighted MRI, fluid-attenuated inversion recovery (FLAIR) MRI (Figure 2) and diffusion-weighted MRI for assessments of peritumoural oedema and changes in cellular density.^39,40 Traditional T₂-weighted images are sensitive to local inhomogeneity of the magnetic field, a phenomenon known as the ‘magnetic susceptibility effect’,⁴¹ which describes the tendency of fat, water and fluids to produce hyperintense signals whereas blood appears dark (hypointense). Definitions of the most-common terminologies used in vessel-calibre MRI are provided in Supplementary Table 1 online.

Figure 2.

Vessel-calibre MRI. a | Example anatomical CE T₁-weighted, FLAIR, macrovessel volume-fraction and microvessel volume-fraction images from MRI assessments of a patient with glioblastoma. Note the sensitivity to high blood-volume fractions in the macrovessel image (total blood volume) compared with the microvessel image. b | Macrovessel and microvessel DSC-MRI in the same patient with glioblastoma, with contrast agent passing through the tissue. Before estimation of perfusion and vessel-calibre MRI parameters, the macrovessel (GE) and microvessel (SE) blood-volume relaxation rate images shown here are computed according to the relative decrease in MRI signal intensity over the course of a sequence of GE MRI and SE MRI images, respectively (images 30–34 in this case), thereby adjusting for baseline intensity values and image sampling time (echo time). The so-called ‘first-pass effect’ denotes the initial and transient passage of a relatively tight bolus of the contrast agent following intravenous administration of the agent and is observed as a peak in the relative relaxation rate signal (peaks for normal tissue approximately at image 32). The contrast agent first-pass effect is particularly dominant in large vessels in the macrovessel images because of high contrast agent concentrations combined with high sensitivity to the magnetic susceptibility effect induced by the agent. After the initial first-pass contrast agent passage, repeated passages of what remains of the circulating and increasingly dispersed contrast agent bolus can be appreciated as smaller and dissipating signal oscillations. c | Macrovessel and microvessel volume fractions are estimated from the area under the GE and SE first-pass curves (top left), respectively. The concept of vessel-calibre MRI stems from the relationship between increased macrovessel volume fractions for increasing vessel calibres, and microvessel volume fractions, which are selectively sensitive to small (radius <10 µm) vessel calibres (bottom left). Mean vessel density maps (Box 1) are thus derived from the quotient of macrovessel and microvessel volume fractions, whereas the vessel-calibre index also accounts for water diffusion and the absolute blood volume fraction. Abbreviations: CE, contrast-enhanced; DSC-MRI, dynamic susceptibility-contrast MRI; FLAIR, fluid-attenuated inversion recovery; GE, gradient-echo; SE, spin-echo.

Formalized imaging-based response criteria are usually based on the Response Evaluation Criteria in Solid Tumors (RECIST) guidelines⁴²—a National Cancer Institute (NCI)-funded initiative—that are applicable to data derived from MRI, CT and most other modalities used for noninvasive in vivo imaging. The aim of the RECIST criteria is to provide simple and objective measures of therapeutic response based on assessment of the post-treatment change (or lack of change) in the longest in-plane diameters of target lesions located throughout the body (Supplementary Table 2). A slightly different approach is used for the evaluation of tumours in the CNS, termed the MacDonald criteria,⁴³ which focus on indicators of response related to the extent (area) of contrast enhancement on imaging, corticosteroid dose, and clinical status. In 2010, the MacDonald criteria were updated by the Response Assessment in Neuro-Oncology (RANO) working group¹¹ to include changes in the non-enhancing lesion on FLAIR MRI; however, the definition of increased FLAIR hyperintensity is subjective and was not clearly defined in these guidelines (Supplementary Table 2), leading to variability in application of the RANO criteria.

Limitations of the current approach

The RECIST and RANO working groups acknowledged the inherent limitations associated with imaging-based response criteria focusing on tumour size alone;^42,44–46 nevertheless, these criteria, which predominantly rely on radiographic measures of the response of tumours to therapy, are frequently criticized.^47–49 The controversy surrounding these guidelines includes debate over the definition of true radiographic response and whether response criteria should be based solely on anatomical tumour load or in conjunction with metrics that better reflect vascular or metabolic changes in the tumour.^47,50,51 In particular, the dynamic nature of the perfusion of vascularized lesions cannot be characterized accurately using traditional measurements of contrast-agent-induced signal enhancement alone,⁹ whereas potential monitoring of the EPR effect will require quantitative measures of the permeability rate-constant.³⁵ Moreover, not all cancer types are expected to follow similar definitions of response; for example, obtaining repeated measurements in adherence with the RECIST criteria is difficult for localized or locally recurrent cancers of the prostate because of the location and small size of such tumours, whereas increased levels of prostate-specific antigen (PSA) are closely correlated with imminent radiographic progression and are, therefore, used widely as the preferred indicator of tumour status.⁵² Whether the RECIST criteria reflect a meaningful end point of response to many specific antiangiogenic therapies also remains unclear. For example, although sunitinib treatment in hepatocellular carcinoma seems to cause tumour necrosis or cysts, no subsequent tumour shrinkage was observed in one study.⁵³

Furthermore, manually defined response criteria that are dependent solely on tumour size are prone to reader bias attributable to inconsistencies in the interpretation of irregular tumour edges, regional variations in contrast enhancement within surgical cavities and permeability changes resulting from radiotherapy and antiangiogenic therapy, as well as the lack of consensus in the type and dose of contrast agent used.^12,54,55 In addition, variations in positioning of the patient’s head in the scanner between examinations can have a substantial influence on the reader’s assessment of the data. Indeed, repeated assessment of therapeutic response using the RANO criteria was shown to return an intra-reader error rate of 22%, which might have resulted in patients without true tumour progression being incorrectly identified as having progressive disease, and could potentially have led to discontinuation of treatment or their removal from a clinical trial.⁵⁶

A transient, therapy-induced increase in radiographic oedema and contrast enhancement following radiotherapy, termed ‘pseudoprogression’, has been the cause of much research and debate, especially in relation to brain tumours.⁵⁷ Radiotherapy has a profound effect on tumour blood-vessel morphology, inducing vasodilation, increased vascular permeability and oedema,^57,58 that can resemble radiographic characteristics of tumour progression. However, unlike true disease progression, pseudoprogression is typically associated with reversion of MRI indications of tumour progression after a few weeks or months without any change in therapy. Of note, up to 30% of patients with intra-axial brain tumours show signs of pseudoprogression,^11,57 and distinguishing these treatment-related changes from active tumour growth is critically important for patient management and the administration of clinical trials. By contrast, blockade of VEGF using antiangiogenic therapies can result in reduced tumour-vessel permeability and vasogenic oedema that enhances radiographic-response rates, but, unfortunately, this improvement does not always translate into increased survival.^13,45 Thus, an urgent need exists, particularly in the setting of antivascular and antiangiogenic clinical trials, for validated methods and standards for assessment of tumour vascular structure and function beyond the current conventional MRI modalities.^59–62

Perfusion MRI

Perfusion MRI can help address the issues relating to tumour vascularity and conflicting treatment effects observed with conventional MRI. A large number of early phase clinical trials have reported values of blood flow, blood volume, and permeability determined using perfusion MRI (recently reviewed in this journal⁹ and elsewhere^63,64). However, these data are beyond the scope of this Review, and we provide only a brief overview of relevance of this technique to vessel-calibre MRI.

Perfusion imaging techniques are based on the central volume principle,⁶⁵ which states that in a closed flow system with a single inflow orifice and a single outflow orifice, the blood volume of a given tissue is equal to the constant rate of blood flow (volume of blood in tissue per unit time) multiplied by the mean transit time of the blood tracer (that is, the MRI contrast agent) passing through the capillary structure of the tissue. Two main contrast agent-based perfusion MRI techniques are available, depending on whether the acquisition is made using T₁-sensitive or T₂-sensitive imaging protocols.⁶⁶ In T₁-weighted perfusion MRI, the signal intensity increases dynamically as a function of the local concentration of contrast agent; this technique is primarily used for estimating vessel permeability and is generally referred to as dynamic contrast-enhanced (DCE)-MRI. Alternatively, a T₂-weighted perfusion MRI sequence is used for accurate measures of blood-flow rate, blood volume and vessel calibre following a rapid, intravenous injection of contrast agent. More specifically, rapid echo-planar imaging techniques, which enable imaging of the entire volume of interest within a few seconds, are typically employed. Using this approach, a decrease in the dynamic MRI signal intensity is observed following injection of contrast agent, as the echo-planar imaging technique is particularly sensitive to the magnetic susceptibility effect (Figure 2b). Consequently, this imaging technique is referred to as dynamic susceptibility contrast (DSC)-MRI (Supplementary Methods).

Vessel-calibre MRI

Basic concepts

In 1991, Fisel, Rosen and colleagues^67–69 at Massachusetts General Hospital, Boston, USA, pioneered the theoretical foundation for the vessel-calibre MRI technique. In a series of studies,^67–69 these researchers used Monte Carlo simulations to provide evidence that T₂-weighted perfusion MRI images were sensitive to the vessel-size-related scale of the magnetic susceptibility effect created by blood and contrast agents flowing through vessels. This initial work was performed in conjunction with substantial efforts to understand MRI signal alterations resulting from changes in oxyhaemoglobin concentrations in blood.^70,71 These initiatives were followed by comprehensive models for quantification of magnetic susceptibility effects in vivo.^72–76 The first image representative of the use of the vessel-calibre technique in cancer, specifically a rat C6 glioma xenograft model, was reported in 1998 by Dennie and collegues.⁷⁷ 19 days after cell implantation, the gliomas showed evidence of a 90% increase in the average vessel diameter compared with adjacent normal grey matter. The first clinical experience of this technique was reported in 2000 by Donahue and colleagues, who published vessel-calibre images of 15 patients with intracranial tumours; a positive correlation between average vessel calibres and tumour grade was also demonstrated.⁷⁸

Although vessel-calibre MRI was conceptualized >20 years ago and can provide a direct link between tumour vascular status and therapeutic response,⁵ the technique has received limited attention in the literature on perfusion MRI.^{9,63,66,79–81} One important reason for this apparent deficit is the complex image acquisition process used in vessel-calibre MRI, which has hindered the technique from becoming readily accessible to the general clinical community.⁵⁹ The methodology necessitates the acquisition of images using both gradient-echo and spin-echo protocols (Figure 2b). This approach enables relative (unitless) measures of average vessel diameters and average vessel densities for arteries, capillaries, and veins combined to be estimated in each image pixel using the quotient of gradient-echo blood volumes to spin-echo blood volumes (Box 1).⁸² Efforts have also been made to quantify the average vessel calibre in an image voxel (in µm), which is often referred to as the vessel size index (VSI), using a weighted average of vessel calibres and accounting for water diffusion and the absolute blood volume fraction.^82–85 These measurements are possible because of the high sensitivity of the gradient-echo sequence to the magnetic susceptibility effect in both small (radius <10 µm) and large (radius >10 µm) vessel calibres, whereas the spin-echo sequence has an extra imaging component that minimizes the magnetic susceptibility effect for large-calibre vessels and is, therefore, selectively sensitive to detection of small calibre vessels (Supplementary Methods).^73,74 Henceforth, for simplicity, the gradient-echo images will be referred to as ‘macrovessel’ (also referred to as total blood volume^86,87) images and the spin-echo images as ‘microvessel’ images.

Box 1. Parameters used in vessel-calibre MRI.

Mean vessel diameter (mVD)

• mVD = ΔR₂* /ΔR₂ (≈ total blood volume MRI/microvessel blood volume MRI)

Mean vessel density (Q)^‡

• Q = ΔR₂/(ΔR₂*)^2/3

  For standard clinical contrast agent dosages, Q is independent of the contrast agent concentration.

Vessel size index (VSI)^‡

• VSI = [k(V_fD)^1/2] × ([ΔR₂*]/(ΔR₂)^3/2)

^‡Calculations of Q and VSI using a linear dependence on the quotient of macrovessel and microvessel volume fractions (ΔR₂*/ΔR₂) show similar results to the preferred calculations using a proposed ³/₂ power dependence (ΔR₂*/ΔR₂^3/2) attributed to the susceptibility effect (see Supplementary Methods online).^75,83,91 Abbreviations: ΔR₂, change over time in spin-echo relaxation rate; ΔR₂*, change in gradient-echo relaxation rate; D, diffusion coefficient of water in tissue; k, proportionality constant; V_f, blood-volume fraction.

Currently, three distinctly different analytical approaches to vessel-calibre MRI have been described, each having dedicated imaging protocols, caveats and postprocessing routines (Table 1). Firstly, a common approach is the direct assessment of the point-by-point difference in the macrovessel and microvessel DSC-MRI signal-intensity curves following injection of a low-molecular-weight (hydrodynamic diameter of <10 nm) gadolinium-based contrast agent (Figure 2c).^{73–75,77,88,89} Secondly, vessel calibres can be estimated from macrovessel and microvessel imaging readouts before and after injection of super-paramagnetic iron-oxide (SPIO) nanoparticles, which are particularly useful in evaluating permeable cancer blood vessels outside the CNS.^77,85,90,91 Because most ultrasmall superparamagnetic iron-oxide molecules have diameters of approximately 30 nm,⁹² SPIO-based agents are restricted to the intravascular space and might, therefore, more accurately reflect the average vessel calibre compared with contrast agents that leak out from the intravascular to the extravascular space (such as gadolinium-based agents). The relative long half-life of SPIO-based agents (up to approximately 24 hours)⁹² also increases the image signal-to-noise ratio.⁹³ Finally, vessel-calibre MRI in the brain can also be acquired without the use of contrast agents, simply by stimulating the blood-oxygen-level-dependent (BOLD) effect in both macrovessel and microvessel images (so-called BOLD-VSI; Supplementary Methods).^70,94,95 Compared with baseline signal intensities, image contrast is attributable to a change in the level of deoxygenated blood from brain activation stimuli or by normoxic or carbogen gas challenges. As this approach relies on the assessment of deoxygenated blood volume fractions, the method can provide a measure of apparent tissue oxygenation levels,⁹⁶ but is inherently limited to measures of venous vessel calibres.⁹⁵

Table 1.

Vessel-calibre MRI methodology

Imaging protocols	Contrast agent*	Target organs*	Computations	Imaging considerations‡
DSC GE/SE MRI	Gadolinium, SPIO	CNS, non-CNS	ΔR2(t) = (1/TE)ln(SIt/SIt0) ΔR2* = (1/TE*)ln(SIt*/SIt0*)	High temporal resolution Low spatial resolution Nonlinear MRI signal in bulk blood Gadolinium: extravasation correction required VSI overestimation compared with histological assessment
SSCE GE/SE MRI	SPIO, gadolinium, MION	Non-CNS, CNS	ΔR2 = (1/TE)ln(SIpre/SIpost) ΔR2* = (1/TE*)ln(SIpre*/SIpost*) or ΔR2* = R2*post − R2*pre	High spatial resolution Nonlinear MRI signal in bulk blood SPIO: high SNR owing to long half-life SPIO: not approved for clinical use VSI overestimation compared with histological assessments Contrast must remain intravascular or stable
BOLD MRI	Challenge with normoxic/carbogen gas or brain-activation stimuli	CNS	ΔR2 = (1/TE)ln(SIpre/SIpost) ΔR2* = (1/TE*)ln(SIpre*/SIpost*) or ΔR2* = R2*post − R2*pre	High temporal resolution Low geometric and flow-related distortions Low SNR Enables measurement of venous vessel calibres only Overestimation of mean vessel density in capillaries Underestimation of mean vessel density in bulk blood

^*Listed in order of preference.

^‡If clinical MRI field strengths (1.5T or preferably 3T) are used.

Abbreviations: ΔR₂, change in spin-echo relaxation rate; ΔR₂*, change in gradient-echo relaxation rate; BOLD, blood-oxygen-level-dependent; CNS, central nervous system; DSC, dynamic susceptibility contrast; GE, gradient-echo; MION, monocrystalline iron oxide nanocompounds; R₂*_post, gradient-echo relaxation rate image readout after contrast-injection; R₂*_pre, gradient-echo relaxation rate image readout before contrast-injection; SE, spin-echo; SI_post, MRI signal intensity after contrast-injection; SI_pre, MRI signal intensity before contrast-injection; SI_t, signal intensity over time; SI_t0, signal intensity at baseline; SNR, signal-to-noise ratio; SPIO, super-paramagnetic iron oxide particles; SSCE, steady-state contrast-enhanced imaging prior to and after contrast injection; TE, image sampling time (echo time); VSI, weighted mean of vessel size index.

Because vessel calibres can be estimated from macrovessel and microvessel MRI before and after injection of a tracer, or by BOLD-VSI, the vessel-calibre technique can be readily implemented in the clinic using commonly available sequences.^83,97 Alternatively, for a combined simultaneous macrovessel–microvessel DSC-MRI readout,⁸⁹ most if not all major manufacturers of MRI apparatus provide dedicated image protocols upon request through their research and development programmes.

Assumptions underlying the calibre estimates

Vessel-calibre MRI is performed according to certain assumptions that in most clinical scenarios hold true. However, a few considerations specific to estimating vessel calibre in tumours need to be highlighted (Table 1).⁸⁵

First, as noted by Troprès and colleagues in 2001,⁸³ reliable average VSI estimations (in µm) are difficult to obtain and must be interpreted with care, appreciating the complex relationships between macrovessel and microvessel blood volume and diffusion. A correct proportionality constant that accurately scales the MRI-based vessel-calibre estimate with the true vessel calibre and tissue blood volume is not currently available.^84,98 Also, owing to the additional influence of the relaxation rate of deoxygenated blood on the T₂ relaxation rate and signal intensities, the relationship between tissue oxygenation and the absolute vessel-calibre estimate is not straightforward in pathological conditions with impaired oxygen saturation levels, and warrants further investigations in the context of cancer therapy.^73,99 The accuracy of the VSI estimate should, however, improve with the use of higher magnetic-field strengths (>3 Tesla [3T]) and of dedicated ultrafast MRI measures of the arterial inflow,¹⁰⁰ and as better approximations of the relationship between the MRI signal and true blood volume values are established.^83,94 Thus, the current clinical hardware and imaging protocols limit the traditional VSI estimate to vessel radii considered larger than ‘very small’ (>2 µm) but smaller than ‘very large’ (<50 µm).^94,101,102 The range of vessel calibres that can be acceptably defined using BOLD-VSI is probably even narrower because of the inherent small changes in magnetic susceptibility during activation of the target tissue.⁹⁴

Second, SPIO-based agents^90,103 are currently restricted to off-label use and should be applied with care because they have been associated with adverse effects, including allergic reactions, back pain and, in some extreme cases, cardiovascular events.^104,105 Although good agreement (r² ~0.8–0.9) has been observed between gadolinium-based VSI and SPIO-based VSI estimates, similarly high agreements might not be seen in direct comparisons of the macrovessel and microvessel parameter values obtained using the two agents owing to differences in the image acquisition protocols (Table 1).¹⁰⁶

Third, unlike intravascular SPIO agents, gadolinium leaks out of the blood vessels into the interstitium, necessitating permeability compensation. Such compensation can be achieved using time-accelerated imaging techniques combined with multiple macrovessel image readouts,^85,107 preloading of a small gadolinium dose^106,108 or by rigorous post-processing corrections of the DSC-MRI signal curves (Supplementary Methods).^109–111

Finally, whereas contrast-agent-based MRI vessel-calibre estimates typically provide values for static mean vessel calibres, Packard and colleagues used partial pressure of carbon dioxide (PaCO₂) challenges to show that tumours have decreased vascular reactivity, and that the native cerebral vasculature might also be dilated and constricted in response to transient or local changes in the PaCO₂.⁸⁶ BOLD-VSI analyses based on gas challenges is also influenced by vasodilation.¹⁰¹ Increased cerebral vasodilation and flow is observed for hypercapnia (abnormally elevated CO₂ concentrations in the blood), whereas the opposite effect is seen in hypocapnia (low blood CO₂ levels);¹¹² however, the effects of gas-induced vasodilation on blood volume and, therefore, BOLD-VSI are not fully understood,¹⁰¹ although venous volume fractions seem to be relatively unaffected by vasodilation.¹¹³

Validation of vessel-calibre MRI in cancer

Vessel-calibre MRI has been extensively validated against histology as well as other imaging modalities, such as ex vivo micro-CT and in vivo two-photon microscopy, to determine the value of MRI-based vessel-calibre measurements as a sensitive biomarker for cancer imaging (Supplementary Table 3). Absolute average MRI-determined vessel calibres tend to be slightly overestimated compared with histological vessel-calibre measurements, probably because of the incomplete understanding of the relationship between the blood MRI signal and the underlying physiology. Alternatively, the histological data might have been underestimated as a result of vessel collapse during cryosectioning (Supplementary Methods).^83,114 However, this differential between MRI and histological measurements seems to be relatively constant, with a subsequent strong correlation between the ratios of vessel calibres in tumour and reference tissue derived using each method (Table 2; Spearman correlation across all studies = 0.84; P <0.001). Of note, comparison of in vivo imaging and ex vivo histological findings might also be compounded by geometrical differences and histological tissue distortions;⁶⁶ thus, not all studies performed to date have had the required sample size or objectives to statistically quantify comparisons of vessel-calibre estimates.

Table 2.

Vessel-calibre MRI preclinical validation studies

Study	Tumour model	Model	Treatment administered	Validated method used for comparison	MRI-derived NmVD ± SD (number of samples)	NmVD ± SD derived using validated method (number of samples)
Dennie et al. (1998)77	C6 glioma	Rat	None	Histology	1.92 ± 0.24 (4)	1.89 ± 1.19 (3)
Badruddoja et al. (2003)177	9L gliosarcoma	Rat	Dexamethasone or vehicle	Histology	Dexamethasone: 1.91 ± 1.26 (17) Vehicle: 5.37 ± 3.56 (10)	Dexamethasone: 1.88 ± 0.28 (13) Vehicle: 2.97 ± 0.32 (7)
Packard et al. (2003)86	U87 glioma*	Rat	None	Histology	1.52 ± 0.63 (14)	1.63 ± 0.14 (6)
Tropres et al. (2004)178	C6 glioma	Rat	None	Histology	2.70 ± 0.65 (7)	1.87 ± 1.54 (7)
Valable et al. (2008)116	C6 glioma RG2 glioma	Rat	None	Histology	C6 glioma: 2.18 ± 0.50 (12) RG2 glioma: 1.79 ± 0.42 (12)	C6 glioma: 2.68 ± 0.61 (12) RG2 glioma 1.90 ± 0.39 (12)
Beaumont et al. (2009)179	C6 glioma RG2 glioma	Rat	None	Histology	C6 glioma: 2.82 ± 0.34 (14) RG2 glioma: 1.71 ± 0.17 (6)	C6 glioma: 1.59 ± 0.20 (6) RG2 glioma: 1.71 ± 0.15 (6)
Farrar et al. (2010)91	U87 glioma	Mouse	None	Histology Intravital microscopy	MRI (histology samples): 1.64 ± 0.10 (4) MRI (intravital microscopy samples): 1.39 ± 0.17 (7‡)	Histology: 1.67 ± 0.67 (4) Intravital microscopy: 1.37 ± 0.08 (5‡)
Douma et al. (2010)180	LS174T colorectal cancer§	Mouse	None	Ex vivo 3D laser scanning microscopy	2.88‖ (8‡)	1.95‖ (11‡)
Lemasson et al. (2011)102	U87 glioma	Rat	None, carmustine (BCNU), or sorafenib	Histology	No treatment: 1.20 ± 0.19 (4) BCNU: 1.20 ± 0.19 (4) Sorafenib: 1.55 ± 0.23 (4)	No treatment: 1.24 ± 0.15 (4) BCNU: 1.29 ± 0.16 (4) Sorafenib: 1.38 ± 0.17 (4)
Lemasson et al. (2013)181	C6 glioma RG2 glioma	Rat	None	Histology	C6 glioma: 1.62 ± 0.46 (15) RG2 glioma: 1.44 ± 0.48 (12)	C6 glioma: 1.69 ± 0.23 (15) RG2 glioma: 1.30 ± 0.30 (12)

^*Average of baseline hypercarbia and hypocarbia groups.

^‡Separate size-matched tumour samples used for MRI and validation assessments.

^§Tumour core was analysed.

^‖SD not reported.

Abbreviations: BCNU, bischloroethylnitrosourea; NmVD, normalized mean vessel diameters (tumour-to-reference tissue ratio); SD, standard deviation.

Several studies have shown excellent agreement between MRI and histological relative vessel calibres in C6, RG2 and U87 orthotopic glioma xenograft models.^{77,91,115,116} Farrar and colleagues⁹¹ also compared vessel-calibre MRI with in vivo two-photon microscopy measurements in separate tumour-size-matched groups of mice with U87 orthotopic gliomas, showing excellent agreement between the two techniques. Several studies of human breast cancer xenograft models have also found close correlations between MRI-determined vessel calibres and vessel calibres measured by ex vivo micro-CT,¹¹⁷ immunohistology¹¹⁸ and in vivo two-photon microscopy.¹¹⁸ The validity of vessel-calibre MRI in mice bearing xenografts of poorly vascularized melanomas and highly vascular hemangioendotheliomas have been confirmed using both immunohistology and two-photon microscopy.¹¹⁸ In addition, Ungersma et al.¹¹⁹ used a HM7 colorectal tumour model to demonstrate strong correlations between vessel calibres and between vessel densities determined using MRI, histology and micro-CT. Models of bone metastasis from breast cancer cell lines also show the same trend in MRI-derived and histological vessel calibres during antiangiogenic therapy.¹²⁰

Collectively, these studies provide corroborating evidence that MRI-determined vessel calibres can serve as a surrogate marker of the microvasculature of cancers. Although a quantitative correction that enables accurate determination of physiologically correct vessel calibres by MRI is currently unavailable, reasonably accurate vessel-calibre ratios between tumour and normal tissue can be expected using this approach. Nevertheless, for vessel-calibre MRI and perfusion MRI to develop beyond expensive boutique methods, the true role and added value of these imaging techniques in clinical trials will remain unanswered until proper validation has been achieved in large, multicentre studies.^9,64,121

Vessel-calibre MRI in human cancers

In keeping with the substantial increase in preclinical studies on vessel-calibre MRI over the past decade, the number of clinical studies is also increasing, providing data on vessel-calibre MRI in cancer, stroke and hypercapnia (Supplementary Figure 1, Supplementary Table 4). To date, vessel-calibre MRI studies in human cancers have focused on optimization of imaging and acquiring vessel-calibre measurements comparable to those in the literature (Table 3).

Table 3.

Clinical cancer studies including vessel-calibre MRI

Study	Cancer type	Number of patients (women/men)	Treatment administered	Findings
Donahue et al. (2000)78	Glioma* (n = 13) Lymphoma (n = 2)	15 adults (6/9)	NA	Vessel calibre was larger in the tumour than in healthy tissue, and increased with malignancy grade
Lamalle et al. (2003)125	Glioma*	9 patients‡	NA	Vessel calibre was larger in the tumour and lower in necrotic pars than in healthy tissue
Kiselev et al. (2003)126	Glioma* (n = 13) Meningioma (n = 4) Metastasis (n = 3)	20 patients‡	NA	Vessel calibre was larger in the tumour than in healthy tissue, and increased with malignancy grade
Schmainda et al. (2004)124	Glioma* (n = 72) Neurocytoma (n = 1)	73 adults (27/46)	NA	Vessel calibre was larger in the tumour than in healthy tissue, and increased with malignancy grade
Pectasides et al. (2004) 129	Glioma*	9 patients‡	NA	Vessel calibre was more heterogeneous in the tumour than in adjacent tissue
Kiselev et al. (2005)84	Glioma* (n = 2) Meningioma (n = 2)	4 patients‡	NA	Vessel calibre was larger in the tumour than in healthy tissue, and increased with malignancy grade
Batchelor et al. (2007) 130	Recurrent glioblastoma	16 adults158 (9/7)	Cediranib (45 mg/kg per day)	Vessel calibre was markedly reduced at day 1 after treatment
Breyer et al. (2007)127	Glioma*	31 patients‡	NA	Vessel calibre was larger in the tumour than in healthy tissue, and increased with malignancy grade
Hsu et al. (2009)97	Glioma* (n = 5) Meningioma (n = 3)	8 adults (4/4)	NA	Vessel calibre was larger in the tumour than in healthy tissue, and increased with malignancy grade
Sorensen et al. (2009) 131	Recurrent glioblastoma	30 adults158 (13/17)	Cediranib (45 mg/kg per day)	Vessel calibre was markedly reduced at day 1 after treatment
Lüdemann et al. (2009)134	Prostate cancer	13 adults‡	NA	Vessel calibre was larger in the tumour than in healthy tissue
Remmele et al. (2011)85	Pleomorphic sarcoma	1 adult‡	NA	Repeated monitoring by vessel-calibre MRI shown to be feasible
Schmiedeskamp et al. (2012)107	Glioblastoma	1 patient‡	Radiotherapy (dose not reported)	Increased vessel calibres and vascular supply attributed to progressive disease, not pseudoprogression
Zhang et al. (2012)182	Glioma* (n = 14) Meningioma (n = 14)	28 adults (17/11)	NA	Vessel calibre was larger in the tumour than in healthy tissue, and increased with malignancy grade
Fredrickson et al. (2012)103	Colorectal metastasis	3 patients‡	NA	Mean vessel density estimation with SPIO agents was feasible and well tolerated
Xu et al. (2013)128	Glioma* (n = 2) Meningioma (n = 2)	4 patients‡	NA	Vessel calibre was larger in the tumour than in healthy tissue
Emblem et al. (2013)99	Recurrent glioblastoma	30 adults158 (13/17)	Cediranib (45 mg/kg per day)	Vessel calibre was markedly reduced at 1, 28 and 58 days after treatment
Batchelor et al. (2013) 132	Newly diagnosed gliobastoma	40 adults159 (13/27) 14 adults161 (8/6)	Cediranib (30 mg/kg per day) and temozolomide (75 mg/m2) and radiotherapy (2 Gy per day)	Vessel calibre was markedly reduced at days 1, 8, 15, 22, 29, 36 and 43 after treatment159 Vessel calibre decreased in 1/14 patients treated with only temozolomide and radiotherapy161

^*Several glioma subtypes included in the study.

^‡Information on patient age and/or gender not reported.

Abbreviations: NA, not administered; SPIO, super-paramagnetic iron oxide particles.

Clinical studies in CNS cancers

Unsurprisingly, considering the well-developed kinetic models for vascular tissue of the CNS,^122,123 the first human vessel-calibre MRI data in cancer were of intracranial tumours. The first images were published in 2000 by Donahue and co-workers.⁷⁸ These images were succeeded in 2004 by data from a corroborating study that included 72 patients with gliomas,¹²⁴ which demonstrated larger relative vessel calibres and a wider range of calibres with increasing tumour grade.¹²⁴ These findings have been confirmed in small glioma studies by Lamalle et al.¹²⁵ and by Hsu et al.⁹⁷ (n = 9 and 4, respectively), with larger vessel calibres and greater heterogeneity in vessel diameters in glioblastomas compared with lower grade gliomas. In parallel, Kiselev, Breyer, Xu and their colleagues^84,126–128 also presented supporting data from patients with brain lesions across several publications between 2003 to 2013; higher average MRI vessel calibres were reported in 20 intracerebral tumours (84 ± 73 µm) compared with adjacent white matter (32 ± 10 µm) and grey matter (34 ± 8 µm) tissue,¹²⁶ and a wider range of vessel calibres were observed in glioblastomas compared with lower grade gliomas.⁸⁴ These findings have also been confirmed in 31 patients at a higher clinical magnetic-field strength—3T, as opposed to 1.5T.¹²⁷

At Massachusetts General Hospital, prior to administration of targeted therapies, abnormal and dilated vessel calibres were observed using MRI in a range of glioma grades,¹²⁹ recurrent glioblastomas^99,130,131 and newly diagnosed glioblastomas.¹³² An almost threefold increase in average tumour-vessel calibres compared with whole-brain normal-appearing tissue was detected in 30 patients with recurrent glioblastomas (21 ± 6 µm versus 8 ± 1 µm),⁹⁹ and in two separate cohorts comprising 40 and 14 patients with newly diagnosed, untreated glioblastomas (24 ± 15 µm versus 8 ± 2 µm, and 33 ± 20 µm versus 11 ± 5 µm, respectively),¹³² confirming the findings of Kiselev and colleagues.¹²⁶ Owing to refinements to the acquisition methods used in these studies,^99,132 including the use of a pre-dose injection of contrast agent to saturate the tissue and arterial input function normalization,^83,126 the absolute vessel-calibre values were lower compared with those reported by Kiselev et al.¹²⁶ Although this inconsistency emphasizes the need for caution when considering absolute vessel-calibre values (Table 1), data from the Massachusetts General Hospital show low intrapatient variations in MRI-derived tumour-vessel calibres.¹³³ The vessel calibres in normal-appearing tissue are also in good agreement with the findings of a study involving healthy volunteers and the use of an intravascular SPIO contrast agent (12–13 ± 2–3 µm).⁹³

Clinical studies in non-CNS cancers

Lüdemann and colleagues¹³⁴ used MRI with a gadolinium-based contrast agent to compare vessel calibres in untreated prostate cancer tissue from 13 patients to peripheral normal-appearing tissue from the same patients, and observed a trend towards larger vessel diameters in tumour tissues; the prostate vasculature had heterogeneous signatures across patients, with a wide range of vessel diameters in tumours (112 ± 158 µm) and normal-appearing tissues (105 ± 205 µm).¹³⁴ The reproducibility of vessel-calibre MRI has also been evaluated in a pleomorphic sarcoma in the left pubic bone; stability over a 6-min acquisition window (4–5 µm) was high and, more importantly, vessel-calibre estimates were independent of dilution of the contrast agent.⁸⁵ Fredrickson et al.¹⁰³ successfully evaluated the feasibility of SPIO-based mean vessel density estimations using vessel-calibre MRI protocols (Box 1) in patients with colorectal liver metastases. Overall, the tracer kinetics of intravascular SPIO contrast agents and the subsequent uptake in the reticuloendothelial system of the liver and spleen over time does not seem to affect the vessel-calibre estimates in healthy and tumour tissues outside the CNS during the typical image acquisition period and at clinical magnetic-field strengths (≥1.5T).^{85,106,135,136}

Vessel-calibre MRI in clinical trials

Lessons learnt from preclinical studies

Changes in vessel calibres determined using MRI compared with control or pretreatment values have been evaluated for a double-digit number of therapies in small rodent cancer models (Table 4). Although small sample numbers and the imaging considerations of the technique (Table 1) are potential limitations of these studies, their findings indicate that targeted therapies can affect the MRI-based vessel-calibre estimate in four ways.

Table 4.

Preclinical antivascular and antiangiogenic studies that used vessel-calibre MRI

Treatment approach	Cancer model (number of animals)	Change in microvessel calibres	Change in macrovessel calibres	Time of response after therapy
Microtubule stabilization
Patupilone; single 0.8 mg/kg dose	BN472 mammary carcinoma in rats (51)137	↓ (P <0.01)	↓ (P <0.01)	Day 6
RTK inhibition
SU11657 (FLT3 inhibitor); 20 mg/kg or 40 mg/kg per day	9L gliosarcoma in rats (18)87	20 mg/kg: NC 40 mg/kg: ↓ (P <0.05)	20 mg/kg: ↓ (P <0.05) 40 mg/kg: ↓↓ (P <0.05)	Day 1–4 in both groups
Sunitinib (multitarget RTK inhibitor); 60 mg/kg per day	MDA-MB-435 melanoma in mice (10)118	↓ (P <0.05)	↓ (P <0.05)	Day 7
Sunitinib; 60 mg/kg per day	A431 SCC in mice (14)144	↓ (P <0.01)*	NC	Day 4
Sunitinib; 40 mg/kg every 3 days	HT1080 human breast fibrosarcoma in mice (NR)152	NC	NC	Days 4 and 7
Sunitinib (20 mg/kg per day) ± zoledronic acid (bisphosphonate; 40 mg/kg per week), or zoledronic acid alone	Bone metastases from MDA-MB-231 breast cancer cells in rats (34)120	Sunitinib alone, and combination therapy: ↓ (P <0.01)* Zoledronic acid alone: NC	Sunitinib alone, and combination therapy: NC Zoledronic acid alone: NC	Days 5 and 25 in all groups
Sorafenib (multitarget RTK inhibitor) at 7 mg/kg per day ± paclitaxel (mitotic inhibitor; 5 mg/kg per week) or paclitaxel alone	Bone metastases from MDA-MB-231 breast cancer cells in rats (43)146	Sorafenib: ↓↓ (P <0.01)* Paclitaxel: ↓ (P <0.01) Combination therapy: ↓↓ (P <0.01)*	Sorafenib: ↓ (P <0.01) Paclitaxel: ↓ (P <0.05) Combination therapy: ↓ (P <0.01)	Day 25 in all groups
Sorafenib 100 mg/kg per day or carmustine (BCNU; alkylating agent; 10 mg/kg every 2 weeks)	U87-MG glioblastoma in rats (60)102	Sorafenib: ↓ (P <0.01)* Carmustine: NC	Sorafenib: NC Carmustine: NC	Sorafenib: day 5 Carmustine: days 1–14
MMP inhibition
Prinomastat (AG-3340); 150 mg/kg twice daily	HaCaT-RAS-A-5RT3 human skin carcinoma in mice (14)149	↓ (P <0.05)*	NC	Day 6
VEGF inhibition
Bevacizumab; 10 mg/kg per week	Human glioblastoma spheroids in rats (8)183	↓ (P <0.05)*	NC	Day 21
Bevacizumab; 25 mg/kg every 2 days	U87-MG glioblastoma in mice (6)139	NC	↓ (P <0.01)	Days 14 and 22
Bevacizumab; 1 mg every 3 days	HaCaT-RAS-A-5RT SCC in mice (16)144	NC	NC	Day 6
Bevacizumab (5 mg/kg every 4 days) ± irinotecan (topoisomerase 1 inhibitor; 40 mg/kg per week)	U87-MG glioblastoma in rats (16)145	Bevacizumab: ↓ (P <0.05)* Combination therapy: ↓ (P <0.05)*	Bevacizumab: NC Combination therapy: NC	Day 16 in both groups
G6-31; single 5 mg/kg dose	HM7 CRC in mice (9)119	NC	↓ (P <0.01)	Day 2
Neuropilin-1 (NRP1); single 80 mg/kg dose	HM7 CRC in mice (10)119	NC	↓ (P <0.05)	Day 2
VEGFR inhibition/antagonism
Cediranib (AZD2171; 3–6 mg/kg per day)	U87-GFP glioblastoma in mice (24)148	↓↓ (P <0.05)*	↓ (P <0.01)	Days 2–3
Vatalanib (PTK787); 75 mg/kg per day	H1975 NSCLC in mice (12)154	NC	NC	Days 6 and 13
Tandutinib (MLN518; VEGFR inhibitor) at 20 mg/kg twice daily	C6 glioma in mice (30)156	NC	NC	Day 10
DC101 (anti-VEGFR2 antibody); 800 μg per day	HaCaT-RAS-A-5RT3 human skin carcinoma in mice (7)147	↓ (P <0.01)*	NC	Day 7
Vascular disruption
Vadimezan (DMXAA, ASA404); 350 mg/kg per day	GH3 prolactinoma in rats (12)138	↓ (P <0.01)	↓ (P <0.01)	Day 1
Vadimezan (DMXAA, ASA404); 350 mg/kg per day	Orthotopic glioma in rats (6)153	NC	NC	Day 1
Zybrestat (CA4P); single 50 mg/kg dose	C3H mammary carcinomas in mice (43)140	↓ (P <0.01)	↓ (P <0.01)	3 h
ZD6126; 200 mg/kg per day	PC3LN3 prostate carcinoma in mice (6)143	↓ (P <0.05)	↓ (P <0.05)	Day 1
ZD6126; 30 mg/kg or 200 mg/kg per day	SW1222 CRC in mice (15)62	30 mg/kg: NC 200 mg/kg: ↓ (P <0.05)	30 mg/kg: NC 200 mg/kg: ↓ (P <0.05)	Day 1 for both groups
Src kinase inhibition
Saracatinib (AZD0530); 25 mg/kg per day	PC3LN3 prostate carcinoma in mice (14)155	NC	NC	Day 5
PI3K/mTOR inhibition
GDC-0980; 10 mg/kg per day	HM7 CRC in mice (18)184	↓ (P <0.01)*	NC	Day 1

^*An apparent increase in mVD and VSI was observed because of the larger decrease in microvessel calibres compared with macrovessel calibres.

↓, Significant reduction; ↓↓, larger relative decrease in microvessel (radius <10 µm) or macrovessel calibres (total blood volume).

Abbreviations: BCNU, bischloroethylnitrosourea; CRC, colorectal carcinoma; FLT3, FMS-like tyrosine kinase-3; MMP, matrix metalloproteinase; mTOR, mammalian target of rapamycin; NC, no change; NR, not reported; NSCLC, non-small-cell lung carcinoma; PI3K, phosphatidylinositol 3-kinase; RTK, receptor tyrosine kinase; SCC, squamous-cell carcinoma; VEGF, vascular endothelial growth factor; VEGFR(2), VEGF receptor (2).

First, the MRI data might provide evidence of a collective reduction in dilated tumour macrovessel and microvessel calibres suggesting an exhaustive response to therapy across all tumour vessel types. Such an effect was observed with a range of targeted therapies (Table 4), and—according to the VSI estimates (Box 1)—indicates reductions of a similar scale in macrovessel and microvessel blood volumes and calibres.^{118,137–140} However, whereas reductions of macrovessel and microvessel calibres will result in a left-ward shift in the corresponding VSI distribution histogram (that is, towards smaller average vessel diameters),^1,27,141,142 equally scaled reductions of macrovessel and microvessel calibres will result in no apparent change in the MRI-based relative mean vessel diameter estimate (mVD; Box 1)—that is, equal reductions of nominator and denominator.

Second, the MRI data might indicate a larger relative decrease in macrovessel blood volume and calibres compared with the decreases in microvessel parameters, indicating effective targeting of larger (radius >10 µm), abnormal tumour vessels.^87,119,143 This pattern will result in decreases in both the VSI and the mVD estimates.

Third, a larger relative decrease in microvessel blood volume and calibres compared with the change in these measurements of macrovessels might be suggested on the basis of the MRI data obtained.^{120,144–147} This therapeutic response indicates selective targeting of small tumour vessels (radius <10 µm) compared with an apparent resistance of larger tumour vessels.^{144,148–150} It is critically important to note that, according to the equation used to determine the mVD estimate (Box 1), a larger reduction in microvessel blood volume and calibres might in fact result in an apparent increase in the MRI-based average VSI estimate (Table 4), which is not to be mistaken for a true increase in individual vessel calibres.^{6,99,120,148,151}

Fourth, the MRI data might show no apparent change in macrovessel and microvessel blood volume and calibres, indicating no effect of therapy or, alternatively, no effect at the specific dose used.^{144,152–156}

Collectively, responses of tumour-vessel calibres after targeted treatment convey a complex picture dependent on not only tumour type and location,²¹ and the therapeutic agent and dose used,²⁸ but also a dependence on the imaging parameters reported (microvessel calibres, macrovessel calibres or average VSI) as well as the timing of evaluation. Of note, no studies on antiangiogenic and antivascular therapies (Table 4) observe an apparent and transient increase in both microvessel and macrovessel vessel calibres.

Emerging data from trials in glioblastoma

The organizers of clinical trials seem undeterred by the challenges of conventional MRI, with close to 30 reported NIH-funded clinical imaging trials on bevacizumab in patients with glioblastomas alone—of which 10 are recruiting at the time of writing.¹⁵⁷ However, the lack of data derived from more-advanced imaging techniques is striking. To the best of our knowledge, Massachusetts General Hospital is the only institution that has reported on vessel-calibre MRI data obtained during antiangiogenic therapy in clinical studies (Table 3). In two phase II clinical trials of cediranib that enrolled 30 patients with recurrent glioblastomas,¹⁵⁸ and 40 patients with newly diagnosed glioblastomas,¹⁵⁹ our data demonstrated substantial reductions in abnormal macrovessel and microvessel calibres within tumours as early as 1 day after administration of this VEGFR inhibitor (Figure 3).^99,130,132 In the patients with newly diagnosed glioblastomas who responded to therapy, the reduction in tumour-vessel calibres was sustained throughout the 6-week period of cediranib treatment combined with temozolomide chemotherapy (75 mg/m²) and radiotherapy (2 Gy per day).¹³² Furthermore, reduced tumour permeability and increased tumour perfusion were seen in up to half of the patients included in these studies.^99,130,132 These findings are in line with the vascular normalization hypothesis,^26,160 and were also reflected in increased overall survival.¹³² However, after the 6-week period of antiangiogenic and adjuvant therapy, vessel normalization was reversed and relapse of vasogenic oedema was observed in most patients,¹³² suggesting an end to the vascular normalization window. In a contemporary control study¹⁶¹ in patients with newly diagnosed glioblastomas undergoing an identical MRI and chemoradiotherapy regimen, but without cediranib, reduced tumour-vessel calibres were only observed in one out of 14 patients (Supplementary Figure 2).¹³²

Figure 3.

Vessel-calibre MRI during antiangiogenic therapy. Selected MRI-derived images obtained over the course of a phase II trial of cediranib combined with CRT¹⁵⁹ in adult patients with newly diagnosed glioblastomas. From top to bottom: contrast-enhanced T₁-weighted MRI illustrating a permeable tumour vasculature; FLAIR MRI showing the tumour and vasogenic oedema; macrovessel blood-volume fractions; vessel-calibre maps; and volume renderings from 3D anatomical MRI with time-of-flight MRI angiography. Compared with pretreatment baseline images (day −1), cediranib therapy (days 1 and 8) decreased the signal enhancement in the tumour, indicative of reduced permeability of the tumour vasculature, by day 8 (yellow arrow), and oedema was subsequently reduced (yellow arrowhead; day 50). Treatment effects of antiangiogenic and adjuvant combination therapy were also observed after the end of CRT (up to day 400); however, the cancer eventually relapsed and progressed, although this outcome is not clear using standard contrast and FLAIR MRI protocols. Volume fractions and vessel calibres show reductions that are related to the responses observed by conventional contrast-enhanced and FLAIR MRI following therapy (white arrow; day 50), but the vessel-calibre response also reveals unique properties of tissue that do not match the spatial distribution of the volume-fraction map. In contrast to the major vessels identified by standard MRI angiography (bottom panels), the vessel-calibre map depicts changes in the tumour microvasculature that are well below the image resolution achievable using the conventional approach; note the increased vessel calibres in the relapsed tumour region (white arrowhead; day 400), which is not easily appreciated on conventional MRI nor on the volume-fraction map. Abbreviations: CRT, chemoradiation therapy; FLAIR, fluid-attenuated inversion recovery.

These promising findings at our institution are in contrast to randomized phase III trials of cediranib in recurrent glioblastomas¹⁶² and also trials of bevacizumab in newly diagnosed glioblastoma multiforme (AVAglio and RTOG 0825 trials^163,164) that failed to show an overall survival benefit. On the basis of radiological response alone (according to a modified version of the MacDonald criteria) or in combination with improved neurological status, progression-free survival was prolonged after bevacizumab and chemoradiation (temozolomide) therapy in the AVAglio study, but not the RTOG 0825 study.¹⁶⁵ These disappointing effects of antiangiogenic therapy on survival indicate that the use of some targeted therapies is probably not favourable or ideal in unselected groups of patients with cancer and, in light of the findings from Massachusetts General Hospital, also suggests that imaging parameters from conventional MRI are inadequate biomarkers of response.

Suggestions for vessel-calibre MRI in trials

A number of key elements determine the success of a cancer-imaging clinical trial, including selection of an appropriate set of biomarkers for monitoring response to the therapy being studied, optimal imaging time points, and the use of a well-designed image acquisition protocol and correct methods of data analysis. Although advanced MRI is not part of the RECIST criteria,⁴² the NCI Cancer Imaging Program does provide a set of guidelines for the application of advanced imaging modalities in clinical trials, including suggestions for DCE-MRI (for measuring tissue permeability).¹⁶⁶ We support these efforts to provide guidance to researchers, and complement these consensus guidelines with our suggestions for use of vessel-calibre MRI in the initial phases of trial design for studies of antiangiogenic and vascular-targeting agents (Box 2).

Box 2. Suggestions for vessel-calibre MRI in clinical trials of cancer.

Study design

• ■Baseline imaging a short time (ideally 1 day) before initiation of therapy is essential, and an additional assessment a few days prior to this ‘immediate’ pretreatment measurement is recommended for calibration of data
• ■Changes in tumour vasculature following antivascular or antiangiogenic therapy are observed as early as day 1 of therapy and, therefore, imaging in the first days after initiation of therapy is critical
• ■Repeated imaging examinations using the same scanner with standardized imaging protocols are warranted within the first months of targeted therapy to detect the angiogenic switch, indicating the end of a vascular-normalization period
• ■For multicentre studies, enrollment of the full range of patients groups at all institutions is important

Imaging

• ■For gadolinium-based studies, a small amount of contrast agent (around 50% of the standard dose used) should be administered prior to the standard dose (pre-dose), in order to minimize conflicting contrast-agent extravasation effects
• ■Gadolinium-based assessment should also use a sufficiently high contrast-agent dose (≥1.0 mmol/kg) to compensate for the relative low signal-to-noise ratio achieved using microvessel (spin-echo) DSC-MRI
• ■Intravascular contrast agents (superparamagnetic iron oxide) are particularly useful for assessments of anatomical sites outside of the central nervous system; however, potential adverse effects of these agents, such as allergic reactions and back pain, are possible
• ■Diffusion-weighted MRI should be included in the imaging protocol to account for regional differences in diffusion, especially within the tumour
• ■Rapid echo planar imaging sequences can suffer from geometrical distortions; thus, image quality—both signal intensity images and relaxation rate maps—should be examined regularly during the study
• ■Variations in image geometry and imperfect patient positioning are major sources of error in longitudinal studies; all of the main instrument manufacturers provide atlas-based position registration systems that minimize variations between scans
• ■Efforts to reduce image variation should be made during the pilot phase by performing phantom imaging or assessment of healthy control individuals, and image signal variations should be evaluated in homogenous tissue regions across institutions using matched controls

Image analysis and reporting

• ■DSC-MRI data are sensitive to motion and can be minimized by motion-correction algorithms or by orienting the patient and imaging slices so that the direction of motion is in-plane
• ■Correct for contrast-agent extravasation using mathematical correction algorithms
• ■High-resolution anatomical MRI and tumour regions-of-interest should be downscaled to vessel-calibre MRI resolution by automatic normalized mutual information co-registration procedures or similar algorithms
• ■Whole-tumour vessel distribution analysis is less sensitive to operator variations compared with ‘hot-spot’ region selection and yields higher reproducibility
• ■For multicentre studies, centralized tumour-region outlining and automatic analysis routines should be considered
• ■Tumour-vessel calibres during therapy should be reported relative to baseline values to reduce inter-patient variations and to adjust for global systemic vascular effects
• ■Reporting of patient-level macrovessel (ΔR₂*) and microvessel (ΔR₂) blood-volume data in a tabular form enables readers to understand the vascular response to therapy and to compare the published data with their own experimental data

Abbreviations: ΔR2, change in spin-echo relaxation rate; ΔR2*, change in gradient-echo relaxation rate; DSC-MRI, dynamic susceptibility-contrast MRI.

In short, the design of imaging trials is critically dependent on at least one baseline scan as close as possible to initiation of therapy, which should be followed by a second scan within the first days of therapy for assessment of early therapeutic response.²⁹ For studies of targeted therapies, repeated weekly or monthly follow-up imaging assessment is ideal—although counterbalanced by patient burden and costs—to observe the responses and plateaus in the various advanced-MRI parameters used, and to identify the end of the vascular normalization period (Figure 4).⁸ Because of the apparent sensitivity of the technique to changes in vascularization, vessel-calibre MRI seems to have the potential to inform personalized therapy and to enable the optimal window of vascular normalization and potentially, therefore, drug administration to be determined in individual patients.

Figure 4.

Advanced MRI assessments might inform optimal dose scheduling. On the basis of our experience with cediranib,^{99,130,132,176} the initial responses, plateaus and reversals in tumour response that are observed using advanced MRI parameters occur at different timepoints after initiation of therapy. In this example model of tumour-vessel architecture, showing a typical vascular-normalization response after cediranib treatment and CRT in patients with newly diagnosed glioblastomas,¹³² the perfusion estimate (blood flow) peaks after approximately 1 week of therapy (day 8), whereas the maximum reduction in vessel calibres occurs 1 week later (day 15; note the reduced diameter of the example vessels). These effects are preceded by an early, substantial, and prolonged normalization of the abnormal arterio-venous ratio evident in the model of the baseline data (day 1; arteries/arteriole shown in red, with veins in blue), estimated using vessel architectural imaging. Consistent with the vessel-calibre response, the apparent change in abnormal ΔSO₂ levels reaches its minimum at two weeks (day 15; purple colour). After the end of CRT, all of these parameters reverse, thereby indicating an end to the vascular-normalization window. Furthermore, in patients who were deemed unresponsive to therapy, limited evidence of such responses was seen. Collectively, advanced MRI protocols might facilitate the design of improved early-phase trials by informing the optimization of the dosing regimen for antiangiogenic agents and could also potentially support a personalized-medicine approach by enabling therapy to be tailored to individual patients based on parameters indicative of biological and, in particular, vascular responses. In addition, early identification of patients who are unlikely to respond to therapy could help decision making regarding whether therapy should be discontinued, particularly in patients who experience adverse events. Abbreviations: ΔSO₂, relative oxygen saturation levels; CRT, chemoradiation therapy.

To achieve a sufficiently large patient population over a relatively short period of time (typically 1–3 years), presently, most clinical trials are multi-institutional studies. As discussed by others, and although logistically challenging, a multicentre study design is essentially the only way of assessing the real physical process of therapeutic intervention independent of institutional variations.^64,121 With some exceptions,^167,168 vessel-calibre MRI—similar to most advanced imaging methods—suffers from lack of multicentre validation when applied alone in a research setting, or as part of a clinical trial. With careful planning,¹²¹ vessel-calibre MRI is no more difficult to perform during a clinical trial than conventional contrast-enhanced MRI. Before the trial commences, multicentre variability in scanner performance and post-processing routines can be reduced by performing development studies in healthy volunteers or standardized phantom testing.¹²¹ An MRI phantom test usually involves imaging of a magnetic-resonance-compatible polycarbonate plastic object with an array of small spheres or compartments that are filled with substances of different magnetic properties and with known scanning positions for calibration. The Alzheimer’s Disease Neuroimaging Initiative (ADNI) phantom programme¹⁶⁹ is dedicated to this purpose, and has the added advantage of a centralized imaging-standard review process, which can reduce errors that contribute to quantitative imprecision. Indeed, this collaborative agreement was shown to identify MRI system failures that, if not corrected for, might have contributed to up to 25% imprecision in quantitative measures across institutions. Most importantly, the ADNI demonstrates that multicentre imaging standardization is achievable. For vessel-calibre MRI in particular, sufficient image quality with high signal-to-noise should be established early in the trial development processes in a small pilot study, before deciding on a standardized imaging protocol that should be used across multiple centres. Protocols should also capitalize on existing scanner-based and postprocessing software solutions to minimize the effects of imperfect patient positioning in the scanner, stochastic or free-breathing movements, as well as extravasation of gadolinium-based contrast agents (Supplementary Methods).^12,134,170

Future directions

The most-pressing issue for imaging-response criteria for targeted therapies is the need for the introduction and validation of advanced imaging beyond the current RECIST and RANO criteria.^11,42,171 Future updated response criteria should include volumetric measurements, dynamic imaging and molecular imaging. The goals of these efforts are: firstly, smaller, cheaper, and selectively targeted patient populations for clinical trials; secondly, earlier and more-focused response assessments; and, thirdly, faster regulatory approvals of therapies than is currently possible for antiangiogenic and antivascular therapies. We believe that MRI-based measures of vessel calibres in solid tumours complement traditional imaging modalities and should play a part in any updated response criteria for antivascular and antiangiogenic therapies.

Composite biomarkers

Although efforts have been made to automate the vessel-calibre analysis protocol,⁹⁹ MRI biomarkers of such high complexity usually have the drawback of a laborious acquisition process compared with simpler response measures, such as tumour size. However, in comparison to traditional DSC-MRI-based estimations of blood volume, vessel-calibre MRI is no more difficult to perform and, because of its unique identification of microvessel and macrovessel calibre, might help in bringing MRI one step closer to enabling personalized diagnoses for specific tumour subtypes and, thereby, facilitate the development of tailored therapeutic regimens. In conjunction with traditional MRI-based measures of tumour size and tumour-vessel permeability,⁹ of which the latter has revealed consistent reductions of the permeability surface area product K^trans following antiangiogenic therapy,^{9,13,15,17,59,172} vessel-calibre MRI might have utility as a companion indicator for tumour response to targeted nanoparticle-based EPR therapies in the future.^35,37 For example, angiotensin inhibitors, such as losartan, are known to effectively increase the diameters of blood vessels in models of matrix-rich breast tumours,³⁸ thus, vessel-calibre MRI could have value in assessing responses and guiding the use of such agents. However, for such composite biomarkers to have a role in clinical trials, they need to be validated in large, multicentre studies to overcome the ‘curse of dimensionality’, which is the concept that multiple input parameters require large amounts of data to support physiologically and statistically sound conclusions.¹⁷³

Next-generation vessel architectural imaging

Vessel architectural imaging (VAI), which exploits a temporal shift in the macrovessel MRI signal compared with the microvessel signal (Supplementary Methods),^84,99,128 has been introduced as a new concept in vessel-calibre MRI assessment of cancer therapy.⁹⁹ This effect was first observed by Kiselev et al.⁸⁴ in 2005 and was attributed to the high sensitivity of the macrovessel MRI signal for deoxygenated capillary and venous blood volume; thus, the temporal shift might reveal information on the vessel types present (arteries, capillaries, or veins), as well as the relative difference in radii and oxygen saturation levels between these vessel types.^93,95,99,145 The highly intriguing link between this effect and oxygen saturation is in line with the BOLD effect,¹⁷⁴ as well as numerical modelling by Jespersen et al.¹⁷⁵ showing that heterogeneous perfusion from torturous microvessels and capillaries, resulting in varying contrast-agent arrival times and mean transit times, influences the maximum oxygen extraction fraction and the metabolic rate of oxygen use in tissues. In our imaging studies in trials of antiangiogenic therapy with cediranib, VAI revealed that overall survival duration was increased in 30–50% of patients with glioblastoma who had an observed reduction in abnormal tumour vessels, improved tumour perfusion, and normalization of the arteries-to-capillaries-to-veins branching hierarchy and oxygen saturation levels.^99,132 We are currently exploring the use of VAI in other tumours types and organs, including brain metastases, and breast and kidney tumours (Supplementary Figure 3). Importantly, an apparent wealth of information beyond a simple measure of the average vessel calibre can be obtained using VAI, such as tumour vascular structure and physiology.

Conclusions

The identification and interpretation of valid prognostic and predictive biomarkers for cancer diagnosis and assessment of therapeutic response is a major challenge in the field of oncology. Because of its high sensitivity to tumour physiology and vascular structure, vessel calibre is a possible biomarker for MRI-based in vivo cancer imaging, with the potential to provide new information and an increased understanding of the complex nature of tumour vascularity. With increased availability and awareness of the technique, combined with optimization of imaging and scheduling protocols, we postulate that vessel-calibre MRI can have a direct effect on the efficacy of future antivascular and antiangiogenic clinical trials in patients with cancer.

Key points.

• ■Negative phase III trials suggest that administration of antiangiogenic therapies in unselected patient groups does not result in optimal outcomes, owing to diversity in cancer biology and natural history
• ■Antivascular and antiangiogenic therapies are not directly cytotoxic and, therefore, traditional assessment of MRI-based tumour volume change alone can no longer be considered an adequate biomarker of therapeutic response
• ■Intravascular paramagnetic contrast agents induce small and local magnetic field variations in tissue that scale with the blood vessel calibre (the cross-sectional width of the vessel)
• ■MRI is exquisitely sensitive to these magnetic field perturbations and, therefore, provides a means for in vivo assessment of tumour vessel calibre
• ■Depending on drug and dose administration, initial experiences with antivascular and antiangiogenic therapies and vessel-calibre MRI suggest that pruning of abnormal tumour vessels is often non-uniform
• ■Collectively, vessel-calibre MRI and emerging MRI-based assessments, such as vessel architectural imaging, can provide insights into vessel type and oxygenation status—creating new possibilities for clinical trial design and monitoring therapeutic response and outcomes

Review criteria.

PubMed, MEDLINE and ISI Web of Knowledge databases were searched for full-text English-language articles published before 24 January 2014 using the following terms: “MRI AND vessel size” or “MRI AND vessel caliber”. The reference lists of the relevant articles identified were searched for additional papers. In addition, abstracts from ASCO, American Association for Cancer Research (AACR), Radiological Society of North America (RSNA), and International Society for Magnetic Resonance in Medicine (ISMRM) conferences were considered for inclusion if they provided relevant data that had not been published in full-text articles.