Assessing Global Invasion of Newly Diagnosed Glial Tumors with Whole-Brain Proton MR Spectroscopy

Methods

MR Imaging: Brain Volume, V_B, Segmentation

Each subject’s brain volume, V_B, was obtained from his or her T1-weighted sagittal magnetization-preparation rapid gradient echo (TE/TR/TI, 7.0/14.7/300 ms; 128 sections, each 1.5 mm thick; 256 × 256 matrix; 210 × 210 mm² field of view) images by using the MIDAS package (23). Specifically, a “seed” region was placed in the periventricular white matter. Following selection of all pixels at or above the gray matter signal intensity, a brain mask was constructed in 3 steps: (1) morphologic erosion; (2) recursive region growth, which retains pixels connected to the “seed” region; and (3) morphologic inflation to reverse the effect of erosion. Next, the brain masks were truncated at the foramen magnum to incorporate the brain stem and cerebellum but not the cord. Finally, tumor beds, including pixels below gray matter signal intensity, were incorporated in the presurgical masks while resection cavities were excluded from the postsurgical masks. V_B was calculated from the mask by multiplying the number of pixels therein by their volume. Examples of CSF/parenchyma segmentation performance are shown in Figs 2A and -B.

• Download figure
• Open in new tab
• Download powerpoint

Fig 2.

Top, A and B, Segmented preoperative T1-weighted sagittal section for 2 patients, 15 and 5 in the Table, respectively. The red sulci and ventricle outline is the output of the tissue/CSF segmentation process. Patient A had a 41-cm³ grade III tumor, 3.0% of his brain volume, and 37% WBNAA deficit. Patient B had a 93-cm³ grade II tumor, 6.2% of his brain volume, but only half the WBNAA decline (∼20%).

Bottom, A′ and B′, Corresponding whole-head ¹H-MR spectroscopy. The hatched regions indicate the peak-areas used to obtain Q_NAA of Equation (1). Note the overall similarity, indicating cross-sectional reproducibility, and good lipid suppression. Also note that, because localization relies on knowledge that NAA is found only in neuronal cells, it cannot quantify the brain’s contribution to metabolites also present in other tissue types (eg, choline, creatine, and so forth).

Results

Individual controls’ and patients’ pre- and postsurgical WBNAA are compiled in the Table. WBNAA in controls (n = 17) was 12.5 ± 1.4 mmol/L (mean ± SD), compared with 9.2 ± 2.1 mmol/L presurgery (n = 17) and 9.7 ± 1.8 mmol/L postsurgery (n = 14) in patients, as shown in Fig 3. Average resected tumor size (n = 15), calculated from the MR imaging as the difference between pre- and postsurgical brain volumes, was 40.9 cm³ (±20 cm³), or ∼3.2% (±1.5%) of mean presurgical brain volume. Histologically, 9 of the 17 tumors were low-grade (one grade I, 8 Grade II) and 8 were high-grade (4 each grades III and IV). Three patients (one each for grades II, III, and IV) were not included in the postsurgical analysis because either imaging was not clinically indicated following biopsy (n = 2) or the signal intensity quality was too poor for analysis because of excessive motion (n = 1).

• Download figure
• Open in new tab
• Download powerpoint

Fig 3.

First, 2nd (median), and 3rd quartiles (box) and ±95% (whiskers) of WBNAA in the 17 controls, patients day of surgery (n = 17) and 1 day postsurgery (n = 14). Note that pre- and postsurgical WBNAA are significantly different from the controls, but not each other.

Adjusted for differences attributable to age and sex, the WBNAA of controls was significantly higher than that of patients both pre- and postsurgery (P < .0005). For patients, pre- and postsurgical WBNAA were statistically indistinguishable (P = .369) with a 95% confidence interval (CI) for mean change ranging from −0.4 to +1.1 mmol/L, as shown in Fig 3.

Presurgical WBNAA analysis revealed a significant interaction with age in the patients, as shown in Fig 4. Specifically, WBNAA in the 9 patients <40 years of age, 10.3 ± 2.0 mmol/L, was significantly lower than their control group (12.4 ± 1.3 mmol/L; P = .017). This deficit was even greater in the 8 patients >40 years of age, 8.0 ± 1.5 mmol/L versus 12.7 ± 1.6 mmol/L in their controls (P < .0005).

• Download figure
• Open in new tab
• Download powerpoint

Fig 4.

Scatter plot of WBNAA versus age for the patients before surgery and their age- and sex-matched controls. Note that the patients’ WBNAA deficit is significant (P < .0005), even after taking into account age (P = .007) and sex effects (P > .05).

Preoperative WBNAA was lower in patients with high-grade (8.3 ± 1.8 mmol/L) than in those with low-grade (10.0 ± 2.1 mmol/L) tumors, as demonstrated in Fig 5. The trend of this difference toward significance (P = .086; 95% CI of mean difference ranging from −0.3 to +3.8 mmol/L) disappeared when the results were adjusted for the effect of age.

• Download figure
• Open in new tab
• Download powerpoint

Fig 5.

First, 2nd (median), and 3rd quartiles (box) and ±95% (whiskers) of preoperative WBNAA in the 9 patients with low-grade tumors compared with their matched controls (left) and 8 patients with high-grade tumors compared with their matched controls (right). Note preoperative WBNAA was lower in patients with high-grade (8.3 ± 1.8 mmol/L) than in those with low-grade (10.0 ± 2.1 mmol/L) tumors.

Discussion

High-grade glial tumors appear as contrast enhanced regions on conventional MR imaging, reflecting a compromised blood-brain barrier (BBB), surrounded by T1 and T2 abnormalities representing edema and infiltrated viable brain. Because the solid tumor itself is devoid of viable brain, its surgical resection typically causes no major neurologic deficits (25–28). This consideration usually precludes removal of the enveloping regions of signal intensity abnormality (25, 29). Consequently, despite adjuvant chemotherapy with or without conformal radiation, the disease always recurs either (1) at the surgical margin, because of incomplete resection (4–6), or (2) remotely, presumably from cells that have migrated to distal regions of intact BBB (ie, appear normal on conventional imaging) and are immune to chemotherapy as a result (3–6, 30). Indeed, one postmortem study has found cells from a pontine glioma as far away as the conus medullaris (31).

The central cause of poor prognosis in these patients, therefore, is diffuse infiltration. Current state of the art techniques, such as thallium SPECT (8–10), perfusion MR imaging (11–15), and localized ¹H-MR spectroscopy (16, 17), are limited in their ability to detect infiltration into “normal-appearing” brain distal to the primary tumor due to intact BBBs, limited neovascularization, and insufficient metabolically active tumor volumes (20). This obstacle, however, can potentially be overcome through a metabolic assessment of the collective damage infiltrating tumor cells impart on the neurons in their path. This injury occurs through perineuronal satellitosis, shown in Fig 1, release of neurotoxic cytokines and excitotoxic neurotransmitters, ischemia resulting from invaded vasculature, and infiltration along white matter tracts (32–35). Because these processes are diffuse and widespread beyond the tumor mass, extensive (ideally, complete) CNS coverage, provided by the WBNAA technique, is needed to quantify their true extent. Such an assessment could potentially facilitate the formulation of clinical decisions by further elucidating the benefits and risks/costs of aggressive management.

Before surgery, our patients already exhibited an average of 26% lower WBNAA than their controls. Although the degree of NAA loss may seem relatively large in light of the mild to moderate neurologic deficits that typically precede diagnosis, similar declines have been reported in multiple sclerosis without major changes in clinical status (36). Because (1) these patients suffered no other known neurologic insult and (2) the average tumor was small, on average only ∼3% of brain volume, the mechanism that led to such profound NAA decline must have predominantly occurred outside the MR imaging-visible lesion. Because NAA is found only in neurons and their processes, this extent of the WBNAA deficit, therefore, indicates widespread damage. This damage may arise from various processes—eg, perineuronal satellitosis (compare Fig 1), neuronal deafferentation, Wallerian and retrograde degeneration, vasogenic edema, functional diaschisis, secondary vascular changes, and so forth. Most of these processes, however, reflect either direct or indirect consequences of the widespread, diffuse infiltration of malignant cells into normal-appearing tissue (3–6, 30, 31, 37).

Additional indications that the occult diffuse infiltration caused the WBNAA decline and not only, for example, tissue destruction in the lesion, are provided by the postsurgical analysis. First, the significant 22% mean WBNAA loss that persisted following resection indicates that the loss originated mostly outside the resected lesions (which averaged only ∼3% of total brain volume). Second, because the NAA in the tumor is nearly depleted (16, 17), excision should increase the WBNAA because Q_NAA remains the same while V_B decreases (Equation [2]). Accordingly, the cohort’s mean WBNAA was greater post- than presurgery by ∼4%, commensurate with the average excised volume, as shown in Fig 3.

If invasion was the primary mechanism leading to neuronal, and therefore WBNAA, loss, it stands to reason that less-invasive tumors would lead to the smallest WBNAA loss. Indeed, this assertion is supported by the histology of patient 1 in the Table. This 23-year-old woman had a pathologic diagnosis of dysembryoplastic neuroepithelial tumor, a known noninvasive neoplasm that is theoretically curable by surgery alone (38). Therefore, the fact that her WBNAA was indistinguishable from the controls (compare Fig 4) is consistent with this noninvasive behavior. Because her tumor size was only 32 cm³, <3% of brain volume, any NAA loss restricted to its visible boundaries would have been below the ±6% detection threshold of the technique (20).

The relationships of WBNAA with age, as shown in Fig 4, and tumor grade suggest that older patients with higher-grade tumors experience a larger loss of WBNAA than younger patients with lower-grade lesions. This is not surprising, because it is well known that advanced age is a predictor of poorer outcome and higher tumor grade (39, 40), consistent with our cohort, where all but one patient <45 years of age had (lower) grades I–II tumors, whereas those older had higher grades—III and IV. Thus, age and tumor grade are potentially confounding variables in any examination of our patients. Nevertheless, statistical analysis showed that, even with age as a covariate, WBNAA deficits were significant for the entire patient group. Additional evidence that the observed WBNAA losses in older patients are, at least in part, due to invasion and not just to an age effect is the decline suffered by the younger (<40 years) half of the cohort. They also exhibited significant deficits compared with their controls (compare Fig 4), even though their tumors were of lower (less-invasive) grades and their residual brain plasticity was presumably higher. Sex, on the other hand, appeared to play no role in WBNAA deficits.

The invasion argument above is corroborated further by comparing patient 5 with patient 15 in the Table. Patient 15’s grade III anaplastic oligodendroglioma (Fig 2A, -A′) was 41 cm³ (3.0% of brain volume) and led to a ∼37% WBNAA decline. In contrast, patient 5 (Fig 2B, -B′) suffered only half (∼20%) the WBNAA loss of patient 15 despite a grade II ganglioglioneurocytoma more than twice as large, 93 cm³ (6.2% of brain volume). If the loss was not due to diffuse infiltration, it would have been reasonable to expect the changes in WBNAA to be proportional to the tumor volume rather than its grade.

As demonstrated in Fig 5, the results of these 2 patients can be generalized with a group comparison of patients with low- and high-grade tumors. Patients with high-grade, and presumably more invasive, tumors exhibited a preoperative WBNAA deficit nearly twice that of patients with low-grade tumors. This difference, which trended toward significance, can also be attributed to an age effect described above as our cohort was evenly divided between young patients with grade I and II tumors and older patients with grades III and IV tumors.

Though infiltration is diffuse, one would expect that most tumor cells would be found in the peritumoral region and the resultant neuronal injury should progressively decrease as an inverse function of distance from the tumor. Indeed, this behavior has been observed and reported in several previous studies in gliomas when using 3D localized multivoxel ¹H-MR spectroscopy (41–43). Their observations corroborate the conclusions drawn from the current WBNAA analysis. Although the WBNAA deficit found here may represent, at least in part, a reversible metabolic dysfunction, it is highly unlikely to be true for most of it, in light of the chronic, unrelenting nature of gliomas. Furthermore, note that the WBNAA deficit persisted following surgical resection of the bulk lesion with even a slight 4% recovery, consistent with removal of tissue relatively deplete of NAA, as described above. To attribute the loss to a reversible dysfunction, one would expect a more substantial WBNAA rebound following elimination of the prime insult.

The main limitation of the WBNAA assessment is its lack of spatial specificity—ie, where the losses occurred and what mechanism caused them (20, 21). Unfortunately, there is no definitive in vivo “gold standard” method to establish either. The preponderance of evidence, however, including the relationship of WBNAA with age and tumor grade, suggests that tumor infiltration is a likely culprit. This conclusion is supported by histopathologic examination of the infiltrated brain adjacent to tumor, the glioblastoma multiforme from patient 12 in the Table, which demonstrates significant quantifiable neuronal damage, as shown in Fig 1. Furthermore, though some NAA signal intensity dropout could also occur at the tumor margin, due, for example, to susceptibility effects from hemosiderin deposition, this effect on WBNAA, however, would be minimal because these tumors were small (∼3% of brain volume) and NAA loss was not proportional to tumor volume (compare Fig 2).

In summary, WBNAA could potentially provide valuable noninvasive information on the global amount of neuronal damage currently unavailable to the neurooncology team and would complement histopathologic findings, which sometimes does not accurately reflect true invasiveness, partially because of tumor heterogeneity and sampling error (44, 45). For instance, WBNAA could help guide management when planning surgery (ie, conservative vs aggressive) for a patient with a suspected low-grade lesion by suggesting its true infiltrative and malignant potential. Perhaps, aggressive surgical resection of highly infiltrative low-grade tumors would preclude early recurrence and, more important, malignant transformation. Because of the poor prognosis of high-grade gliomas and the morbidity of their treatments, such assessments could also aid in identifying patients who would most benefit from aggressive management versus those for whom palliation would be more appropriate. Consequently, the main utility of WBNAA could be in customizing treatment. WBNAA could also serve as an efficacy monitor of systemically delivered therapeutics—eg, angiogenesis inhibitors, immunotherapy and gene therapy (46)—for which localization of tumor infiltration is inconsequential.

Conclusion

In this study, a significant WBNAA deficit was found in patients with glial tumors that persisted following surgery and whose magnitude cannot be accounted for by depletion within the lesion alone. This deficit was greater in patients with higher-grade tumors, who, as per the epidemiology of this disease, were also the older individuals, both independent predictors of a poor prognosis. This opens the possibility of a noninvasive assessment of diffuse tumor infiltration. Further investigation is required to delineate potential roles of WBNAA in customizing treatment for individuals with glial tumors.