Quantitative Assessment of Whole-Body Tumor Burden in Adult Patients with Neurofibromatosis

Purpose Patients with neurofibromatosis 1 (NF1), NF2, and schwannomatosis are at risk for multiple nerve sheath tumors and premature mortality. Traditional magnetic resonance imaging (MRI) has limited ability to assess disease burden accurately. The aim of this study was to establish an international cohort of patients with quantified whole-body internal tumor burden and to correlate tumor burden with clinical features of disease. Methods We determined the number, volume, and distribution of internal nerve sheath tumors in patients using whole-body MRI (WBMRI) and three-dimensional computerized volumetry. We quantified the distribution of tumor volume across body regions and used unsupervised cluster analysis to group patients based on tumor distribution. We correlated the presence and volume of internal tumors with disease-related and demographic factors. Results WBMRI identified 1286 tumors in 145/247 patients (59%). Schwannomatosis patients had the highest prevalence of tumors (P = 0.03), but NF1 patients had the highest median tumor volume (P = 0.02). Tumor volume was unevenly distributed across body regions with overrepresentation of the head/neck and pelvis. Risk factors for internal nerve sheath tumors included decreasing numbers of café-au-lait macules in NF1 patients (P = 0.003) and history of skeletal abnormalities in NF2 patients (P = 0.09). Risk factors for higher tumor volume included female gender (P = 0.05) and increasing subcutaneous neurofibromas (P = 0.03) in NF1 patients, absence of cutaneous schwannomas in NF2 patients (P = 0.06), and increasing age in schwannomatosis patients (p = 0.10). Conclusion WBMRI provides a comprehensive phenotype of neurofibromatosis patients, identifies distinct anatomic subgroups, and provides the basis for investigating molecular biomarkers that correlate with unique disease manifestations.

These related syndromes have overlapping clinical features and for years, clinicians struggled to differentiate the types of neurofibromatosis. Clinical criteria were established for NF1 and NF2 in 1987 and for schwannomatosis in 2005. [5,6] These patients share a predisposition to developing benign nerve sheath tumors, including neurofibromas and schwannomas, that are derived from neoplastic Schwann cells. Despite the benign histology of neurofibromas and schwannomas, neurofibromatosis patients have increased mortality due to malignant peripheral nerve sheath tumor (MPNST), glioma, cardiovascular disease, and organ compression by neurofibromas. [7,8] For NF1 patients, the median age at death is 59 years, compared with 74 years for the general population. [9] For NF2 patients, actuarial survival after diagnosis is 85% at 5 years, 67% at 10 years, and 38% at 20 years. [10] Mortality figures for schwannomatosis have not been reported.
The spectrum of tumor involvement of these disorders is highly variable. In practice, clinicians carefully select the body region to image based on the presence of symptoms and knowledge of disease phenotype (e.g., cranial MRI for vestibular schwannomas in NF2 patients). Imaging of the entire body using traditional regional scans is not possible due to the cost (in time and money) of MRI. For this reason, data regarding the prevalence of wholebody disease patterns from large, multicenter cohorts are lacking.
Whole-body MRI (WBMRI) can evaluate the entire body in a relatively short time without the use of ionizing radiation. We performed an international multicenter study of WBMRI to assess tumor burden of internal nerve sheath tumors in NF1, NF2, and schwannomatosis. Our goal was to identify phenotypic similarities among these related neurogenetic disorders, to identify patterns of tumor involvement, and to relate tumor burden to demographic factors and cutaneous disease manifestations.

Whole-body MRI
We performed WBMRI in patients with NF1, NF2, or schwannomatosis. [5,11,12] Inclusion criteria for the study included age $18 years of age; diagnosis of NF1, NF2, or schwannomatosis by clinical criteria; [5,11,12] and ability to provide written informed consent. Exclusion criteria included inability to undergo MRI because of a medical or psychological condition; presence of a metallic implant; need for general anesthesia; pregnancy; or breast-feeding. Patients were drawn from a convenience sample of patients seen at the Neurofibromatosis clinics at Massachusetts General Hospital and University of Hamburg, Eppendorf, Germany.

Determination of tumor burden
WBMRI was performed once per individual as previously described. [13] MRI scans were first reviewed by a board-certified radiologist who identified the location and appearance (circumscribed vs. plexiform) of each tumor based on its MRI appearance. Tumors that were locally circumscribed on MRI were classified as circumscribed and those that were invasive or involved multiple nerves were classified as plexiform ( Figure 1). Pathological diagnosis was not required. Second, each tumor was segmented using computerized volumetry method developed for WBMRI. [13] Third, the study radiologist reviewed the computerized tumor contours. Finally, whole-body tumor burden was determined by recording the number, location, appearance (circumscribed vs. plexiform), and volume of individual tumors for each patient.

Statistical analysis
Whole-body tumor analysis. We calculated descriptive statistics for clinical and demographic factors for each disease group and compared these factors using Kruskal-Wallis test for continuous variables and Chi-square test for categorical variables. We compared whole-body tumor burden among disease groups using Kruskal-Wallis test.
We used whole-body imaging data to identify risk factors associated with internal nerve sheath tumors. We first used logistic regression to analyze the relationship between the presence/ absence of internal tumors and clinical/demographic factors. We then used multivariable linear regression to analyze the relationship between tumor volume and clinical/demographic factors in patients with at least one internal tumor. In our analysis, we included clinical and demographic factors previously associated with paraspinal neurofibromas in NF1 and with disease severity in NF2 [14][15][16]; we included additional clinical factors to explore their relationship with tumor burden. For multivariable analyses, we included age, gender, inheritance pattern (sporadic vs. familial), number of café-au-lait macules (0 vs. 1-5 vs. 6-15, vs. .15), and history of skeletal abnormalities as covariates in all patients. In NF1 patients only, we included the presence of gliomas, number of cutaneous neurofibromas (0, 1-9, 10-100, 101-500, .500) [17] and number of subcutaneous neurofibromas (0, 1-9, 10-100, 101-500, .500) as covariates. In NF2 patients only, we included the presence of meningiomas and cutaneous schwannomas as covariates. Tumor volumes were natural log-transformed prior to multivariable regression analyses.
Distribution analysis. We calculated the absolute and relative tumor volume in six anatomic regions (head/neck, chest, abdomen, pelvis, arms, and legs) in patients with internal tumors. We then quantified the distribution of tumor volume across body regions for each patient (i.e., regionality of tumor burden) using the Gini coefficient. [18] A low Gini coefficient indicates a more even distribution across body parts, with a value of 0 corresponding to uniform distribution of tumor volume across body regions; higher Gini coefficients indicate a more unequal distribution with a value of 1 corresponding to concentration of tumor volume to a single body region. We then identified specific body regions preferentially affected by tumors. We adjusted our analysis to account for known differences in the size of each body part. [19] In this analysis, the null hypothesis was that the percentage of whole-body tumor volume in each body part would be equivalent to the percentage of whole-body volume of each body part. To determine the average volume of each body part, we measured the volume of body parts in 5 male and 5 female patients selected at random. We then used a bootstrap method to test for significance of the Manhattan distance between the predicted distribution and the observed average distribution.
We next used unsupervised cluster analysis to sort patients into groups based on anatomic predisposition. Tumor volumes across all regions were combined and scaled for each patient; scores were then analyzed by agglomerative hierarchical clustering. We partitioned the resulting dendrogram and assessed possible associations between the resulting groups and the following variables by a series of Chi-square tests: diagnosis, number of café-au-lait macules, presence of skeletal abnormalities, presence of gliomas, inheritance pattern, presence of cutaneous tumors, and gender. We assessed the variable age using Kruskal-Wallis test.
All statistical calculations were performed with SAS software (version 9.2, SAS Institute Inc, NC, USA) and the statistical programming language R. [20] The distribution and cluster analyses were designed post-hoc. The study was approved by the institutional review boards at Massachusetts General Hospital (Partners Human Research Committee); University of Hamburg, Eppendorf, Germany (Ethics Committee of the Chamber of Physicians, Hamburg); and the Department of Defense (Human Research Protection Office). Written informed consent was obtained from all participants.
In patients with internal tumors on WBMRI, tumor count ranged from 1 to 69; the distribution of tumor count per patient showed an exponential decline (Figure 2A). In patients with at least one tumor, the median number of tumors did not differ by diagnosis (Table 3). In logistic regression analysis, the presence of internal nerve sheath tumors was correlated with decreasing number of café-au-lait macules in NF1 patients (P = 0.003) and with history of skeletal abnormalities in NF2 patients (P = 0.09) ( Figure 3). All other clinical factors had P-values .0.10, the level of significance used for this exploratory correlative analysis.
In patients with internal tumors, whole-body tumor volume per patient ranged from 1.2 ml to 9106.1 ml; the distribution of total tumor volume showed a progressive decline in frequency as tumor volume increased ( Figure 2B). The median whole-body tumor volume for all groups was 83.0 ml and differed among diagnosis groups (p = 0.02, Table 3). Although 41% (528/1286) of lesions were plexiform in appearance, these tumors contributed 78% of the total tumor volume: the median tumor volume was 29.4 ml per plexiform tumor and 6.7 ml per circumscribed tumor. In multivariable analysis, increased tumor volume was correlated with female gender (P = 0.05) and with presence of subcutaneous neurofibromas (P = 0.03) in NF1 patients, with the absence of cutaneous schwannomas in NF2 patients (P = 0.06), and with increasing age in schwannomatosis patients (P = 0.10). All other clinical factors had P-values .0.10.

Distribution analysis of internal nerve sheath tumors
The median Gini coefficient was 0.84 (range, 0.26 to 1.0), suggesting that tumor volume within patients was not evenly distributed across body parts. The majority of patients (122/144, 85%) had high Gini coefficients ($0.67) indicating that tumor volume was concentrated in a limited region. Seventeen patients (12%) had moderate Gini coefficients (0.34-0.66), and just 5 patients (3%) had low Gini coefficients (#0.33) indicating even distribution of tumor volume across body parts. The complete collection of WBMRI with tumor volumes and Gini coefficients can be viewed at www.wholebodymri.org.
Clustering was associated with age (p = 0.06) and with inheritance pattern (sporadic vs. familial) (P = 0.1). Patients were then dichotomized according to median age at diagnosis (#20 years vs. .20 years), median age at WBMRI (#40 years vs. .40 years), and inheritance pattern for comparison of relative tumor volume across body regions. There was no significant difference between sporadic and familial patients or between patients diagnosed before or after 20. However, patients #40 years had a higher proportion of tumor volume in the pelvis and legs compared those .40 ( Figure 5). Clinical variables which did not differ significantly among clusters included diagnosis, numbers of café-au-lait macules, skeletal abnormalities, gender, number of cutaneous neurofibromas, and presence of gliomas (p.0.1).

Discussion
In this study, we used WBMRI to examine internal tumor burden in three closely related tumor-suppressor syndromes. Our findings corroborate that WBMRI can detect internal tumors in a far higher percentage of patients than conventional, regional imaging methods. [21] For example, the prevalence of plexiform or spinal nerve sheath tumors on regional MRI or CT scan ranges from 16%-39% in NF1 patients. [17,[22][23][24][25] Using WBMRI, we found that 60% of such patients have internal nerve sheath tumors. As expected, patients without a known history of internal tumor had high rates of lesions on WBMRI, indicating that asymptomatic tumors are common in this population [25].
Because WBMRI can detect even relatively small or asymptomatic tumors in all body regions, it provides a more comprehensive picture of tumor burden in patients and, for the first time, allows for analysis of tumor distribution across body parts. We used novel approaches to describe this distribution, including Gini coefficient to measure the regionality of tumor burden and clustering analysis to sort patients into groups based on tumor predilection. Our results show that internal nerve sheath tumors, like dermal neurofibromas, are not randomly or evenly distributed across body parts (see www.wholebodymri.org). Instead, particular body regions appear to be preferentially affected, while other body regions are relatively spared.
There are multiple processes that might explain why the pelvis and head/neck are particularly affected by internal tumors or why younger patients have a greater proportion of internal tumor volume in the pelvis and legs. Genetic mosaicism is well documented in neurofibromatosis and may be diagnosed when patients display disease features that are restricted to portions of the body (e.g., segmental findings). [26,27] Alternatively, there may be tissue-specific biological factors in the affected regions that are permissive for tumor formation. It is increasingly clear that neurofibroma formation requires a microenvironment containing bone marrow-derived cells that are heterozygous at Nf1. [28] These biological factors could be important during development (e.g., for congenital lesions like plexiform neurofibromas) or post-natally (e.g., for non-congenital lesions like vestibular schwannomas). Genetically engineered mouse (GEM) models of NF1 and NF2 develop nerve sheath tumors in restricted locations that differ from humans. [29][30][31] Subsequent investigations of molecular biomarkers in both GEM and human tissue may help identify biological factors involved in these unique disease manifestations.
Overall, tumor burden and patterns of regionality were similar for the three tumor-suppressor syndromes. We observed a high prevalence of nerve sheath tumors in schwannomatosis patients (71%), NF1 patients (60%), and NF2 patients (42%). There was no difference in the median number of tumors but there was greater median tumor volume in NF1 patients than in NF2 or schwannomatosis patients. This difference was due to the increased prevalence of larger plexiform tumors in this population. The overall similarity in tumor burden among these conditions is striking given the diverse functions of the NF1, NF2, and SMARCB1 tumor-suppressor genes. Inactivation of NF1 leads to upregulation of RAS signaling [32]; inactivation of NF2 leads to dysregulation of cell surface receptors and intercellular signaling, and to disinhibition of the E3 ubiquitin ligase CRL4(DCAF1) [33,34]; and inactivation of SMARCB1 leads to dysregulation of the SWI-SNF chromatin remodeling complex [35]. Current laboratory models do not explain the precise interaction between these pathways, but can be used to explore whether these separate pathways converge upon a final common pathway in Schwann cells.
Having an accurate phenotype of patients not only provides a basis for future research investigations, but has important implications for clinical management of NF patients. Our data identifies clinical risk factors for internal nerve sheath tumors in the neurofibromatoses; this information builds upon established genetic factors that influence tumor burden, such as NF1 gene microdeletions (increased tumor burden) [36] and mosaicism for the NF2 gene (decreased tumor burden). [27] In NF1 patients, decreasing number of cafe-au-lait macules correlated with the presence, but not volume, of internal neurofibromas (Figure 3). This finding, which was reported recently, [14] was not explained by increasing age since this variable was included in the multivariate model. An inverse relationship between the presence of neurofibromas and café-au-lait macules -the two cardinal features of NF1 -has been reported in rare variants of NF1. Patients with spinal neurofibromatosis have multiple spinal neurofibromas without café-au-lait macules whereas patients with a 3-bp inframe deletion in exon 17 of the NF1 gene have café-aulait macules without neurofibromas. [37] Our data suggests that this inverse relationship may apply to a broader subset of patients than previously recognized. Further research should examine differential effects of germline NF1 mutations on melanocytes and Schwann cells since both are derived from neural crest cells. Increasing numbers of subcutaneous tumors correlated with increasing neurofibroma volume in NF1 patients, corroborating previous reports [38] and presumably reflecting their shared biologic underpinnings with internal tumors. We also found that women with NF1 have greater tumor burden, on average, than men. Although the biological basis for this finding is not known, some studies have suggested that sex hormones may stimulate tumor growth. For example, tumor number and size increase in women during pregnancy, [17,39] and in women using contra-ception with high doses of synthetic progesterone. [40] In addition, laboratory studies have implicated progesterone in neurofibroma progression. [41] Additional studies on the effect of sex hormones on tumor formation and growth are warranted to clarify this finding.
Our data identify scoliosis and decreasing numbers of cutaneous schwannomas as risk factors for internal schwannomas in NF2. While the presence of internal schwannomas correlated with scoliosis, it did not correlate with established markers of disease  severity such as spinal tumors or meningiomas. [15,16] For this reason, we suspect that scoliosis may result from spinal tumors rather than being a predisposing factor. Unexpectedly, wholebody tumor volume in NF2 patients was inversely correlated with the number of cutaneous schwannomas. [15] If this finding is confirmed, it further highlights the importance of different microenviroments in tumor formation. In schwannomatosis patients, no risk factors were identified for the presence of internal schwannomas although increasing age correlated with whole-body tumor volume. This finding may explain, in part, the late age at onset of symptoms compared to NF1 and NF2 [5]. The limitations of our study include the use of a sample of convenience at two large referral centers and the lack of children in the study population. For this reason, our study population does not represent all patients with neurofibromatosis or schwannomatosis. Additionally, coverage of the legs and arms by WBMRI may be incomplete in individuals who are tall or heavy. Our study did not include serial WBMRI scans of individual patients and therefore cannot determine the rate by which tumor count or volume changed with time. Future longitudinal studies of WBMRI in NF patients should quantify changes in tumor burden over time to better understand tumor progression.
In conclusion, in this large prospective international study, we found high rates of internal nerve sheath tumors in neurofibromatosis patients and showed that tumors were non-randomly distributed across body parts. These results provide valuable information about risk factors for internal nerve sheath tumors and raise new biological questions for future research. The addition of WBMRI to careful phenotyping represents a powerful approach to studying hereditary tumor predisposition syndromes and other complex genetic syndromes in humans.