Invasion patterns in brain metastases of solid cancers

Abstract

Brain metastases are a frequent complication in oncology and affect up to 40% of patients with metastatic cancer.¹ While the incidence of brain metastases has shown a constant increase over the last decades, treatment options remain limited and rely mainly on local approaches like surgery, radiosurgery, or whole brain radiotherapy (WBRT).^2–4 Better understanding of the pathobiology of brain metastases may lead to novel treatments.

Brain metastases are usually regarded as growing in a well-delineated fashion within the brain parenchyma.⁵ This notion is based mainly on their neuroradiological presentation with relatively sharp demarcation of contrast-enhancing areas and a generally better delineation than that of malignant gliomas. However, the histological patterns of invasion in brain metastases have so far not been addressed in comprehensive studies, although infiltrative behavior has occasionally been noted.^6,7 Clinically, infiltrative behavior with unclear resection margins is regularly noted by neurosurgeons, and high local recurrence rates after surgery and radiosurgery have been reported for such cases.^8,9

In general, cancer cells grow and invade solid tissues in different ways such as expansive growth, multicellular migration, and individual cell migration.¹⁰ Migration and invasion require complex regulation of specific molecules, including adhesion molecules (eg, integrins, cadherins), cytoskeletal components (eg, actomyosin), proteolytic enzymes (eg, matrix metalloproteases [MMPs]), and others.^10–13 However, the types of invasive behavior of tumor cells have been described mostly in models of non-CNS tissues (eg, skin), and it is unknown whether similar mechanisms are active in the brain, with its distinct microenvironment. The CNS microenvironment differs from that of other solid organs. The brain parenchmya is composed of highly specialized cells (neurons, astrocytes, oligodendrocytes, microglia), and its extracellular matrix (ECM) has a distinct composition. It lacks constituents usually found in solid organs, such as fibronectin and collagen, but it is rich in proteoglycans, tenascin, laminin, heparin/chondroitin/dermatan sulfates, and hyaluronic acid.¹⁴

In this study we systemically characterized the invasion patterns of brain metastases and their correlation with the expression of several adhesion molecules in a series of autopsy specimens. Surgery specimens are not suitable for such studies, since in most cases they include no or only little well-preserved brain tissue around the resection margin and are thus not sufficient for investigation of the invasion front and the interaction of cancer cells with the brain parenchyma.

Materials and Methods

Results

Invasion Patterns of Brain Metastases

Histomorphology

We delineated 3 distinct invasion patterns: 29/57 (50.9%) brain metastases showed a distinct, well-demarcated border to the surrounding brain parenchyma (well-demarcated group); 10/57 (17.5%; Fig. 1A) brain metastases showed distinct perivascular protrusions of multicellular tumor cell formations from the main tumor mass into the brain parenchyma (vascular co-option group; Fig. 1B); and 18/57 (31.6%) brain metastases showed a diffuse infiltration of single tumor cells into the surrounding brain parenchyma (diffuse infiltration group; Fig. 1C). Generally, the invasion pattern was consistent throughout major parts (>90% of the border) of the tumor/brain border in individual metastases.

Open in a separate window

Fig. 1.

Distinct invasion patterns in human brain metastases. Top row: example of well-demarcated invasion (patient 5, NSCLC, Supplemental Table S1): (Aa) H&E slide (magnification ×1.25); (Ab) invasion front of well-demarcated brain metastasis (magnification × 400); (Ac) macroscopic sample of well-demarcated brain metastasis; (Ad) MRI contrast-enhanced T1-weighted sequence sample of well-demarcated brain metastasis. Middle row: example of vascular co-option (patient 13, melanoma, Supplemental Table S1): (Ba) H&E slide (magnification × 1.25); (Bb) CD34 immunohistochemistry of vascular co-option (magnification × 400); (Bc) macroscopic sample of brain metastasis growing via vascular co-option; (Bd) MRI contrast-enhanced T1-weighted sequence sample of brain metastasis growing via vascular co-option. Bottom row: example of diffuse invasion (patient 2, SCLC, Supplemental Table S1): (Ca) H&E of brain metastasis (magnification × 1.25); (Cb) CD18 cytokeratin staining showing diffusely infiltrating tumor cells; (Cc) macroscopic sample of diffuse infiltrating brain metastasis; (Cd) MRI contrast-enhanced T1-weighted sequence sample of diffuse infiltrating brain metastasis. Color figures available from authors upon request.

In 30/57 (52.6%) cases, multiple distinct brain metastases of the same patient were available for investigation, and the invasion type was generally highly congruent among the lesions (P < .001, chi-square test; Table 3).

Table 3.

Concordance between first, second, and third brain metastases

First Brain Metastasis (n = 57)	Well-demarcated (n = 29)		Vascular Co-option (n = 10)		Diffuse Infiltration (n = 18)		Chi-square Test
	n	%	n	%	n	%
Second brain metastasis (n = 30)
Well-demarcated	14	100	0	0	0	0
Vascular co-option	1	17	5	80	0	0	P < .001
Diffuse infiltration	0	0	1	10	9	90
Third brain metastasis (n = 10)
Well-demarcated	3	60	1	20	1	20
Vascular co-option	0	0	3	100	0	0	P = .023
Diffuse infiltration	0	0	0	0	2	100

Open in a separate window

Median maximal measurable invasion distance of tumor cells from the border of the main tumor mass was 68.7 µm (range 12.5–125 µm) in the vascular co-option group and 56.2 µm (range 12.5–450 µm) in the diffuse infiltration group. The maximal measurable invasion distance was not different between the vascular co-option group and the diffuse infiltration group (P = .486, t-test).

Correlation with primary tumor type

Brain metastases of melanoma tended to grow via vascular co-option more often than other primary tumors (4/8). The most frequent primary tumor in the well-demarcated as well as in the diffuse infiltration group was lung cancer. Brain metastases of small cell lung cancer (SCLC) showed frequently a diffuse infiltrative growth (2/3), whereas non-SCLC (NSCLC) grew rather well demarcated (13/24, 54.2%). Squamous NSCLC was more common in the well-demarcated group (4/6), whereas adenocarcinoma NSCLC was equally represented in the well-demarcated (5/12, 41.7%) and the diffuse infiltration group (5/12, 41.7%). See Table 2 for correlation of invasion patterns with primary tumor type.

Correlation with treatment

Complete information on applied therapies after diagnosis of brain metastases was available for 55/57 (96.5%) patients. No statistically significant association was observed between first-line or other brain metastasis treatment and invasion patterns (Table 2). In the diffuse infiltration group, a higher proportion of patients had received WBRT at any time point (4/18; 22%) compared with the well-demarcated group (2/28, 7.1%) or the vascular co-option group (1/10). Further, a higher proportion of patients had received chemotherapy during their course of disease in the diffuse infiltration group (8/17, 47.1%) than in the well-demarcated (8/28, 28.6%) or the vascular co-option group (1/10).

Correlation with clinical characteristics

Survival times from first diagnosis of brain metastases were available in 37 patients. Median survival from diagnosis of brain metastases was 2.0 months in the well-demarcated group (n = 19), 1.8 months in the vascular co-option group (n = 5), and 1.8 months in the diffuse infiltration group (n = 13). There was no statistically significant correlation of invasion pattern with survival time from diagnosis of brain metastases in this small cohort (P = .945, log-rank test). Patients in the well-demarcated group had more often a singular brain metastasis at first diagnosis of brain metastases (52%) than the vascular co-option (33.3%) or diffuse infiltration group (35.7%; P = .483, chi-square test). Extracranial metastases were present in 19/57 (33.3%) patients. No difference in the presence of extracranial metastases was observed between the 3 invasion patterns (P = .781, chi-square test). In the cohort of all 56 patients, there was no significant association of invasion pattern with survival time (P = .825, log-rank test).

Correlation with macroscopic pathology findings and neuroradiology

Illustrative correlations of macroscopic pathology and neuroradiological findings with histological invasion patterns are shown in Fig. 1. The low number of available macroscopic photographs (n = 4) and premortal neuroradiological images (n = 5) precluded systematic correlation with histological findings.

Integrin Expression

General description

Alpha-v integrins showed strong membranous expression on tumor, vascular, and stromal cells in variable fractions of cases. In general, the majority of specimens showed homogeneous αv integrin expression patterns throughout the tumor tissue, except for αvβ8 expression, which was absent in the majority of specimens (Supplemental Table S1). However, regional accentuation of integrin expression was observed in some specimens. Accentuated expression in perivascular tumor cells was observed in 11/57 (19.3%), 9/57 (15.8%), and 4/57 (7.0%) cases for the αv subunit, αvβ6, and αvβ5, respectively. Perinecrotic overexpression of αvβ6, αv subunit, and αvβ5 was found in 4/57 (7.0%), 2/57 (3.5%), and 2/57 (3.5%) cases, respectively (Fig. 2).

Open in a separate window

Fig. 2.

Integrin expression patterns: (A) αv subunit expression (patient 5, NSCLC, Supplemental Table S1); (B) αvβ5 expression with accentuation around a vessel (patient 5, NSCLC, Supplemental Table S1); (C) αvβ5 expression in a brain metastasis growing via vascular co-option (patient 13, melanoma, Supplemental Table S1); (D) αvβ6 expression with accentuation around necrosis, the stars at upper left and lower right mark the necrosis (patient 5, NSCLC, Supplemental Table S1). Color figures available from authors upon request.

Analyzing expression on vascular structures, we observed αvβ5 integrin expression on all (57/57, 100%) vessels, including tumoral and peritumoral vessels as well as the vascular structures of the surrounding brain parenchyma. Alpha-vβ3 expression was not observed on the vascular structures of the surrounding brain parenchyma, except randomly on some larger vessels of the meninges. Prominent expression of αvβ3 was observed on angiogenic, sprouting vessels with multilayered endothelium within the tumor and in the peritumoral area. In 30/57 (52.6%) specimens, immunoreactivity for αvβ3 of the angiogenic tumor vessels was observed, and in 29/57 (50.9%) specimens, angiogenic vessels in the peritumoral area showed specific immunoreactivity for αvβ3. Specific immunoreactivity for the β3 subunit was observed in angiogenic tumor vessels of 46/57 (80.7%) specimens and in angiogenic vessels in the peritumoral area of 45/57 (78.9%) specimens (Fig. 3). The expression detected for the αvβ3 complex was generally lower than for the β3 cytoplasmic domain. The β3 chain is on 2 integrin complexes, αvβ3 and glycoprotein IIbIIIa (the platelet fibrinogen receptor). Tissue localization suggested that the staining of β3 was not due to platelet deposits or aggregates. The αvβ3 antibody used (EM22703) preferentially binds particular ligated conformation of αvβ3, while the cytoplasmic-β3 antibody (EM00212) does not discriminate.¹⁸ This may explain the difference in staining results between these antibodies.

Open in a separate window

Fig. 3.

Integrin expression in vascular structures: (A) αvβ3 expression on tumor vessel, arrow marks a vessel in the peritumoral area without αvβ3 expression (patient 5, NSCLC, Supplemental Table S1); (B) αvβ3 expression in glomeruloid-like vessels of the peritumoral area, the star at lower right marks the tumor tissue, the cross marks the surrounding cerebellum (patient 8, SCLC, Supplemental Table S1); (C) αvβ3 expression in a vessel with vascular co-option (patient 13, melanoma, Supplemental Table S1); (D) CD34 immunohistochemistry of the case shown in (B), illustrating the peritumoral sprouting vessels (patient 8, SCLC, Supplemental Table S1). Color figures available from authors upon request.

Fibrous tumoral stroma was observed in 32/57 (56.1%) specimens and showed expression of the αv subunit (32/32, 100%), αvβ3 (2/32, 6.3%), αvβ5 (28/32, 87.5%), and αvβ6 (6/32, 18.8%).

Relative overexpression of αv integrins at the invasion front was not consistently found, but only in some specimens (αv subunit: 8/57, 14.0%; αvβ5: 8/57, 14.0%; αvβ6: 8/57, 14.0%).

Correlation with invasion patterns

Median H-score of αvβ6 was significantly higher in the well-demarcated group (median 90, range 0–300) than in the vascular co-option group (median 0, range 0–120) and the diffuse infiltration group (median 30, range 0–120; P = .033, Kruskal–Wallis test). No correlation of invasion pattern and median H-score of αv subunit, αvβ3, αvβ5, αvβ8, or β3 subunit was observed.

Correlation with therapy

Brain metastases previously treated with SRS presented more frequently with αvβ5 expression in the tumor cells (6/7) than did specimens without prior SRS (21/49, 42.9%; P = .034, chi-square test). Furthermore, the median αvβ5 H-score was significantly higher in brain metastases with prior SRS (60 vs 0; P = .05, Mann–Whitney U-test). No correlation between SRS treatments and αv subunit, αvβ3, αvβ6, αvβ8, or β3 subunit expression was observed. Prior WBRT or chemotherapy did not show a significant correlation with expression of any of the integrin subunits investigated in this study.

Discussion

Brain metastases are an increasing challenge in oncological practice, as survival in many types of solid cancers is increased by novel treatment strategies. The dominant treatment strategies for brain metastases are local and include surgery and radiosurgery. The benefit from local treatments is likely to be heavily affected by the degree to which macroscopically focal disease is truly focal on a microscopic level. Accordingly, the brain metastasis/brain interface may assume major prognostic significance.

Here, we delineate 3 distinct invasion patterns of brain metastases: well-demarcated growth, vascular co-option, and diffuse infiltration. We found a high fraction of cases showing invasive growth via vascular co-option (18%) or single-cell infiltration (32%). These surprising findings challenge the general notion that brain metastases predominantly grow in an expansive and well-delineated fashion, but they are in good agreement with previous results from experimental studies, smaller and less comprehensive investigations on human tissue samples, and clinical observations.

In half of our cases, we observed expansive growth of an outwardly extending tumor mass within the brain parenchyma. Expression levels of αvβ6 were significantly higher in this group of well-demarcated tumors. Integrin αvβ6 is not expressed in healthy adult epithelia but is upregulated in cancer and has been shown to modulate invasion and inhibit apoptosis.²³ However, the exact role of αvβ6 in cancer pathobiology and in particular in brain metastases remains to be determined.

The vascular basement membrane may act as a guiding track for perivascular growth of cell collectives. Our results indicate that this invasion behavior is not only present in mesenchymal and epithelial tissues but also occurs in the distinct microenvironment of the CNS. In line with previous studies, we observed vascular co-option most commonly in melanoma brain metastases; however, we found it also in other tumor types, such as NSCLC adenocarcinoma.^24–26

Single-cell infiltration into the brain parenchyma of brain-metastatic tumor cells has previously been reported to be characteristic for SCLC.⁷ Our data show that this invasion pattern is also not uncommon in brain metastases of other tumor types, including NSCLC adenocarcinoma/squamous cell carcinoma, breast cancer, and melanoma. The high fraction of brain metastasis cases showing infiltrative growth has implications for local therapy options and highlights the need for including a safety margin beyond the neuroradiologically visible tumor borders.⁶ In our study, depth of invasion into the CNS parenchyma from the main tumor mass reached up to 450 µm. Of note, SRS uses no margins for treatment, but at 0.4 mm from the prescription isodose line, nearly a full dose is delivered, and therefore the magnitude of invasion has no clear consequence on SRS practice. The selection of patients in need of an extended local treatment approach is challenging, as the current neuroradiological techniques cannot precisely visualize the invasion distance of a given tumor. However, we previously demonstrated that the extent of peritumoral brain edema might function as a surrogate marker for infiltrative tumor growth, as little brain edema was significantly more common in infiltrative brain metastases and correlated with impaired patient survival times.⁸ Further prospective studies need to address the prognostic implications of brain metastasis invasion patterns in more detail.

We did not observe a statistically significant correlation of treatment modality with invasion pattern in our cohort. However, a higher proportion of patients in the diffuse infiltration group had received prior radiation or chemotherapy. As our sample size is not adequate for firm statistical conclusions, we cannot exclude that radio- or chemotherapy may select for or produce tumor cells with a higher infiltrative potential, similarly to some observations in primary tumors and gliomas.²⁷ Interestingly, brain metastasis lesions previously treated with SRS showed higher expression of αvβ5, a finding that is well in line with previous reports showing that this molecule is essential for tumor growth in preirradiated stroma.²⁸ Further, we observed a high consistency of invasion behavior between the growth patterns of different brain metastases in individual patients. This again may indicate that intrinsic molecular features of metastases originating from a given primary tumor correlate with certain invasion patterns.

Our results have to be interpreted with caution because the small sample sizes often do not allow firm statistical conclusions. Herein, we concentrated on the descriptive presentation of our results. However, it has to be taken into account that autopsy samples of brain metastases are very rare, and our series displays a rather large cohort compared with previously published studies on autopsy specimens.^6,7

We noted prominent expression of αv integrins in many brain metastasis cases. This underscores the important function of this class of molecules in metastatic cancer. Currently, several integrin inhibitors are under clinical development, and promising activity was shown in some of the most frequent primary tumors of brain metastases such as NSCLC and melanoma.^29–31 Interestingly, the anti–αv-integrin antibody intetumumab reduced brain metastasis outgrowth in mice after intracarotid infusion of brain-seeking human epidermal growth factor receptor 2–positive breast cancer cells.³² Thus, clinical trials specifically investigating the potential of integrin inhibitors for prophylaxis and treatment of brain metastases seem warranted.

Supplementary Material

Supplementary material is available online at Neuro-Oncology (http://neuro-oncology.oxfordjournals.org/).

Funding

The costs for this project were covered by the research budget of the Medical University of Vienna and the University of Tübingen.

Conflict of interest statement. S.L.G. is an employee at Merck-Serono and has a patent application referring to the anti-integrin antibodies used here. M.W. has received research support and honoraria for lectures and service on its advisory board from Merck-Serono. All other authors declare no conflict of interest.

Supplementary Material

Acknowledgments

References