Human and nonhuman primate meninges harbor lymphatic vessels that can be visualized noninvasively by MRI

1Recent reports described the existence of a network of true lymphatic vessels within the mammalian dura mater that runs alongside blood vessels, notably the superior sagittal and transverse sinuses. The dural lymphatic vessels display typical immunohistochemical markers that identify lymphatic vessels elsewhere in the body. They provide an alternate conduit for drainage of immune cells and cerebrospinal fluid (CSF) from the brain, beyond previously described pathways of flow: via arachnoid granulations into the dural venous sinuses, and via the cribriform plate into the ethmoid region. Although early reports, based on injections of India ink into the cisterna magna of the rat, suggested that the dural pathway accounts for only a minority of the drainage, the more recent studies, which are based on injections of fluorescent tracers and in vivo microscopy, indicate that the dural system may be substantially more important for drainage of macromolecules and immune cells than previously realized.

2Whether a similar network of dural lymphatics is present in primates remains unknown. Moreover, noninvasive visualization of the dural lymphatics – a necessary first step to understanding their normal physiology and potential aberrations in neurological diseases – has not been reported. We therefore verified pathologically the existence of a dural lymphatic network in human and nonhuman primates (common marmoset monkeys) and evaluated two magnetic resonance imaging (MRI) techniques that might enable its visualization in vivo. First, the T2-weighted fluid-attenuation inversion recovery (T2-FLAIR) pulse sequence, which is the clinical standard for detecting lesions within the brain parenchyma, is highly sensitive to the presence of gadolinium-based contrast agents in the CSF. Second, ‘black-blood’ imaging sequences, which are typically used for measurement of vascular wall thickness or detection of atherosclerotic plaque, are tuned to darken the contents of blood vessels (even when they contain a gadolinium-based contrast agent), but in the process the images may highlight vessels with other contents and flow properties. For comparison, we also acquired a postcontrast T1-weighted Magnetization Prepared Rapid Acquisition of Gradient Echoes (MPRAGE) MRI sequence, which is widely implemented for structural brain imaging and depicts avid enhancement of dura mater and blood vessels, but which would not be expected to discriminate lymphatic vessels.

3Cerebral blood vessels have a highly regulated blood-brain barrier, protecting the neuropil from many contents of the circulating blood. Under physiological conditions, the blood-brain barrier prevents gadolinium-based chelates in standard clinical use from passing into the Virchow-Robin perivascular spaces and parenchyma, so that these structures do not enhance on MRI. On the other hand, dural blood vessels lack a blood-meningeal barrier, enabling leakage of circulating fluids and small substances, including gadolinium-based compounds. This explains the thin, though often incomplete, dural enhancement that is seen on conventional T1-weighted MRI scans under physiological conditions (Figure 1), as well as its abnormal diffuse or localized thickening in a variety of pathological conditions. Using high-resolution (in-plane resolution 270 × 270 μm or finer) T2-FLAIR and T1-weighted black-blood MRI images, obtained after the intravenous injection of standard FDA-approved contrast material (gadobutrol), we were able to visualize the collection of interstitial gadolinium within dural lymphatic vessels (maximum apparent diameter ~1 mm) in 5/5 human healthy volunteers and 3/3 common marmoset monkeys (Figure 1). Our results suggest that in the dura, similar to many other organs throughout the body, small intravascular molecules extravasate into the interstitium and then, under a hydrostatic pressure gradient, collect into lymphatic capillaries through a loose lymphatic endothelium.

4To further test this hypothesis, meningeal lymphatics were also assessed using a second gadolinium-based contrast agent, gadofosveset, a blood-pool contrast agent suitable for angiography. Gadofosveset binds reversibly to serum albumin, increasing its molecular weight from 0.9 to 67 kDa. Under physiological conditions, albumin has a low transcapillary exchange rate into the interstitial compartment, estimated to be on the order of 5% per hour, which explains the propensity of gadofosveset to remain within blood vessels. In both species, gadofosveset did not reveal dural lymphatics, especially on T1-black blood images (Figure 2 and Figure 2—figure supplement 1). As expected, on T1-weighted MPRAGE images, gadofosveset provided superior intravascular enhancement, in both meningeal and parenchymal blood vessels, compared to gadobutrol (Figure 2).

5On 3D-rendering of subtraction MRI images (Videos 1–2, Figure 1—figure supplement 1), dural lymphatics are seen running parallel to the dural venous sinuses, especially the superior sagittal and straight sinuses, and alongside branches of the middle meningeal artery. The topography of the meningeal lymphatics fits with the previously described network in rodents as well as our neuropathological data (Figures 3 and 4). It is worth noting that the lymphatics visualized by MRI are large slow-flow lymphatic ducts, whereas blind-ending and small lymphatic capillaries, clearly seen by histopathology (Figure 3 and Figure 3—figure supplement 1), are unlikely to be revealed by MRI. The induction of experimental autoimmune encephalomyelitis (EAE) did not affect detection of dural lymphatic vessel in either of the two animals that we tested (not shown).

6To support our in vivo imaging results, we further investigated the existence and topography of lymphatics in coronal and longitudinal sections of human and marmoset dura mater. To accomplish this, we tested a variety of putative lymphatic endothelial markers and found that selective double immunostaining for D2-40 podoplanin/CD31 and for PROX1/CD31 was the most effective strategy in discriminating lymphatic vs. venous blood vessels in dura samples – a challenging task since lymphatics sprout from transdifferentiation of venous endothelium and persistently share some endothelial markers. A branched network of lymphatics was clearly seen within the dura mater. On D2-40 podoplanin/CD31 double staining, we identified a total of 93 human dural lymphatics; most were collapsed, explaining the large range of maximum transverse diameters (range = 7–842 μm, mean = 125 μm, standard deviation = 161 μm). The density of dural lymphatics was higher around the venous sinuses than in more lateral areas of the dura, and higher within the meningeal layer than the periosteal layer of the dura. As expected, red blood cells were not seen within lymphatics. In marmosets, direct comparison between in vivo MRI and histopathology was performed. As shown in Figure 4 and Figure 4—figure supplement 1, the three dural vessels detected on coronal postcontrast T2-FLAIR and on subtraction images colocalize with three clusters of dural cells expressing the full panel of lymphatic endothelial markers (LYVE-1, D2-40 podoplanin, PROX1, COUP-TFII) and CCL21, a chemokine implicated in lymphatic transmigration.

7In the ongoing debate about the precise localization of lymphatics within the meninges (either completely within the dura or ‘shared’ between the dura and the arachnoid), our pathological data clearly show that at least some lymphatics are contained entirely within the dura (Figures 3 and 4). On limited evaluation, we were unable to visualize lymphatics within the leptomeninges, but additional dedicated studies, ideally with nonconventional tissue-preparation methods, are warranted to fully explore this possibility. A comprehensive map of the meningeal lymphatic network would have implications for unraveling the ways in which the meningeal lymphatics participate in waste clearance and in immune cell trafficking within the central nervous system.

8In inflammatory pathological conditions, cellular migration toward the dural lymphatics might be profoundly enhanced by specific signaling and lymphatic plasticity. Noteworthy in this context are clusters of extravascular CD3+ lymphocytes and CD68+ phagocytic meningeal macrophages that we observed in the dura of several multiple sclerosis autopsies (not shown), confirming intense immune cell trafficking and communication. Indeed, without proper normative comparison, we cannot rule out that the extensive presence of small lymphatics observed in multiple sclerosis dura samples is the result of inflammation-mediated lymphoangiogenesis. On the other hand, lymphatic dysfunction might impair waste clearance in neurodegenerative diseases and aging, in line with the recently captured deposition of β-amyloid in human dura in elderly people.

9Differently from experiments implementing injections of tracers within brain structures, here we aimed primarily to image dural lymphatic vessels in human and nonhuman primates, but we could not prove whether dural lymphatic vessels drain immune cells, CSF, or other substances from the brain to deep cervical lymph nodes, nor could we assess any link with the glymphatic system. Such an analysis would probably require injection of specific MRI-detectable tracers and acquisition of MRI time-series (not thoroughly explored in the current work), an important future research direction for the nonhuman primate work.

10Overall, our data clearly and consistently demonstrate the existence of lymphatic vessels within the dura mater of human and nonhuman primates. Together with recent studies in rodents, our results show that the meningeal lymphatic system is evolutionarily conserved in mammals and confirm, after exactly two centuries, what the Italian anatomist Paolo Mascagni speculated were lymphatic vessels at the surface of human brain. The ability to image the meningeal lymphatics noninvasively immediately suggests the possibility of studying potential abnormalities in human neurological disorders.

11We carried out human studies under a protocol (NCT02504840) approved by NIH Institutional Review Board. Informed consent was obtained from all participants. Formalin-fixed human brains were attained at autopsy after obtaining consent from the next of kin. Animal studies were performed under a protocol approved by the Institutional Animal Care and Use Committee.

12We studied five healthy volunteers (three women, age range 28–53 years) and obtained scans on a 3-tesla MRI unit (Skyra, Siemens Healthcare, Erlangen, Germany), using the body coil for radiofrequency transmission and a 32-element phased array coil for reception.

13Prior to injection of gadolinium-based contrast agent, the following high-resolution MRI sequences were collected:

14Whole-brain T1-Magnetization Prepared Rapid Acquisition of Gradient Echoes (MPRAGE, sagittal 3D turbo-fast low angle shot [TFL] sequence, acquisition matrix 256 × 256, isotropic resolution 1 mm, 176 slices, repetition time [TR]/echo time [TE]/inversion time [TI]=3000/3/900 ms, flip angle 9, acquisition time 5 min 38 s);

15Limited T2-weighted Fluid Attenuation Inversion Recovery (FLAIR, coronal 2D acquisition over the superior sagittal sinus, field-of-view 256 × 256, 22 slices, reconstructed in-plane resolution 0.25 mm x 0.25 mm, 42 contiguous 3 mm slices, TR/TE/TI = 6500/93/2100 ms, echo train length 17, bandwidth 80 Hz/pixel, acquisition time 5 min), optimized for detection of gadolinium-based contrast agent in the subarachnoid space;

16Black-blood scan (coronal 2D acquisition, Sampling Perfection with Application optimized Contrasts using different flip angle Evolution [SPACE] sequence, field-of-view 174 × 174, matrix 320 × 320, reconstructed in-plane resolution 0.27 × 0.27 mm, 64 contiguous 0.5 mm sections, TR/TE = 938/22 ms, echo train length 35, bandwidth 434 Hz/pixel, acquisition time 7 min 50 s). A series of 2 or three overlapping coronal acquisitions were acquired to cover most of the cerebral hemispheres;

17Whole-brain T2-FLAIR scan (coronal 3D acquisition, SPACE sequence, field-of-view 235 × 235, matrix 512 × 512, reconstructed in-plane resolution 0.46 × 0.46 mm, 176 1 mm sections, TR/TE/TI = 4800/354/1800 ms, nonselective inversion pulse, echo-train length 298, bandwidth 780 Hz/pixel, acceleration factor 2, acquisition time 14 min);

18Whole-brain T1-SPACE (axial 3D acquisition, acquisition matrix 256 × 256, isotropic resolution 0.9 mm, 112 sections, TR/TE = 600/20 ms, flip angle 120, echo-train length 28, acquisition time 10 min).

19We repeated these acquisitions after injection of gadobutrol (Gadavist, Bayer HealthCare, Whippany, NJ) in all five participants. In 2 of the participants, we also repeated the entire protocol prior to and following intravenous injection of gadofosveset (Ablavar, Lantheus Medical Imaging, N Billerica, MA). Injections followed the manufacturer’s suggested dosing (0.1 mmol/kg for gadobutrol, 0.03 mmol/kg for gadofosveset).

20We studied three healthy adult common marmosets (Callithrix jacchus) (one female, two males, age range 4–11 years). After the baseline MRI, experimental autoimmune encephalomyelitis (EAE) was induced in 2 marmosets with 0.2 mg of fresh-frozen human white matter homogenate as previously described. Marmosets were placed in a sphinx position within the magnet, and scans were obtained on a 7-tesla MRI unit (Avance AVIII, Bruker BioSpin, Billerica, MA, USA). Data acquisition was performed in transmit-only/receive-only mode using a homemade linear birdcage coil (120 mm inner diameter) as a radiofrequency transmitter and a homemade, 8-channel radio-frequency surface-array receiver coil assembly placed over the head of the animal. Prior to injection of contrast agent, we collected:

21Whole-brain T1-weighted MPRAGE scan (coronal 3D acquisition, Modified Driven Equilibrium Fourier Transform [MDEFT] sequence, in-plane voxel size 0.15 mm x 0.15 mm, 36 contiguous 1 mm sections, TR/TE/TI = 12.5/4/1200 ms, flip angle 12 degrees, 2 segments of 1800 ms, acquisition time 7 min);

22Whole-brain T2-FLAIR scan (coronal 2D acquisition, Rapid Acquisition with Relaxation Enhancement [RARE] sequence, voxel size 0.15 mm x 0.15 mm, 36 contiguous 1 mm sections, TR/TE/TI = 10,000/36/2500 ms, flip angle 90–180 degrees, 2 averages, acquisition time 13 min).

23We performed the same scans following intravenous injection of gadobutrol and gadofosveset, in two different MRI sessions. Injections used single (0.1 mmol/kg for gadobutrol, 0.03 mmol/kg for gadofosveset) or triple the recommended human dosing (0.3 mmol/kg for gadobutrol, 0.09 mmol/kg for gadofosveset).

24Scanner-generated DICOM images were converted into NIFTI files for postprocessing. Using MIPAV software (https://mipav.cit.nih.gov), precontrast scans were rigidly registered to postcontrast scans, and subtraction images were created for anatomical identification of dural lymphatic vessels. 3D skull-stripped subtraction images were then imported into Osirix software for maximum intensity projection (MIP) 3D rendering. The same postprocessing was performed for scans acquired after gadofosveset injection, and a direct comparison between the two gadolinium-based contrast agents was made.

25Neuropathological evaluation focused on human and marmoset brain dura mater samples. Multiple human dura samples were obtained from 2 formalin-fixed brains (60- and 77-year-olds with longstanding progressive multiple sclerosis) and from a 33-year-old with refractory epilepsy undergoing anterior temporal lobectomy (Figure 3—source data 1). Primate samples were obtained from 3 adult common marmosets (2 with EAE) [Figure 4—source data 1]. After general anesthesia and transcardial perfusion of 4% paraformaldehyde, marmoset brains were extracted and stored in 10% formalin. For primate brains, MRI-matched histological sections were achieved via 7T MRI of the fixed brain and subsequent gross sectioning with an individualized, MRI-designed, 3D-printed cutting box, as previously described.

26Coronal and longitudinal 3 to 7 μm paraffin sections of the dura mater were stained with hematoxylin and eosin (H and E). Immunohistochemical and immunofluorescent analysis for classical lymphatic endothelial cell markers [lymphatic vessel endothelial hyaluronan receptor 1 (LYVE-1), podoplanin (D2-40), prospero homeobox protein 1 (PROX1), COUP transcription factor 2 (COUP-TFII) and CCL21], were performed on representative slides. Selective double staining for lymphatic vs. vascular endothelial markers (D2-40/CD31 and PROX1/CD31) were also executed on representative sections and implemented for histological quantification; the maximum diameter was recorded for each lymphatic vessel, and descriptive statistics were computed. We also stained for lymphocytes (CD3) and monocytes/macrophages (CD68). Five-μm paraffin sections of the human skin were used as positive controls for lymphatic endothelium cell marker assessment (Figure 3—figure supplement 2). For comparison, 10 paraffin human brain tissue blocks, selected for the presence of nearly intact leptomeningeal architecture, were implemented to assess the presence of lymphatics within the subarachnoid space and cerebral parenchyma.

27Source, antibody type, and dilution are indicated sequentially as follows: LIVE-1 (Abcam, UK, rabbit polyclonal, 1:200); podoplanin D2-40 (AbD Serotec, Hercules, CA, mouse monoclonal, 1:50); PROX-1 (AngioBio, San Diego, CA, rabbit polyclonal, 1:50); COUP TFII (R&D Systems, Minneapolis, MN, mouse monoclonal, 1:200); CCL21 (Abcam, rabbit polyclonal, 1:200); CD31 (Abcam, rabbit polyclonal, 1:50); CD68 KP1 (Thermofisher, Waltham, MA, mouse monoclonal, 1:100); CD3 (Dako, Santa Clara, CA, rabbit polyclonal, 1:200); goat anti-mouse Alexafluor 488 IgG (Invitrogen, Carlsbad, CA, 1:250) and goat anti-rabbit Alexafluor 594 IgG antibodies (Invitrogen, 1:250).

28Immunostaining was performed on 3 to 7 μm-thick consecutive paraffin sections with antibodies to LYVE1, D2-40 podoplanin, PROX1, COUP-TFII, CCL21, CD31, CD68, CD3 and visualized with 3,3'-diaminobenzidine (DAB). Briefly, after deparaffinization, sections were rinsed in triphosphate-buffered saline (TBS) for 10 min each, processed for antigen retrieval, and treated in 3% hydrogen peroxide for 10 min. Sections were blocked with a protein-blocking solution for 20 min and incubated with the primary antibodies overnight at 4°C or 1 hr at room temperature, then rinsed and incubated with secondary antibodies for 30 min. The immunoreaction was visualized with DAB. After washing, sections were counterstained with 10% hematoxylin. Stained sections were visualized and digitized (Axio Observer Z.1, Carl Zeiss Microscopy, NY, USA). Selective double staining for lymphatic and vascular endothelium markers (D2-40 podoplanin/CD31, PROX1/CD31), as well double staining for D2-40 podoplanin/PROX1,were also executed on representative sections. Double staining was performed using, in sequence, DAB horseradish peroxidase (HRP) and Vector Blue alkaline phosphatase (AP) methods. As negative control, sections of human skin, human dura mater, and marmoset brain were stained without the primary antibodies; no background staining and/or nonspecific binding was noted (Figure 3—figure supplement 2).

29After deparaffinization, sections were treated using the same method as described above. Sections were incubated with a cocktail of antibodies to D2-40 podoplanin and CD31 for 2 hr at room temperature, then rinsed and incubated with secondary antibodies with Alexafluor 488 goat anti-mouse (Invitrogen, dilution 1:250) and Alexafluor 594 goat anti rabbit IgG antibodies (Invitrogen, dilution 1:250), respectively, for 30 min at room temperature. After washing, sections were mounted with DAPI Immuno Mount. Stained sections were visualized and digitized (Axio Observer Z.1, Carl Zeiss Microscopy, NY, USA).