Authors: Marina Wakid, Daniel Almeida, Zahia Aouabed, Reza Rahimian, Maria Antonietta Davoli, Volodymyr Yerko, Elena Leonova-Erko, Vincent Richard, René Zahedi, Christoph Borchers, Gustavo Turecki, Naguib Mechawar
Categories: Study Protocol, Blood-Brain Barrier, LC-MS/MS, Microvessels, Neurovascular unit, Postmortem, RNA sequencing
Source: Brain, Behavior, & Immunity - Health
The neurovascular unit, comprised of vascular cell types that collectively regulate cerebral blood flow to meet the needs of coupled neurons, is paramount for the proper function of the central nervous system. The neurovascular unit gatekeeps blood-brain barrier properties, which experiences impairment in several central nervous system diseases associated with neuroinflammation and contributes to pathogenesis. To better understand function and dysfunction at the neurovascular unit and how it may confer inflammatory processes within the brain, isolation and characterization of the neurovascular unit is needed. Here, we describe a singular, standardized protocol to enrich and isolate microvessels from archived snap-frozen human and frozen mouse cerebral cortex using mechanical homogenization and centrifugation-separation that preserves the structural integrity and multicellular composition of microvessel fragments. For the first time, microvessels are isolated from postmortem ventromedial prefrontal cortex tissue and are comprehensively investigated as a structural unit using both RNA sequencing and Liquid Chromatography with tandem mass spectrometry (LC-MS/MS). Both the transcriptome and proteome are obtained and compared, demonstrating that the isolated brain microvessel is a robust model for the NVU and can be used to generate highly informative datasets in both physiological and disease contexts.
Keywords: Neurovascular unit, Blood-Brain Barrier, Microvessels, Postmortem, RNA sequencing, LC-MS/MS
To meet the metabolic needs of the 86 billion neurons in the human brain, an elaborate 400 mile-long microvascular network (Iadecola, 2017; Kisler et al., 2017) supplies blood flow to the deep structures of the cerebral hemispheres and gatekeeps blood-brain barrier (BBB) properties. Extensive research efforts have underscored the BBB as a highly selective cellular system. Ultrastructurally, the BBB consists of continuous non-fenestrated brain microvascular endothelial cells (BMECs) that precisely regulate movement between the blood and brain interface through the expression of specialized solute carriers and efflux transporters (Daneman, 2012; Betz and Goldstein, 1978; Betz et al., 1980; Cordon-Cardo et al., 1989; Thiebaut et al., 1989; Loscher and Potschka, 2005a; Mittapalli et al., 2010; Zlokovic, 2008). BMECs, along with astrocytic endfeet and mural cells (pericytes or smooth muscle cells), positioned at the vascular basement membrane, are the cell types that comprise the neurovascular unit (NVU) (McConnell et al., 2017) and work in concert to implement coordinated vascular responses to central and peripheral signals. Such responses include the continuous delivery of oxygen and glucose to neurons (Mintun et al., 2001; Hoge et al., 1999; Fox et al., 1988; De Vivo et al., 1991; Farrell and Pardridge, 1991; Gerhart et al., 1989), homeostatic maintenance of the brain (Jeong et al., 2006; Gendelman et al., 2009; Wilhelm et al., 2007; Mokgokong et al., 2014; Smith and Rapoport, 1986), the regulation of cerebral blood flow (Mathiesen et al., 1998; Caesar et al., 2003; Iadecola, 1993; Roy and Sherrington, 1890; Fergus and Lee, 1997) and clearance of interstitial fluid (Verheggen et al., 2018; Deane et al., 2009; Iliff et al., 2012; Iliff et al.). An underappreciated lens through which to investigate disease, NVU dysfunction contributes to cognitive decline in Aβ and tau pathology (Iturria-Medina et al., 2016; Montagne et al., 2015; Sweeney et al., 2015; Arvanitakis et al., 2016; Toledo et al., 2013; Rosenberg, 2014), traumatic brain injury (Stein et al., 2002; Schwarzmaier et al., 2010; Dietrich et al., 1994; del Zoppo and Mabuchi, 2003; Schroder et al., 1998; von Oettingen et al., 2002; Shapira et al., 1993; Baldwin et al., 1996; Hicks et al., 1997; Baskaya et al., 1997), perivenous myelin lesions presented in multiple sclerosis (Gaitan et al., 2013; Buch et al., 2021; Geraldes et al., 2020; Al-Louzi et al., 2022; Khan et al., 2010; Doepp et al., 2011), as well as multiphasic changes in BBB permeability after stroke (Liu et al., 2018; Lin et al., 2008; Strbian et al., 2008; Durukan and Tatlisumak, 2009; Pillai et al., 2009). Disease states may arise when BBB function can no longer match the needs of the central nervous system (CNS), which confers dire consequences for the ability of the BBB to communicate both with cells within the brain parenchyma and with cells in the periphery. Indeed, the BBB acts as the interface between the brain and peripheral systems through which neuroimmune interactions occur (Quan and Banks, 2007)and is highly responsive to immune activity encroaching the brain. Such interactions include, but are not limited regulation of major efflux transporter P-glycoprotein 1 (ABCB1), endothelial Toll-like receptor and NOD-like recepto activation by TNF-α signaling (Erickson and Banks, 2018; Nagyoszi et al., 2010, 2015), as well as modulation of Na–K–Cl cotransporter, which is critical for cerebral ionic homeostasis, by IL-6 secretion from astrocytes (Sun et al., 1997). The BBB itself secretes substances that interact with the neuroimmune system, including cytokines, prostaglandins, and nitric oxide (Fabry et al., 1993; Mandi et al., 1998; McGuire et al., 2003; Reyes et al., 1999). These substances may be constitutively expressed or pathologically induced, as shown by interleukin-8 in HIV infection (Hofman et al., 1999), invasion of autoreactive CD4^+^ T cells in multiple sclerosis (Traugott et al., 1983), microglial release of TNFα with cocaine exposure (Lewitus et al., 2016), and nitric oxide release in Alzheimer's disease (Dorheim et al., 1994). Because of its bipolar nature, dysfunction of the BBB can arise from, and be further aggravated by, either the CNS or peripheral compartments. Evidence of neurovascular dysfunction has similarly been observed in bipolar disorder (Kamintsky et al., 2020), schizophrenia (Kirkpatrick and Miller, 2013; Axelsson et al., 1982; Campana et al., 2023; Goldwaser et al., 2022), major depressive disorder (Torres-Platas et al., 2014; Gal et al., 2023; Najjar et al., 2013), and Parkinson's disease (Al-Bachari et al., 2020; Fowler et al., 2021).
Our understanding of neurovascular development and function has been advanced largely by mouse models. Functional characteristics of the BBB are regulated at the transcriptomic level and, in recent years, different methodologies have been employed to investigate the neurovascular transcriptome. Single-cell sequencing studies have leveraged transgenic-reporter claudin-5-GFP (Vanlandewijck et al., 2018; He et al., 2018), Tie2–eGFP (Zhang et al., 2014), and Pdgfrb-eGFP (He et al., 2016a) mouse lines in conjunction with fluorescence-activated cell sorting (FACS) to generate highly informative transcriptomic datasets of mouse BMECs and other vascular cell types, whereas other studies have carried out RNA sequencing of BMECs isolated from Rosa-tdTomato; VE-Cadherin-CreERT2 mouse models of stroke, multiple sclerosis, traumatic brain injury and seizure (Munji et al., 2019). While past mouse data have provided precious insight into defining core NVU gene expression and underscores the relevance of transcriptomic profiling for better understanding neurovascular function (and dysfunction), recent breakthroughs demonstrate that there are numerous species-specific differences between mouse and human neurovasculature, including solute carrier and efflux transporter expression (Munji et al., 2019; Garcia et al., 2022; Yang et al., 2022). Such findings reveal the partial utility of animal models for studying disease of the human CNS. Due to the scarcity of well-preserved human brain tissue available for research, transcriptomic profiling in human brain samples has been considerably more limited, leaving the investigation of vascular cells neglected in favour of non-vascular cell types, such as neurons and oligodendrocytes. In addition, single-cell or single-nucleus sequencing used to profile expression in all cell populations yield very low populations of endothelial cells and pericytes from human adult and embryonic cortex samples depiste vessel density ranging between 361 and 811 vessels/mm^2^ (Klein et al., 1986; Wu et al., 2004; Weber et al., 2008) at an overall endothelial cell density of 4504 ± 2666 cells/mm^2^ (Ventura-Antunes et al., 2022). Such techniques seem to deplete vascular cells/nuclei for reasons that are not understood and have impeded analysis of human neurovascular transcriptomes (Nagy et al., 2020; Velmeshev et al., 2019; Mathys et al., 2019; Grubman et al., 2019; Jakel et al., 2019). Recently, detailed transcriptome-wide atlases of human and mouse brain vascular nuclei were generated by two independent groups (Garcia et al., 2022; Yang et al., 2022), both addressing the underrepresentation of vascular cell types and elaborating on species-specific differences in NVU gene expression. Such progress and tools deepen our understanding of human NVU function, yet certain limitations that persist challenge further progress in the single-cell and single-nucleus sequencing remains inaccessible to many due to high costs and lack of bioinformatic expertise. Moreover, there is heavy reliance on brain banks for well-characterized frozen human brain tissue, which also requires considerable adaptation of techniques initially optimized for fresh tissue and creates even further disparity in how mouse and human brain tissue are utilized, even with the same experimental question in mind. The biological and bioinformatic biases these experimental decisions create and their extent are unknown. Understanding the molecular mechanisms of NVU dysfunction can be achieved by examining gene expression changes in brain microvessels in different disease contexts; and the limited number of existing human neurovascular datasets motivates transcriptomic characterization of more human samples. While it is understood that BMECs perform the BBB function and that other vascular-associated cell types critically regulate that function, the study of microvessels as a preserved unit provides greater insight into the neurovasculature in a manner that dissociated cell types cannot. To this aim, we use a singular, standardized protocol to enrich and isolate microvessels from archived snap-frozen human and frozen mouse cerebral cortex using mechanical homogenization and centrifugation-separation that is gentle enough to dissociate brain tissue while preserving the structural integrity and multicellular composition of microvessel fragments.
The common issue of multiple or contaminating cell types in samples used for tissue-derived RNA sequencing has been largely eliminated by single-cell workflows. However, single-cell and single-nucleus workflows introduce other significant measurements typically suffer from large fractions of observed zeros, possibly due to technical limitations or randomness (Hicks et al., 2018; Bacher and Kendziorski, 2016). Moreover, tissue dissociation and storage biases can induce unwanted transcriptomic alterations and cell type composition differences (Denisenko et al., 2020). Because of this, bulk and single-cell sequencing are complementary strategies in which the former approach warrants a versatile and effective method for isolating the NVU from human brain. Several approaches attempting to investigate the isolated NVU have been developed in the past, albeit with major FACS-sorting with antibodies against PECAM1/CD31 and CD13 (targeting the endothelial cell membrane and pericyte cell membrane, respectively) require fresh brain tissue (Yang et al., 2022), as do other iterations of microvessel isolation for cell culture expansion (Navone et al., 2013; van Beijnum et al., 2008). Opting instead for selective capture of endothelial and other vascular-associated cells from frozen human brain by laser capture microdissection (LCM) (Kinnecom and Pachter, 2005; Mojsilovic-Petrovic et al., 2004; Harris et al., 2008; Song et al., 2020a) demands considerable optimization if microvessels are to be used for high-throughput applications downstream (Almeida and Turecki, 2022). Finally, past attempts at microvessel isolation from frozen brain homogenates have only yielded samples suitable for qPCR and Western blot (Bourassa et al., 2019), and more comprehensive knowledge obtained from high-throughput techniques is currently lacking from such studies. Critically, microvessels isolated using the described method are in high yield, possess all major vascular-associated cell types, and maintain their in situ cellular structure, making them suitable for characterization using high-throughput techniques. The advantages of this simple protocol are it does not require the experimental setup needed by single-nucleus sorting, nor does it require transgenic mice (Lee et al., 2019), enzymatic dissociation (Crouch and Doetsch, 2018; Lee et al., 2019; Spitzer et al., 2023), or fresh brain tissue (Crouch and Doetsch, 2018; Lee et al., 2019; Spitzer et al., 2023). Importantly, this is the first standardized microvessel isolation method demonstrated to work with snap-frozen brain tissue and that is compatible across high-throughput downstream applications, removing unknown biases introduced by the use of varied isolation methods.
We have successfully applied the described protocol to postmortem ventromedial prefrontal cortex (vmPFC) tissue from individuals having died suddenly with no neurological or psychiatric disorder as well as mouse forebrain tissue, as proof of concept that the same procedure can be used in both species. To demonstrate the utility of microvessels isolated from postmortem vmPFC tissue, we processed 5 samples of extracted total RNA and 3 samples of extracted total protein using RNA sequencing and liquid chromatography with tandem mass spectrometry (LC-MS/MS), respectively. Moreover, we sorted endothelial nuclei from isolated microvessels using fluorescence-activated nuclei sorting (FANS) as proof of concept that specific neurovascular cell types may be further purified if needed. Bioinformatic processing and analysis of human transcriptomic and proteomic data indicated that isolated samples showed major enrichment for BMEC, pericyte, SMC, and astrocytic endfeet components at both the mRNA and protein level, generating the first multi-omic datasets from human brain microvessels.
This study was approved by the Douglas Hospital Research Ethics Board and written informed consent from next of kin was obtained for each individual included in this study. For each individual, the cause of death was determined by the Quebec Coroner's Office and medical records were obtained. Samples were obtained from Caucasian individuals having died suddenly with no neurological or psychiatric disorder (Table 1). Postmortem brain tissues were provided by the Douglas–Bell Canada Brain Bank (www.douglasbrainbank.ca). Frozen grey matter samples were dissected from the vmPFC (Brodmann area 11) by expert brain bank staff stored at −80 °C. The postmortem interval (PMI) is a metric for the delay between an individual's death, the collection and processing of the brain. To assess RNA quality, RNA integrity number (RIN) was measured for brain samples using tissue homogenates, with an average value of 5.34. A total of 5 subjects were subjected to RNA-sequencing and 3 subjects were subjected to Liquid Chromatography with tandem mass spectrometry (LC-MS/MS).
Note: All experiments involving the use of human brain samples must be performed in accordance with the relevant institutional and national regulations. Use of postmortem tissues was approved by the Institutional Review Board of the Douglas Hospital.
Male C57BL/6J mice (n = 2) aged between 120 and 126 days of age were bred, housed, and cared for in accordance with the Canadian Council on Animal Care guidelines (CCAC; http://ccac.ca/en_/standards/guidelines), and all methods were approved by the Animal Care Committee from the Douglas Institute Research Center under protocol number 5570. Mice were housed in standard conditions at 22 ± 1 °C with 60% relative humidity, and a 12-h light-dark cycle with food and water available ad libitum (Isingrini et al., 2017). Following pertinent guidelines and regulations, the mice were anesthetized via intraperitoneal injection of ketamine (10 mg/ml)/xylazine (1 mg/ml) and transcardially perfused with cold PBS 1X. The frontal cortices were removed and immediately frozen in liquid nitrogen and then stored at −80 °C.
Prepare fresh Homogenization Buffer (1M sucrose + 1% BSA dissolved in DEPC-treated water) and keep at 4 °C until thoroughly chilled. If transcriptomic techniques are to be used downstream to microvessel isolation, it is recommended to use DEPC-treated water as opposed to Millipore water, as early work done by our group has demonstrated that DEPC treatment preserves an approximate 10% of transcripts that are otherwise lost when processing samples for RNA sequencing (Supplementary Table 1). DEPC treatment functions as a cost-effective RNAse inhibitor when homogenizing brain tissue in large volumes of buffer.
If microvessel detection parallel to isolation is desired, BCIP/NBT (5-bromo-4-chloro-3'-indolyphosphate and nitro-blue tetrazolium) can be used as a chromogenic substrate for endothelial enzyme alkaline phosphatase. SIGMAFAST™ BCIP®/NBT tablet (Sigma-Aldrich, Missouri, United States) should be crushed and then dissolved in the fresh Homogenization Buffer according to the manufacturer's instructions (1 tablet per 10 ml of solution). Crushing the tablet first encourages faster dissolution in the highly viscous buffer. Within the brain, alkaline phosphatase is localized to cerebral blood vessels (Shimizu, 1950; Leduc and Wislocki, 1952; Bourne, 1958; Becker et al., 1960; Bannister and Romanul, 1963; Romanul and Bannister, 1962; Ball et al., 2002). BCIP is hydrolyzed by the alkaline phosphatase expressed exclusively in endothelial cells to form a blue intermediate that is then oxidized by NBT to produce a dimer, leaving an intense insoluble purple dye.
Prepare 100 mM TRIS at pH 7.8 with 5% final volume of sodium dodecyl sulfate (SDS).
By virtue of a preserved basement membrane, the structures isolated using the described method contain all neurovascular-associated cell types and, therefore, are referred to as “microvessels”. The following protocol describes the specific steps used to isolate and enrich microvessels with retained in situ morphology, which is achieved by using semi-automated dissociation of microdissected brain tissue into a homogenate followed by low-speed centrifugation (schematic overview in Fig. 1). These steps have been optimized to isolate microvessels from 100 mg of frozen brain tissue, processing a maximum of 4 samples at a time.
Fig. 1 Stage After the vmPFC is dissected from a coronal slice containing the frontal lobe, 100 mg of vmPFC tissue is more precisely microdissected using a sterile blade or scalpel. Afterwards, the vmPFC tissue sample is dissociated in Homogenization Buffer using the GentleMACS™ Dissociator. Stage Following dissociation, the sample is centrifuged at 3200 g for 30 mins in order to pellet dissociated microvessels. Additional downstream applications can be applied using microvessel-enriched pellets as shown in stages 3–6 by BCIP/NBT detection, immunofluorescent visualization, total RNA extraction, and total protein extraction, respectively. Image generated using BioRender.
Timing ∼5 mins per sample
Note: Postmortem human tissue can contain transmissible pathogens. Take appropriate precautions, including wearing PPE, and seek medical attention if the scalpel breaks skin.
Timing 45 mins – 1h
Note: Experimental objective must be decided at this step. If either transcriptomic, proteomic investigation, or immunofluorescent visualization is desired, then prepare Homogenization Buffer without BCIP/NBT tablet. If detection of microvessels is to be performed using BCIP/NBT substrate, BCIP/NBT must be prepared in the Homogenization Buffer, as described above.
Fig. 2 a-g) Overview of experimental steps taken to collect and resuspend pelleted microvessels (BCIP/NBT staining is omitted). a) 100 mg of vmPFC tissue submerged in 2 ml of Homogenization Buffer. b) 100 mg of vmPFC tissue after homogenization. c) Lysate topped-up to 10 ml with Homogenization Buffer. d) Lysate transferred to 15 ml falcon tube. e-f) After centrifugation at 3200g for 30 min, a microvessel-enriched pellet forms at the bottom of the tube. g) microvessel-enriched pellet is resuspended in 500 ul of PBS. h-i) Output of experimental steps when BCIP/NBT staining is used. h) Example of BCIP/NBT-stained microvessel pellet, which lends a purple colour to the pellet. i) BCIP/NBT-stained microvessels trapped within the meshwork of a cellular strainer.
Snap gentleMACS™ C Tube into the Dissociator and run the rotating paddle on program Lung 02.01 (Fig. 1). For the first half of the Lung 02.01 program, the speed of the rotating paddle is gradually increased to its maximum in a clockwise direction, and then in an anti-clockwise direction for the second half of the program. The duration of the program is 37 seconds.
After tissue homogenization is complete, return gentleMACS™ C Tube to ice and pipette an additional 8 ml of cold Homogenization Buffer into the tube, topping up the homogenate to 10 ml. Gently invert to mix and collect homogenate (Fig. 2b–c).
Using a serological pipette, transfer the 10 ml of homogenate to a chilled 15 ml falcon tube, including any foam produced during paddle rotation (Fig. 2d).
Gently invert to mix homogenate one more time before centrifugation in order to prevent the formation of a foamy seal atop the homogenate when left sitting. Centrifuge homogenates at 3200 g for 30 mins at 4 °C. Once centrifugation is complete, a microvessel-enriched pellet will form at the bottom of the falcon tube (Fig. 2e). See Troubleshooting Step 2 in Supplementary Table 3.
Carefully vacuum-aspirate the supernatant (which may include an upper layer of clumped dissociated myelin, similar to milk skin on top of boiled milk) without disturbing the microvessel-enriched pellet (Fig. 2e–f).
Gently resuspend the pellet in 400μl of cold PBS and pipette 50 ul of resuspended pellet into each well of an 8-well chamber slide (Nunc™ Lab-Tek™ II Chamber Slide™ System, Thermo Scientific™, Massachusetts, United States) (Fig. 2g). Leave the chamber slide open-faced in a 37 °C oven overnight. Once the PBS has evaporated, microvessels will dry flush to the surface of the slide.
Timing 35–40 mins
Yield and stability of microvessel isolation can be assessed using chromogenic substrate BCIP/NBT to visualize endothelial enzyme alkaline phosphatase. When detecting microvessels with BCIP/NBT, the process of staining microvessels occurs throughout homogenization and centrifugation steps (steps 5–11) (Fig. 2h).
Timing 4–4.5 hours, spread across 3 days (2 overnight incubations)
Greater immunophenotypic characterization of isolated microvessels can be carried out following the resuspension of additional microvessel-enriched pellets. Primary antibodies raised against canonical expression markers for BMECs (Laminin and PECAM1), tight junctions (CLDN5), pericytes (PDGFRβ), smooth muscle cells (Vimentin), and astrocytic endfeet (AQP4) are utilized, along with appropriate fluorophore-conjugated secondary antibodies to characterize collected microvessels thoroughly.
Note: Immunofluorescent visualization must omit any steps in which the microvessel-enriched pellet is filtered through a cellular strainer because microvessels cannot be released from the strainer and, therefore, cannot be mounted onto a microscope slide.
Timing 30–40 mins
Total RNA extraction from isolated microvessels begins with resuspension of the microvessel-enriched pellet (section 3.2, step 10), followed by filtration through a 35 μm cellular strainer in which microvessels become trapped. From these entrapped microvessels, total RNA is immediately extracted using the Single Cell RNA Purification Kit (Norgen Biotek Corp., Ontario, Canada). After total RNA is successfully extracted, a stopping point is possible during which extracted RNA is frozen and stored at −80 °C.
Gently resuspend the pellet (section 3.2, step 10) in 500 μl of cold PBS and gradually pipette through a 35 μm Strainer Cap for FlowTubes™ (Canada Peptide, Quebec, Canada) using vacuum-aspiration underneath to encourage filtration. The result is intact microvessels trapped within the strainer mesh, where smaller cellular debris and free-floating nuclei have passed through. See Troubleshooting Step 3 in Supplementary Table 3.
Using a flat-ended spatula and point-tip forceps, swiftly push and pull out the strainer mesh, removing it from its plastic frame (Fig. 2i).
26.Immediately submerge the mesh into 100 μl of RL buffer in a 1.5 ml Eppendorf tube, according to step 1A of the manufacturer's protocol.
27.Transfer 100 μl of fresh 70% ETOH, pipette 10 times to wash through the mesh (you may briefly vortex for good measure).
28.Discard the mesh and follow steps according to the manufacturer's instructions, including on-column DNase digestion.
Quantify RNA using the Agilent TapeStation 2200 or other quantification system of choice (for total RNA concentration and RIN, see Supplementary Table 2). Freeze and store RNA sample at −80 °C.
STOPPING POINT: tubes of total RNA should be frozen and stored at −80 °C.
Timing 1 hour 10 mins–1 hour 15 mins
Total protein has been successfully extracted from either the microvessel-enriched pellet or microvessels stained through a cellular sieve, with one modification to Homogenization Buffer preparation required in which cOmplete™ EDTA-free Protease Inhibitor Cocktail (Roche, Basel, Switzerland) and Phosphatase Inhibitor Cocktail 2 (Sigma-Aldrich, Missouri, United States) are added according to the manufacturer's instructions and based on final volume prepared. With the procedure identical to when using strained microvessels (kept on a strainer), we elaborate protein extraction from frozen microvessel-enriched pellets (pellets stored at −80 °C after section 3.2, step 10). After total protein is successfully extracted, a stopping point is possible during which extracted protein is frozen and stored at −80 °C.
STOPPING POINT: tubes of total protein should be frozen and stored at −80 °C
Note: Total protein and total RNA extraction from the same microvessel-enriched pellet has not been validated. Two microvessel-enriched pellets per sample should be prepared, one for RNA and one for protein extraction.
Timing 2 hours 5 mins preparation, ∼1–2 hours sorting
If further isolation of BMECs from the microvessel-enriched pellet is desired, FANS-sorting of ETS-related gene (ERG)^+^ BMECs from the obtained microvessel-enriched pellet can be performed. ERG is a transcription factor whose expression in normal physiological conditions is found exclusively in endothelial nuclei, making it a highly specific pan-endothelial nuclear marker (Nikolova-Krstevski et al., 2009; Miettinen et al., 2011; He et al., 2018; Garcia et al., 2022; Yang et al., 2022). FANS of BMECs can be carried out following the resuspension of the microvessel-enriched pellet (section 3.2, step 10).
Microvessel-enriched pellets yielded an average of 8.54 ug/ul of total RNA per sample (Supplementary Table 2). Libraries were then constructed using the SMARTer Stranded Total RNA-Seq Kit v3 - Pico Input Mammalian (Takara Bio Inc., Shiga, Japan), which features integration of unique molecular identifiers (UMIs). Libraries were constructed using 10 ng of RNA as input, 2 minutes of fragmentation at 94 °C (Applied Biosystems Corporation, model ProFlex PCR), 5 cycles of amplification at PCR1 (addition of Illumina adapters and indexes), 12 cycles of amplification at PCR2 (final RNA-seq library amplification) and clean-up of final library using NucleoMag NGS Clean-up and Size Select beads (Takara Bio Inc., Shiga, Japan). Libraries were then quantified at the Genome Quebec Innovation Centre (Montreal, Quebec) using a KAPA Library Quantification kit (Kapa Biosystems, USA), and average fragment size was determined using a LabChip GX (PerkinElmer, USA) instrument. Libraries were sequenced on the NovaSeq 6000 system (Illumina, Inc., California, United States) using S4 flow cells with 100bp PE sequencing kits.
RNA sequencing of microvessel libraries yielded an average of ∼72 million reads per library, which were then processed following our in-house bioinformatic pipeline. Briefly, UMI extraction based on fastq files was performed using the module extract of umi_tools (v.1.1.2) (Smith et al., 2017). Reads were then aligned to the Human Reference Genome (GRCh38) using the STAR software v2.5.4b (Dobin et al., 2013) with Ensembl v90 as the annotation file and using the -twopassMode Basic --outSAMprimaryFlag AllBestScore --outFilterIntronMotifs RemoveNoncanonical --outSAMtype BAM SortedByCoordinate --quantMode TranscriptomeSAM GeneCounts. Resultant bam files were then sorted and indexed using SAMtools (v.1.3.1) (Li et al., 2009), and duplicate reads with the same UMI were removed using the dedup module of umi_tools (v.1.1.2) (Smith et al., 2017). Different metrics,including the fraction of the exonic, intronic, and intergenic reads were calculated using the CollectRnaSeqMetrics module of Picard (version 1.129; Picard2019toolkit, Broad Institute, GitHub repository). The expected counts and the transcripts per million (TPMs) were generated using RSEM (v1.3.3; reverse strand mode) (Li and Dewey, 2011). The number of reads, alignment percentages, genomic contamination, and duplication rate for each sample are shown in the Supplementary data.
Two approaches were used for computational deconvolution of RNA sequencing data. The first approach was performed using the web tool BrainDeconvShiny (https://voineagulab.shinyapps.io/BrainDeconvShiny/), which implements the best-performing algorithms and all cell type signatures for brain, as well as goodness-of-fit calculations based on benchmark work conducted by Sutton et al. (2022) (Sutton et al., 2022). UMI counts for each gene were converted to transcripts per million (TPMs) to account for the varying length of gene and sequencing depth of each sample, facilitating comparisons across samples. Genes with zero TPMs were removed; 34370 genes from the original 58303 passed this QC criteria and were then used as input into the BrainDeconvShiny tool. Deconvolution was performed the first approach used average expression in control samples from the Velmeshev et al. (2019) (Velmeshev et al., 2019) single nuclei dataset (raw data available through the Sequence Read Archive, accession number PRJNA434002; analyzed data available at https://autism.cells.ucsc.edu) as the reference signature for annotated cell types and CIBERSORT v1.04 algorithm to deconvolute sample profiles and estimate cell type composition. The second approach used the MultiBrain (MB) composite signature (Sutton et al., 2022) generated by averaging the expression signatures of five datasets for five cell types (neurons, astrocytes, oligodendrocytes, microglia, and endothelia). MB was used as cell type signatures for deconvolution of our dataset using CIBERSORT v1.04.
The second approach to deconvolution was performed in-house. Postmortem NVU single-nucleus data generated on the 10X Genomics Chromium system was accessed from Yang et al. (2022; raw sequencing data are accessible on GEO using the accession code GSE163577) (Yang et al., 2022) and used as the reference signature. Seurat (Stuart et al., 2019) was used to pre-process raw count expression data, removing genes with less than 3 cells or cells with less than 200 expressed genes. 23054 genes from a total of 23537 and 141468 nuclei from a total of 143793 passed these QC criteria. Counts per million (CPM) values were averaged across nuclei of each cell type to generate the Yang signature input for the CIBERSORTX tool (Newman et al., 2019).
Extracted proteins were reduced with 20 mM tris(2-carboxyethyl)phosphine (TCEP) at 60 °C prior to alkylation with 25 mM iodoacetamide at room temperature for 30 minutes in the dark. An equivalent of 10 μg of total protein was used for proteolytic digestion using suspension trapping (S-TRAP). Briefly, samples were acidified with phosphoric (1.3% final concentration) and then diluted 6-fold in STRAP loading buffer (9:1 water in 100 mM TEAB, pH 8.5). Samples were loaded onto S-TRAP Micro cartridges (Protifi LLC, Huntington, NY) prior to centrifugation at 2000g for 2 minutes and washed three times with 50 μl of STRAP loading buffer. Proteins were digested with trypsin (Sigma Corporation, Kanagawa, Japan) at a 10 enzyme to substrate enzyme-to-substrate ratio for 2 hours at 47 °C. Peptides were sequentially eluted in 100 mM TEAB, 0.1% formic acid in water, and 50% acetonitrile and lyophilized to dryness prior to labelling with TMT 10plex reagents according to the vendor's specifications (Thermo Fisher Scientific, Massachusetts, United States).
Labelled peptides were pooled and again lyophilized to dryness, and then reconstituted in 5 mM ammonium formate and fractionated offline by high pH reversed-phase separation using a Waters Xbridge Peptide BEH C18 column (2.1 × 150mm, 2.5 μm) (Waters Corp., Massachusetts, United States) and an Agilent 1290 LC system (Agilent Technologies, California, United States). Binary gradient elution was performed at a flow rate of 400 μL/minute using mobile phase A) 5 mM ammonium formate adjusted with ammonium hydroxide to pH 10, and B) 100% acetonitrile using the following 0 min, 0% B; 2min, 0% B; 2.1min, 5% B; 25min, 30% B; 30min, 80% B; 32min, 80% B; 2 min post-run, 0% B. Fractions were collected every 30 seconds and the first and last 7 fractions were concatenated such that even and odd samples were pooled separately, resulting in 20 fractions in total.
Samples were analyzed by data- dependent acquisition (DDA) using an Easy-nLC 1200 online coupled to a Q Exactive Plus (both Thermo Fisher Scientific, Massachusetts, United States). Samples were first loaded onto a pre-column (Acclaim PepMap 100 C18, 3 μm particle size, 75 μm inner diameter x 2 cm length) in 0.1% formic acid (buffer A). Peptides were then separated using a 60-min binary gradient ranging from 3 to 40% B (84% acetonitrile, 0.1% formic acid) on the analytical column (Acclaim PepMap 100 C18, 2 μm particle size, 75 μm inner diameter x 25 cm length) at 300 nL/min. MS spectra were acquired from m/z 350-1500 at a resolution of 70,000, with an automatic gain control (AGC) target of 1 × 10^6^ ions and a maximum injection time of 50 ms. The 15 most intense ions (charge states +2 to +4) were isolated with a window of m/z 1.2, an AGC target of 2 × 104, and a maximum injection time of 64 ms and fragmented using a normalized higher-energy collisional dissociation (HCD) energy of 28. MS/MS spectra were acquired at a resolution of 17,500 and the dynamic exclusion was set to 30 s.
DDA MS raw data was processed with Proteome Discoverer 2.5 (Thermo Scientific, Massachusetts, United States) and searched using Sequest HT against a FASTA file containing all reviewed protein sequences of the canonical human proteome without isoforms downloaded from Uniprot (https://www.uniprot.org). The enzyme specificity was set to trypsin with a maximum of 2 missed cleavages. Carbamidomethylation of cysteine was set as static modification and methionine oxidation as variable modification. The precursor ion mass tolerance was set to 10 ppm, and the product ion mass tolerance was set to 0.02 Da. The percolator node was used, and the data was filtered using a false discovery rate (FDR) cut-off of 1% at both the peptide and protein level. The Minora feature detector node of Proteome Discoverer was used for precursor-based label-free quantitation.
Human brain microvessels were isolated from 5 frozen vmPFC grey matter samples microdissected from healthy individuals who died from peripheral diseases or natural events (Table 1). Because the same basement membrane that maintains the integrity of the endothelium also ensheaths astrocytic endfeet as well as pericytes or smooth muscle cells, it is impossible to isolate solely BMECs and, therefore, isolated microvessels also contain microvessel-associated cell types. Notably, the described method allows for isolation of microvessels from other brain regions (data not shown) and has additionally been performed using the dorsolateral prefrontal cortex (dlPFC, Brodmann area 8/9), the primary visual cortex (Brodmann area 17), as well as the hippocampus (Brodmann area 28).
Following isolation of human brain microvessels, chromogenic staining of resuspended pellets using BCIP/NBT (5-bromo-4-chloro-3'-indolyphosphate and nitro-blue tetrazolium), a substrate for endothelial enzyme alkaline phosphatase, demonstrated isolation and enrichment of predominantly microvessels from vmPFC tissue samples (Fig. 3a–b), with similar success when isolating microvessels from mouse cortex (Fig. 3c). The cytoarchitecture of brain microvessels is both complex and comprised of several cell types; thus, following the isolation of brain microvessels, we aimed to characterize the structure and morphological integrity of isolated microvessels. Immunophenotypic characterization of several NVU markers revealed expression of vimentin (VIM; anti-Vimentin antibody RV202, Abcam), laminins (LAM; anti-Laminin antibody L9393, Sigma-Aldrich), claudin 5 (CLDN5, anti-Claudin 5 antibody ab15106, Abcam), platelet-derived growth factor receptor beta (PDGFRβ, anti-PDGFRβ monoclonal antibody G.290.3, Thermo Fisher Scientific) and aquaporin 4 (AQP4, anti-Aquaporin 4 antibody [4/18], Abcam). More precisely, vimentin (Fig. 4a–b), a regulator of actin cytoskeleton primarily in smooth muscle (Chang and Goldman, 2004) and to a lesser extent endothelial cells (Boraas and Ahsan, 2016) and pericytes (Bandopadhyay et al., 2001), as well as laminins (Fig. 4c), the major basement membrane component responsible for signal transduction via interaction with cell surface receptors (Aumailley and Smyth, 1998), are expressed continuously and homogeneously across the entire length of the endothelial surface. CLDN5, a major functional constituent of tight junctions (Greene et al., 2019), was also stained with no apparent discontinuity, suggesting that endothelial tight junctions were preserved (Fig. 4d). Pericyte coverage of BMECs, immunolabeled by regulator of angiogenesis and vascular stability PDGFRβ (Winkler et al., 2010), was detected adhering to the surface of microvessels (Fig. 4e). These results suggest that the overall in situ brain microvessel structure is conserved after isolation.
Fig. 3 a-c) Isolated microvessels are enriched from brain tissue following the described protocol. a-c) Brightfield images of chromogenically stained microvessels using BCIP/NBT substrate demonstrated isolation and enrichment of predominantly microvessels from postmortem vmPFC tissue samples. b) example of preserved microvessel morphology and integrity isolated from postmortem vmPFC tissue. c) Similar microvessel enrichment was observed when the same protocol was carried out using mouse cortex.
Fig. 4 a-b) Endothelial and smooth muscle cells immunolabelled for vimentin (green). c) Extracellular matrix laminins expressed by vascular-related cell types are immunolabelled (red). d) Endothelial cell tight junctions immunolabelled for CLDN5 (red). e) Pericytes immunolabelled for PDGFRβ (green). f) AQP4 expressed by the vessel-facing astrocyte membrane is immunolabelled (magenta) and laminins (green). g) Oligodendrocyte marker MBP (red) shows an absence of immunoreactivity in microvessel preparations; similarly, h) neuronal marker NeuN (red) shows an absence of immunoreactivity in microvessel preparations. Nuclei were stained with DAPI (blue) in all micrographs.
Astrocytes serve multiple essential functions in supporting normal brain physiology (Kimelberg and Nedergaard, 2010). This is, in part, due to the extension of astrocytic endfeet that surround approximately 99% of the cerebrovascular surface (Mathiisen et al., 2010) which, in conjunction with pericytes (Winkler et al., 2011), regulate expression of molecules that form the BBB tight junction, enzymatic, and transporter proteins (Abbott et al., 2006, 2010; Wolburg et al., 2009). Although astrocytes were not co-isolated with microvessels, their perivascular endfeet are ensheathed within the same vascular basement membrane as endothelial cells and pericytes, making it possible that astrocytic endfeet remained attached after tissue homogenization. Because of this, we further observed astrocyte vascular coverage of isolated microvessels. AQP4, a water channel protein essential for the maintenance of osmotic composition and volume within the interstitial, glial, and neuronal compartments (Nagelhus and Ottersen, 2013; Papadopoulos and Verkman, 2013), is expressed at the vessel-facing astrocytic membrane and superimposes the walls of isolated microvessels (Fig. 4f).
Finally, we examined possible contamination of our microvessel preparations by other cell types found within the brain. Immunolabelling for myelin basic protein (MBP; anti-Myelin Basic Protein antibody, BioLegend) was not detected (Fig. 4g), nor did immunolabelling for neuronal nuclear protein (NeuN; anti-NeuN antibody clone A60, Sigma-Aldrich) reveal the presence of neuronal elements (Fig. 4h). Thus, neurons and oligodendrocytes were consistently not observed to be co-isolated with microvessels. Together, these results indicate that the described protocol allows for the isolation and enrichment of structurally preserved brain microvessel fragments that are comprised of BMECs, astrocytic endfeet, pericytes, and tight junction proteins.
To estimate the enrichment of our microvessel preparations, we performed computational deconvolution using the BrainDeconvShiny tool (https://voineagulab.shinyapps.io/BrainDeconvShiny/) and calculated TPMs as input (Fig. 5a–b) . To demonstrate stability of outcome regardless of resources used, different iterations of deconvolution were performed using the CIBERSORT v1.04 algorithm to estimate cell type composition and either the single-nucleus dataset from Velmeshev et al. (2019; VL) (Velmeshev et al., 2019) or the MultiBrain (MB) dataset from Sutton et al. (2022) (Sutton et al., 2022) as the reference signature. Regardless of the approach used, majority endothelial gene expression was estimated, with an average of 94.92% using the VL dataset (Fig. 5a) and an impressive 86.91% using the MB dataset (Fig. 5b), which is a composite signature generated by quantile-normalising and averaging five previously published datasets. Some contamination from neurons (1.29% and 4.66%, respectively), as well as negligible contamination from oligodendrocytes, was observed. When using either dataset as the reference signature, a limited presence of astrocytic genes was observed (1.16% and 6.11%, respectively), which may represent the contribution of astrocytic endfeet that cover the length of the neurovasculature. To estimate the multicellular composition of our microvessels at a finer resolution, computational deconvolution was performed a third time using the reference single-nucleus data generated by Yang et al. (2022), in which the different neurovascular cell type signatures were determined (Fig. 5c). Averaged CPMs across nuclei of each cell type were used as input for the CIBERSORTX tool (Newman et al., 2019), which estimated an average composition of 44.02% capillary and 37.62% SMC, along with much lower estimations for pericyte, arterial, venous, astrocyte and perivascular fibroblast genes (Fig. 5c). Although differentially distributed along the arteriovenous axis, both SMCs and pericytes are embedded within the vascular basement membrane (McConnell et al., 2017) and, therefore, it is unlikely that our isolated method preferentially selects one cell type over the other. Because of this, and the known similarity in molecular signature between SMCs and pericytes (Chasseigneaux et al., 2018; Muhl et al., 2020), as well as the high percentage of captured capillary segments (in which pericytes are predominantly observed) (Sweeney et al., 2016; Gonzales et al., 2020; Hartmann et al., 2021; Alarcon-Martinez et al., 2018), the surprisingly low estimation of pericyte genes may represent a limitation in comparing single-nucleus and bulk tissue datasets to one another. Critically, an average of 90.4% of the total TPMs across samples were assigned to NVU-constituent cell types.
Fig. 5 a-b) Using the BrainDeconvShiny tool, different iterations of computational deconvolution were performed to demonstrate stability of outcome when using our microvessel isolation preparations. a) The average expression from every control sample from the VL dataset was calculated and used as cell type signatures for deconvolution of our dataset using CIBERSORT v1.04. b) The MB (MultiBrain) is a composite signature generated by Sutton et al. (2022) by averaging the expression signatures of five datasets for five cell types (neurons, astrocytes, oligodendrocytes, microglia, and endothelia). MB was used as cell type signatures for deconvolution of our dataset using CIBERSORT v1.04. c) In-house analysis where the average expression from every control sample from the Yang dataset was calculated and used as cell type signatures for deconvolution of our dataset using CIBERSORTX. Heattables were generated using the gt package in R.
To explore our transcriptomic data, TPMs were averaged across the 5 subjects, and the top 10% of most highly expressed genes were designated (a total of 3437 genes) and used as input for over-representation analysis (ORA) using the enrichR package in R, selecting the “Descartes Cell Types and Tissue 2021” database to identify gene sets that are statistically over-represented (Fig. 6a–b and Supplementary Table 4). The threshold value of enrichment was selected by a p-value <0.05 and, as shown, over-represented genes were dramatically enriched for vascular-related terms, such as “Vascular endothelial cells in Cerebellum” and “Vascular endothelial cells in Cerebrum”. Genes behind enriched terms were extracted and the expression of a subset of known brain endothelial, pericyte, astrocytic, and smooth muscle genes in isolated microvessels were examined (Fig. 6c–i). As expected, microvessels had increased expression of canonical endothelial genes such as CLDN5, CDH5, SLC2A1, ABCB1, VWF, and MFSD2A, with the highest expression in endothelial genes ACTG1, B2M, BSG, EEF1A1, HLA-B, HLA-E, SPARCL1, TMSB10, and VIM (Fig. 6a).
Fig. 6 a-b) Enrichment analysis of averaged 10% of most highly expressed genes (3437 genes) returns predominantly vascular-related terms. Over-representation analysis (ORA) using the enrichR package in R and the “Descartes Cell Types and Tissue 2021” database was used to identify gene sets that are statistically over-represented. The threshold value of enrichment was selected by a p-value <0.05, indicating that over-represented genes were significantly enriched for vascular-related terms. a) Count for genes in our dataset that are present in returned gene sets. b) Ratio for genes in our dataset that are present in returned gene sets, determined by the total number of genes in each set. c-i) Isolated microvessels have increased expression of canonical neurovascular-related genes. c) Bar plots showing TPMs for endothelial-defining genes, displayed according to general expression range. Highest expression found in endothelial genes B2M, BSG, FLT1, IFITM3, MT2A, SLC2A1, VIM, and VWF. d) Bar plot showing TPMs for pericyte-defining genes. Highest expression was detected in pericyte genes CALD1, FN1, IGFBP7, RGS5, and SPARCL1. e) Bar plot showing TPMs for smooth muscle cell-defining genes. Highest expression was detected in smooth muscle genes ACTG1, ACTN4, MYL6, PTMA, and TAGLN. f) Bar plot showing TPMs for tight junction-defining genes and g) Bar plot showing TPMs for adherens junction-defining genes. Several genes encoding for junctional proteins are found in the top 10% of most highly expressed genes, including CLDN5, CTNNB1, CTNND1, OCLN, JAM1, TJP1, and TJP2. h) Bar plot showing TPMs for astrocyte-defining genes. Astrocytic gene expression was predominantly limited to markers of astrocytic processes or endfeet, namely CLU, GFAP, and GLUL. i) TPMs from neurovascular-related genes were summarized according to cell type expression, demonstrating an overrepresentation of endothelial, smooth muscle cell, and pericyte genes. j) Overlap between top 10% of most highly expressed genes from our RNA sequencing data and immune genes found within the Immunome Database. Bar plots were generated using the ggplot package and Venn diagram was generated using the ggVennDiagram package in R.
Additionally, there was enrichment for other neurovascular cell types, as suggested by canonical pericyte genes PDGFRβ, MCAM, RGS5, AGRN, and NOTCH3 (Fig. 6b), as well as canonical smooth muscle genes ACTA2, MYL6, MYL9, TAGLN, and LGALS1 (Fig. 6c). Highest expression was detected in pericyte genes CALD1, FN1, IFGBP7, RGS5, SPARC, and SPARCL1 (Fig. 6b); as well as smooth muscle genes ACTG1, ACTN4, CALD1, MYL6, and PTMA (Fig. 6c). Complimentary to immunophenotypic characterization of isolated microvessels, several genes whose products are involved in junctional complex maintenance and organization (tight junctions, Fig. 6f; adherens junctions, Fig. 6g) are found in the top 10% of most highly expressed genes, for e.g., CLDN5, CTNNB1, CTNND1, ITGB1, JAM1, OCLN, TJP1, and TJP2; with additional vascular makers expressed at lower transcripts per million present throughout the entire dataset. Expression of astrocytic genes was observed to a lesser extent, several which have been previously validated as markers of astrocytic processes or endfeet (Boulay et al., 2017; Derouiche and Geiger, 2019; Sakers et al., 2017), namely EZR, GJA1, RDX, SLC1A2, and SLC1A3 (Fig. 6h). To better visualize the proportion of expression contributed by these genes, TPMs were summarized by cell type over the total number of TPMs (Fig. 6i), demonstrating an overrepresentation of endothelial, smooth muscle cell, and pericyte genes in samples.
Intriguingly, results from ORA reveal genes that are of interest in numerous disease contexts. Enriched in the described dataset is gene CLDN5, an indispensable junctional protein for the correct organization of tight junctions and maintenance of BMEC integrity (Greene et al., 2019), which was previously reported to be downregulated in the nucleus accumbens of depressed suicides (Menard et al., 2017) by altered epigenetic regulation via histone deacetylase 1 (HDAC1) (Dudek et al., 2020). Enriched pericyte genes TXNIP, RUNX1T1, ITGA1 and, DOCK9 are also reported to be differentially expressed in Schizophrenia (Puvogel et al., 2022). Similarly enriched in this dataset are angiogenic growth factors EGFL7, FLT1, VWF*,* and antigen-presentation machinery B2M and HLA-E*,* all of which are upregulated in a subpopulation of angiogenic BMECs from subjects with Alzheimer's disease (Lau et al., 2020), suggesting a compensatory angiogenic and immune response in AD pathogenesis. Likewise, endothelial genes PICALM, INPP5D, ADAMTS1, and PLCG2 that are found in the top 10% of the microvessel dataset are differentially expressed in Alzheimer's disease (Yang et al., 2022). Recent breakthroughs in deciphering the underlying etiology of Huntington's disease reveal aberrant downregulation of endothelial ABCB1, ABCG2, SLC2A1, and MFSD2A as well as mural PDGFRB, SLC20A2 and FTH1 (Garcia et al., 2022) – mutations in which are known to cause HD-like syndromes with primary pathology localized in the basal ganglia (Chinnery et al., 2007; Tadic et al., 2015). Indeed, human brain microvessel datasets integrated with others, like those from experimental models, will expedite our understanding of neurovascular contributions to mood disorders and neurodegenerative disease, and even further propagate hypothesis generation for established vascular diseases, such as white matter vascular dementia in which each cell type of the NVU exhibits a specific disease-associated expression signature (Mitroi et al., 2022).
During neuroinflammation, the BBB endothelium reconfigures its landscape of adhesion molecules, cytokines, chemokines, and reactive oxygen species, combined with reduced expression of junctional molecules. These changes prime for bidirectional interaction between the neuroimmune and peripheral immune compartments and enable increased recruitment of circulating leukocytes across the BBB. The top 10% of most highly expressed genes were also assessed for overlap with a curated immune-related gene list (Immunome Database accessed at https://www.innatedb.com/redirect.do?go=resourcesGeneLists, data originally provided by http://structure.bmc.lu.se/idbase/immunome/) (Breuer et al., 2013; Ortutay et al., 2007), sharing 145 out of 824 validated genes involved in immunological processes (Fig. 6j), including notable immune factors CX3CL1, IFNGR1, CD74, IL4R, CXCL2, CXCL12, CD81, IRF1, MIF, as well as HLAs (HLA-A, HLA-B, HLA-C, HLA-E, HLA-F, HLA-DRA, HLA-DPA1, HLA-DPB1). Moreover, adhesion molecules known to mediate cell-cell adhesion at the BBB are also present, CD44, previously implicated in monocyte transmigration (He et al., 2016b) and T-cell-endothelial cell interaction (Flynn et al., 2013), MCAM, which mediates recruitment of pathogenic CD4^+^ T lymphocytes (Charabati et al., 2023) as well as T helper (TH) 1 cells (Breuer et al., 2018), and ICAM2, which is also critical for T helper (TH) 1 cell diapedesis (Laschinger et al., 2002). Other well-characterized immune, adhesion, and trafficking molecules in our data are listed in the Supplementary data.
As a means to assess our isolation method, our top 10% of most highly expressed genes were juxtaposed to validated neurovascular cell type-defining markers, as designated by Garcia et al. (2022) (Garcia et al., 2022) based on sequencing of 4992 and 11,689 vascular nuclei from ex vivo and postmortem brain tissue, respectively. Results indicated substantial overlap between the three datasets for all NVU-constituent cell types (Fig. 7), including arteriole-defining genes (Fig. 7a), capillary-defining genes (Fig. 7b), venule-defining genes (Fig. 7c), arteriolar SMC-defining genes (Fig. 7d), pericyte-defining genes (Fig. 7e), venular SMC-defining genes (Fig. 7f), and perivascular fibroblast-defining genes (Fig. 7g). Interestingly, overlap was consistently greater between the two postmortem datasets.
Fig. 7 a-g) Overlap between the top 10% of most highly expressed genes from our RNA sequencing data and published NVU dataset. Validated neurovascular cell type markers were obtained from Garcia et al. (2022) postmortem and human ex vivo single-nucleus sequencing datasets (found under Supplementary Table 2) and compared for potential overlap with the top 10% of most highly expressed genes from our data. Results indicate a significant overlap between the three datasets for all NVU-related cell types. a) Arteriole-defining genes. b) Capillary-defining genes. c) Venule-defining genes. d) Arteriolar SMC-defining genes. e) Pericyte-defining genes. f) Venular SMC-defining genes. g) Fibroblast-defining genes. Venn diagrams were generated using the ggVennDiagram package in R.
While transcript level may show positive correlations with protein level, protein abundance should not necessarily be inferred from RNA sequencing counts (Nie et al., 2006; Vogel and Marcotte, 2012); hence, interrogation of the highly dynamic proteome is needed to speculate the functional consequences of changes in protein expression. We sought to interrogate NVU-specific proteomic signatures using total protein extracted from isolated microvessels. Microvessel-enriched pellets were prepared from 3 frozen vmPFC grey matter samples microdissected from healthy individuals who died of peripheral diseases or natural events (Table 1). Although resuspended and strained pellets yield sufficient protein needed to perform LC-MS/MS (validated, data not shown), straining was omitted in favour of protein extraction directly from pellets to maximize microvessel material input for proteomic interrogation. Using Tandem Mass Tag (TMT) isobaric labeling and sample fractionation of peptides, global, relative quantitation of a total of 1638 individual proteins were detected from microvessel-enriched pellets (quality parametres shown in Supplementary Fig. 2). Importantly, there was significant overlap between transcriptomic and proteomic output, with 1635/1638 (99.8%) of detected proteins likewise identified in the transcriptomics data (albeit no corresponding transcripts were detected for proteins F9, DCD, and SERPINB12); and with 961/1638 (58.7%) proteins found in the top 10% of most highly expressed transcripts, resulting in an overall 23% overlap between high expressors in both datasets (Fig. 8a).
Fig. 8 a-i) There is substantial overlap between transcriptomic and proteomic data output. a) 1635/1637 (99.9%) of proteins likewise identified in the transcriptomics data, and with 1024/1637 (62.6%) proteins found in the top 10% of most highly expressed genes. b) Several canonical BMEC markers were detected with high normalized average abundance. c) Over-representation analysis using the enrichR package in R and the “Descartes Cell Types and Tissue 2021” database was used to identify statistically over-represented sets. Genes corresponding to proteins detected during LC-MS/MS were used as input and the threshold value of enrichment was selected by a p-value <0.05. Output indicated that over-represented proteins were significantly enriched for vascular-related terms. a) Count for peptides in our dataset that are present in returned sets. b) Ratio for peptides in our dataset that are present in returned sets, determined by the total number in each set. d) Bar plot showing normalized abundance for endothelial cell-defining proteins. e) Bar plot showing normalized abundance for pericyte-defining proteins. f) Bar plot showing normalized abundance for smooth muscle cell defining-proteins. g) Bar plot showing normalized abundance for tight junction defining-proteins. h) Bar plot showing normalized abundance for adherens junction-defining proteins. i) Bar plot showing normalized abundance for astrocyte and astrocytic endfeet-defining proteins. Bar plots generated using the ggplot package and Venn diagram generated using the ggVennDiagram package in R.
Proteins known to be expressed by vascular-associated cell types and perivascular extracellular matrix were positively identified in all 3 samples. Several canonical endothelial markers were detected with high peptide abundance, including vimentin, several protein subunits of laminin (LAMA3, LAMA5, LAMB2, LAMC1, LAMC3), OCLN, TJP1, TJP2, VWF, VWA1, BSG, PECAM1, Monocarboxylate transporter 1 (SLC16A1), broad substrate specificity ATP-binding cassette transporter (ABCG2), ESAM, CLDN5, CDH5, ATP-binding cassette sub-family B member 1 (ABCB1), ATP-binding cassette sub-family D member 3 (ABCD3), and Protocadherin-1 (PCDH1) (normalized average abundance for these proteins are listed in Fig. 8b). As expected, enrichment terms returned by ORA (using corresponding gene names as input) were predominantly vascular-associated (Fig. 8c and Supplementary Table 5), and normalized abundance of known brain endothelial, pericyte, astrocytic and smooth muscle cell proteins were high (Fig. 8d–i).
The BBB poses a major pharmacological barrier as BMECs express a vast array of enzymes and transport systems that facilitate brain uptake processes of essential nutrients and neuroactive agents across the BBB (Hediger et al., 2004), controlling the rate and extent to which drugs are able to reach the brain parenchyma via the transcellular pathway (Ballabh et al., 2004). There is a pressing need for improved knowledge surrounding the expression and functionality of these systems at the human BBB as the majority of data comes from either in vitro cell culture or animal studies, making in vitro to in vivo or interspecies scaling less reliable. Analyses revealed high abundance of SLC2A1/GLUT1, a transmembrane protein responsible for the facilitated diffusion of glucose (Mueckler and Thorens, 2013), and the two glutamate transporters SLC1A2/EAAT2 and SLC1A3/EAAT1 in brain microvessels. Transporters SLC7A5/LAT1 and SLC3A2/4F2hc, which supply the brain with large neutral amino acids (Yanagida et al., 2001; Nakamura et al., 1999; Nicklin et al., 2009), and SLC16A1/MCT1 and SLC16A2/MCT8, which are involved in the transport of monocarboxylates (Vijay and Morris, 2014) and T3 thyroid hormone (Trajkovic et al., 2007) at the BBB, respectively, are also found in the described proteomic dataset. Also found are ABCB1/P-glycoprotein and breast cancer-related protein ABCG2/BCRP, the principal ABC efflux transporters (Begley, 2004; Loscher and Potschka, 2005b; Sun et al., 2003) that limit entry of drug candidates, toxic compounds, as well as xenobiotics from the central nervous system (Agarwal et al., 2010, 2011a, 2011b; Chen et al., 2009; de Vries et al., 2007a; Polli et al., 2009). Due to major species-specific differences in transporter expression profiles, there is utility in obtaining absolute protein amounts as they may elucidate the contribution of each transporter in facilitating the entry of endogenous substances and nutrients like glucose, glutamate, and amino and fatty acids into the brain, in addition to drugs and other xenobiotics that exploit these mechanisms (Hindle et al., 2017).
The use of frozen brain tissue demands sorting of target nuclei as opposed to intact cells and, therefore, requires the use of nuclear fluorescent tags to facilitate the isolation of endothelial nuclei. Expression of transcription factor ETS-related gene (ERG) has been previously validated as an endothelial-specific nuclear marker (Miettinen et al., 2011; Haber et al., 2015; Kim et al., 2013; Birdsey et al., 2015) and is found in the top 10% most highly expressed genes from isolated microvessels (TPM shown in Fig. 9a). As an additional testament to the versatility in downstream use of the described method, resuspended microvessels subjected to a 2-h incubation with anti-ERG antibody under rotation is sufficient for vascular disassembly and to expose the ERG epitope for endothelial nuclei immunolabelling (Fig. 9b) without the need for enzymatic digestion as previously described (Crouch and Doetsch, 2018; Pastrana et al., 2009; Codega et al., 2014). Critically, a 2-h incubation in combination with the appropriate gating strategy adjusted to collect endothelial nuclei, is sufficient for the purification of endothelial nuclei from frozen postmortem brain tissue (data not shown). The potential to isolate ERG^+^ endothelial nuclei from postmortem microvessels under different physiological conditions (Plane et al., 2010; Ohab et al., 2006; Yamashita et al., 2006; Kojima et al., 2010; Bardehle et al., 2013; Greenberg, 2014; Paul et al., 2012) is an advantage of our approach and addresses a growing interest in the use of primary BMECs, circumventing phenotypical or gene expression changes induced in primary BMEC cultures by prolonged adherence steps used in other isolation protocols (Durafourt et al., 2013; De Groot et al., 2000; Goldeman et al., 2020; Lyck et al., 2009).
Fig. 9 Isolated microvessels can be used to further sort endothelial nuclei. a) ERG is a top 10% most highly expressed gene in isolated microvessels. b) Dissociated endothelial nuclei immunolabelled with ERG conjugated to Alexa-647 antibody prior to FANS protocol. Resuspended microvessels subjected to a 2-h incubation with anti-ERG antibody under rotation is sufficient for immunolabelling of endothelial nuclei prior to FANS sorting.
We describe a singular, standardized protocol to enrich and isolate microvessels from archived snap-frozen human brain tissue with the ability to apply the same protocol to frozen mouse cerebral cortex. Integral to this method are three factors that confer gentility and 1) the correct molarity of the sucrose-based buffer to separate and cushion microvessels during centrifugation-separation, 2) centrifugation-separation of microvessels from the rest of the tissue homogenate occurs in a 10 ml volume, which facilitates the formation of a microvessel-enriched pellet as heavier structures reach the bottom of the 15 ml falcon tube, and 3) the limited number of wash steps that minimize eventual damage done to the integrity of microvessel fragments. Microvessel enrichment and stability were assessed by chromogenic staining of microvessels using BCIP/NBT substrate and, through immunophenotypic characterization, we show that isolated microvessel fragments are comprised of NVU cellular components including BMECs, astrocytic endfeet, pericytes, as well as tight junction protein complexes. The demonstrated gentility and simplicity of the approach, in turn, confer versatility in the high-throughput techniques that can be utilized downstream to microvessel isolation, as demonstrated here with RNA sequencing and LC-MS/MS.
This protocol should have excellent reproducibility in isolating intact microvessels from any region of the adult human brain. It should be noted, however, that the current version of the protocol may need adjustments if myelin content of brain samples used is high, as dissociated myelin (which travels to the top of the homogenate; section 2.2, step 10) may interfere with the formation of a microvessel-enriched pellet (which travels to the bottom of the homogenate; section 2.2, step 10). As reported, microvessels isolated using the described method are in high yield wihout other contaminating cell types. However, the protocol enriches for but does not purify microvessels, so a limited proportion of contamination from non-vascular cell types might be found in collected pellets. Limited expression of neuronal markers is observed during computational deconvolution of RNA sequencing data; however, whether a minority of neurons are indeed co-enriched with microvessels remains unclear, as co-enrichment of neurons was not observed during immunophenotypic characterization of isolated microvessels. While some co-enrichment is consistent with the technically unavoidable capture of some parenchymal cells when utilizing a method suitable for frozen postmortem brain tissue (which necessitates as few and as gentle steps as possible), the human BBB atlases generated by Yang et al. (2022) and Garcia et al. (2022) reveal that “canonically neuronal” genes are also expressed in vascular-associated cell types. Therefore, it is possible that counts for such genes indeed originate from vascular-associated cells and not neurons. Additionally, representation of all vascular-associated cell types might not be uniform, as different cell types may vary across brain regions and even brain subregions, and different cell types may be differentially susceptible to steps taken during RNA or protein extraction. Nonetheless, in the absence of harsh experimental steps, it is reasonable to assume that the proportions of vascular-associated cell types found within isolated microvessels represent the natural multicellular composition of vasculature within the brain. During the development of this protocol, several brain tissue samples were used possessing a range of values for PMI, RIN, pH, and other metrics, including age at death, manner of death, freezer storage time, and prior medication exposure. Although not systematically assessed, success in isolating and enriching microvessels from different tissue samples was consistent and robust across variations in such quality metrics (as demonstrated here by the wide range of age and PMI across subjects shown in Table 1). We attribute such reproducibility to the protective nature of the vascular basement membrane that may protect the vessel structure and its contents against changes experienced by the brain parenchyma during the postmortem interval. Despite this, it is recommended to utilize postmortem samples with quality metrics PMI, RIN, and pH that indicate moderate to good quality, which is a fundamental prerequisite to investigating complex brain disorders using molecular profiling techniques. Inevitably, studies that make use of isolated human microvessels will encounter high inter-subject variation in both RNA and protein yield as well as experimental read-out. Large cohorts with matched subjects (if studying disease and/or sex differences) would be required to power a study investigating how brain vascular gene or proteomic expression varies with factors such as brain region, age, or disease. Although sensitivity of mass spectrometers has been greatly improved over the years (Hahne et al., 2013; Hebert et al., 2018; Shishkova et al., 2016, 2018; Timp and Timp, 2020), as well as peptide separation for untargeted proteome analysis (Shishkova et al., 2016, 2018; Toth et al., 2019), the literature consistently shows that high output from mass spectrometry relies on a balance between sample quantity as well as sample complexity. Because protein extracted frommicrovessels, as opposed to bulk tissue, could be considered a sample lacking complexity, it is reasonable to assume that this relays to lower output, which may be further impacted by several low signal-to-noise events that result in unidentified peptides (Griss et al., 2016) and a more limited dynamic range of peptide detection (Timp and Timp, 2020).
Our successfully generated datasets promise a more complete approach to investigating BBB function and disease, where one arm of a study may leverage experimental animal or cell culture models to identify pertinent biological mechanisms, and the other arm may utilize microvessels isolated from human brain to provide insight into species-specific differences and other limitations of experimental models. Species differences in disease-affected neuronal and myeloid gene expression have been characterized by single-cell or single-nucleus sequencing (Cosacak et al., 2022; Friedman et al., 2018; Kamath et al., 2022), but such differences in neurovascular gene expression have only just begun to be comprehensively analyzed, and solely in Huntington's disease and Alzheimer's disease (Garcia et al., 2022; Yang et al., 2022). Despite limited data, it has become evident there are striking differences, with one study reporting 142 mouse-enriched genes, including Vtn, Slco1c1, Slc6a20a, Atp13a5, Slc22a8, and 211 human-enriched genes, including SLCO2A1, GIMAP7, and A2M (Song et al., 2020b). Yang et al. (2022) (Yang et al., 2022) further reported hundreds of species-enriched genes in BMECs and pericytes (Yang et al., 2022), finding that BMECs and pericytes exhibit the greatest transcriptional divergence in several vascular solute transporters (for e.g., GABA transporter SLC6A12) and genes of disease and pharmacological importance (Yang et al., 2022). This observation was corroborated by Garcia et al. (2022) (Garcia et al., 2022), who further detailed that species-specific DEGs were strongly enriched for marker genes of vascular-associated cell types, indicating that cell type identity markers were among those that vary the most between species (Garcia et al., 2022). Breakthroughs can similarly be made using the described method, with the added benefit that microvessels kept as a structurally intact unit provide insight into the neurovasculature in a manner that nuclei or dissociated cells not only is unwanted technical-related depletion of vascular-associated cells avoided, but cytoplasm, which carries much higher amounts of mRNA and protein, as well as interstitium are also preserved. This is not trivial, as work done by others comparing microglia single cells versus single nuclei from postmortem tissue reveals that certain populations of genes are depleted in nuclei compared to whole cells. Those depleted were previously implicated in microglial activation, such as APOE, SPP1, CST3, and CD74, totalling 18% of previously identified microglial-disease-associated genes (Thrupp et al., 2020). While there is undeniable benefit in characterizing disease-related changes at the cell type level, the neurovasculature is so interconnected that dysfunction is typically observed in all vascular-associated cells, conferring whole-unit dysfunction that becomes difficult to parse when looking at its individual nuclei. We put forward the ATP-binding cassette (ABC) transporter family as an interesting candidate that can be further explored using microvessels isolated from postmortem brain. The human genome carries 48 different ABC transporters (Gil-Martins et al., 2020; Morris et al., 2017; Robey et al., 2018), several of which are expressed in the CNS, primarily at the BBB (Gil-Martins et al., 2020). Dysfunction of ABC transporters, at expression and/or activity level, has been repeatedly associated with neurological disease (Gil-Martins et al., 2020); and of note are the functionally important yet redundant ABCB1 and ABCG2, both of which have been observed in our transcriptomic and proteomic datasets. For example, ABCB1 is culpable in the neuroinflammatory mechanisms of mutliple sclerosis, where ABCB1 expression and activity are significantly decreased by mechanisms involving CD4^+^ T cells (Kooij et al., 2010), which is just one aspect of the broad barrier impairment that permits lymphocytes activated in the periphery to infiltrate the CNS (Cashion et al., 2023; Ortiz et al., 2014). Moreover, numerous drug candidates show potential anticancer effects against different brain cancer cell lines in vitro and yet, their efficacy in vivo and in clinical trials has been considerably more modest, in large part due to ABCB1/ABCG2-mediated efflux at the BBB. Both ABCB1 and ABCG2, which are expressed on the luminal membrane of BMECs (Biegel et al., 1995; Cooray et al., 2002; Eisenblatter and Galla, 2002; Zhang et al., 2003), present a double-edged sword at the blood-brain on the one hand, ABCB1/ABCG2-mediated efflux is vital for protecting the brain; on the other hand, several anticancer drugs have been identified as substrates of ABCB1 and/or ABCG2 (de Vries et al., 2007b; Agarwal and Elmquist, 2012; Traxl et al., 2019), and their function restricts brain uptake of anticancer drugs, significantly limiting their efficacy in the treatment of primary and metastatic brain tumors (de Vries et al., 2007b; Agarwal and Elmquist, 2012; Juliano and Ling, 1976; Doyle et al., 1998; Marchetti et al., 2008; Agarwal et al., 2011c; de Gooijer et al., 2018; Sorf et al., 2018; Schinkel et al., 1994). Efforts to better understand ABCB1/ABCG2 expression and circumvent the restriction of drug uptake have been considerable. Previous studies point to a lower ABCB1/ABCG2 ratio (Dehouck et al., 2022; Uchida et al., 2011), whereas our findings exhibit a higher ABCB1/ABCG2 ratio. This discrepancy may be an insight into the interindividual differences that give rise to variable drug responses, a common clinical challenge. Several synonymous single nucleotide polymorphisms (SNPs) that affect function and substrate binding have been identified in both transporters (Kimchi-Sarfaty et al., 2007; Fung and Gottesman, 2009; Dickens et al., 2013; Furukawa et al., 2009; Delord et al., 2013; El Biali et al., 2021). The individuals studied here were all of French-Canadian origin, a group with a distinct genotype, and so certain SNPs that affect mRNA stability or transporter activity may, in turn, modulate translational output of ABCB1 relative to ABCG2, as has been observed with ABCB1 SNPs (Wang et al., 2005). It should be noted, however, that these studies (including ours) comprised of small cohorts and warrant further investigations with larger sample sizes.
In sum, the isolated brain microvessel is a robust model for the NVU and can be used to generate a variety of highly dimensional datasets. The availability of characterized human neurovascular transcriptomes and proteomes can aid in identifying potential roles for BMECs and pericytes in the pathogenesis of neurological and psychiatric disorders and additionally aid in assessing the expression of molecules with potential relevance to drug delivery and novel therapies (e.g., SLCs, ABCs, and large molecule receptors) across the human brain vasculature.
None.
The authors have no competing interests to declare. This study was funded by an ERA NET Neuron grant and a CIHR Project grant to N.M. M.W. and R.R. respectively received scholarship and fellowship support from the FRQ-S. The Douglas-Bell Canada Brain Bank is supported in part by platform support grants from the Réseau Québécois sur le Suicide, les Troubles de l’Humeur et les Troubles Associés (FRQ-S), Healthy Brains for Healthy Lives (CFREF), and Brain Canada. The present study used the services of the Molecular and Cellular Microscopy Platform (MCMP) at the Douglas Research Centre. The authors would like to thank the expert help of Douglas-Bell Canada Brain Bank staff members (J. Prud'homme, M. Bouchouka, D. Mirault, V. Lariviere, A. Baccichet), the technology development team at the McGill University and Genome Quebec Innovation Centre, and the Segal Cancer Proteomics Centre at the Lady Davis Institute. The authors would also like to thank Dr. Ghazal Fakhfouri for rearing and providing the mice used in this study.
The following are the Supplementary data to this
Data will be made available on request.
Data will be made available on request.