Comparative transcriptomics of choroid plexus in Alzheimer’s disease, frontotemporal dementia and Huntington’s disease: implications for CSF homeostasis

The choroid plexus (CP) is a CNS secretory tissue within the cerebroventricular system consisting of a vascular stroma surrounded by epithelium [1, 2]. Although the primary function of choroidal tissue is to produce and regulate cerebrospinal fluid (CSF), it also importantly provides a permeability-regulating blood–CSF barrier (BCSFB) [3]. Other additional roles of CP relate to CNS wound repair [4], sex hormone modulation of BCSFB-CNS [5], catabolite detoxification [6], ion regulation [7], a selective leukocyte gate [8], and CSF–brain neuroimmune homeostasis, including interferon actions [9‐13]. Recently, the BCSFB tissue has been examined for unique CP changes in diverse disorders: mitochondrial diseases [14], multiple sclerosis/experimental autoimmune encephalitis [15, 16], schizophrenia [17], acute response to peripheral immune challenge [18], normal pressure hydrocephalus [19], and Alzheimer’s disease (AD) [20, 21].

Analyzing the transformed CP tissue composition and pathophysiologic functions in neurodegeneration elucidates specific metabolic/secretory processes underlying CSF–CNS disease pathogenesis [22]. BCSFB alterations such as choroid epithelial cell atrophy, stromal fibrosis, vascular thickening, tight junction (claudin-5) downregulation, and basement membrane thickening are associated with AD pathology [20]. These changes in the epithelial–stromal nexus likely affect secretion and transport, resulting in diminished CSF turnover and modified neuroimmune regulation. Neuroimmune phenomena in the CP and/or CSF include adjustments in the level of proteins (e.g., neurotrophins, interferons and growth factors), cytokines, and certain immune cells [12, 22]. Oxidative stressors in AD and other dementias may also differentially impact CP’s ability to synthesize/transport proteins/hormones, and to regulate cellular/CSF metabolites such as methionine/homocysteine [23], Aβ/tau [24], and creatine/creatinine [25].

This investigation at the Brown University Medical School, in collaboration with Merck & Co., analyzed gene expression in CP tissues from late-stage Alzheimer patients, for comparison with control subjects (Ctrl) and two other diseases: frontotemporal dementia (FTD) and Huntington’s disease (HuD). Our working hypothesis anticipated: (i) common denominators of altered CP expression in the three diseases, as well as (ii) differential expression patterns due to disease-specific alterations in neural metabolites, that by ‘homeostatic feedback signaling’ via volume transmission from brain to CSF to CP, could uniquely modulate gene expression at the BCSFB.

Investigating CP tissue gene expression in various CNS diseases likely informs on diverse BCSFB adjustments to neurodegeneration. Bergen et al. [26] focused on gene expression changes by CP epithelial cells in AD. In this study, we analyze CP tissue responses (epithelium plus stroma) by providing profiles of mRNA changes. The altered expression profiles in AD, FTD and HuD are discussed in relation to restorative homeostatic mechanisms, as well as to chronic BCSFB damage and disrupted CSF–brain homeostasis.

We first compared the genome-wide differences in gene expression between diseased and Ctrl CP. p-value distributions from T-test comparisons, between the Ctrl group and each of the neurodegenerative disease groups (AD, HuD, FTD), revealed significant effects on the CP transcriptome in each of the 3 diseases (Fig. 1a, Additional file 2: Table S2). The AD group had the highest number of differentially expressed probe sets likely due, at least in part, to the higher number of AD subjects compared to HuD and FTD. 3935 (7%) out of the 57,060 probe sets on the array were differentially expressed [p < 0.01, false discovery rate (FDR) = 14%] between AD and Ctrl subjects, while 1287 (FDR = 44%) and 2136 (FDR = 27%) probe sets were regulated with p < 0.01 in HuD and FTD, respectively (Fig. 1a, b, Additional file 2: Table S2). Despite the limited statistical power and resulting high false discovery rates in the HuD and FTD comparisons, there was a large degree of overlap in the genes identified in each of the comparisons (Fig. 1b, c, Additional file 2: Table S2). Almost all probe sets differentially expressed (p < 0.01) in the AD samples had significant or a trend toward differential expression in the same direction in the other two disease groups (Fig. 1c, Additional file 2: Table S2). In Additional file 4: Table S4 are listed the top 10 most upregulated and 10 most downregulated genes from the AD vs. Ctrl comparison; 80% of these findings were confirmed by multiple probe sets when available on the array.

In order to validate these findings with different subjects and gene expression platforms, we compared the results in whole CP to those obtained by Bergen et al. in laser-dissected CP epithelial cells from control and AD subjects [26]. We anticipated that the changes reported in CP epithelial cells would also be evident in the whole CP samples but to a lesser degree given the presence of additional cell types such as stroma and immune cells in the whole tissue samples. Indeed, the 36 genes reported as differentially expressed in AD CP epithelium by microarray, and in some cases [17] also by quantitative PCR (Bergen et al. Tables 1 and 2 in [26]), were regulated similarly in AD whole CP tissue in our study (r² ~ 0.7), but to a lesser magnitude (by ~ 35%, as indicated by linear regression in Fig. 1d). 34 of the 36 differentially expressed genes reported by Bergen et al. [26] were modulated in the same direction from Ctrl in both studies, with 20 obtaining significance (p < 0.05) in the current AD whole CP tissue comparison. Similarly the expression values in whole CP from FTD and HuD also correlated with those reported for AD CP epithelium in reference # [26] (r² = 0.8 and 0.6, respectively, Fig. 1e, f).

In this study, among the 3935 probe sets differentially expressed (p < 0.01) in AD compared to Ctrl subjects, 2332 were upregulated and 1603 downregulated (Fig. 1b, Additional file 2: Table S2). The differentially expressed genes were examined for overlap with ~ 2000 GeneGo and Ingenuity pathways. Ninety-two pathways were enriched (Bonferroni corrected p-value, i.e., e-value, < 0.01) among the upregulated genes. These enrichments represented primarily immune-related pathways, including acute phase response, cytokine and interferon signaling, NFkB, and cell adhesion, as well as growth factor, JAK-STAT and mTOR signaling pathways, PPAR signaling and protein/nucleic acid salvage pathways (Fig. 2, Additional file 3: Table S3a). Pathway enrichment among downregulated genes was less extensive (12 pathways with e < 0.01), including genes involved in methionine degradation and protein translation (Fig. 2, Additional file 3: Table S3b).

Many differences between AD and Ctrl were also observed in HuD vs. FTD (Fig. 1, Additional file 2: Table S2). However, there were also some significant differences between AD vs. HuD + FTD, with 902 (1.6%) probe sets significant at p < 0.01 (63% FDR) (Fig. 3). Figure 3a displays the genes up and down-regulated in AD more than in the combined HuD + FTD group; whereas Fig. 3b presents the opposite, i.e., genes regulated more in HuD + FTD than in AD. The 513 probe sets uniquely upregulated in AD: AD vs. Ctrl (p < 0.01) and AD vs. HuD + FTD (p < 0.05) were enriched (e < 0.1) predominately in interleukin and VEGF signaling genes (Additional file 3: Table S3c). There were 272 probe sets uniquely downregulated in AD: AD vs. Ctrl (p < 0.01) and AD vs. HuD + FTD (p < 0.05) but were not significantly (e < 0.1) enriched in any queried pathway. The 112 probe sets uniquely upregulated in HuD + FTD, that is, HuD + FTD vs. Ctrl (p < 0.01), and HuD + FTD vs. AD (p < 0.05), were enriched (e < 0.1) in cadherin-mediated cell adhesion (Additional file 3: Table S3d). The 115 probe sets uniquely downregulated in AD vs. Ctrl (p < 0.01), and AD vs. HuD + FTD (p < 0.05), were not significantly (e < 0.1) enriched in assessed biological pathways.

The emphasis here is on gene sets and associated biological pathways. Still, it is instructive to focus on several genes/proteins currently of great interest in CP pathophysiology. Altered tight junction protein claudin-5 in CP is associated with BCSFB breaching [26], while interferon has a protein-signaling role that couples choroidal-cerebral neuroimmune interactions [13]. Claudin-5 was downregulated 1.6-, 1.9- and 2.5-fold in AD, FTD and HuD, respectively, reaching significance in FTD (respective p values of 0.071, 0.028 and 0.059). Multiple interferon signaling genes were upregulated in AD: IFI-TM1 (p = 0.0008), IFN-AR1 (p = 0.006), IFN-AR2 (p = 0.0007) and IFN-GR2 (p = 0.0002). The complete lists of differentially-expressed genes, as well as enriched biological pathways, are provided in Additional file 2: Table S2, Additional file 3: Table S3, Additional file 4: Table S4.

Comparing gene expression in CP of AD subjects to that of Ctrls and other neurodegenerative diseases reveals important biological functions altered by dementia generation and progression. The CP as a dynamic interface between blood, CSF and brain, is able to monitor distortions and homeostatically respond, e.g., by the JAK-STAT pathway as well as cytokine and protein-signaling molecules [12, 13]. These homeostatic adjustments impact neural viability. This intimates that CP beneficial adjustments, as well as BCSFB malfunctioning, are pertinent to AD progression. The strategic role of CSF to safeguard brain is manifested by CP upregulation of many genes in response to neurodegeneration ([26] and this investigation); and by its ability to protect basic CSF composition (e.g., K, pH and vitamin homeostasis) even in advanced AD [7]. Nevertheless, the CP in AD incurs structural damage [20] and distorted epithelial metabolism and transport (e.g., Aβ, cytokine (e.g., TNFα) and methionine/homocysteine [24, 29]). The injured epithelium likely depends on enhanced mTOR expression to facilitate cellular repair and/or replacement. Accordingly, the state of CP–CSF viability in ongoing neurodegeneration is a balance between debilitating and restorative events at the BCSFB [29‐31].

In this study we have shown numerous transcriptional alterations in the CP of subjects with neurodegenerative disease. Given that the BCSFB makes adjustments in CSF composition, the prolific disease-induced expression changes in CP fits previous homeostasis modeling [1, 30]. Most transcript changes were common to AD, FTD and HuD, while fewer genes were modulated differently between AD and HuD + FTD. It is instructive to compare the present observations with an earlier transcriptional investigation of the BCSFB in AD [26]. Whereas we studied homogenates of entire CP, including epithelium and stroma, Bergen et al. studied single epithelial cells captured by laser microscopy [26]. Thus, Bergen et al. provided data on epithelium-specific changes in AD (larger-fold changes, see Fig. 1d–f), while our dataset also includes potential pathophysiological interactions between CP stroma and epithelium. Indeed, robust inflammatory responses were not reported by Bergen et al. [26], suggesting deductively that the immune-reactive pathophysiology occurs primarily within CP stroma.

Some limitations of our study include the small number of patients in the HuD and FTD groups, and the lack of direct confirmation of specific genes by RT-PCR. Future studies with a larger N value for HuD and FTD will increase the statistical power for disease comparisons. For post-mortem tissue, RNA stability is challenging with autopsy specimens collected at various PMIs; however, we carefully assessed RNA integrity, using only samples passing stringent criteria. Moreover, our emphasis was on significantly-enriched pathways (affecting multiple genes within a given pathway) rather than specific gene targets, with possible individual false discoveries. Any residual blood elements in specimens would unlikely explain differences in tissue mRNA among the three disease groups. The somewhat younger Ctrl group may be advantageous in avoiding potentially confounding issues with clinically-silent early dementia in an older ‘Ctrl’ cohort, otherwise presumed normal.

We used GeneGo (http://​www.​genego.​com) and Ingenuity (http://​www.​ingenuity.​com) sets for pathway analysis. In AD compared to Ctrls, this revealed upregulated inflammation genes: acute phase response, cell adhesion and cytokine, interferon, JAK-STAT signaling (for translating extracellular signals into transcriptional responses). Notable downregulated pathways were methionine degradation and protein translation; both are implicated in AD pathology. Claudin-5 expression was downregulated, consistent with the enhanced leakiness of the BCSFB [31] encountered in neurodegenerative diseases [26] and pathophysiology models [29]. Amyloid beta peptide (Aβ) damages CP tight junctions by activating matrix metalloproteineases, thereby increasing paracellular permeability [32]. The functional significance of such altered CP pathways, for disease outcome, awaits elucidation.

Prominent in AD was upregulated inflammation-related signaling. These inflammation signatures differ from cortex-associated microglial infiltration [33, 34]. Key marker genes of cortical inflammation-APOE, TREM2, TYROBP-did not upregulate in AD CP. Rather, acute phase response genes dominated the upregulation: multiple cytokine and interferon receptors, JAK-STAT signaling components, MAPK, NFκB signaling and cell adhesion. Cytokines and growth factors in disease-associated reactions in BCSFB come from brain [35], blood, or CP itself. The CP responds biochemically and transcriptionally to circulating cytokines, central injury and systemic diseases [18, 36, 37]. Localized CP immunoreactions (e.g., inflammation-resolving leukocyte trafficking) may benefit brain by sensing ‘injury signals’ flowing from brain to CSF to CP, then feeding back to make homeostatic neural adjustments [35].

At certain stages of advancing neuroinflammation (caused by brain-residing pro-inflammatory microglial responses to Aβ loads), the CP receives plasma interferon-γ as a signal to promote homeostatic transport of anti-inflammatory monocyte-derived macrophages into CSF for resolving parenchymal inflammation [12]. Disease-induced disruption of this neuroimmune interferon adjustment at the BCSFB [38] could compromise the ability of the CSF–brain to thwart AD exacerbation. The CP competently adapts to AD stress [39] by maintaining an immunosuppressant profile of factors, e.g., VEGF and TGFβ1 in CSF, to help manage brain inflammation after neuronal injury [40].

Increased expression of LRP-1, a choroidal Aβ transporter, agrees with mouse AD modeling [24]. Upregulated LRP-1 in the apical membrane expedites Aβ removal from CSF [24]. Augmented reabsorptive clearance of Aβ at the BCSFB aids the CNS because cerebral capillaries in AD extrude less Aβ [41]. This compensatory Aβ removal by CP counters the disabled microvessels [42]. Titers of inflammatory cytokines and choroidal proteins, in CSF and blood, present in different degrees in AD [43]. Activated astrocytes and microglia congregate in Aβ plaques [44]. The manner in which CP inflammatory-signaling molecules modify AD pathogenesis is heterogeneous. Acute inflammation may beneficially promote CSF clearance of affected cells and Aβ aggregates, protecting neurons. However, persistently-elevated CP–CSF cytokines and sustained activation of microglia adversely affect neurons. An effective CP will balance the beneficial vs. detrimental effects of CSF cytokine changes in AD, FTD and HuD.

Amyloid beta induces cytokine production; and astrocytes activated by Aβ, release inflammatory factors that sustain Aβ production. Clearly the CP–CSF, using soluble signals and upregulated cellular adhesion factors, appropriately distributes certain T cell phenotypes to CSF [12]; such leukocyte penetration into CSF helps to control neuroinflammation and Aβ levels in AD brain. The dynamic relationship between pro-inflammatory and anti-inflammatory cytokines in CP–CSF impacts neuroinflammation processes and AD pathology.

Peroxisome-proliferator-activated receptor (PPAR) signaling genes, including PPARδ and its obligate heterodimer RXRα, were enriched in AD CP. PPAR/RXRs are neuroprotective in AD and Aβ therapies due to anti-inflammatory and endothelial actions [43]. PPAR activation, through endogenous or synthetic ligands, likely protects CP by increasing antioxidant capacity and improving energy supply; this maintains fuel for the Na pump [45] and CSF secretion [46], and increases expression of Aβ transporters [47]. The novel GFT1803 agent (a pan-PPAR agonist that activates all 3 PPAR isoforms) attenuates Aβ loading-induced damage and neuroinflammation [48]. PPAR thus deserves attention as a potential pathway for restoring CP–CSF integrity in AD in order to counter neurodegeneration.

Significant expression differences were also observed between AD and FTD + HuD. The VEGF signaling pathway (including VEGFA and VEGF receptors FLT1 and FLT4) displayed significant upregulation in AD but not in FTD or HuD. This agrees with our previous findings of increased VEGF within AD CP [1]. VEGF is required for maintaining endothelial cell fenestration in CP capillaries [49], an important microstructural feature for delivering plasma substances into the choroidal interstitium for epithelial processing.

Cadherin, on the other hand, was upregulated in FTD and HuD but not AD. Cadherin is a superfamily of cellular adhesion molecules (CAM), that maintain tissue structure and boundaries between cells and organelles. CAM binding also modifies gene expression. Cell–cell adhesions mediate specific immune actions [50], of which there is a plethora in CP of FTD and HuD patient specimens. A cadherin family member prominent in CP is γ-protocadherin (γ-Pcdh), expressed at the apical membrane [51]. Mutation of γ-Pcdh causes ventricular collapse. Keep et al. proposed an immune and CSF dynamics role for CP γ-Pcdh [52], that co-expresses with the NaBCN2 Na transporter supporting CSF secretion. This gene may function in CP ion transport-CSF formation by way of apical–microskeletal membrane interactions with NaBCN2 that regulate ion trafficking [32]. Moreover, Kolmer immune cells, attached to CP apical surface [53], may have an altered function in neurodegenerative diseases when cadherin is upregulated.

We hypothesize that the downregulated expression observed in this study reflects failing metabolic pathways involved in choroid cell and CSF homeostasis. Reduction in methionine-degradation genes is intriguing given that excessive homocysteine, a product of methionine metabolism, is a risk factor for AD [54‐57]. Methionine loading increases brain homocysteine, Aβ and phospho-tau in mouse models [58]. Decreased expression in AD CP of the methionine-degrading gene may relate to elevated homocysteine levels in CSF [59]. The impact of augmented CSF homocysteine on raising brain Aβ and tau hints that additional methionine gene studies on CP transcription factors and metabolism in neurodegenerative diseases are needed.

We also determined a decreased expression of protein translation genes, including multiple eukaryotic translation initiation factors (EIF genes) and ribosomal proteins. CP has a major role in producing and secreting CSF proteins, e.g., transthyretin that stabilizes Aβ conformation. In AD there is decreased choroidal synthesis of transthyretin [24], lowering its CSF concentration [60]. Moreover, the heat stress glucose regulatory proteins 78 and 94 in human AD CP are diminished [39], implicating suboptimal glucose or calcium homeostasis. Altered heat stress proteins at the AD BCSFB deserve examination for impact on cerebral metabolism.

Expression of mTOR associates with controlling cell growth and proliferation [61], possibly a factor as damaged choroid epithelial cells need replacement. Our finding of increased fatty acid oxidation and upstream mTOR signaling (Fig. 2), with juxtaposed downregulated protein translation, fits existing concepts suggesting altered energy metabolism in AD onset and progression. While increased PPAR activity downstream of mTOR fits the compensatory adaptation to retain CP resiliency, there is a disconnect between upregulated mTOR and the downregulated protein translation machinery typically induced by mTOR. This suggests a break in normal mTOR signaling (see Fig. 2) that could undermine CP function or resiliency when challenged with neurodegeneration. This is significant because of CP’s pivotal role in providing brain with supportive factors and immune cells that migrate across BCSFB into CSF–brain. Studies need to assess the role of the dynamic CP transcriptome in providing resiliency to the BCSFB, in order to retain CSF homeostatic reserve for staving off neurodegeneration.

The AD transcript findings reported herein for bulk CP tissue compare favorably with and expand prior analysis in laser-captured epithelial cells [26]. Such concurrence is remarkable given the different tissue sampling, measurement platforms and patient cohorts. Highlights of our investigation include upregulated genes linked to inflammation and interferon neuroimmune homeostasis, as well as to JAK-STAT and mTOR; and downregulated genes for methionine degradation, protein translation and claudin-5 (tight junction). CP is a complex homeostatic tissue. The BCSFB undergoes deleterious, but sometimes functional and adaptive, changes in dementia-related pathophysiology. The enriched JAK-STAT and mTOR pathways (Fig. 2), respectively, are likely instrumental in promoting adaptive transcriptional responses and epithelial repair/replacement when CP is harmed by injuries associating with neurodegeneration. Analyzing biological pathway mechanisms expedites specific pharmacologic targeting.

Future transcriptome work with larger cohorts should delineate gene expression by demographic endpoints, Braak staging, Aβ plaque score, disease duration, ApoE genotype, co-morbidity, and other disease characteristics. The transcriptome distinctions here precisely describe CP–CSF function in, and response to, certain neuropathologies: AD vs. FTD vs. HuD. This categorical approach provides crucial knowledge on the BCSFB role in pathogenesis; and hopefully should improve prophylaxis of various neural diseases. The goal: To identify exact CP targets to exploit when implementing pharmacologic/genetic therapies to alleviate CSF–brain metabolic distortions in dementia.