Blood-brain barrier dysfunction in aging induces hyperactivation of TGFβ signaling and chronic yet reversible neural dysfunction

See allHide authors and affiliations

Science Translational Medicine  04 Dec 2019:
Vol. 11, Issue 521, eaaw8283
DOI: 10.1126/scitranslmed.aaw8283

Breaking barriers in aging and disease

The blood-brain barrier (BBB) regulates the communication between the vasculature and the brain. Aging and neurological disorders have been associated with BBB defects. Now, Milikovsky et al. and Senatorov et al. studied the consequences of BBB impairments in aging and disease. Milikovsky et al. found that in patients with epilepsy or Alzheimer’s disease, as well as in aging mice, BBB impairments were spatially associated with transient electroencephalographic abnormalities. Senatorov et al. extended the study at the molecular level, showing that BBB breakdown triggered transforming growth factor–β (TGFβ) signaling in astrocytes and cognitive impairments in aging rodents. Similar abnormalities were also found in brain tissue from aging individuals. TGFβ inhibition in aged mice reversed the pathological phenotype.


Aging involves a decline in neural function that contributes to cognitive impairment and disease. However, the mechanisms underlying the transition from a young-and-healthy to aged-and-dysfunctional brain are not well understood. Here, we report breakdown of the vascular blood-brain barrier (BBB) in aging humans and rodents, which begins as early as middle age and progresses to the end of the life span. Gain-of-function and loss-of-function manipulations show that this BBB dysfunction triggers hyperactivation of transforming growth factor–β (TGFβ) signaling in astrocytes, which is necessary and sufficient to cause neural dysfunction and age-related pathology in rodents. Specifically, infusion of the serum protein albumin into the young rodent brain (mimicking BBB leakiness) induced astrocytic TGFβ signaling and an aged brain phenotype including aberrant electrocorticographic activity, vulnerability to seizures, and cognitive impairment. Furthermore, conditional genetic knockdown of astrocytic TGFβ receptors or pharmacological inhibition of TGFβ signaling reversed these symptomatic outcomes in aged mice. Last, we found that this same signaling pathway is activated in aging human subjects with BBB dysfunction. Our study identifies dysfunction in the neurovascular unit as one of the earliest triggers of neurological aging and demonstrates that the aging brain may retain considerable latent capacity, which can be revitalized by therapeutic inhibition of TGFβ signaling.


Aging involves cognitive deterioration, which poses an increasing health care burden on today’s society with its prolonged life expectancy. Despite calls for a better understanding of brain aging and new therapeutic targets, the underlying mechanisms that cause decline in neural function in aging are still not understood. We approached this topic by exploring the causal involvement of the blood-brain barrier (BBB) pathology in age-related brain dysfunction. The BBB is a tightly regulated interface composed of specialized endothelial cells, pericytes, and astrocytic end feet that form a protective sheath around brain capillaries (1, 2). By restricting the free diffusion of blood-borne molecules, the BBB establishes a sequestered brain microenvironment, including the precisely balanced ionic concentrations needed for neural activity, the compartmentalization of brain-specific growth factors and signaling molecules, and the immune-privileged brain environment (1, 2). Thus, the BBB is a fundamental and essential component of healthy brain function.

Alarming observations of widespread BBB breakdown in aging patients were first reported in the 1970s (3), raising the possibility that vascular leakiness and infiltration of toxic blood-borne molecules into the brain could cause neural impairments and contribute to diseases such as dementia (2, 47). This theory has been primarily supported by human clinical evidence showing that BBB breakdown in aging individuals is strongly correlated with cognitive decline and Alzheimer’s disease (AD) (712). Furthermore, this association has been localized to functional brain subregions: BBB breakdown specifically in the hippocampus is associated with mild cognitive impairment (MCI) (13), suggesting that BBB dysfunction could be a cause of impairment within impacted tissues. However, these human studies have also yielded contradictory results, are complicated by variability in population and study methodology (14, 15), and are inherently correlative. There have been very few rodent studies that assess BBB status in aging (16), and whether and how BBB breakdown affects brain function in aging remain unclear. A mechanistic understanding of the biological consequences of BBB breakdown is critical to discern whether the association between BBB dysfunction and cognitive decline is spurious, correlative, or causal. Furthermore, mechanistic insights would have the potential to reveal druggable targets for the intractable health problems of age-related cognitive decline and dementia. An improved understanding of vascular contributions to cognitive decline and dementia has been designated as one of the highest priority research areas for potential breakthroughs in AD and related dementias (ADRDs) (17).

Although little is known about the consequences of BBB breakdown in aging, there are several other disease contexts that involve BBB dysfunction, which could reveal relevant candidate mechanisms. In particular, traumatic brain injury (TBI) not only causes severe BBB breakdown (1820) but also involves secondary symptoms of cognitive impairment and increases risk for dementia. In rodent models of TBI and leaky BBB, blood-borne proteins that infiltrate into the brain cause a robust injury response by activating the transforming growth factor–β (TGFβ) signaling pathway. Key mediators of this response include the serum protein fibrinogen, which carries and releases latent TGFβ in the brain (21), and albumin, which binds to the TGFβ receptor and activates signaling (22). In both cases, astrocytes act as the primary responders that detect the blood-borne ligands and transduce TGFβ signaling (2123). In turn, activated astrocytes release inflammatory cytokines and more TGFβ1 (22, 24, 25), form glial scars (21), and remodel neural circuits to cause hyperexcitability and dysfunction (22, 24, 2628). These studies point to the TGFβ signaling pathway as a candidate mechanism that could play a role in pathological outcomes following BBB breakdown in aging. To test this hypothesis, we used genetic and pharmaceutical interventions to test the causality of BBB dysfunction and TGFβ signaling in progressive neural dysfunction across the life span of naturally aging mice.


Progressive BBB dysfunction and albumin extravasation in the hippocampus starts in middle age

To establish the time course of age-related BBB decline, we quantified extravasation of serum albumin, an established marker of BBB permeability (2931), which also plays a mechanistic role in triggering TGFβ signaling (22). We focused our analysis on the hippocampus, a key brain region associated with age-related memory decline (13). Albumin was effectively absent from the hippocampus of young mice but was first detectable in the aging hippocampus starting as early as 12 months (“middle age”) and consistently elevated in aging up to 2 years, near the end of the life span (Fig. 1A and fig. S1A). These results indicate that BBB dysfunction appears earlier than has been appreciated, placing it among the earliest known indicators of aging in the rodent brain. The onset of BBB dysfunction at around 12 months corresponds with other early biological hallmarks of aging, including the typical onset of reproductive senescence in female mice, as well as the earliest appearance of MCIs in various behavior tasks (3237). To confirm age-related BBB dysfunction, we raised a separate cohort of aged mice and used an alternate method to quantify BBB permeability based on detecting leakage of a fluorescent tracer, Evans blue (EB), into the brain after intravenous injection (38). Compared to 2- to 4-month-old young mice, 12- and 21- to 24-month-old aged mice showed elevated extravascular EB tracer localized to hippocampal cells (fig. S1, B and C), thus reproducing our finding of age-related BBB breakdown by an independent method.

Fig. 1 Progressive BBB dysfunction in aging mice is associated with elevated TGFβ signaling and aberrant network activity.

For all measures, groups were compared by Bonferroni post hoc test. Sample sizes are n = 4 (3 months), n = 5 (12 months), n = 3 (18 months), and n = 6 (21 months). (A) Representative images from immunofluorescence-stained brain sections used to quantify progression of BBB decline in the aging mouse hippocampus at 3, 12, 18, and 21 months old. Astrocytes are labeled by GFP expressed under the pan-astrocytic promoter Aldh1L1 (Rep-aldh1L1-eGFP), albumin is labeled by immunostaining, and cell nuclei are labeled by 4′,6-diamidino-2-phenylindole (DAPI) (blue). Scale bar, 100 μm. (B) Representative image from a brain section of an 18-month-old mouse immunostained to investigate colocalization of albumin and pSmad2 (arrows) in Rep-aldh1L1-eGFP astrocytes. Scale bar, 30 μm. (C) Percentage of cells colabeled for albumin and astrocyte reporter in the hippocampus of 3-, 12-, 18-, and 21-month-old mouse brain (ANOVA, P < 0.0001). (D) Percentage of cells colabeled for pSmad2, albumin, and astrocyte reporter (ANOVA, P < 0.0001). (E) Western blot analyzing TGFβ1 and pSmad2, outputs of the TGFβ signaling pathway, and actin in hippocampus from young (n = 5) and old (n = 5) mice (pSmad2, P = 0.019; TGFβ, P = 0.018, t test with Welch’s correction). (F) At 3, 12, 18, and 24 months, seizures were induced in mice by PTZ injection to assay hyperexcitability [n = 13 (3 months), n = 10 (12 months), and n = 8 (18 and 24 months)]. Seizure severity in mice was quantified by fitting linear regression slopes to measures of progression through each stage of seizure in the modified Racine scale. Decreased linear regression slopes indicate faster progression through all stages of seizure severity (one-way ANOVA, P = 0.001 with Bonferroni posttest). (G) Latency to mortality caused by severe seizures was measured (one-way ANOVA, P < 0.0001 with Bonferroni posttest). (H) A representative trace of ECOG recording showing a PSWE with slow wave activity less than 5 Hz within a 10-s window (marked with *). (I) Number of PSWEs per day was counted in young (n = 5) and old mice (n = 18) (Mann-Whitney test, P = 0.019). For all tests, *P < 0.05, **P < 0.01, and ****P < 0.001.

Age-dependent albumin accumulation is cell-type specific

Albumin is known to accumulate in astrocytes when it extravasates through a dysfunctional BBB (22, 24, 39). We therefore used a transgenic mouse line that comprehensively labels astrocytes via the Aldh1L1 (aldehyde dehydrogenase 1 family member L1) promoter (Rep-Aldh1L1) (40, 41). Expression of this marker was stable and did not change across the life span (fig. S1D). To estimate the specificity of albumin uptake in different cell types, we performed immunostaining for markers of microglia (Iba1), oligodendrocytes (CAII), and neurons (NeuN) and quantified the percentage of all albumin-positive cells for each cell type in the aged (18 to 24 months) mouse hippocampus. Albumin colocalized predominantly with Aldh1L1-positive astrocytes, accounting for approximately 60% of albumin-labeled cells, while each of the other cell types colabeled approximately 20% of all albumin-positive cells (fig. S1E). Concurrent with the time course of overall albumin extravasation and BBB dysfunction, substantial astrocytic uptake of albumin was first detected at 12 months and further increased with age up to 21 months old (P < 0.0001; Fig. 1C).

Age-dependent activation of aberrant TGFβ signaling in astrocytes

Albumin endocytosis into astrocytes is mediated by binding to the TGFβ receptor (TGFβR) composed of subunits TGFβRI and TGFβRII. Albumin binding, in turn, activates the TGFβRI ALK5 (22, 28) to induce phosphorylation of the Smad2 protein (pSmad2) and carry out signal transduction of the ALK5-TGFβ signaling cascade (22). Furthermore, this activation also increases the production of TGFβ1 in astrocytes (28) and activation of latent TGFβ1 protein from extracellular matrix (26), yielding an increase in the canonical ligand of TGFβR and therefore amplification of the TGFβ cascade. Hence, we next investigated the relationship between albumin uptake and TGFβ signaling in the aged mouse brain by quantifying immunolabeled phosphorylated Smad2 protein (pSmad2). Concurrent with the time course of albumin extravasation, aging mice showed a progressive increase in the amount of pSmad2 colocalized with albumin-positive astrocytes (Fig. 1, B and D). Activation of the TGFβ pathway was further quantified by Western blot, showing increased concentration of pSmad2 in the hippocampus of old mice compared to young and increased concentrations of active TGFβ1, a positive feedback output of the TGFβ pathway (Fig. 1E).

Network hyperexcitability in aged mice

On the basis of the finding that BBB dysfunction and TGFβ signaling cause hyperexcitability after head injury (22, 24, 2628, 39), we hypothesized that similar hyperexcitability may be triggered by BBB decline and contribute to cognitive impairment in aging mice. Hippocampal hyperexcitability is an early biomarker of MCI in humans that precedes progression to AD (42, 43) and is also an early marker of disease progression in rodent AD models (44, 45). Thus, we used the pentylenetetrazol (PTZ) seizure assay to investigate the time course for onset and the progression of hyperexcitability in aging mice. In this assay, injection of PTZ, a noncompetitive γ-aminobutyric acid receptor blocker, induces seizures that have been shown to be a reliable readout of underlying hippocampal hyperexcitability associated with aging and AD models (45) (mice with higher hyperexcitability exhibit lower induced seizure threshold). Compared to 3-month-old mice, aged mice showed increased severity of induced seizures (quantified based on rate of progression through the Racine seizure severity scale; fig. S1F), beginning at the 12-month-old time point (Fig. 1F). Furthermore, old mice were highly vulnerable to mortality from severe seizures, with a shorter latency to mortality (Fig. 1G).

Aberrant paroxysmal slow wave events in aged mice

Next, we sought to measure and characterize hyperexcitability recording telemetric electrocorticography (ECOG) using epidural electrodes implanted in young (3 months old) and old (18 to 24 months old) mice over a period of 5 days in the home cage. We found that aged mice showed an increase in the relative power of slow wave activity (<5 Hz) (fig. S1G), similar to electroencephalogram (EEG) slowing described in patients with dementia (4648), which is thought to reflect dysfunctional neural networks. Detailed analysis of this aberrant ECOG signal revealed that the slow wave activity was not continuous, but rather manifested in discrete, transient paroxysmal slow wave events (PSWEs) (median frequency <5 Hz; Fig. 1H), which were elevated in aged mice relative to young mice (Fig. 1I). These PSWEs, characterized by a median power of less than 5 Hz over 10 consecutive seconds, were further identified also in patients with AD and epilepsy and in the 5x familial AD (5xFAD) mouse model (49) and similar to paroxysmal ECOG events that have been observed in epileptogenic animals that have hippocampal hyperexcitability (50). Given the potential association between hyperexcitability and seizures, we also used an automated seizure detection algorithm (24, 50) to search for spontaneous seizures in the ECOG recordings and found that aged mice had no seizure events and were not epileptic. Furthermore, the PSWEs were elevated throughout the circadian cycle in old mice and not associated with the inactive sleep period (fig. S1H) (49). Thus, the PSWEs constitute distinct subclinical paroxysmal events that are associated with hyperexcitability. During analysis, we observed that approximately half of the aged mice showed a phenotype of high PSWEs, whereas others showed a relatively low number of PSWEs, similar to young mice, suggesting heterogeneity in the incidence of network dysfunction in aging mice. A blind Gaussian mixture model algorithm to cluster PSWE frequency (per minute) confirmed a best fit for the aged mouse population as two distinct clusters of high- and low-occurrence groups (hPSWE and lPSWE, respectively) (fig. S1, I to K).

Together, our initial studies showed that BBB breakdown and astrocytic TGFβ signaling occur at early stages of mouse brain aging, concurrent with hyperexcitability and neural network dysfunction. Next, we performed comparative human studies to investigate whether similar BBB dysfunction and TGFβ signaling are also present in aging human subjects.

The aging human brain shows progressive BBB dysfunction associated with astrocytic TGFβ signaling

In our rodent studies, we directly measured the progression of albumin extravasation and TGFβ signaling across the life span. BBB dysfunction has also been reported in aging humans including patients with MCI and ADRD, using indirect measures such as cerebrospinal fluid (CSF) sampling or imaging, which can be technically challenging and may produce conflicting results (15). Thus, we sought to further quantify age-related BBB dysfunction in human subjects, expanding on results reported elsewhere (8, 10, 12, 13). We used dynamic contrast-enhanced magnetic resonance imaging scanning (DCE-MRI) to quantify BBB permeability in 113 healthy human subjects with an age range of 21 to 83 years old (Fig. 2A and table S1). We first established normal brain permeability based on the permeability values in healthy, young subjects (age 21 to 40), setting an upper value for “normal permeability” at the 95th percentile (95% of brain voxels in the averaged healthy young brain were below this value). On the basis of this threshold, we constructed permeability maps for each subject and quantified the percentage of “leaky” voxels in the whole brain (Fig. 2A). This analysis revealed a linear increase in BBB permeability in aging (fig. S2A). To estimate the overall prevalence of BBB dysfunction, we further categorically classified individuals as either BBB-intact (BBB-I) or BBB-disrupted (BBB-D; permeability in more than 5% of brain volume). By age 60, nearly half of the population was affected by BBB-D (fig. S2B).

Fig. 2 Aging patients show decline in the integrity of the BBB and increased TGFβ signaling.

(A) Top: Representative images from DCE-MRI scans of a young subject (30 years old) and an old subject (70 years old) using an MRI-sensitive contrast agent, Gd-DTPA (diethylenetriamine pentaacetic acid), which does not cross the intact BBB. Intensity of signal reflects relative BBB permeability [in arbitrary units (a.u.)]. Bottom: Average permeability maps from all young (ages 20 to 40) and old (ages 60 to 80) subjects (n = 105). (B) Percentage of voxels classified as disrupted BBB in the right hippocampus in young and aged subject groups as under (A). Permeability was calculated for each brain voxel using in-house MATLAB script (MathWorks, USA), as described in Materials and Methods. (C) Representative images of immunofluorescent staining of postmortem human hippocampal tissue from young (31.3 ± 5 years) and old (70.6 ± 5.6 years) individuals, showing albumin in the aging hippocampus, localized in astrocytes (GFAP) and colocalized with pSmad2, indicated by arrows. Nuclei are visualized by DAPI stain. Scale bars, 50 μm. (D) Quantification of cells immunostained with astrocyte marker GFAP, albumin, and pSmad2 [t test, P = 0.0324; n = 3 (young) and n = 10 (old)]. (E) Scatterplots and best-fit linear regressions of TGFβ expression levels [transcripts per million (TPM)] from the hippocampus of human subjects aged 20 to 70 years old (n = 123 subjects), accessed from the GTEx project expression database. Aging is significantly correlated with increasing hippocampal expression of TGFβ1 (Pearson correlation, r = 0.195, P = 0.031) and TGFβ3 (r = 0.184, P = 0.042) isoforms but not TGFβ2 (r = −0.028). For all tests, *P < 0.05, **P < 0.01.

BBB breakdown, specifically in the hippocampus, has been associated with the incidence of MCI (13). Furthermore, hippocampal atrophy and, in particular, asymmetrical atrophy are early biomarkers that predict the transition from cognitively normal to MCI to ADRD (5156). Thus, we used anatomical localization to specifically assess BBB permeability in the left and right hippocampus and found that aging individuals showed asymmetric BBB breakdown localized to the right hippocampus (Fig. 2B).

We next sought to complement the indirect MRI approach by directly assessing BBB dysfunction and its association with astrocytic TGFβ signaling in aging human brains. We examined postmortem tissue from young (31.3 ± 5 years old) and aging (70.6 ± 5.6 years) human subjects with no history of brain disorder (table S2) and directly measured colocalization of albumin extravasation and TGFβ signaling in astrocytes as performed in our rodent studies. We found high concentration of the serum protein albumin in the old hippocampus, which was absent in the young (Fig. 2C). Albumin was detected in astrocytes [identified by the astrocytic marker glial fibrillary acidic protein (GFAP)] and colocalized with pSmad2, the primary signaling protein of the canonical TGFβ signaling cascade. The number of albumin-pSmad2–colabeled astrocytes was increased in old versus young individuals (Fig. 2D).

This immunostaining indicated that BBB dysfunction and albumin induce TGFβ signaling in the aging human hippocampus, as observed in our aging rodent studies. In turn, this would be predicted to cause positive feedback and elevated concentration of TGFβ. To assess this, we obtained human brain transcriptome data from the publicly available Genotype-Tissue Expression (GTEx) project (57), with data from 123 human subjects ranging in age from 20 to 70 years old, and investigated brain transcript expression of the major TGFβ isoforms (TGFβ1, TGFβ2, and TGFβ3) in the human hippocampus. These data showed that hippocampal expression of the ligands TGFβ1 and TGFβ3 increases with age in human subjects (Fig. 2E).

These studies showed that BBB dysfunction and TGFβ signaling are early indicators of aging in both rodents and humans and are associated with symptomatic hyperexcitability in rodents. Next, we used gain-of-function and loss-of-function interventions to determine whether albumin-induced TGFβ signaling plays a causal role in age-related symptoms of neural dysfunction and cognitive impairment.

Infusion of albumin into young brains causes hyperexcitability, PSWEs, and memory impairments

To test whether TGFβ signaling, as induced by BBB dysfunction, is sufficient to cause symptomatic pathology associated with aging, we infused albumin (iAlb) or control artificial CSF (aCSF) into the brain ventricles of healthy, young adult rats and mice via osmotic mini-pumps (fig. S3A). This method of albumin infusion is an established model for BBB dysfunction and blood protein extravasation and robustly triggers TGFβ signaling (26, 28, 50). After infusion, the exogenous albumin diffused readily into the ipsilateral hippocampus and was taken up by astrocytes within 48 hours of infusion (fig. S3, A and B). We then assessed outcomes in young rodents (Fig. 3A) to determine whether iAlb is sufficient to cause aged-like symptoms of network and cognitive dysfunction. We conducted the hyperexcitability PTZ seizure assay in young mice 48 hours after iAlb. The iAlb mice showed increased seizure severity and mortality induced by PTZ compared to aCSF controls (Fig. 3, B and C, and fig. S3C), fully reprising the hyperexcitable seizure vulnerability observed in naturally aged mice. Furthermore, we directly investigated network dysfunction via recordings of ECOG activity in young iAlb rats and found that iAlb caused a symptomatic slowing of ECOG activity and an increased PSWE count (fig. S3, D to G, and Fig. 3D) similar to the aberrant activity observed in aged rodents. This increased count of PSWEs was observed only in recordings from the ipsilateral hemisphere receiving iAlb infusion but not in the contralateral hemisphere, indicating specificity of the aberrant neural activity to the tissue affected by iAlb (fig. S3, F and G).

Fig. 3 Induction of TGFβ signaling in young rodents causes aberrant network activity, vulnerability to induced seizures, and cognitive impairment.

(A) Timeline of outcome measures (end points) taken in cohorts of young adult rats and mice after intracerebroventricular albumin infusion (iAlb). (B) Young mice received iAlb or aCSF control infusion via ALZET micro-osmotic pumps into the right lateral cerebral ventricle for 48 hours, followed by PTZ seizure induction (n = 4 in each group). Seizure severity in mice was quantified by fitting linear regression slopes to measures of progression through each stage of seizure in the modified Racine scale. Decreased linear regression slopes indicate faster progression through all stages of seizure severity (t test with Welch’s correction, P = 0.0497). (C) Time to mortality (minutes) was measured following PTZ injection (t test, P = 0.004). (D) ECOG activity was recorded from rats following 1-month iAlb (n = 8) or aCSF (n = 5) control via ALZET micro-osmotic pumps into the right lateral cerebral ventricle. Presented are numbers of PSWEs per day in the ipsilateral hemisphere receiving infusion and the contralateral hemisphere (Mann-Whitney test, P = 0.007). (E) Young mice received iAlb or aCSF control infusion via ALZET micro-osmotic pumps into the right lateral cerebral ventricle for 7 days (aCSF, n = 9; iAlb, n = 10) and then tested in the MWM task 1 month later [repeated-measures ANOVA, P < 0.0001 (main effect of learning over time) and P = 0.0093 (main effect of group), with Bonferroni posttest on day 9 showing significant differences between aCSF and iAlb, P < 0.001]. For all tests, *P < 0.05, **P < 0.01, and ***P < 0.005.

We next investigated whether iAlb is sufficient to cause cognitive impairment in young mice. We implanted mice with iAlb (or aCSF control) osmotic pumps for 1 week. One month later, we tested mice in the Morris water maze (MWM) spatial memory task. Mice that received iAlb infusion were impaired in memory performance over 9 days of MWM training compared to control mice with aCSF implant (Fig. 3E). Together, these data show that iAlb is sufficient to induce TGFβ signaling and confer a dramatic “old age” phenotype in young rodents, including aberrant neural ECOG activity, hyperexcitable seizure vulnerability, and cognitive impairment.

Genetic knockdown of astrocytic TGFβ signaling reverses pathological outcomes in aging mice

To test the causal role of astrocytic TGFβ signaling in age-related impairments, we generated a transgenic mouse line (aTGFβR2/KD) expressing inducible Cre recombinase under the astrocyte-specific glial high-affinity glutamate transporter (GLAST) promoter. This enables conditional knockout of the floxed (fl) TGFβR with temporal precision, specifically in astrocytes (Fig. 4A), allowing us to interrogate the role of astrocytic TGFβ signaling in mediating pathological outcomes. Treatment with tamoxifen induced efficient recombination in approximately 40% of hippocampal astrocytes but not in neurons (fig. S4, A to C) and reduced amount of TGFβR (fig. S4D), thus effectively causing knockdown (KD) to inhibit, but not fully abolish, TGFβ signaling. We aged cohorts of aTGFβR2/KD mice to early (12 to 16 months) and late (17 to 24 months) stages of aging and then induced TGFβR2 KD in astrocytes to test their role in age-related hyperexcitability and cognitive dysfunction (Fig. 4B). After induction, aged KD mice showed a decrease in hippocampal pSmad2 protein and reduction in expression of TGFβR2 (fig. S4, E and F).

Fig. 4 Knockdown of astrocytic TGFβ reverses neurological outcomes in aging mice.

(A) Schematic of the transgenic aTGFβR KD system. The astrocytic GLAST promoter drives expression of Cre recombinase in astrocytes. After induction with tamoxifen injection, activated Cre excises the TGFβR gene at inserted floxed (fl) loxP sites. LacZ reporter expression provides a readout of Cre activity. (B) Experimental timeline. KD was induced in early (12 to 16 months) and late (17 to 24 months) aged mice, and T-maze testing was performed 35 days later. At 65 days after induction, mice were tested for vulnerability to PTZ-induced seizures. (C) Linear regression slopes measuring progression through each stage of seizure in the modified Racine scale measured following PTZ injection in 12- to 16-month-old aTGFβR KD (Tgfbr2fl/fl) as compared to control heterozygous mice (Tgfbr2fl/+) and mice given oil control rather than tamoxifen induction (Tgfbr2+/+) mice [one-way ANOVA, P = 0.0019, with Bonferroni posttest; n = 5 (Tgfbr2+/+) and n = 6 (Tgfbr2fl/+ and Tgfbr2fl/fl)]. Latency to mortality was measured in minutes after PTZ induction (one-way ANOVA, P = 0.022, with Bonferroni posttest). (D) Linear regression slopes measuring progression through each stage of seizure in the modified Racine scale measured after PTZ injection in 17- to 24-month-old mice with aTGFβ KD mice [t test with Welch’s correction, P = 0.002; n = 5 (Tgfbr2fl/+) and n = 9 (Tgfbr2fl/fl)] and time to mortality (t test with Welch’s correction, P = 0.004). (E and F) Percentage of correct choices in T-maze aTGFβR KD mice and heterozygous controls at early aging [12 to 16 months old, Mann-Whitney test, P = 0.0273; n = 21 (Tgfbr2fl/+) and n = 20 (Tgfbr2fl/fl)] and late aging assessments [t test, P = 0.035; n = 12 (Tgfbr2fl/+) and n = 11 (Tgfbr2fl/fl)]. (G) An additional cohort of 12- to 16-month-old mice was tested in T maze, and hippocampi were dissected for Western blot analysis of TGFβ signaling (pSmad2) to assess individual outcomes. Correlation of pSmad2 amount to percentage of correct choices in T maze across individuals of both genotypes is presented (Tgfbr2fl/+ and Tgfbr2fl/fl) (Pearson’s correlation, r = −0.598, P = 0.024; n = 14). For all tests, *P < 0.05, **P < 0.01, and ***P < 0.005.

In early aging (12 to 16 months), mice with homozygous induced KD of the aTGFβR2 (fl/fl) were protected against symptomatic hyperexcitability, showing low vulnerability to PTZ-induced seizures and mortality (Fig. 4C and fig. S4G) that was similar to young mice. In contrast, 12- to 16-month-old control mice that were heterozygous for floxed TGFβR2 (fl/+) (and hence retained intact astrocytic TGFβ signaling) or that were homozygous for floxed TGFβR2 but injected with vehicle instead of tamoxifen (hence no KD induction; aTGFβR2+/+) showed increased seizure vulnerability and mortality compared to young and aTGFβR2 KD mice (Fig. 4C). The KD intervention was also effective at late aging time points: 17- to 21-month-old aTGFβR2 heterozygote (fl/+) controls showed typical age-related vulnerability to induced seizures, whereas aTGFβR2 KD (fl/fl) displayed low vulnerability and mortality to PTZ challenge (Fig. 4D and fig. S4H). These results show that genetic KD of astrocytic TGFβ signaling is sufficient to reverse symptoms of hyperexcitability in the PTZ assay at both early and late stages of aging.

Genetic KD of astrocytic TGFβ signaling reverses cognitive impairments in aging mice

To assess the role of astrocytic TGFβ signaling in age-related cognitive decline, we tested aged transgenic mice for spontaneous alternation in the T maze, a hippocampal working memory task (58) that is impaired in aging rodents (59). The task is optimal for assessing aging rodents because it can be performed rapidly without extensive training and is sensitive to mild impairments in hippocampal function (60, 61), and yet it is relatively unaffected by motor and vision impairments that may confound aging mice in traditional tasks such as MWM (58). At both early and late aging time points, aTGFβR2 KD mice made more correct choices in the T-maze task, indicating improved working memory compared to heterozygous controls (P = 0.0273 for the early aging group and P = 0.035 for the late aging group; Fig. 4, E and F).

Whereas aTGFβR2 KD was effective at improving cognitive function in both early and late stages, we also found greater heterogeneity in cognitive scores in the early (12 to 16 months) aging group, as would be expected for an early stage of aging in which some mice may have transitioned into cognitive impairment, while others remain cognitively healthy. To investigate this individual variability, we performed T maze in an additional cohort of early aging (12 to 16 months old) aTGFβR2 KD and heterozygous controls and collected dissected hippocampi to quantify the relationship individual cognitive scores and pSmad2, the molecular marker of TGFβ signaling. Across both heterozygous (fl/+) and homozygous (fl/fl) genotypes, pSmad2 concentrations were negatively correlated with T-maze performance (P = 0.024; Fig. 4G), providing further evidence for the role of TGFβ signaling in cognitive impairment.

Together, these results show that targeted inhibition of the TGFβ signaling pathway, via induced KD in astrocytes, is sufficient to reverse the outcomes of seizure vulnerability and cognitive impairment in a hippocampal spatial working memory task in old mice and that cognitive outcomes in a heterogeneous “mildly impaired” early aging cohort are correlated with the individual extent of TGFβ signaling.

A small-molecule TGFβR1 kinase inhibitor blocks iAlb-induced effects in the young brain

Our findings in mice and human brains support an evolutionarily conserved role of astrocytic TGFβ signaling in the pathogenesis of age-related neurological vulnerability, further indicating a therapeutic potential in targeting TGFβR. Thus, we next tested the efficacy of a small-molecule TGFβR1 kinase inhibitor, IPW (62). IPW has a promising clinical profile, including the ability to cross the BBB and good stability after oral dosing (fig. S5, A and B), making it suitable for once-per-day dosing to achieve inhibition of TGFβR signaling (62). We first tested IPW in young mice with TGFβ signaling induced by iAlb, treating them with daily intraperitoneal injections (20 mg/kg) of IPW for 2 days after pump implant. IPW treatment reduced pSmad2 amounts measured in the dissected hippocampus of treated mice compared to mice treated with vehicle control (Fig. 5, A and B). When examining the vulnerability to PTZ, iAlb mice treated with vehicle showed high seizure vulnerability, replicating our previous results, whereas IPW treatment effectively reversed this vulnerability, reducing both seizure severity and mortality to the values seen in the control aCSF-infused mice (Fig. 5, C and D, and fig. S5B). These data supported the likelihood for IPW efficacy in naturally aged mice, since it showed not only excellent target engagement but also efficacy in reducing symptomatic hyperexcitability induced by iAlb.

Fig. 5 IPW reduces TGFβ signaling and seizure vulnerability in young mice infused with albumin.

(A) Representative immunofluorescent images from iAlb mice after treatment with IPW or vehicle injections during 7 days of intracerebroventricular albumin infusion. Scale bar, 100 μm. Dotted box indicates region of inset image; arrows indicate examples of astrocytes colabeled with pSmad2. (B) Percentage of cells that positively immunoassayed for pSmad2 and colocalization of pSmad2 and the transgenic astrocytic reporter following 7 days injection of IPW or vehicle control in mice that received intracerebroventricular infusion of albumin (Alb + IPW, n = 6; Alb + veh, n = 6) or aCSF (aCSF + veh, n = 6) (one-way ANOVA with Neuman-Keuls multiple comparison test: % pSmad2, P = 0.027; % pSmad2+rep+, P = 0.0048). (C) Linear regression slopes measuring progression through each stage of seizure in the modified Racine scale measured after PTZ injection in young mice that received intracerebroventricular infusion of albumin for 2 days, with daily injections of IPW or vehicle control (Alb + IPW, n = 7; Alb + veh, n = 6) or intracerebroventricular infusion of aCSF with daily injections of vehicle (aCSF + veh, n = 6) [one-way ANOVA, P = 0.004 (main effect of treatment), with Bonferroni posttest]. (D) Survival time in minutes after PTZ injection was quantified in the same cohort (one-way ANOVA with Tukey’s posttest, P = 0.003). For all tests, *P < 0.05, **P < 0.01.

Drug inhibition of TGFβ signaling reverses molecular and functional brain aging in mice

On the basis of these validation studies, we then tested IPW as an intervention against TGFβ signaling in mice aged to 2 years old, near the end of the life span. Immunofluorescent analysis of hippocampal sections from aged mice revealed that 5 days of treatment with IPW (20 mg/kg, intraperitoneal) reduced the number of astrocytes colabeled with pSmad2 (Fig. 6A). Similarly, hippocampal Western blot showed that 5 days of IPW treatment reduced the amount of pSmad2 in aged mice, thus restoring TGFβ signaling extent similar to that of young mice (Fig. 6B). Furthermore, IPW treatment reduced the downstream output TGFβ1 (Fig. 6C).

Fig. 6 IPW reverses TGFβ signaling, aberrant neural activity, seizure vulnerability, and cognitive impairment in aged mice.

(A) Aged Rep-Aldh1L1 mice were treated for 7 days with IPW (20 mg/kg, intraperitoneal; n = 6) or vehicle (old, n = 5), followed by immunofluorescent staining for pSmad2 and albumin (Alb). Presented are fractions of astrocytes (rep+) that colabeled with pSmad2 (t test, P = 0.037) and the fraction of cells triple-labeled with pSmad2, Alb, and rep (t test, P = 0.014). (B and C) Protein extracted from the hippocampus of young (n = 4), old (n = 4), and old + IPW (n = 4 and 5 days) mice was ran on Western blot and analyzed by densitometry for amounts of pSmad2 and TGFβ and normalized for the control protein glyceraldehyde-3-phosphate dehydrogenase (GAPDH) [pSmad2: ANOVA, P = 0.0001 (main effect), with Bonferroni posttest; TGFβ1: ANOVA, P < 0.0001 (main effect), with Bonferroni posttest]. (D) Old mice received 7 days of IPW (n = 9) or vehicle treatment (n = 9) and were assessed for PTZ-induced seizure by progression through the Racine scale (t test, P < 0.0001) and onset of seizure induced mortality (t test, P < 0.0001). (E) Number of PSWEs per day is presented from continuous ECOG recording over 15 days and was used to investigate effects of IPW on aberrant network activity. During 5 days of baseline recording, clustering analysis was used to classify symptomatic old mice as affected by high numbers of PSWEs (hPSWE, n = 6) or asymptomatic old mice with low numbers of PSWEs (lPSWE, n = 6) that were similar to the level of young mice (n = 5). In the subsequent 5 days of recording, daily IPW injections were administered to all mice (Dunn’s multiple comparisons test, P = 0.0098). After IPW dosing was halted, five more days of recording were collected (Dunn’s multiple comparisons test baseline versus washout, P = 0.0002). (F) Percentage of correct choices in the T-maze test of old mice injected with IPW (n = 14 for 7 days) or vehicle control (n = 12) (t test, P = 0.028). (G) Percent time spent with a novel object in the novel object memory task of old mice injected with IPW (n = 8 for 7 days) or vehicle control (n = 10) (t test, P = 0.017). For all tests, *P < 0.05, **P < 0.01, ***P < 0.005, and ****P < 0.001.

To test whether IPW inhibition of TGFβ signaling could reverse the symptoms of neural hyperexcitability in aged mice, we treated 24-month-old mice with IPW for 7 days and then performed the PTZ assay. As in aTGFβR2 KD genetic intervention, mice treated with IPW showed lower seizure severity and mortality compared to aged control mice treated with vehicle (Fig. 6D and fig. S6A). To further investigate the efficacy of IPW on aberrant neural activity, we conducted a longitudinal experimental design in which young and old mice from the ECOG cohort were continuously recorded for 5 days of baseline, followed by 5 days of IPW treatment, and then 5 days of “washout” (no further dosing). This design was intended to assess not only the acute efficacy of IPW for treating symptoms of aberrant ECOG activity but also whether any treatment effects persist after dosing is halted.

In the longitudinal experimental design, we thus analyzed these subgroups separately to assess the efficacy of IPW treatment on the hPSWE phenotype. In hPSWE mice, treatment with IPW markedly reduced the number of PSWEs, restoring a profile of ECOG activity similar to that of young mice (Fig. 6E). In contrast, IPW treatment had no effect on lPSWE mice or on young mice, which did not show any aberrant high PSWE activity. Furthermore, treatment of aged mice with vehicle control showed no efficacy on reducing PSWEs (fig. S6, B and C). Beyond the treatment phase, the efficacy of IPW on reducing hPSWEs also persisted through the end of the washout period. Although IPW concentration was not measured at this end point, pharmacokinetic measurements show clearance concentration of IPW at 48 hours (fig. S5B). Although we cannot explicitly rule out the possibility of IPW accumulating to higher concentration over the experimental 5-day dosing period with our current data, we still expect the most likely scenario that IPW concentration would have decreased below effective concentration in the washout period, indicating that inhibition of TGFβ signaling may mediate a long-lasting change in the underlying hyperexcitability of neural circuits.

Given the effects of IPW on inhibiting TGFβ signaling and reversing associated outcomes of hyperexcitability and ECOG neural network dysfunction, we next assessed functional cognitive outcomes. Aged mice were treated for 7 days with IPW or vehicle control and then assessed in two cognitive behavioral tasks performed over consecutive days: spontaneous alternation in T maze and the novel object task, which is also sensitive to age-related memory decline (63, 64). After 7 days of IPW treatment, aged mice showed improvement in both cognitive tasks, relative to vehicle-treated controls (Fig. 6, F and G), demonstrating that IPW is effective in improving cognitive impairment in aged mice. Together, these studies show that IPW inhibition of chronic TGFβ signaling in aged mice can rapidly restore a “youthful” profile of network activity and cognitive capacity.


Aging is often accompanied by cognitive decline even in the absence of dementia or measurable neurodegeneration (65, 66). Unlike transgenic models for artificially inducing age-like disease, our investigations focused on naturally aging mice, allowing us to observe the relative sequence of biological changes associated with brain aging. We found that BBB dysfunction and consequent albumin extravasation appear as early as middle age in mice and in humans, placing it among the earliest known hallmarks of the aging brain. Consistent with our findings, relatively subtle changes in neural and synaptic function have been widely observed in humans and other mammals as one of the first signs of neurological aging (6568), and these changes in neurotransmission are associated with hippocampal hyperexcitability that is thought to be one of the earliest events in the progression of MCI (43, 44, 69). However, the regulatory pathways that may trigger or control these changes are unknown. We found that microvascular BBB dysfunction allows for the extravasation of serum albumin into the brain and hyperactivation of TGFβ signaling (fig. S7), similar to the activation of TGFβ signaling that has been shown in head injury models (22). Activation of this signaling cascade, in turn, causes symptoms associated with aberrant neural function (fig. S7).

A limitation is that these results should not be interpreted as implying that albumin is the only serum factor involved in triggering pathology after BBB dysfunction, as fibrinogen has also been implicated in activation of TGFβ signaling (21), and there may be other, yet undiscovered blood-borne signals. We used gain-of-function and loss-of-function experimental designs to demonstrate that TGFβ signaling is a causal mechanism underlying pathological outcomes associated with age-related BBB dysfunction. Infusion of albumin, a model of BBB dysfunction, was sufficient to cause hyperexcitability, aberrant neural activity, and cognitive impairment in young rodents. We chose this model based on extensive previous studies showing that albumin binds TGFβR and robustly activates TGFβ signaling (2224, 28, 50, 70). Thus, the loss-of-function interventions, including a genetic KD of TGFβR targeted specifically to astrocytes and a small-molecule TGFβR inhibitor, are critical in further establishing the specificity and causality of the TGFβ mechanism. Both interventions ameliorated all age-related pathological outcomes that we assessed.

We used telemetric ECOG to directly record abnormal neural network activity during aging. We found slowing of EEG/ECOG activity, consistent with other reports in the context of aging (4648). We further showed that this slowing of activity in mice is characterized as discrete, paroxysmal transient events (PSWEs), occurring frequently and spontaneously against a backdrop of “normal” ECOG activity. These PSWEs, which have characteristics that are similar to but distinct from epileptic seizures, may constitute “silent” or “subclinical” epileptiform activity, as has been reported in patients with dementia (7173). In the accompanying paper (49), we identified PSWEs in patients with AD and/or epilepsy. The data in both papers support the suggestion, proposed elsewhere (7476), that there may be common mechanistic links between age-related dementia and epilepsy, which has a remarkably high incidence in the elderly (7779). We emphasize that elevated TGFβ signaling, triggered by BBB dysfunction, provides a simple, parsimonious model for how this dysfunction may arise in aging. We found that TGFβ inhibition reversed aberrant ECOG activity, increased seizure threshold, and improved cognitive outcomes in aged mice, suggesting efficacy in the myriad symptoms that would be expected to arise from a dysfunctional neural network.

Could inhibition of TGFβ signaling, as a strategy to counteract the detrimental consequences of age-related BBB dysfunction, hold therapeutic potential? One of the major challenges (and causes of failure) in treating progressive neurological diseases is that patients decline over time and thus may accumulate irreversible damage by the time of diagnosis. Considering that BBB dysfunction begins relatively early in aging, this might call for a scenario in which chronic preventive treatment with TGFβ inhibitors is required to avoid future damage. However, we found that 1 week of acute treatment reversed the pathological outcomes in aged mice, including elevated TGFβ signaling, aberrant ECOG activity, seizure vulnerability, and cognitive dysfunction. Our findings suggest that the aging brain may retain considerable cognitive capacity, which may be chronically suppressed (but not irreversibly lost) by BBB leakiness and its inflammatory fallout.

By uncovering a foundational mechanism linking BBB decline to neural dysfunction, our study raises several critical questions: What causes BBB decline itself? Is BBB decline causal to, concurrent with, or an outcome of (or independent from) other well-known mechanisms of aging such as inflammation, reactive oxidation stress and metabolic failures, proteasome senescence, DNA damage, etc. (8085)? Our work focusing on microvascular integrity and interactions within the neurovascular unit can be placed in context of some of these mechanisms. For example, activation of astrocytes and gliosis has been shown to be a key step in many different aging diseases, with astrocytes playing potent roles in controlling neuroinflammation and neural functions including synaptic plasticity, senescence, and neurodegeneration (8690); we show that BBB dysfunction may be an early step causing or contributing to activation of astrocytes and the ensuing inflammatory response. Similarly, several previous studies have reported increased brain TGFβ signaling in aging (9196) and suggested that it could be a primary regulatory factor inducing aged neural phenotypes, although it was unknown what may trigger this increase in TGFβ signaling in aged individuals. We show that extravasation of albumin through the leaky BBB and astrocytic uptake may be one of the earliest steps that induce this age-related TGFβ cascade. Ultimately, our mechanistic findings provide a guiding framework for translation into the human clinical context, including large-scale epidemiological studies that are needed to establish the relationship between BBB status, other known biomarkers, and disease outcomes. In particular, this research offers new hope in two key unmet areas: early detection (via MRI of BBB status) and a new avenue for disease-modifying treatment that is mechanistically distinct from other canonical dementia targets, many of which have failed in clinical trials.


Study design

The aim of this study was to use experimental gain-of-function (iAlb) and loss-of-function (TGFβR KD and IPW) interventions to investigate the causal role of TGFβ signaling in age-related BBB pathology and the efficacy of therapeutic intervention. Outcomes were assessed using molecular (immunofluorescent staining and Western blot), electrophysiological (ECOG), and behavioral (PTZ and cognitive tasks) measures of symptomatic pathology in rodents. Translatability of these findings was further supported by complementary measures of BBB permeability and TGFβ signaling in human subjects (via DCE-MRI and postmortem histology). The experimental designs and methodology of outcome measures were chosen before initiating each experiment, except in the case of the ECOG experiments, in which post hoc analysis of ECOG signal led to the discovery of PSWEs as a possible indicator of aberrant network activity. For all experiments, subjects were randomly assigned to experimental groups, and data were collected under blinded experimental conditions with the exception of the immunofluorescent albumin counts in Fig. 1A, in which differences between age groups were visually apparent in the tissue even under naïve conditions such that complete blinding was not possible. Study design and sample sizes used for each experiment are provided in the figure legends.

Animal care and transgenic mice

All animal procedures were approved by the institutional animal care committees. Animals were housed with a 12-hour light/12-hour dark cycle with food and water available ad libitum. Aldh1L1-eGFP (enhanced green fluorescent protein) mice were bred from STOCK Tg(Aldh1l1-eGFP)OFC789Gsat/Mmucd (identification number 011015-UCD); obtained from the Mutant Mouse Resource and Research Center (MMRRC) at University of California at Davis, a National Institutes of Health (NIH)–funded strain repository; and donated to the MMRRC by N. Heintz [The Rockefeller University, Gene Expression Nervous System Atlas (GENSAT)]. These FVB/N mice were crossed to a C57BL/6 genetic background. The resulting strain exhibited constitutive astrocytic expression of eGFP protein under the astrocytic promoter Aldh1L1. Triple transgenic aTGFβR/KD mice were bred from strains purchased from the Jackson Laboratory to generate mice that express CreERT under the astrocytic promoter GLAST, with a floxed exon 4 of TGFβR2 (Tgfbr2fl) and a transgenic lacZ reporter gene inhibited by a floxed neomycin cassette. Tamoxifen induction thus induces activation of astrocytic CreERT, resulting in a null TGFβR2 allele (tgfbr2null) and lacZ expression (R26R−/−). The parental strain STOCK Tg(Slc1a3-cre/ERT)1Nat/J mice were outcrossed with B6;129-Tgfbr2tm1Karl/J and B6.129S4-Gt(ROSA)26Sortm1Sor/J mice to produce males, whereas B6;129-Tgfbr2tm1Karl/J and B6.129S4-Gt(ROSA)26Sortm1Sor/J mice were outcrossed to produce females. The resulting GLAST-CreERT; tgfbr2fl/+ males were bred with tgfbr2fl/+; R26R−/− and tgfbr2fl/fl; R26R−/− females to produce triple transgenic offspring. Subsequent generations were incrossed to produce experimental triple transgenic mice of genotypes GLAST-CreERT; tgfbr2fl/fl; R26R−/−, GLAST-CreERT; tgfbr2fl/+; R26R−/−, GLAST-CreERT; tgfbr2fl/fl; R26R−/+, and GLAST-CreERT; tgfbr2fl/+; R26R−/−. All mice were genotyped via polymerase chain reaction (PCR) analysis of tissue biopsy samples.

Human subjects

The human imaging protocol was approved by the Soroka University Medical Center Helsinki Institutional Review Board, and written informed consent was given by all participants. BBB status was assessed by DCE-MRI in n = 105 subjects with an age range of 21 to 83 years old.

For staining of human brains, postmortem hippocampus was obtained from young (n = 3; mean age, 31.3 ± 5 years) and old patients (n = 10; mean age, 70.6 ± 5.6 years). All participants gave informed and written consent, and all procedures were conducted in accordance with the Declaration of Helsinki and approved by the University of Bonn Ethics Committee.

Statistical analysis

All graphs are plotted showing the mean and standard error (SE). Two sample comparisons were performed by Student’s t test or Mann-Whitney test, and multiple group comparisons were conducted by analysis of variance (ANOVA) or Kruskal-Wallis test, followed by post hoc testing to compare individual groups when a main effect was detected. Multiple correction comparisons were used as described in the figure legends. Seizure progression in the PTZ experiments was analyzed by two-way ANOVA, and linear regression was used to calculate regression slopes. Differences in regression slope were also compared by ANOVA. In mouse ECOG experiments, different subgroups (hPSWE and lPSWE) were observed after data collection. These subgroups were formally classified using an unbiased Gaussian mixed model, and inferential statistics were performed on the subsequent groups. For all inferential statistics, two-tailed tests were used, and significance thresholds were set at P < 0.05. Summaries of each statistical test including sample size and P values are provided in the figure legends for each data plot and also in table S3. For rodent behavior experiments, animals were removed from the study cohort if they were unable to complete the requisite task according to the following predefined criteria: For MWM, mouse shows inability to swim (unable to maintain swim speed or buoyancy), and for T maze, mouse does not leave stem to complete arm choice in greater than three trials. For TGFβ1 Western blot analysis after IPW treatment (Fig. 6C), two samples were excluded because of gel loading error. Exclusion of these data did not alter statistical significance (results of ANOVA were significant with or without excluded values).


Fig. S1. Age-related BBB function, astrocyte reporter expression, and neurological function in mice.

Fig. S2. Age-related BBB function in neurologically intact human subjects.

Fig. S3. Characterization of the induced albumin intracerebroventricular infusion rodent model.

Fig. S4. Characterization of the astrocytic TGFβR KD mouse line.

Fig. S5. Pharmacokinetics of the small-molecule TGFβR inhibitor IPW.

Fig. S6. IPW reverses seizure vulnerability in aged mice.

Fig. S7. A scheme representation of TGFβ family of ligand and receptors.

Table S1. Age and gender of human subjects who received DCE-MRI scans.

Table S2. Gender and age of subjects providing postmortem brain tissue.

Table S3. Summary of statistical tests used in each experiment.

Table S4. PCR primers used to perform genotyping.

Table S5. Components of PCR reactions.

Table S6. PCR thermocycling conditions.

Table S7. Primers for quantitative PCR.

Data file S1. Tabular data points for experiments with a sample size of n < 20 (provided as a separate Excel file).

References (97102)


Acknowledgments: We thank K. Patten, A. Citri, H. Soreq, and I. Goshen for critical reading of the manuscript. We thank S. Lee for graphic design of the schematic diagram. Funding: This research was supported by NIH grants R01NS066005 and R56NS066005, a Bakar Foundation Fellowship, the Archer Foundation Award (D.K.), the Borstein Foundation award (D.K.), the European Union’s Seventh Framework Program (FP7/2007–2013, grant agreement 602102, EPITARGET), the Israel Science Foundation (717/15) (A.F.), NIH grant R01AG042679 (A.D.), the Binational Israel-USA Science Foundation, the Milstein Family Foundation (D.K. and A.F.), NSF GRFP fellowships (V.V.S. and A.R.F.), Siebel Fellowship (V.V.S.), and NIH NRSA fellowship F31AG054147 (A.R.F.). Author contributions: V.V.S. and A.R.F. devised and performed experiments and analysis. D.Z.M. and J.O. analyzed ECOG data. R.S.-A. and A.C. performed MRI and analyzed data. N.J., G.C., E.M., J.M.L., H.J.R., A.M., M.K.P., C.R.E., R.P., Y.S., and H.G. performed molecular biology experiments. H.V.H. performed rodent behavior experiments. H.S. and P.H.S. performed analysis of human transcription data. R.V. developed MRI methods. A.B. supervised postmortem tissue collection. B.H. supervised synthesis and validation of IPW and pharmacokinetic experiments. M.A.R. provided oversight and technical support for ECOG experiments. A.D. provided aged mice and supervised procedures for rearing and care of aged mice. A.F. and D.K. directed the project. V.V.S., A.R.F., A.F., and D.K. wrote the manuscript. Competing interests: D.K. and A.F. are inventors on patents US9468649B2 (Methods of treating epilepsy with transforming growth factor beta inhibitors) and US20160367530A1 (Methods of treating neurological disorders). D.K. is a member of the advisory board of Minerva Technologies Ltd. D.K., A.F., and B.H. are cofounders and shareholders of Mend Neuroscience. A.F. is the founder of Emagix Inc. B.H. is an equity stakeholder in Innovation Pathways and Allomek Therapeutics and a consultant for Capulus Therapeutics. A.R.F. and V.V.S. have stock options in a company working in an area related to the subject matter of this manuscript. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. Aldh1L1-eGFP mice were bred from STOCK Tg(Aldh1l1-eGFP)OFC789Gsat/Mmucd (identification number 011015-UCD); obtained from the Mutant Mouse Resource and Research Center (MMRRC) at University of California at Davis, an NIH-funded strain repository; and donated to the MMRRC by N. Heintz (The Rockefeller University, GENSAT). IPW was obtained under a material transfer agreement with Innovation Pathways Ltd.

Stay Connected to Science Translational Medicine

Navigate This Article