Liu WS, You J, Chen SD, Zhang Y, Feng JF, Xu YM, Yu JT, Cheng W.
Plasma proteomics identify biomarkers and undulating changes of brain aging.
Nat Aging. 2025 Jan;5(1):99-112. Epub 2024 Dec 9
PubMed.
I commend the authors for their multimodal imaging approach to brain age estimation. Making use of 1705 imaging-derived phenotypes as parameters for estimating brain age, this study used information from T1- and T2-weighted structural MRI, functional MRI, susceptibility weighted imaging (T2star), and diffusion MRI skeleton measurement. This multimodal approach should provide enhanced accuracy for prediction of brain age and BAG estimation, relative to previous approaches.
Brain age gap (BAG) is a useful metric because it predicts one’s age-adjusted vulnerability to a range of neurological disorders, such as AD, dementia, and stroke. Although brain age may indeed represent a latent and unifying phenotype that underlies multiple neurologic and psychiatric conditions, little is known about the factors that may directly promote accelerated brain aging independent of clinically defined CNS disease. In addition to identifying 13 proteins in blood that are associated with BAG, the authors used a Mendelian randomization approach to identify a causal role for Brevican (BCAN)—a CNS-specific neural proteoglycan involved in synaptic plasticity—as protective against accelerated brain age. This paper highlights the therapeutic potential of BCAN, suggesting a need for continued mechanistic research into what appears to be a protective CNS protein.
Other proteins linked to BAG were also causally implicated in related phenotypes, such as cortical surface area (KLK6) and diffusion-based white-matter integrity (GFAP). The implication is that these proteins may promote brain aging via their effect on gray- and white-matter structure, respectively.
This paper also took two unique approaches to BAG biomarker characterization. First, the authors identified biologically relevant (enriched for ECM, leukocyte adhesion, anoikis, etc.) clusters of proteins representing six distinct age-based trajectories from the 427 proteins nominally associated with BAG. While this is interesting, it is not clear how specific these features are to brain age or the set of BAG-associated proteins. Might we see the same patterns if we plotted these proteins against chronological age, or if analyses were not restricted to BAG-associated proteins?
With respect to the three waves of BAG-associated protein changes defined using the DE-SWAN differential expression-sliding window approach, the authors found the most prominent peaks at ages 57, 70, and 78. It’s worth noting that this is distinct from the set of peaks at ages 34, 60, and 78 found in a similar paper (Lehallier et al., 2019) which examined age effects on the full proteome rather than restricting analyses to proteins associated with brain aging. Thus, proteins linked to accelerated brain age may have a set of nonlinear dynamics distinct from that of the broader proteome (shifted backward). Notably, the authors found many of the proteins differentially expressed at the earliest stage (age 57) were associated with adaptive immune markers, potentially implicating adaptive immunity’s early influence on brain aging.
References:
Lehallier B, Gate D, Schaum N, Nanasi T, Lee SE, Yousef H, Moran Losada P, Berdnik D, Keller A, Verghese J, Sathyan S, Franceschi C, Milman S, Barzilai N, Wyss-Coray T.
Undulating changes in human plasma proteome profiles across the lifespan.
Nat Med. 2019 Dec;25(12):1843-1850. Epub 2019 Dec 5
PubMed.
This new study by Liu et al. provides a clear and compelling example of how plasma proteomics can be leveraged to identify novel targets for brain disorders. Using U.K. Biobank data from nearly 11,000 individuals, the authors developed a multimodal brain-aging model that took advantage of available brain imaging-derived phenotype (IDP) data and machine learning approaches to calculate “brain age,” which was then used to estimate a brain age gap (BAG) against the subjects’ chronological age. Comparing available plasma proteomic data from a subset of these individuals with the subjects’ brain age gap, the authors identified proteins potentially related to altered brain aging. Similar approaches have recently been published, including one that examined organ age gaps (including BAG), plasma proteomics, and their relationship to neurological disease and another that examined associations between BAG and AD (Oh et al., 2023; Lee et al., 2022; Millar et al., 2023).
Interestingly, Liu et al. find 13 proteins that are significantly associated with BAG, of which three, TIMP4, Brevican, and ADAM22, are key proteins associated with extracellular matrix (ECM) function. Six proteins persist upon repeat validation, of which two are ECM-related. Brevican, in particular, is among the most compelling proteins linked to brain aging in the study, with associations found for dementia (including AD) and other neurological disorders, as well as its causal association with BAG identified by Mendelian randomization. Of note, while the number of subjects, protein coverage, and the platform used differed from the current study, Oh et al. also identified an extracellular matrix signature linked to brain aging (Oh et al., 2023).
These results from Liu et al. are exciting, as several recent studies support the concept that components of the brain ECM change with age and may regulate brain aging or AD in humans and mouse models (Oh et al., 2023; Dammer et al., 2022; Johnson et al., 2022; Tewari et al., 2022). Our work supports this possibility; we find that one of the TIMPs, TIMP2, declines rapidly in the plasma with age and appears to be involved in regulating plasticity in the mouse hippocampus through interactions with the brain ECM, a network that itself changes with age (Castellano et al., 2017; Ferreira et al., 2023). Whether these changes are causally linked to how the brain ages and creates conditions for AD is unclear, but this is an intriguing direction for future work. The origins of plasma protein changes associated with BAG would also be a fruitful direction. While associations of protein changes with specific brain structure changes were characterized, along with changes in their brain expression pattern by cell type, it will be interesting to explore whether these changes solely reflect altered brain structure or whether the changes can arise from the altered aging of a specific peripheral organ that produces high levels of these proteins that ultimately act on the affected brain.
Overall, the authors convincingly demonstrate the utility of their approach in identifying novel plasma biomarkers associated with premature brain aging that can be leveraged to develop early interventions. The ease of plasma collection, including longitudinal monitoring, and the expanding proteomic coverage offered by emerging platforms highlight the potential of models like these to support biomarker discovery and monitoring targeted therapies for neurological disorders.
References:
Oh HS, Rutledge J, Nachun D, Pálovics R, Abiose O, Moran-Losada P, Channappa D, Urey DY, Kim K, Sung YJ, Wang L, Timsina J, Western D, Liu M, Kohlfeld P, Budde J, Wilson EN, Guen Y, Maurer TM, Haney M, Yang AC, He Z, Greicius MD, Andreasson KI, Sathyan S, Weiss EF, Milman S, Barzilai N, Cruchaga C, Wagner AD, Mormino E, Lehallier B, Henderson VW, Longo FM, Montgomery SB, Wyss-Coray T.
Organ aging signatures in the plasma proteome track health and disease.
Nature. 2023 Dec;624(7990):164-172. Epub 2023 Dec 6
PubMed.
Lee J, Burkett BJ, Min HK, Senjem ML, Lundt ES, Botha H, Graff-Radford J, Barnard LR, Gunter JL, Schwarz CG, Kantarci K, Knopman DS, Boeve BF, Lowe VJ, Petersen RC, Jack CR Jr, Jones DT.
Deep learning-based brain age prediction in normal aging and dementia.
Nat Aging. 2022 May;2(5):412-424. Epub 2022 May 9
PubMed.
Millar PR, Gordon BA, Wisch JK, Schultz SA, Benzinger TL, Cruchaga C, Hassenstab JJ, Ibanez L, Karch C, Llibre-Guerra JJ, Morris JC, Perrin RJ, Supnet-Bell C, Xiong C, Allegri RF, Berman SB, Chhatwal JP, Chrem Mendez PA, Day GS, Hofmann A, Ikeuchi T, Jucker M, Lee JH, Levin J, Lopera F, Niimi Y, Sánchez-González VJ, Schofield PR, Sosa-Ortiz AL, Vöglein J, Dominantly Inherited Alzheimer Network, Bateman RJ, Ances BM, McDade EM.
Advanced structural brain aging in preclinical autosomal dominant Alzheimer disease.
Mol Neurodegener. 2023 Dec 19;18(1):98.
PubMed.
Dammer EB, Ping L, Duong DM, Modeste ES, Seyfried NT, Lah JJ, Levey AI, Johnson EC.
Multi-platform proteomic analysis of Alzheimer's disease cerebrospinal fluid and plasma reveals network biomarkers associated with proteostasis and the matrisome.
Alzheimers Res Ther. 2022 Nov 17;14(1):174.
PubMed.
Johnson EC, Carter EK, Dammer EB, Duong DM, Gerasimov ES, Liu Y, Liu J, Betarbet R, Ping L, Yin L, Serrano GE, Beach TG, Peng J, De Jager PL, Haroutunian V, Zhang B, Gaiteri C, Bennett DA, Gearing M, Wingo TS, Wingo AP, Lah JJ, Levey AI, Seyfried NT.
Large-scale deep multi-layer analysis of Alzheimer's disease brain reveals strong proteomic disease-related changes not observed at the RNA level.
Nat Neurosci. 2022 Feb;25(2):213-225. Epub 2022 Feb 3
PubMed.
Tewari BP, Woo AM, Prim CE, Chaunsali L, Patel DC, Kimbrough IF, Engel K, Browning JL, Campbell SL, Sontheimer H.
Astrocytes require perineuronal nets to maintain synaptic homeostasis in mice.
Nat Neurosci. 2024 Aug;27(8):1475-1488. Epub 2024 Jul 17
PubMed.
Castellano JM, Mosher KI, Abbey RJ, McBride AA, James ML, Berdnik D, Shen JC, Zou B, Xie XS, Tingle M, Hinkson IV, Angst MS, Wyss-Coray T.
Human umbilical cord plasma proteins revitalize hippocampal function in aged mice.
Nature. 2017 Apr 19;
PubMed.
Ferreira AC, Hemmer BM, Philippi SM, Grau-Perales AB, Rosenstadt JL, Liu H, Zhu JD, Kareva T, Ahfeldt T, Varghese M, Hof PR, Castellano JM.
Neuronal TIMP2 regulates hippocampus-dependent plasticity and extracellular matrix complexity.
Mol Psychiatry. 2023 Sep;28(9):3943-3954. Epub 2023 Nov 2
PubMed.
This interesting study provides valuable insights into plasma proteomic changes associated with the brain age gap (BAG), highlighting brevican and GDF-15 as markers of brain aging physiology manifesting in multimodal structural brain imaging. As described in our recent Lancet paper, brain aging unfolds through phases of growth, homeostasis, and allostasis, with BAG capturing the interplay of genetic, environmental, pathological aging-related physiology, and disease processes within these processes (Jones and Topol, 2022).
While it is often emphasized that amyloid-related physiology must precede clinical symptoms, it is equally true that some age-related physiology must precede amyloid. The proteomic waves identified in this study may reflect metabolic shifts linked to aging-related network changes, consistent with the cascading network failure model (Jones et al., 2017). This model posits that network-level stress and compensation contribute to systemic changes, ultimately leading to protein misfolding pathology.
If replicated, these findings suggest that these proteomic measures could serve as systemic markers of network-driven metabolic stress. Their integration with imaging biomarkers, such as FDG-PET, could refine models of brain aging by capturing the early physiology that bridges aging, health, and disease (Lee et al., 2022).
References:
Jones DT, Lee J, Topol EJ.
Digitising brain age.
Lancet. 2022 Sep 24;400(10357):988.
PubMed.
Lee J, Burkett BJ, Min HK, Senjem ML, Lundt ES, Botha H, Graff-Radford J, Barnard LR, Gunter JL, Schwarz CG, Kantarci K, Knopman DS, Boeve BF, Lowe VJ, Petersen RC, Jack CR Jr, Jones DT.
Deep learning-based brain age prediction in normal aging and dementia.
Nat Aging. 2022 May;2(5):412-424. Epub 2022 May 9
PubMed.
Jones DT, Graff-Radford J, Lowe VJ, Wiste HJ, Gunter JL, Senjem ML, Botha H, Kantarci K, Boeve BF, Knopman DS, Petersen RC, Jack CR Jr.
Tau, amyloid, and cascading network failure across the Alzheimer's disease spectrum.
Cortex. 2017 Dec;97:143-159. Epub 2017 Oct 3
PubMed.
Aging is associated with many ailments such as cardiovascular, cancer, and neurodegenerative diseases. In this article, the authors focus on the aging brain, specifically, and how the difference between biologic aging and chronologic aging of the brain, defined as brain age gap, or BAG, can affect the proteome of the brain and plasma. Proteome-wide association analysis across 4,696 participants of 2,922 proteins identified 13 significantly associated with BAG, implicating stress, regeneration and inflammation. Six of them were validated: BCAN, GDF15, TIMP4, KLK6, ADGRG1, and LGALS4.
The authors correlate changes in these proteins with brain structures identified by multimodal brain imaging. They detected undulating changes in the plasma proteome across brain aging, and profiled brain age-related change peaks at 57, 70, and 78 years, implicating distinct biological pathways during brain aging.
Brevican (BCAN) and growth differentiation factor 15 (GDF15) showed the most significant and multiple associations wih dementia, stroke, and movement disorders. Brevican is one of the most abundant chondroitin sulfate proteoglycans in the adult brain. It is highly specific to the brain and increases as the brain develops (Yamaguchi, 1996).
GDF15, a stress-regulated hormone, is upregulated in response to environmental stress, and is involved in energy homeostasis, insulin resistance, and mitochondrial dysfunction. GDF15 is ubiquitously expressed.
Courtesy of Lehallier et al., Nat Med, 2019.]
I find of particular interest and relevance to the current paper the findings of Lehallier et al., who studied the plasma proteomic changes that occur across the lifespan (Lehallier et al., 2019). They measured 2,925 plasma proteins from 4,263 people ranging from young adults to nonagenarians (18–95 years old) and developed a novel bioinformatics approach that uncovered marked nonlinear alterations in the human plasma proteome with age. Waves of changes in the proteome in the fourth, seventh, and eighth decades of life reflected distinct biological pathways. They describe a 46-protein aging signature that is conserved in humans and mice, containing known aging proteins such as, interestingly, GDF15.
So, why the difference in the timing of the three peaks between the two studies? Cheng et al. correlated the plasma proteomic changes to BAG, while no such attempt was made by Lehallier et al. Can we speculate that changes in the brain start to occur later that those in peripheral organs? That neural cells that are post mitotic are more resilient because they must last longer? Both Brevican and GDF15 fulfill crucial and protective roles in the brain and may become targets for the development of neuroprotective drugs for neurodegenerative diseases.
References:
Yamaguchi Y.
Brevican: a major proteoglycan in adult brain.
Perspect Dev Neurobiol. 1996;3(4):307-17.
PubMed.
Lehallier B, Gate D, Schaum N, Nanasi T, Lee SE, Yousef H, Moran Losada P, Berdnik D, Keller A, Verghese J, Sathyan S, Franceschi C, Milman S, Barzilai N, Wyss-Coray T.
Undulating changes in human plasma proteome profiles across the lifespan.
Nat Med. 2019 Dec;25(12):1843-1850. Epub 2019 Dec 5
PubMed.
This paper by Liu and colleagues presents a novel and interesting study of protein associations with the brain age gap (BAG). The BAG approach aims to characterize biological aging of the brain based on neuroimaging features, and it has recently become quite popular with the ease of access to machine learning techniques and the availability of large neuroimaging datasets. In this paper, the authors performed protein-wide association (PWAS) and Mendelian randomization analyses of BAG using multimodal neuroimaging (n > 10,000) and plasma proteomic (n > 4,000) data from healthy adults in the U.K. Biobank. By identifying proteins associated with BAG, including brevican (BCAN), which appears to have a causal protective relationship, this study begins to characterize potential biological pathways and mechanisms that may give rise to brain aging as observed in MRI. These results build upon our understanding of brain aging, but also raise several interesting questions, which may hopefully be addressed in future research.
The authors take a multimodal approach to model BAG, including not only T1-weighted structural MRI—by far the most popular single modality for brain age models—but also T2-weighted, susceptibility-weighted, diffusion, and both resting-state and task functional MRI. While this broad approach allows for a comprehensive characterization of BAG, it also raises some interesting questions regarding heterogeneity among imaging modalities and aging trajectories. Studies from our own group and others suggest that different imaging modalities capture complementary signals of biological aging, reflecting distinct patterns of brain aging, rather than a single uniform trajectory (e.g., Millar et al., 2023; Skampardoni et al., 2024; Smith et al., 2020; Yang et al., 2024). Are the BAG-related proteins and pathways identified in this study consistent across individual imaging modalities and unique aging patterns or, rather, are distinct biological pathways at work?
Additionally, it is important to keep in mind that these associations were identified in analyses of healthy individuals from the U.K. Biobank—excluding participants with neuropsychiatric or other disorders. Notably, BAG has been found to be elevated across a wide range of these conditions, including Alzheimer’s disease, schizophrenia, HIV, and many others, despite massive heterogeneity in the severity, localization, and quality of changes in the brain (for review, see Cole and Franke, 2017). An important question then is whether elevated BAG in these disparate clinical populations is associated with common biological pathways, similar to those identified here, or if distinct mechanisms may play additional roles.
Finally, a parallel line of work has also recently gained momentum in modeling biological age from proteomic measures, identifying proteins that are predictive of age (e.g., Lehallier et al., 2019; Melendez et al., 2024; Oh et al., 2023). Given differences in the proteomic platforms and modeling approaches across studies, it is not clear whether the BAG-associated proteins identified here are consistent with those that have been identified in other proteomic age clocks. Careful, direct comparison of BAG-associated and general age-related proteins might be useful in identifying biological pathways that are relevant for brain aging and distinguishing them from other age-related changes.
References:
Cole JH, Franke K.
Predicting Age Using Neuroimaging: Innovative Brain Ageing Biomarkers.
Trends Neurosci. 2017 Dec;40(12):681-690. Epub 2017 Oct 23
PubMed.
Lehallier B, Gate D, Schaum N, Nanasi T, Lee SE, Yousef H, Moran Losada P, Berdnik D, Keller A, Verghese J, Sathyan S, Franceschi C, Milman S, Barzilai N, Wyss-Coray T.
Undulating changes in human plasma proteome profiles across the lifespan.
Nat Med. 2019 Dec;25(12):1843-1850. Epub 2019 Dec 5
PubMed.
Melendez J, Sung YJ, Orr M, Yoo A, Schindler S, Cruchaga C, Bateman R.
An interpretable machine learning-based cerebrospinal fluid proteomics clock for predicting age reveals novel insights into brain aging.
Aging Cell. 2024 Sep;23(9):e14230. Epub 2024 Jun 24
PubMed.
Millar PR, Gordon BA, Luckett PH, Benzinger TL, Cruchaga C, Fagan AM, Hassenstab JJ, Perrin RJ, Schindler SE, Allegri RF, Day GS, Farlow MR, Mori H, Nübling G, Dominantly Inherited Alzheimer Network, Bateman RJ, Morris JC, Ances BM.
Multimodal brain age estimates relate to Alzheimer disease biomarkers and cognition in early stages: a cross-sectional observational study.
Elife. 2023 Jan 6;12
PubMed.
Oh HS, Rutledge J, Nachun D, Pálovics R, Abiose O, Moran-Losada P, Channappa D, Urey DY, Kim K, Sung YJ, Wang L, Timsina J, Western D, Liu M, Kohlfeld P, Budde J, Wilson EN, Guen Y, Maurer TM, Haney M, Yang AC, He Z, Greicius MD, Andreasson KI, Sathyan S, Weiss EF, Milman S, Barzilai N, Cruchaga C, Wagner AD, Mormino E, Lehallier B, Henderson VW, Longo FM, Montgomery SB, Wyss-Coray T.
Organ aging signatures in the plasma proteome track health and disease.
Nature. 2023 Dec;624(7990):164-172. Epub 2023 Dec 6
PubMed.
Skampardoni I, Nasrallah IM, Abdulkadir A, Wen J, Melhem R, Mamourian E, Erus G, Doshi J, Singh A, Yang Z, Cui Y, Hwang G, Ren Z, Pomponio R, Srinivasan D, Govindarajan ST, Parmpi P, Wittfeld K, Grabe HJ, Bülow R, Frenzel S, Tosun D, Bilgel M, An Y, Marcus DS, LaMontagne P, Heckbert SR, Austin TR, Launer LJ, Sotiras A, Espeland MA, Masters CL, Maruff P, Fripp J, Johnson SC, Morris JC, Albert MS, Bryan RN, Yaffe K, Völzke H, Ferrucci L, Benzinger TL, Ezzati A, Shinohara RT, Fan Y, Resnick SM, Habes M, Wolk D, Shou H, Nikita K, Davatzikos C.
Genetic and Clinical Correlates of AI-Based Brain Aging Patterns in Cognitively Unimpaired Individuals.
JAMA Psychiatry. 2024 May 1;81(5):456-467.
PubMed.
Smith SM, Elliott LT, Alfaro-Almagro F, McCarthy P, Nichols TE, Douaud G, Miller KL.
Brain aging comprises many modes of structural and functional change with distinct genetic and biophysical associations.
Elife. 2020 Mar 5;9
PubMed.
Yang Z, Wen J, Erus G, Govindarajan ST, Melhem R, Mamourian E, Cui Y, Srinivasan D, Abdulkadir A, Parmpi P, Wittfeld K, Grabe HJ, Bülow R, Frenzel S, Tosun D, Bilgel M, An Y, Yi D, Marcus DS, LaMontagne P, Benzinger TL, Heckbert SR, Austin TR, Waldstein SR, Evans MK, Zonderman AB, Launer LJ, Sotiras A, Espeland MA, Masters CL, Maruff P, Fripp J, Toga AW, O'Bryant S, Chakravarty MM, Villeneuve S, Johnson SC, Morris JC, Albert MS, Yaffe K, Völzke H, Ferrucci L, Nick Bryan R, Shinohara RT, Fan Y, Habes M, Lalousis PA, Koutsouleris N, Wolk DA, Resnick SM, Shou H, Nasrallah IM, Davatzikos C.
Brain aging patterns in a large and diverse cohort of 49,482 individuals.
Nat Med. 2024 Oct;30(10):3015-3026. Epub 2024 Aug 15
PubMed.
With great respect to the authors and others in the field, I urge the field to reject the nomenclature of "brain aging," or "biological aging," as distinct from "age." We are all, including the authors of this paper, trying hard to understand why people are likely to suffer cognitive impairment as they get older, and why some people experience cognitive decline at a young age. We also know that many people grow old without cognitive impairment. It is inaccurate language to describe someone who is young but has brain atrophy, or other abnormal brain imaging according to an MRI-based model, as "biologically old." The risk of prostate disorders skyrockets with age, too. We would not characterize a multimodal risk predictor for prostate cancer, or perhaps even nonspecific dysuria, as describing someone's "prostate age." Instead, I would propose we use the most accurate terms we can.
The work described in this paper is important. Perhaps a title could be "Plasma Proteomics Predicts a Multimodal Model of Abnormal Brain MRI."
By using the word "age" to mean "brain dysfunction," we risk summoning societal biases against old people. We should use the word "age" to mean just that—the amount of time elapsed since birth.
This is a monumental exercise in bioinformatics. I should probably let it rest there since it’s not my area of expertise. But, in this as in many things, I’m a bit of a contrarian so let me give my overall impression of the work from the standpoint of a lay reader.
As a cell biologist, I am unsure how much we truly learn from the work. I am worried about the exclusion criteria. They are not well articulated. Excluding people with underlying conditions is an understandable decision in a study of healthy aging, but it carries a risk of biasing the sample. I also worry about the sample being >97 percent white-Caucasian. The choice to use the U.K. Biobank is understandable, but also diminishes the broader application of the findings to the human population. Finally, I worry about using serum markers to gain insight into brain function. The authors are responsibly cautious in this regard, but offer no experiments to validate the strength of the connection between blood and brain with regard to their new findings.
My other issue is that the paper mostly contains a series of interesting correlations. Causality cannot be inferred from the data, and the supporting literature cited, while compelling, is far from proof of mechanism. Take brevican, for example. The authors' ideas are well argued but, based on their data, my own hypothesis would be that brevican is a symptom of aging (and cognitive decline), not a cause. That would make it possibly valuable as a diagnostic marker, but I would need to see much more to be convinced that it has therapeutic potential. Even as a diagnostic tool, the violin plots in the supplement suggest that while the correlation is true at a population level, there may be less utility at the level of an individual, which, after all, is where medical practice is applied.
One last comment. Aging is a phenomenon that is built into every eukaryote on the planet. It is not well understood at a biological level despite its importance. Because of this wide applicability, I would have asked the authors to either test their ideas in other species or change the title of the paper by inserting the word “human” in front of the word “brain” in their title. And for a journal entitled “Nature Aging,” I am surprised that the editors didn’t flag this themselves.
Comments
National Institute on Aging
I commend the authors for their multimodal imaging approach to brain age estimation. Making use of 1705 imaging-derived phenotypes as parameters for estimating brain age, this study used information from T1- and T2-weighted structural MRI, functional MRI, susceptibility weighted imaging (T2star), and diffusion MRI skeleton measurement. This multimodal approach should provide enhanced accuracy for prediction of brain age and BAG estimation, relative to previous approaches.
Brain age gap (BAG) is a useful metric because it predicts one’s age-adjusted vulnerability to a range of neurological disorders, such as AD, dementia, and stroke. Although brain age may indeed represent a latent and unifying phenotype that underlies multiple neurologic and psychiatric conditions, little is known about the factors that may directly promote accelerated brain aging independent of clinically defined CNS disease. In addition to identifying 13 proteins in blood that are associated with BAG, the authors used a Mendelian randomization approach to identify a causal role for Brevican (BCAN)—a CNS-specific neural proteoglycan involved in synaptic plasticity—as protective against accelerated brain age. This paper highlights the therapeutic potential of BCAN, suggesting a need for continued mechanistic research into what appears to be a protective CNS protein.
Other proteins linked to BAG were also causally implicated in related phenotypes, such as cortical surface area (KLK6) and diffusion-based white-matter integrity (GFAP). The implication is that these proteins may promote brain aging via their effect on gray- and white-matter structure, respectively.
This paper also took two unique approaches to BAG biomarker characterization. First, the authors identified biologically relevant (enriched for ECM, leukocyte adhesion, anoikis, etc.) clusters of proteins representing six distinct age-based trajectories from the 427 proteins nominally associated with BAG. While this is interesting, it is not clear how specific these features are to brain age or the set of BAG-associated proteins. Might we see the same patterns if we plotted these proteins against chronological age, or if analyses were not restricted to BAG-associated proteins?
With respect to the three waves of BAG-associated protein changes defined using the DE-SWAN differential expression-sliding window approach, the authors found the most prominent peaks at ages 57, 70, and 78. It’s worth noting that this is distinct from the set of peaks at ages 34, 60, and 78 found in a similar paper (Lehallier et al., 2019) which examined age effects on the full proteome rather than restricting analyses to proteins associated with brain aging. Thus, proteins linked to accelerated brain age may have a set of nonlinear dynamics distinct from that of the broader proteome (shifted backward). Notably, the authors found many of the proteins differentially expressed at the earliest stage (age 57) were associated with adaptive immune markers, potentially implicating adaptive immunity’s early influence on brain aging.
References:
Lehallier B, Gate D, Schaum N, Nanasi T, Lee SE, Yousef H, Moran Losada P, Berdnik D, Keller A, Verghese J, Sathyan S, Franceschi C, Milman S, Barzilai N, Wyss-Coray T. Undulating changes in human plasma proteome profiles across the lifespan. Nat Med. 2019 Dec;25(12):1843-1850. Epub 2019 Dec 5 PubMed.
View all comments by Keenan WalkerIcahn School of Medicine at Mount Sinai
This new study by Liu et al. provides a clear and compelling example of how plasma proteomics can be leveraged to identify novel targets for brain disorders. Using U.K. Biobank data from nearly 11,000 individuals, the authors developed a multimodal brain-aging model that took advantage of available brain imaging-derived phenotype (IDP) data and machine learning approaches to calculate “brain age,” which was then used to estimate a brain age gap (BAG) against the subjects’ chronological age. Comparing available plasma proteomic data from a subset of these individuals with the subjects’ brain age gap, the authors identified proteins potentially related to altered brain aging. Similar approaches have recently been published, including one that examined organ age gaps (including BAG), plasma proteomics, and their relationship to neurological disease and another that examined associations between BAG and AD (Oh et al., 2023; Lee et al., 2022; Millar et al., 2023).
Interestingly, Liu et al. find 13 proteins that are significantly associated with BAG, of which three, TIMP4, Brevican, and ADAM22, are key proteins associated with extracellular matrix (ECM) function. Six proteins persist upon repeat validation, of which two are ECM-related. Brevican, in particular, is among the most compelling proteins linked to brain aging in the study, with associations found for dementia (including AD) and other neurological disorders, as well as its causal association with BAG identified by Mendelian randomization. Of note, while the number of subjects, protein coverage, and the platform used differed from the current study, Oh et al. also identified an extracellular matrix signature linked to brain aging (Oh et al., 2023).
These results from Liu et al. are exciting, as several recent studies support the concept that components of the brain ECM change with age and may regulate brain aging or AD in humans and mouse models (Oh et al., 2023; Dammer et al., 2022; Johnson et al., 2022; Tewari et al., 2022). Our work supports this possibility; we find that one of the TIMPs, TIMP2, declines rapidly in the plasma with age and appears to be involved in regulating plasticity in the mouse hippocampus through interactions with the brain ECM, a network that itself changes with age (Castellano et al., 2017; Ferreira et al., 2023). Whether these changes are causally linked to how the brain ages and creates conditions for AD is unclear, but this is an intriguing direction for future work. The origins of plasma protein changes associated with BAG would also be a fruitful direction. While associations of protein changes with specific brain structure changes were characterized, along with changes in their brain expression pattern by cell type, it will be interesting to explore whether these changes solely reflect altered brain structure or whether the changes can arise from the altered aging of a specific peripheral organ that produces high levels of these proteins that ultimately act on the affected brain.
Overall, the authors convincingly demonstrate the utility of their approach in identifying novel plasma biomarkers associated with premature brain aging that can be leveraged to develop early interventions. The ease of plasma collection, including longitudinal monitoring, and the expanding proteomic coverage offered by emerging platforms highlight the potential of models like these to support biomarker discovery and monitoring targeted therapies for neurological disorders.
References:
Oh HS, Rutledge J, Nachun D, Pálovics R, Abiose O, Moran-Losada P, Channappa D, Urey DY, Kim K, Sung YJ, Wang L, Timsina J, Western D, Liu M, Kohlfeld P, Budde J, Wilson EN, Guen Y, Maurer TM, Haney M, Yang AC, He Z, Greicius MD, Andreasson KI, Sathyan S, Weiss EF, Milman S, Barzilai N, Cruchaga C, Wagner AD, Mormino E, Lehallier B, Henderson VW, Longo FM, Montgomery SB, Wyss-Coray T. Organ aging signatures in the plasma proteome track health and disease. Nature. 2023 Dec;624(7990):164-172. Epub 2023 Dec 6 PubMed.
Lee J, Burkett BJ, Min HK, Senjem ML, Lundt ES, Botha H, Graff-Radford J, Barnard LR, Gunter JL, Schwarz CG, Kantarci K, Knopman DS, Boeve BF, Lowe VJ, Petersen RC, Jack CR Jr, Jones DT. Deep learning-based brain age prediction in normal aging and dementia. Nat Aging. 2022 May;2(5):412-424. Epub 2022 May 9 PubMed.
Millar PR, Gordon BA, Wisch JK, Schultz SA, Benzinger TL, Cruchaga C, Hassenstab JJ, Ibanez L, Karch C, Llibre-Guerra JJ, Morris JC, Perrin RJ, Supnet-Bell C, Xiong C, Allegri RF, Berman SB, Chhatwal JP, Chrem Mendez PA, Day GS, Hofmann A, Ikeuchi T, Jucker M, Lee JH, Levin J, Lopera F, Niimi Y, Sánchez-González VJ, Schofield PR, Sosa-Ortiz AL, Vöglein J, Dominantly Inherited Alzheimer Network, Bateman RJ, Ances BM, McDade EM. Advanced structural brain aging in preclinical autosomal dominant Alzheimer disease. Mol Neurodegener. 2023 Dec 19;18(1):98. PubMed.
Dammer EB, Ping L, Duong DM, Modeste ES, Seyfried NT, Lah JJ, Levey AI, Johnson EC. Multi-platform proteomic analysis of Alzheimer's disease cerebrospinal fluid and plasma reveals network biomarkers associated with proteostasis and the matrisome. Alzheimers Res Ther. 2022 Nov 17;14(1):174. PubMed.
Johnson EC, Carter EK, Dammer EB, Duong DM, Gerasimov ES, Liu Y, Liu J, Betarbet R, Ping L, Yin L, Serrano GE, Beach TG, Peng J, De Jager PL, Haroutunian V, Zhang B, Gaiteri C, Bennett DA, Gearing M, Wingo TS, Wingo AP, Lah JJ, Levey AI, Seyfried NT. Large-scale deep multi-layer analysis of Alzheimer's disease brain reveals strong proteomic disease-related changes not observed at the RNA level. Nat Neurosci. 2022 Feb;25(2):213-225. Epub 2022 Feb 3 PubMed.
Tewari BP, Woo AM, Prim CE, Chaunsali L, Patel DC, Kimbrough IF, Engel K, Browning JL, Campbell SL, Sontheimer H. Astrocytes require perineuronal nets to maintain synaptic homeostasis in mice. Nat Neurosci. 2024 Aug;27(8):1475-1488. Epub 2024 Jul 17 PubMed.
Castellano JM, Mosher KI, Abbey RJ, McBride AA, James ML, Berdnik D, Shen JC, Zou B, Xie XS, Tingle M, Hinkson IV, Angst MS, Wyss-Coray T. Human umbilical cord plasma proteins revitalize hippocampal function in aged mice. Nature. 2017 Apr 19; PubMed.
Ferreira AC, Hemmer BM, Philippi SM, Grau-Perales AB, Rosenstadt JL, Liu H, Zhu JD, Kareva T, Ahfeldt T, Varghese M, Hof PR, Castellano JM. Neuronal TIMP2 regulates hippocampus-dependent plasticity and extracellular matrix complexity. Mol Psychiatry. 2023 Sep;28(9):3943-3954. Epub 2023 Nov 2 PubMed.
View all comments by Joseph CastellanoMayo Clinic
This interesting study provides valuable insights into plasma proteomic changes associated with the brain age gap (BAG), highlighting brevican and GDF-15 as markers of brain aging physiology manifesting in multimodal structural brain imaging. As described in our recent Lancet paper, brain aging unfolds through phases of growth, homeostasis, and allostasis, with BAG capturing the interplay of genetic, environmental, pathological aging-related physiology, and disease processes within these processes (Jones and Topol, 2022).
While it is often emphasized that amyloid-related physiology must precede clinical symptoms, it is equally true that some age-related physiology must precede amyloid. The proteomic waves identified in this study may reflect metabolic shifts linked to aging-related network changes, consistent with the cascading network failure model (Jones et al., 2017). This model posits that network-level stress and compensation contribute to systemic changes, ultimately leading to protein misfolding pathology.
If replicated, these findings suggest that these proteomic measures could serve as systemic markers of network-driven metabolic stress. Their integration with imaging biomarkers, such as FDG-PET, could refine models of brain aging by capturing the early physiology that bridges aging, health, and disease (Lee et al., 2022).
References:
Jones DT, Lee J, Topol EJ. Digitising brain age. Lancet. 2022 Sep 24;400(10357):988. PubMed.
Lee J, Burkett BJ, Min HK, Senjem ML, Lundt ES, Botha H, Graff-Radford J, Barnard LR, Gunter JL, Schwarz CG, Kantarci K, Knopman DS, Boeve BF, Lowe VJ, Petersen RC, Jack CR Jr, Jones DT. Deep learning-based brain age prediction in normal aging and dementia. Nat Aging. 2022 May;2(5):412-424. Epub 2022 May 9 PubMed.
Jones DT, Graff-Radford J, Lowe VJ, Wiste HJ, Gunter JL, Senjem ML, Botha H, Kantarci K, Boeve BF, Knopman DS, Petersen RC, Jack CR Jr. Tau, amyloid, and cascading network failure across the Alzheimer's disease spectrum. Cortex. 2017 Dec;97:143-159. Epub 2017 Oct 3 PubMed.
View all comments by David JonesADvantage
Aging is associated with many ailments such as cardiovascular, cancer, and neurodegenerative diseases. In this article, the authors focus on the aging brain, specifically, and how the difference between biologic aging and chronologic aging of the brain, defined as brain age gap, or BAG, can affect the proteome of the brain and plasma. Proteome-wide association analysis across 4,696 participants of 2,922 proteins identified 13 significantly associated with BAG, implicating stress, regeneration and inflammation. Six of them were validated: BCAN, GDF15, TIMP4, KLK6, ADGRG1, and LGALS4.
The authors correlate changes in these proteins with brain structures identified by multimodal brain imaging. They detected undulating changes in the plasma proteome across brain aging, and profiled brain age-related change peaks at 57, 70, and 78 years, implicating distinct biological pathways during brain aging.
Brevican (BCAN) and growth differentiation factor 15 (GDF15) showed the most significant and multiple associations wih dementia, stroke, and movement disorders. Brevican is one of the most abundant chondroitin sulfate proteoglycans in the adult brain. It is highly specific to the brain and increases as the brain develops (Yamaguchi, 1996).
GDF15, a stress-regulated hormone, is upregulated in response to environmental stress, and is involved in energy homeostasis, insulin resistance, and mitochondrial dysfunction. GDF15 is ubiquitously expressed.
Courtesy of Lehallier et al., Nat Med, 2019.]
I find of particular interest and relevance to the current paper the findings of Lehallier et al., who studied the plasma proteomic changes that occur across the lifespan (Lehallier et al., 2019). They measured 2,925 plasma proteins from 4,263 people ranging from young adults to nonagenarians (18–95 years old) and developed a novel bioinformatics approach that uncovered marked nonlinear alterations in the human plasma proteome with age. Waves of changes in the proteome in the fourth, seventh, and eighth decades of life reflected distinct biological pathways. They describe a 46-protein aging signature that is conserved in humans and mice, containing known aging proteins such as, interestingly, GDF15.
So, why the difference in the timing of the three peaks between the two studies? Cheng et al. correlated the plasma proteomic changes to BAG, while no such attempt was made by Lehallier et al. Can we speculate that changes in the brain start to occur later that those in peripheral organs? That neural cells that are post mitotic are more resilient because they must last longer? Both Brevican and GDF15 fulfill crucial and protective roles in the brain and may become targets for the development of neuroprotective drugs for neurodegenerative diseases.
References:
Yamaguchi Y. Brevican: a major proteoglycan in adult brain. Perspect Dev Neurobiol. 1996;3(4):307-17. PubMed.
Lehallier B, Gate D, Schaum N, Nanasi T, Lee SE, Yousef H, Moran Losada P, Berdnik D, Keller A, Verghese J, Sathyan S, Franceschi C, Milman S, Barzilai N, Wyss-Coray T. Undulating changes in human plasma proteome profiles across the lifespan. Nat Med. 2019 Dec;25(12):1843-1850. Epub 2019 Dec 5 PubMed.
View all comments by Carmela AbrahamWashington University in St. Louis
This paper by Liu and colleagues presents a novel and interesting study of protein associations with the brain age gap (BAG). The BAG approach aims to characterize biological aging of the brain based on neuroimaging features, and it has recently become quite popular with the ease of access to machine learning techniques and the availability of large neuroimaging datasets. In this paper, the authors performed protein-wide association (PWAS) and Mendelian randomization analyses of BAG using multimodal neuroimaging (n > 10,000) and plasma proteomic (n > 4,000) data from healthy adults in the U.K. Biobank. By identifying proteins associated with BAG, including brevican (BCAN), which appears to have a causal protective relationship, this study begins to characterize potential biological pathways and mechanisms that may give rise to brain aging as observed in MRI. These results build upon our understanding of brain aging, but also raise several interesting questions, which may hopefully be addressed in future research.
The authors take a multimodal approach to model BAG, including not only T1-weighted structural MRI—by far the most popular single modality for brain age models—but also T2-weighted, susceptibility-weighted, diffusion, and both resting-state and task functional MRI. While this broad approach allows for a comprehensive characterization of BAG, it also raises some interesting questions regarding heterogeneity among imaging modalities and aging trajectories. Studies from our own group and others suggest that different imaging modalities capture complementary signals of biological aging, reflecting distinct patterns of brain aging, rather than a single uniform trajectory (e.g., Millar et al., 2023; Skampardoni et al., 2024; Smith et al., 2020; Yang et al., 2024). Are the BAG-related proteins and pathways identified in this study consistent across individual imaging modalities and unique aging patterns or, rather, are distinct biological pathways at work?
Additionally, it is important to keep in mind that these associations were identified in analyses of healthy individuals from the U.K. Biobank—excluding participants with neuropsychiatric or other disorders. Notably, BAG has been found to be elevated across a wide range of these conditions, including Alzheimer’s disease, schizophrenia, HIV, and many others, despite massive heterogeneity in the severity, localization, and quality of changes in the brain (for review, see Cole and Franke, 2017). An important question then is whether elevated BAG in these disparate clinical populations is associated with common biological pathways, similar to those identified here, or if distinct mechanisms may play additional roles.
Finally, a parallel line of work has also recently gained momentum in modeling biological age from proteomic measures, identifying proteins that are predictive of age (e.g., Lehallier et al., 2019; Melendez et al., 2024; Oh et al., 2023). Given differences in the proteomic platforms and modeling approaches across studies, it is not clear whether the BAG-associated proteins identified here are consistent with those that have been identified in other proteomic age clocks. Careful, direct comparison of BAG-associated and general age-related proteins might be useful in identifying biological pathways that are relevant for brain aging and distinguishing them from other age-related changes.
References:
Cole JH, Franke K. Predicting Age Using Neuroimaging: Innovative Brain Ageing Biomarkers. Trends Neurosci. 2017 Dec;40(12):681-690. Epub 2017 Oct 23 PubMed.
Lehallier B, Gate D, Schaum N, Nanasi T, Lee SE, Yousef H, Moran Losada P, Berdnik D, Keller A, Verghese J, Sathyan S, Franceschi C, Milman S, Barzilai N, Wyss-Coray T. Undulating changes in human plasma proteome profiles across the lifespan. Nat Med. 2019 Dec;25(12):1843-1850. Epub 2019 Dec 5 PubMed.
Melendez J, Sung YJ, Orr M, Yoo A, Schindler S, Cruchaga C, Bateman R. An interpretable machine learning-based cerebrospinal fluid proteomics clock for predicting age reveals novel insights into brain aging. Aging Cell. 2024 Sep;23(9):e14230. Epub 2024 Jun 24 PubMed.
Millar PR, Gordon BA, Luckett PH, Benzinger TL, Cruchaga C, Fagan AM, Hassenstab JJ, Perrin RJ, Schindler SE, Allegri RF, Day GS, Farlow MR, Mori H, Nübling G, Dominantly Inherited Alzheimer Network, Bateman RJ, Morris JC, Ances BM. Multimodal brain age estimates relate to Alzheimer disease biomarkers and cognition in early stages: a cross-sectional observational study. Elife. 2023 Jan 6;12 PubMed.
Oh HS, Rutledge J, Nachun D, Pálovics R, Abiose O, Moran-Losada P, Channappa D, Urey DY, Kim K, Sung YJ, Wang L, Timsina J, Western D, Liu M, Kohlfeld P, Budde J, Wilson EN, Guen Y, Maurer TM, Haney M, Yang AC, He Z, Greicius MD, Andreasson KI, Sathyan S, Weiss EF, Milman S, Barzilai N, Cruchaga C, Wagner AD, Mormino E, Lehallier B, Henderson VW, Longo FM, Montgomery SB, Wyss-Coray T. Organ aging signatures in the plasma proteome track health and disease. Nature. 2023 Dec;624(7990):164-172. Epub 2023 Dec 6 PubMed.
Skampardoni I, Nasrallah IM, Abdulkadir A, Wen J, Melhem R, Mamourian E, Erus G, Doshi J, Singh A, Yang Z, Cui Y, Hwang G, Ren Z, Pomponio R, Srinivasan D, Govindarajan ST, Parmpi P, Wittfeld K, Grabe HJ, Bülow R, Frenzel S, Tosun D, Bilgel M, An Y, Marcus DS, LaMontagne P, Heckbert SR, Austin TR, Launer LJ, Sotiras A, Espeland MA, Masters CL, Maruff P, Fripp J, Johnson SC, Morris JC, Albert MS, Bryan RN, Yaffe K, Völzke H, Ferrucci L, Benzinger TL, Ezzati A, Shinohara RT, Fan Y, Resnick SM, Habes M, Wolk D, Shou H, Nikita K, Davatzikos C. Genetic and Clinical Correlates of AI-Based Brain Aging Patterns in Cognitively Unimpaired Individuals. JAMA Psychiatry. 2024 May 1;81(5):456-467. PubMed.
Smith SM, Elliott LT, Alfaro-Almagro F, McCarthy P, Nichols TE, Douaud G, Miller KL. Brain aging comprises many modes of structural and functional change with distinct genetic and biophysical associations. Elife. 2020 Mar 5;9 PubMed.
Yang Z, Wen J, Erus G, Govindarajan ST, Melhem R, Mamourian E, Cui Y, Srinivasan D, Abdulkadir A, Parmpi P, Wittfeld K, Grabe HJ, Bülow R, Frenzel S, Tosun D, Bilgel M, An Y, Yi D, Marcus DS, LaMontagne P, Benzinger TL, Heckbert SR, Austin TR, Waldstein SR, Evans MK, Zonderman AB, Launer LJ, Sotiras A, Espeland MA, Masters CL, Maruff P, Fripp J, Toga AW, O'Bryant S, Chakravarty MM, Villeneuve S, Johnson SC, Morris JC, Albert MS, Yaffe K, Völzke H, Ferrucci L, Nick Bryan R, Shinohara RT, Fan Y, Habes M, Lalousis PA, Koutsouleris N, Wolk DA, Resnick SM, Shou H, Nasrallah IM, Davatzikos C. Brain aging patterns in a large and diverse cohort of 49,482 individuals. Nat Med. 2024 Oct;30(10):3015-3026. Epub 2024 Aug 15 PubMed.
View all comments by Peter MillarBrigham and Women's Hospital, Harvard Medical School
With great respect to the authors and others in the field, I urge the field to reject the nomenclature of "brain aging," or "biological aging," as distinct from "age." We are all, including the authors of this paper, trying hard to understand why people are likely to suffer cognitive impairment as they get older, and why some people experience cognitive decline at a young age. We also know that many people grow old without cognitive impairment. It is inaccurate language to describe someone who is young but has brain atrophy, or other abnormal brain imaging according to an MRI-based model, as "biologically old." The risk of prostate disorders skyrockets with age, too. We would not characterize a multimodal risk predictor for prostate cancer, or perhaps even nonspecific dysuria, as describing someone's "prostate age." Instead, I would propose we use the most accurate terms we can.
The work described in this paper is important. Perhaps a title could be "Plasma Proteomics Predicts a Multimodal Model of Abnormal Brain MRI."
By using the word "age" to mean "brain dysfunction," we risk summoning societal biases against old people. We should use the word "age" to mean just that—the amount of time elapsed since birth.
View all comments by Andrew SternUniversity of Pittsburgh School of Medicine
This is a monumental exercise in bioinformatics. I should probably let it rest there since it’s not my area of expertise. But, in this as in many things, I’m a bit of a contrarian so let me give my overall impression of the work from the standpoint of a lay reader.
As a cell biologist, I am unsure how much we truly learn from the work. I am worried about the exclusion criteria. They are not well articulated. Excluding people with underlying conditions is an understandable decision in a study of healthy aging, but it carries a risk of biasing the sample. I also worry about the sample being >97 percent white-Caucasian. The choice to use the U.K. Biobank is understandable, but also diminishes the broader application of the findings to the human population. Finally, I worry about using serum markers to gain insight into brain function. The authors are responsibly cautious in this regard, but offer no experiments to validate the strength of the connection between blood and brain with regard to their new findings.
My other issue is that the paper mostly contains a series of interesting correlations. Causality cannot be inferred from the data, and the supporting literature cited, while compelling, is far from proof of mechanism. Take brevican, for example. The authors' ideas are well argued but, based on their data, my own hypothesis would be that brevican is a symptom of aging (and cognitive decline), not a cause. That would make it possibly valuable as a diagnostic marker, but I would need to see much more to be convinced that it has therapeutic potential. Even as a diagnostic tool, the violin plots in the supplement suggest that while the correlation is true at a population level, there may be less utility at the level of an individual, which, after all, is where medical practice is applied.
One last comment. Aging is a phenomenon that is built into every eukaryote on the planet. It is not well understood at a biological level despite its importance. Because of this wide applicability, I would have asked the authors to either test their ideas in other species or change the title of the paper by inserting the word “human” in front of the word “brain” in their title. And for a journal entitled “Nature Aging,” I am surprised that the editors didn’t flag this themselves.
View all comments by Karl HerrupMake a Comment
To make a comment you must login or register.