📄 medicine

A shared injury–homeostasis gradient in the human CSF proteome across neurological disease

By analyzing a meta-atlas of over 10,000 cerebrospinal fluid samples across diverse neurological conditions, researchers identified a universal injury–homeostasis gradient characterized by opposing complement-plasma inflammation and neuronal-synaptic homeostasis that tracks disease severity and can be approximated by a simple two-protein biomarker model.

Original authors: Alireza Mansouri, Nicholas Mikolajewicz, Cecile Riviere-Cazaux, Kyle Tuohy, Sruthi Ranganathan, Rahul Kumar, Alexandra Miller, Manmeet Ahluwalia, Paul Boutros, Chetan Bettegowda, Terry Burns, Thomas K

Published 2026-08-11

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Alireza Mansouri, Nicholas Mikolajewicz, Cecile Riviere-Cazaux, Kyle Tuohy, Sruthi Ranganathan, Rahul Kumar, Alexandra Miller, Manmeet Ahluwalia, Paul Boutros, Chetan Bettegowda, Terry Burns, Thomas Kislinger

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine your brain is a bustling, high-security city, constantly buzzing with activity. To keep this city running smoothly, it's surrounded by a protective moat called cerebrospinal fluid (CSF). This isn't just water; it's a rich, complex soup that washes over the brain's streets, carrying messages, nutrients, and waste. Scientists have long been trying to read the "newspaper" written in this fluid to understand what's happening inside the city. If the city is healthy, the newspaper has a calm, orderly tone. But if there's a riot, a fire, or a construction project gone wrong (like a disease), the newspaper changes. The challenge has been that every time scientists looked at this fluid for a different disease—like Alzheimer's, brain tumors, or infections—they seemed to be reading a completely different language. It was like trying to compare a weather report from London with a traffic update from Tokyo; the words were different, making it hard to see the big picture.

This is where the new study steps in, acting like a master translator. The researchers wanted to know: Is there a universal "headline" that appears in the CSF of people with any neurological problem? They gathered a massive library of data, combining information from 35 different studies involving over 10,000 people. By using advanced computer tools to sift through thousands of proteins (the tiny building blocks that make up the fluid's message), they discovered that despite the different diseases, there is actually a shared, underlying story. They found that the brain's fluid doesn't just get "messy" randomly; it shifts along a specific, predictable path. This path is a tug-of-war between two states: a state of "homeostasis" (where the brain is calm, connected, and working in harmony) and a state of "injury" (where the brain is under attack, inflamed, and scrambling to repair itself).

The team, led by Alireza Mansouri and Nicholas Mikolajewicz, didn't just find this pattern; they mapped it out in incredible detail. They started by defining a "core" list of 747 proteins that are reliably found in the CSF of humans, no matter which lab tested them or which disease the patient had. Think of this as the "standard alphabet" of the brain's language. Using this alphabet, they identified 13 distinct "meta-programs"—groups of proteins that work together like a choir singing a specific song. Some of these songs are about immune defense, others about brain cell connections, and some about the barriers that keep the brain safe.

The most exciting discovery, however, is the "injury–homeostasis gradient." When the researchers looked at how these 13 programs behaved across all the different diseases, they realized they all lined up on a single sliding scale. On one end of the scale is the "Homeostasis Pole," where the fluid is rich in proteins that keep neurons happy and connected. On the other end is the "Injury Pole," where the fluid is flooded with proteins related to inflammation, blood leakage, and the immune system's emergency response.

What's fascinating is that this scale isn't just for one disease. Whether a patient has a brain tumor, a severe infection, multiple sclerosis, or Alzheimer's, their CSF profile slides along this same line. The more severe the disease or the more active the damage, the further the profile slides toward the "Injury Pole." For example, patients with high-grade brain tumors or acute infections were found deep in the injury zone, while healthy controls and people in the early, pre-symptomatic stages of disease were closer to the homeostasis side. Even more surprisingly, the researchers found that they could approximate this complex, deep-proteomic score using just two proteins: C5 (a marker of inflammation and injury) and NRCAM (a marker of healthy brain connections). It's like realizing that instead of reading the entire newspaper to know if the city is in trouble, you only need to check two specific headlines to get a very accurate picture of the situation.

This study suggests that instead of looking for a unique "fingerprint" for every single disease, we might be able to use this shared gradient to understand the general state of the brain's health. It offers a new way to measure how "injured" a brain is, regardless of what caused the injury. While the paper notes that this is a candidate model that needs further testing to become a standard bedside tool, it provides a powerful new framework. It turns the chaotic noise of thousands of different diseases into a single, understandable story: the brain is constantly balancing between staying calm and fighting a fire, and now, we have a better ruler to measure that balance.

Technical Summary: A Shared Injury–Homeostasis Gradient in the Human CSF Proteome Across Neurological Disease

Problem Statement
Cerebrospinal fluid (CSF) is the biofluid most proximal to the central nervous system (CNS), reflecting both local CNS biology and regulated exchange with the circulation. While deep proteomic studies have expanded the detectable CSF proteome from ~2,600 to over 3,300 proteins and identified disease-specific biomarkers in conditions like Alzheimer's disease and multiple sclerosis, these advances have remained largely disease- and cohort-specific. Consequently, it remains unclear which CSF proteins are reproducibly measurable across heterogeneous studies, which coordinated protein states are transferable between cohorts, and whether diverse neurological disorders share a common higher-order proteomic structure beyond diagnosis-specific signatures.

Methodology
The authors assembled a meta-atlas comprising 35 CSF proteomic datasets (2016–2026), including two unpublished datasets, totaling 12,627 enrolled samples and 10,560 quality-controlled samples. These datasets span 15 countries, 11 disease types, and two primary proteomic platforms: mass spectrometry (MS; n=32) and affinity proteomics (n=3).

The study employed a multi-stage analytical framework:

Core Proteome Definition: Using a "coverage-aware detection" (CAD) framework on 21 MS studies, the authors stratified 12,606 proteins into five detection tiers (A–E) based on pooled detection rates and confidence intervals, defining a 747-protein "core" mass-spectrometry-accessible proteome (Tiers A and B).
Meta-Program Discovery: To identify coordinated protein variation, the authors independently derived proteomic programs within each cohort using non-negative matrix factorization (NMF), independent component analysis (ICA), and scale-free shared-nearest-neighbour (SSN) networks. They filtered 28,795 candidate programs for robustness and recurrence, resulting in 13 consensus meta-programs (MPs).
Axis Derivation: The activities of the 13 meta-programs across 10,560 samples were subjected to principal component analysis (PCA). The first principal component (PC1) was identified as a dominant axis of variation.
Validation and Modeling: The authors validated the transferability of meta-programs using leave-one-study-out analyses. They further developed a two-protein model (C5 and NRCAM) to approximate the injury–homeostasis axis and benchmarked it against routine CSF indices (total protein, leukocyte count, albumin quotient).
Clinical Correlation: Axis scores were correlated with clinical outcomes (survival, tumor burden, disease progression) across neuro-oncological, neurodegenerative, and neuroinflammatory cohorts.

Key Contributions and Results

The 747-Protein Core Proteome: The study defines a reproducible, mass-spectrometry-accessible core of 747 proteins (337 strict-core, 410 broad-recurrent). This core is consistently detected regardless of CSF sampling site (lumbar vs. cranial) and generalizes to held-out cohorts with high precision-recall performance (AUPRC = 0.823). The core is biologically structured into six classes (e.g., coagulation/complement, neuronal adhesion, metabolic) and is enriched for plasma- and barrier-associated proteins, while underrepresenting intracellular nuclear/mitochondrial proteins.
Thirteen Consensus Meta-Programs: The analysis identified 13 transferable meta-programs that span major biological domains: neuronal/synaptic, metabolic/catabolic, extracellular matrix (ECM), barrier/epithelial, immune/inflammatory, and plasma/complement. These programs generalize across independent cohorts, recovering held-out meta-program activity with a median Spearman's $\rho$ of 0.69.
The Injury–Homeostasis Axis: The dominant finding is a single axis of variation (PC1) that explains 55.6% of the variance in meta-program activity. This axis opposes a "homeostatic" pole (enriched for neuronal-synaptic proteins like NRCAM) against an "injury" pole (enriched for complement, plasma, and inflammatory proteins like C5).
- Diagnosis-Spanning Nature: The axis is not specific to a single disease. Acute infections, structural injuries, and high-grade neoplasms cluster at the injury pole; reference controls and presymptomatic groups cluster at the homeostatic pole; and chronic inflammatory/neurodegenerative disorders occupy intermediate positions.
- Clinical Severity: Higher axis scores correlate with adverse clinical outcomes, including treatment non-response, shorter overall survival, and greater tumor burden in CNS lymphoma, leptomeningeal disease, and glioma. In neurodegenerative diseases (AD, DLB, FTD, PD) and multiple sclerosis, the axis tracks disease severity and progression.
The C5–NRCAM Surrogate Model: The authors demonstrate that the complex injury–homeostasis axis can be approximated by a two-protein model consisting of Complement Component 5 (C5) and Neuronal Cell Adhesion Molecule (NRCAM).
- In the Riviere-Cazaux cohort, this model recapitulated the axis score with $R^2 = 0.869$ .
- In the larger, heterogeneous Bader cohort, the model achieved $R^2 = 0.764$ , outperforming routine CSF indices (total protein, leukocytes, QAlb), which yielded $R^2 = 0.547$ .
- A random-effects meta-analysis across 34 cohorts confirmed a pooled $R^2$ of 0.812 for the C5–NRCAM model.

Significance and Claims
The paper claims to establish a generalizable framework and vocabulary for cross-cohort benchmarking and biological interpretation of CSF proteomics. By moving from single-disease marker discovery toward a transferable architecture of CSF proteomic states, the study reveals that diverse neurological disorders share a common higher-order structure driven by an injury–homeostasis gradient.

The authors posit that this gradient represents a dynamic response to CNS injury, where the "injury" pole reflects barrier disruption, complement activation, and plasma ingress, while the "homeostasis" pole reflects neuronal integrity and synaptic function. They suggest that the C5–NRCAM model offers a potential bridge between deep proteomic profiling and feasible bedside assessment, serving as a transdiagnostic state biomarker to position patients along a continuum from neuronal homeostasis to active inflammatory injury. This approach complements disease-specific biomarkers by providing a measure of shared inflammatory and tissue-injury processes across the CNS.

The study acknowledges limitations, including the use of relative normalized abundances (precluding absolute concentration inference), the loss of isoform-specific information due to gene-symbol harmonization, and the observational nature of the data, which cannot determine causality between complement-associated injury and neuronal loss. Nevertheless, the atlas provides a reproducible foundation for future cross-disease biological interpretation.

Technical Summary: A Shared Injury–Homeostasis Gradient in the Human CSF Proteome Across Neurological Disease

More like this