Biomolecular map of albumin — Sharma et al. 2023
Bibliographic info
- Authors: Neha Sharma†, Sushmita Pandey†, Manisha Yadav, Babu Mathew, Vasundhra Bindal, Nupur Sharma, Gaurav Tripathi, Sadam H. Bhat, Abhishak Gupta, Rakhi Maiwall, Shvetank Sharma, Shiv Kumar Sarin*, Jaswinder Singh Maras* (†equal contribution)
- Journal: Journal of Hepatology, September 2023, vol. 79, pp. 677–691
- DOI: 10.1016/j.jhep.2023.04.018
- Institution: Institute of Liver and Biliary Sciences (ILBS), New Delhi, India
Key question
Does the multi-omics landscape of albumin-bound molecules (the “albuminome”) distinguish non-survivors from survivors in acute liver failure, and can it be used as a validated early-mortality predictor?
Study design
| Cohort | ALF (S) | ALF (NS) | HC | Disease controls |
|---|---|---|---|---|
| Training | 8 | 32 | 5 | — |
| Test cohort 1 | 53 | 107 | 20 | — |
| Test cohort 2 | — | — | — | 50 SAH |
| Total | 61 | 139 | 25 | 50 |
- ALF definition: jaundice + hepatic encephalopathy within 4 weeks + INR >1.5; no prior CLD or alcohol use
- Analytical workflow: albumin purified from baseline plasma → PTM characterisation + ABiC measurement → 4-arm multi-omics (proteomics, metaproteomics, metabolomics, lipidomics on the albumin-bound fraction) → ML validation
Concept: the albuminome
The authors define the albuminome as the ensemble of albumin-bound biomolecules: proteins, bacterial peptides, metabolites, and lipids that co-purify with serum albumin. Since low-MW molecules bind albumin to increase their half-life, the albuminome acts as a compartmentalised reservoir reflecting systemic pathophysiology. The key insight is that even one drop of blood suffices (albumin is ~60% of plasma protein).
Albumin PTM and functionality findings
Cys34 redox fractions (HMA/HNA1/HNA2)
| Parameter | HC | S | NS | HC→NS trend |
|---|---|---|---|---|
| HMA (%) | highest | intermediate | lowest | ↓ |
| HNA1 (%) | lowest | intermediate | highest | ↑ (p=0.0016) |
| HNA2 (%) | lowest | intermediate | highest | ↑ (p=0.001) |
| AOS (HNA/HMA ratio) | lowest | intermediate | highest | ↑ (p=0.001) |
HNA2 and AOS significantly higher in NS vs S vs HC (p<0.05), consistent with progressive irreversible oxidation of Cys34 across the severity spectrum.
Oxidative/glycative damage markers
| Marker | NS vs S | NS vs HC |
|---|---|---|
| AGE (arbitrary units) | ↑ (p=0.0003) | ↑ (p=0.0001) |
| AOPP (μg/mol) | ↑ (p=0.0001) | ↑ (p=0.0001) |
| IMA (absorbance units) | ↑ (p=0.0001) | ↑ (p=0.0001) |
| IMAr (IMA/albumin ratio) | ↑ (p=0.0003) | ↑ (p=0.0001) |
| ABiC (%) | ↓ (p=0.0001) | ↓ (p=0.0001) |
ABiC (albumin binding capacity) is lowest in NS, reflecting maximal functional impairment.
Proteomics (albumin-bound proteins)
- 233 proteins total identified in the albumin-bound fraction
- 56 DEPs (differentially expressed proteins) NS vs S (FC >1.5, p<0.05, FDR<0.01)
- Top 5 proteins enriched in non-survivors: QSOX1, IGKV1-16, IKBIP, JUP, ALDH4A1
- QSOX1: quiescin sulfhydryl oxidase 1 (redox enzyme)
- JUP: junction plakoglobin (cell junction)
- ALDH4A1: aldehyde dehydrogenase 4 (proline catabolism, mitochondrial)
- AUC 0.997 for the top-5 protein panel predicting non-survival
Metaproteomics (bacterial peptides bound to albumin)
- Enriched phyla in NS: Proteobacteria, Firmicutes, Actinobacteria
- Key taxa enriched in NS: Listeria, Clostridium, Pseudomonas, Enterobacteriaceae
- AUC 0.999 for top-5 bacterial peptide panel predicting non-survival
- Bacterial taxa in NS correlated with triglycerides (TG[4:0/12:0/12:0]) and phosphatidylserine (PS[39:0]) at r² >0.7
Metabolomics (albumin-bound metabolites)
- 17 KEGG pathways enriched in NS
- Key pathways: arginine/proline metabolism, bile acid, mitochondrial breakdown
- Top 5 metabolites for NS prediction:
- Nicotinic acid
- L-acetyl carnitine
- L-carnitine
- Pregnenolone sulfate
- N-(3-hydroxybutanoyl)-L-homoserine lactone
- AUC 0.98 (95% CI: 0.95–1.0) for non-survival prediction; POD >90%
- Test cohort 1 validation (LC-MS + 5 ML algorithms): AUC 0.936, sensitivity 95%, specificity 90%, overall accuracy >92%
- KM survival: POD metabolome cut-off >70% → significantly higher early mortality (<30 days; log-rank <0.05)
- Cox regression: HR 5.81 (POD metabolome), best predictor; outperforms MELD score
Lipidomics (albumin-bound lipids)
- 270 albumin-bound DELs in ALF vs HC (196 up, 74 down)
- 240 DELs in NS vs S (189 up, 51 down; FC >1.5, p<0.05, FDR<0.01)
- Top 5 lipid species panel: AUC 0.99 for non-survival
- Upregulated lipid classes in NS: neutral glycerolipids, phosphatidylserine, phosphatidylmethanol
- PLS-DA clearly segregates ALF-NS from ALF-S and HC
Multi-omics integration
Cross-correlation of all 4 arms yielded 4 clusters (r² >0.7, p<0.05):
- Cluster 1 (↑ in NS): bacterial taxa correlated with metabolites (porphobilinogen, nicotinic acid) + TG + PS lipids → points to gut-liver axis dysbiosis + mitochondrial failure
- Cluster 2 (↑ in NS): immunoglobulin variable chains + metabolites (mitochondrial failure pathway) + lipids; Serum amyloid A1 with multiple lipid species
- Clusters 3 & 4: downregulated in NS (includes phyla Firmicutes/Actinobacteria correlating with ↓ PS, sphingosine, TG, PMe)
- Principal component analysis PC1 (30.9%) and PC2 (12.7%) clearly segregate NS
Multi-omics correlation with albumin function
All DEP/DEM/DEL/DEMP signatures correlated with:
- Clinical parameters and severity indices (MELD)
- Albumin modification markers (HNA2, AOS, AGE, AOPP, IMA)
- Albumin functionality (ABiC)
- In contrast, HMA (native albumin) correlated positively with AbiC
Aetiology robustness
Aetiologies compared (HEV vs DILI): no change in albumin-bound multi-omics signatures across aetiologies in survivors vs non-survivors → signatures are aetiology-independent.
Key conclusions
- In ALF, albumin is hyperoxidised (↑HNA2) and glycated (↑AGE/AOPP), with substantially reduced binding capacity (↓ABiC)
- The albuminome distinguishes NS from S with near-perfect AUC across all four omics arms (0.997 proteomics, 0.999 metaproteomics, 0.99 lipidomics, 0.98 metabolomics)
- The 5-metabolite panel (nicotinic acid, L-acetyl carnitine, L-carnitine, pregnenolone sulfate, N-(3-hydroxybutanoyl)-L-homoserine lactone) is the best single panel: AUC 0.98, validated >92% in test cohort, HR 5.81 by Cox regression, outperforms MELD
- All signatures are aetiology-independent (HEV vs DILI)
- Even one drop of blood is sufficient for albumin-based multi-omics profiling
Connections to ALBOM and our work
- HNA2/HMA framework is the same as the top-down isoform HMA/HNA2 equivalence used in el-balkhi-2025 — provides independent ALF validation
- Confirms Cys34 irreversible oxidation (HNA2 ↑) as a pan-liver-disease severity marker across cirrhosis AND acute liver failure
- Albuminome concept directly extends the “albumin as a recorder” framework central to el-balkhi-2025
- ABiC (albumin binding capacity) parallels the “effective albumin” concept from bernardi-2023 and baldassarre-2021-ealb
- The gut dysbiosis → albumin-bound bacterial peptides → lipid dysregulation axis (Cluster 1) is a novel ALF-specific mechanism not seen in CLD studies
Relevance to PlasmaMatch
Top-5 NS proteins (QSOX1, IGKV1-16, IKBIP, JUP, ALDH4A1) are albumin-bound, not albumin itself — these are potential candidates to add to the PlasmaMatch library if top-down MS of individual proteins is performed.