🧬 Aging Wiki

Systemic proteome adaptions to 7-day complete caloric restriction in humans

8 min read

Systemic proteome adaptions to 7-day complete caloric restriction in humans

Pietzner M, Uluvar B, … O’Rahilly S, Jensen J, Langenberg C · Nature Metabolism 6(4):764–777 · 2024 · DOI: 10.1038/s42255-024-01008-9

The first in-depth temporal map of the human plasma proteome across a 7-day water-only fast, combining broad-capture proteomics with proteogenomic (pQTL → disease) integration. The headline conceptual advances: (1) the systemic, multi-organ response does not begin until after ~3 days of complete fasting — far longer than any commonly practiced intermittent-fasting window; (2) the fasting proteome signature is largely independent of weight loss; and (3) it is dominated by non-metabolic, extracellular-matrix (ECM) remodeling rather than a purely energy-centric switch. See prolonged-fasting for the intervention-level synthesis.

Design

  • n = 12 apparently healthy, non-smoking adults (5 women, 7 men); mainly (all but one) white-European ancestry. Exclusion: any known disease, body fat <15% (F) / <12% (M).
  • Intervention: stringent 7-day fast, absolute caloric restriction with ad libitum water only.
  • Sampling: morning (6–9 am) fasting blood drawn 2 days before, daily during the 7-day fast, and for 3 days after resuming ad libitum food intake. DXA at days −4/−1 (baseline mean), day 7, and day 10.
  • Proteomics: 2,923 protein targets via 2,941 assays across 117 samples on the Olink Explore 3072 platform (proximity extension assay, NPX/log2 units). ~80% (n=2,334) of targets had >75% of values above the limit of detection. All NPX values scaled to pre-fasting mean/SD, so changes read in standard-deviation (SD) units.
  • Statistics: linear mixed models (timepoint = fixed effect, participant = random effect); Benjamini–Hochberg FDR at q<5%. Proteogenomic integration used external large-scale pQTL + GWAS data with colocalization (posterior probability, PP).

Physiological response (clinical chemistry + body composition)

  • Mean weight loss 5.7 kg (SEM ±0.8) over 7 days (≈1.9 BMI units; baseline 77.5 kg, BMI 25.4); weight stayed 3.1 ± 0.6 kg below baseline after 3 days of refeeding.
  • Weight loss split between lean mass (−3.6 ± 0.49 kg, p<5.5×10⁻¹²) and fat mass (−1.6 ± 1.3 kg, p<9.9×10⁻¹³). Subcutaneous fat fell (−0.21 ± 0.07 kg, p<5.1×10⁻⁸); visceral fat did not (−0.07 ± 0.09 kg, p=0.12); no bone-mass loss (0.008 ± 0.014 kg, p=0.42).
  • Lean-mass loss was almost completely reversed (−0.69 ± 0.49 kg) within 3 days of refeeding, whereas fat loss was sustained (−1.85 ± 0.34 kg) — i.e., the acute lean-mass drop is largely a fluid/protein-pool phenomenon, not durable muscle catabolism.
  • Expected glucose→lipid fuel switch within 2–3 days: plasma glucose fell, free fatty acids rose and plateaued, and 3-hydroxybutyrate rose continuously throughout the fast (ketogenesis). Urinary nitrogen fell only late in the fast, consistent with increased renal urea reabsorption to spare protein.

Core proteomic findings

  • 1,044 targets (35.9%) changed significantly (FDR q<5%) over the study; 144 changed by >2 SD (22 up, 122 down).
  • The systemic response is delayed: only n=6 proteins were significant at 24 h and n=54 at 48 h (p<4.7×10⁻⁵); the bulk of change appeared only after 3 days. The count of decreasing proteins grew exponentially after day 3; increasing proteins plateaued after day 3. 66 proteins remained significantly different from pre-fasting levels 3 days after refeeding.
  • Canonical feeding/fasting hormones behaved as expected or surprisingly muted: leptin fell sharply (max −2.39 SD at day 4, p<7.7×10⁻³⁴); thyrotropin fell (min −1.41 SD at day 2). Ghrelin (+0.28 SD, p=0.66), adiponectin (−0.46 SD at day 7, p=0.01), and BDNF (−0.47 SD, p=0.57) changed little.
  • Largest movers (>4 SD), proposed as novel markers of prolonged (≥3 day) fasting: follistatin (FST) +5.62 SD at day 3 (p<3.2×10⁻⁴⁴); pcsk9 4.75 SD extreme at day 7 (p<4.1×10⁻⁴⁷); NBL1 −5.20 SD at day 7 (p<8.4×10⁻¹⁵). Only 3 proteins showed significant sex-differential effects. contradictory-evidence — PCSK9 direction: the manuscript’s page-4 text labels PCSK9 a “4.75 s.d. units increase,” but the same paper assigns PCSK9 to the declining cluster 4 (below), and Commissati 2025 independently found PCSK9 decreased (−1.49 fold) during a comparable fast with “no discrepancies” vs this study at the 7-day endpoint. Low-insulin/SREBP-2 biology also predicts a decrease. The “increase” wording appears to be a manuscript-internal error; PCSK9 most likely decreases during prolonged fasting (magnitude ~4.75 SD, extreme at day 7).
  • Pathway enrichment among changed proteins: IGF signaling (insulin-igf1), cytokine signaling, lipoprotein metabolism, plus less-described changes in complement/coagulation, protein glycosylation, cell adhesion, and neutrophil degranulation.
  • Tissue origin (Human Protein Atlas): changed proteins were enriched for those specifically expressed in liver, pancreas, adipose tissue, and intestine — a genuinely multi-organ response.

Nine temporal trajectory clusters

NAvMix clustering of the 1,034 changing targets yielded 9 clusters — 3 increasing, 6 decreasing — mapping distinct dynamics:

  • Increasing (clusters 1–3): hunger signaling (agouti-related peptide), lipid regulators (LDL-receptor, ANGPTL4), IGF-binding proteins, protein-degradation enzymes (cathepsins B/L), transient complement activation (p<1.6×10⁻⁵), and exocrine-pancreas proteins persistently elevated even after refeeding (REG1B, REG3G).
  • Early-declining (cluster 4): the glucose→lipid switch — PYY, PCSK9, lipoprotein lipase (lpl), APOA4, ECM organization (collagens, hyaluronidase-1), iron metabolism, plus strong sustained changes in EPHA1 (epithelial motility) and BPIFA2.
  • Late-declining (clusters 5–7): changes not appearing until day 2–3, enriched for ECM-receptor interaction (p<1.1×10⁻⁸), elastic-fiber proteins (p<4.8×10⁻⁴), and vascular-wall cell-surface interaction — interpreted as structural ECM degradation in the vasculature plausibly contributing to lean-mass loss. Includes fetuin-B, chemerin (RARRES2), and bone-metabolism proteins (SFRP4, osteoglycin). Senescence marker GLB1 and angiogenesis proteins (AMOT, SPON1, THBS2) overshot baseline after refeeding.
  • Compensatory (cluster 8): end-of-study exocrine-pancreas proteins (REG3A, elastases).
  • Stress response (cluster 9): ‘cellular response to stress’ (PSME2) + nucleotide metabolism, returning to baseline by day 7 even before refeeding; intracellular origin suggests leakage from erythroid/blood cells adapting to fasting.

Fasting is distinct from weight loss

Modeling proteome change against weight change vs against 3-hydroxybutyrate: weight and 3-OHB trajectories were only weakly correlated (r = −0.20). Most proteins (n=452) associated with the fasting signal (3-OHB) rather than with weight change (n=49). This dissociates prolonged fasting from short-term weight loss — but the authors stress early (≤48 h) changes were subtle relative to the profound post-3-day changes, so the dissociation is driven by the late, duration-dependent response.

Two proteins tracked weight most tightly (declining with loss, rebounding with regain): pulmonary surfactant protein D (SFTPD; beta=0.08, p<7.0×10⁻¹⁷**), genetically linked to COVID-19 critical illness, and IL-7 receptor (IL7R; beta=0.12, p<9.3×10⁻¹³)**, linked to autoimmune disease (multiple sclerosis, asthma) — candidate mediators of adiposity-associated disease risk.

The fasting signature was strongly enriched for ECM proteins (p<3.6×10⁻⁷). The single most-associated protein was tenascin-R, a brain-specific ECM protein integral to perineuronal nets (beta=−0.73, p<2.4×10⁻³⁷). Rising plasma 3-hydroxybutyrate was strongly associated with lower plasma levels of brain-specific ECM members (brevican, vitrin) and neurotrophic factors (neurotrophin-4) — offering a candidate molecular path for the centuries-old efficacy of ketogenic/fasting therapy in drug-resistant epilepsy.

Proteogenomic disease map

Linking the 1,044 fasting-responsive proteins to germline pQTLs and disease GWAS established putatively causal links for 212 proteins across ~500 diseases/health measures. Early-changing (<2-day) proteins showed little disease linkage except lipid genes (PCSK9, LPL, ANGPTL4); late, ≥3-day-changing proteins carried the diverse disease associations. For 652 of 1,044 predicted protein–disease links (52.2%), prolonged fasting moved the protein in a direction that compensates for genetically predicted adverse risk; for the remainder, fasting moved proteins toward higher predicted risk.

Worked examples:

  • SWAP70 ↔ rheumatoid arthritis (colocalization PP=99.1%): the rs4910499 A-allele (MAF 36.9%) raises plasma SWAP70 (beta=0.28, p<4.0×10⁻²⁷) and RA risk (OR 1.08, p<6.7×10⁻¹⁰). Fasting lowered SWAP70 (p<6.0×10⁻⁵, ~1.5 SD drop peaking day 6) — a candidate partial explanation for RA symptom/pain relief reported during prolonged fasting.
  • HYOU1 ↔ coronary artery disease (PP=95.1%): the rs1177562 T-allele (MAF 40.3%) raises HYOU1 (beta=0.20, p<2.8×10⁻¹⁶) and CAD risk (OR 2.02 per SD genetically predicted HYOU1, 95% CI 1.54–2.64, p<3.0×10⁻⁷); also linked to higher adiposity/fat mass/liver enzymes and lower HDL. HYOU1 (an HSP70-family ER chaperone accumulating under hypoxia) fell during fasting — potential compensation for CAD risk; mechanism unestablished.
  • INHBC ↔ triglycerides: rs61352607 G-allele (beta=0.65, p<3.4×10⁻¹³⁹) raises triglycerides (PP 96.0%); INHBC rose with triglycerides during fasting (beta=0.49, p<3.7×10⁻⁵). Liver-expressed; possible novel endocrine TG regulator.
  • Bone: 7 proteins genetically linked to higher bone-mineral-density declined during fasting (TNFRSF11A, SFRP4, HIP1R, RGMA, PGF, GALNT3, ATXN3), but bone-promoting markers also declined (AXL, GDF15/GFRAL, ADAM12, ANGPTL7) — net effect ambiguous, and total bone mass was unchanged over the 7 days (p=0.42).
  • Adverse-direction example: coagulation factor XI rose, aligning with its genetically predicted higher thrombotic-event risk — a putative harm signal during prolonged fasting.

Limitations (authors’ + extracted)

  • n=12, no control group, homogeneous (white-European, young, healthy) — restricts generalizability; cachexia/severe-starvation states untested.
  • Proteogenomic ‘compensation’ predictions are exploratory; the frequency/intensity of fasting needed to translate into clinical benefit is unknown, and the paper explicitly warns against reading fasting as a “sole solution” for the diseases discussed.
  • Platform coverage is finite (no direct brain/tissue proteomics); diurnal rhythm in the response may be partly missed by once-daily sampling.

Cross-references

Graph View

Backlinks