henhaobio
Sign in

bioRxivImmunologyPreprint

Longitudinal proteomic module profiles vary across human monocyte differentiation and polarization conditions

Roberto Navarro QuiroziD, Yiris Diaz-Olmos, Noelia Geribaldi-DoldániD, Cecilia Fernandez-Ponce, Eloína Zarate PeñataiD, Katherine EscorciaiD, Ibeth Karina Luna-Rodríguez, Andrea Jaruffe Pinilla, Yesit Bello Lemus, Lisandro Pacheco-LugoiD, Leonardo Pacheco Londono, Antonio Acosta-Hoyos, Nataly Galan FreyleiD, Elkin Navarro QuiroziD

PDFDOI

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (9 Sept 2026) · CC BY 4.0

In reanalyzed PXD016245 monocyte-differentiation proteomes, eight predefined protein-module scores retain a joint condition association (incremental R^2 0.313663, permutation p = 0.0002) beyond source block and day.

Comparisons across monocyte culture conditions can confound source-preparation variation and shared differentiation time with condition effects, while serial or pooled samples inflate apparent replication. Extending a visual account of the source proteome, this reanalysis specifies covariates, module representation and a trajectory-preserving permutation null, and states what a small structured dataset can and cannot support.

Key results

  1. ShownWithin PXD016245, adding condition and condition-by-day terms to a source-block plus categorical-day model explained an extra 31.37% of total eight-module score variation, satisfying both prespecified criteria (p ≤ 0.05 and incremental R^2 > 0.02).

    Incremental R^2 = 0.313663, pseudo-F = 4.257567 (15 tested and 57 residual df), permutation p = 0.0002 from 9,999 within-block, trajectory-preserving permutations (only one sampled assignment at least as extreme); reduced model 40.64% vs full model 72.00% of total variation; 80 observations. · 2.4 Primary multivariate model and permutation…; 3.3 Condition and its temporal component account…

  2. ShownDropping any single source block preserved the joint condition association, but estimates shifted by preparation, being largest when block B was removed; these analyses reuse frozen scores standardized with all 84 samples.

    Leave-one-block-out with 60 observations each: incremental R^2 0.315012 (omit P), 0.316772 (omit A), 0.324852 (omit C), 0.389790 (omit B); pseudo-F 2.946285 to 5.089541; all permutation p = 0.002000. These overlapping subsets are not independent replications. · 2.5 Secondary analyses and sensitivity checks; 3.5 The association persists across source block…

  3. ShownSeparate block-adjusted tests per sampling day detected condition differences at days 4, 6, 8 and 10 but not day 2, and the effect fraction dipped between days 4 and 6; a residual-permutation interaction sensitivity also supported condition-dependent temporal patterns.

    Day 2: incremental R^2 0.122569, pseudo-F 1.503224, q = 0.103600. Days 4, 6, 8, 10: incremental R^2 0.551451, 0.480102, 0.610152, 0.614912 (q = 0.000333, 0.002125, 0.000333, 0.000333; Benjamini-Hochberg over five days, 16 observations each). Freedman-Lane interaction sensitivity: pseudo-F 1.613093, incremental R^2 0.095072, p = 0.0004. · 2.5 Secondary analyses and sensitivity checks; 3.4 Later sampling days show adjusted…

  4. ShownEvery one of the eight modules was individually associated with condition after multiplicity correction, and the joint association survived dropping any single module and rescaling modules to unit variance, though dispersion differences remain untested.

    Module tests: q = 0.001600 to 0.004300 (Benjamini-Hochberg, eight tests); within-module incremental R^2 0.133295 (glycolysis/pentose-phosphate) to 0.481138 (differentiation-associated). Leave-one-module-out: all p = 0.0002, change −0.044889 to 0.018480. Unit-variance scaling: incremental R^2 0.339754, pseudo-F 4.403193, p = 0.0002. · 3.5 The association persists across source block…

  5. SuggestedDescriptive profiles showed glycolysis/pentose-phosphate scores rising and signed redox scores falling from day 2 to 10 in all conditions, while day-10 mitochondrial scores were lowest in M2 and the three-protein resolution-associated score highest in M1 cultures.

    Glycolysis/pentose-phosphate rose and redox fell in all 16 block-condition pairs (osteoclast glycolysis mean paired change 1.994, range 1.686 to 2.385). Day-10 mitochondrial means: M2 −0.099 vs M1 0.846, dendritic-cell 1.065, osteoclast 0.906. M1 resolution score: −0.442 to 1.128 (terminal 0.884 to 1.520) vs day-10 M2 0.221, dendritic-cell −0.277, osteoclast 0.013. · 3.2 Shared early-to-terminal changes coexist…

Methods, data and reproducibility

Deposited MaxQuant LFQ protein-group intensities were log2-transformed, z-scored per feature and averaged into eight signed predefined module scores; a Euclidean multivariate linear model tested condition plus condition-by-day terms beyond source block and categorical day with 9,999 within-block trajectory-preserving permutations.

  • PXD016245 serial bulk proteomes of monocyte-derived M1, M2, dendritic-cell and osteoclast cultures, days 2 to 10; three individual-donor plus one 40-donor pooled preparation · human · n = 80 observations; 16 block-condition trajectories · PXD016245
  • Deposited MaxQuant LFQ protein-group matrix filtered of reverse, contaminant and site-only groups; 84 sample columns incl. four day-0 monocytes used only for standardization · human · n = 6,029 protein groups (346 of 6,375 rows removed) · PXD016245
  • Eight predefined protein modules: glycolysis/pentose-phosphate, mitochondrial, lipid, redox, inflammasome/cytokine, resolution, differentiation, proteostasis · human · n = 54 eligible assay features; 55 feature-module memberships
  • Archived Metabolomics Workbench metabolite-score comparison (three scores), audited and withdrawn from biological interpretation · ST001835
Methods & tools
MaxQuant 1.5.3.30 (original LFQ search, match-between-runs), Orbitrap Fusion mass spectrometer (source acquisition), PRIDE/ProteomeXchange, Metabolomics Workbench, Permutational multivariate linear model (Euclidean distance-based partition, pseudo-F), Freedman-Lane residual permutation test, Benjamini-Hochberg false-discovery correction, NumPy PCG64 generator (seed 20260615), Python 3.10.20, NumPy 2.2.6, pandas 2.3.3, SciPy 1.15.3
Reagents & environment
Python 3.10.20, NumPy 2.2.6, pandas 2.3.3, SciPy 1.15.3; permutations via NumPy PCG64, seed 20260615; frozen 80-row input SHA-256 1fb757edd9bbb8e2354547ab733fc04a7888cad661b19e488a6e13c4664a8e66
Reproducibility gaps
No code URL stated; the Supplementary Data 3 verifier reproduces only the primary endpoint from frozen scores, not score construction or secondary analyses; exact FASTA release (HUMAN170307.fasta) not established.
Limitations (6)
  • The four source blocks mix three individual-donor preparations with one preparation pooled from 40 donors, whose mapping to block labels is unresolved; the 80 serial observations are not independent donor replicates and cannot support population-wide claims. · stated by the authors
  • Module scores rest on partial protein coverage (resolution panel limited to LRP1, APOE and TGFB1; MERTK, AXL, IL10 and XBP1, ATF4, DDIT3 absent), duplicate G6PD and PDIA3 features and signed averaging, so they do not measure pathway activity, flux or function. · stated by the authors
  • No external validation: the archived ST001835 metabolomic comparison was withdrawn after audit found nonspecific metabolite mappings and unmodeled donor/time dependence, and leave-one-block-out subsets overlap rather than replicate. · stated by the authors
  • Restricted dispersion testing and confidence intervals were not produced, so the effect cannot be attributed purely to centroid shifts, and condition-linked acquisition or culture artifacts cannot be excluded by the permutation design. · stated by the authors
  • The time-adjusted plan was locked only after earlier descriptive and pooled analyses of the same dataset, and bulk serial cultures cannot separate changes in cell composition or survival from within-cell remodeling. · stated by the authors
  • Features with up to 50% missing LFQ values were scored without imputation, so module scores for different samples can average different feature subsets; condition-linked missingness could shape scores, and no analysis of this is reported. · noted in this summary

Read this if you reanalyze public serial proteomes of monocyte-derived macrophage, dendritic-cell or osteoclast cultures and want a worked example of block-preserving permutation testing with explicit scope limits.

Citations
0

Abstract

Human monocyte differentiation involves changes in multiple protein programs, but comparisons across culture conditions can conflate source variation with differentiation time. We reanalyzed public proteomic measurements arranged in four source blocks across M1-polarizing, M2-polarizing, dendritic-cell and osteoclast conditions at days 2, 4, 6, 8 and 10. The original experiment used three individual-donor preparations and one preparation pooled from 40 donors; these are not four individual donors. Eight predefined protein-module scores formed an 80-observation matrix. A multivariate model tested joint condition and condition-by-time terms beyond source block and categorical time, using 9,999 permutations that preserved complete block-specific culture trajectories. The joint condition terms accounted for an additional 31.37% of total module-score variation (pseudo-F = 4.257567; permutation p = 0.0002). The association persisted after omission of each source block, with incremental R-squared of 0.315-0.390. Secondary day-specific tests detected differences at days 4-10, whereas day 2 did not meet the false-discovery criterion. Descriptive trajectories showed a shared increase in glycolysis/pentose-phosphate scores alongside condition-dependent mitochondrial, redox and proteostasis profiles. At day 10, M2-labelled cultures had the lowest mean mitochondrial score, whereas M1-labelled cultures had the highest three-protein resolution-associated score. All eight module tests survived multiplicity correction, and no single-module omission abolished the multivariate association. These findings describe condition-associated configurations within four heterogeneous source preparations. Equal weighting of source blocks does not estimate a mean across individual donors. The analysis does not establish pathway activity, single-cell trajectories, population-level replication or external biological validation.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISPinto '20Palmer '25Purcu '21Oates '22Chevrier '25Anders '20
62% semantic matchImmunology
S. Pinto, H. Kim, Y. Subbannayya+3 more·bioRxiv·Feb 27, 2020