Longer lipids with age: a causal change or a shift in lipid composition?

The October 2026 paper combines cross-species associations, genetic perturbation, supplementation and membrane models. It already distinguishes long-species accumulation from short-species depletion and analyzes classes separately. Our addition is sample-level estimands, fixed panels and measurement boundaries. 3 4

01Define length before interpreting it

Lipids serve storage, membrane and signaling functions. Signals depend on separation, ionization, standards and structural annotation. Comparability within a class does not imply that signals across classes are calibrated, additive molar quantities. 1 11

PC(36:2) generally describes total carbon across two tails; CL(72:8) has four. Both can average 18 carbons per chain. Pooling 36 and 72 as “chain length” also pools chain count. Sum composition does not identify every tail, double-bond position or isomer. Sphingoid backbones and fatty acids, or ether-linked and acyl chains, are not interchangeable. 2

Our main proxy is limited to clearly annotated ester glycerolipids, dividing total carbon by known chain count. It does not reconstruct full structures. Within one class, division by a positive constant cannot change Pearson correlation. Thus chain-count correction cannot itself reverse the author's within-class correlations. Weighting, species inclusion and the estimand can change the answer.

A deliberately chosen four-species example illustrates the distinction: carbons 14,16,18,20; baseline amounts 1,100,1000,1; fold changes 2,4,0.5,8. Length and log2 change have a positive correlation (r approximately 0.227), while the weighted mean falls from 17.817 to 17.130. This is a finite definition example, not observed aging. A pure PC36-to-CL72 replacement likewise raises total carbon without changing the 18-carbon chain mean. Neither example supplies real cross-class molar fractions or human parameters.

02Data gates: samples, pairing and versions

We obtained the 2024 per-sample tables for ten mouse tissues, human muscle aging and short-term activity. The 2026 paper already used them, so this is reanalysis. Age comparisons involve male mice aged 3 versus 24 months and humans aged 20–30 versus 65–80. The older human group combines different fitness/activity groups; the source columns do not recover those covariates. Age is not the only difference. 6 7 8 10

The mouse table shares an eight-column template, but BAT AGED2 and spleen YOUNG1/AGED1 are entirely empty. Actual young/old comparisons are therefore 8/7 for BAT, 7/7 for spleen, and 8/8 elsewhere. The reporting summary allows preparation losses. We did not fill empty columns with zero or claim 8/8 everywhere. Tissues share animals and are not ten independent studies.

The female aging table has 5/17 samples and the male table 7/23. Activity data contain 12 identifiable participants and 35 columns, with one SIT column absent. Exact pairing yields 11 EXERCISE–SIT, 11 SITLESS–SIT and 12 EXERCISE–SITLESS pairs. The original experiment was a randomized crossover; these tables do not provide each person's period/order, so our paired comparison does not reconstruct the original period model. 9 10

An additional 2021 adipose dataset contains 48 male mice aged 2 or 24 months, four per age at six daily time points. BAT/WAT came from the same animals. It is not listed among the 2026 integrated inputs, but shares authors/platform and is not independent-laboratory certification. We retain all times and both tissues. One Results statement uses ZT16 where Methods/table use ZT18; we follow original sample columns. The study explicitly disallows absolute interclass quantification. 5 11 12

The 2026 main text remains subscription-only. We actually read public legends, reporting summary, code and supplements. The university archive exposes a 2025 discussion chapter; the underlying Chapter 6 is embargoed until 2027. Discussion cannot substitute for final Methods. The reporting summary says at least three species per class, whereas v1.0 code sets five. We use the explicit five-species code gate and retain untested classes, without claiming complete pipeline reproduction. 3 4 5 27 28

03A shared panel retains some positive directions and limits generality

Within each tissue/cohort, we fixed uniquely named ester species with finite positive values in every actual sample, then required at least five species and variable carbon counts per class. Missing values were not imputed. Original peak processing, responses and structures were not revalidated. Across 15 tissue/cohort units, 63 species overlap; 54 meet the class gate: PC 14, PE 11, PI 11, PG 7, LPC 11. These are not 15 independent biological replications. The shared-panel comparison was added after reading local-panel results: identities were fixed by input availability before shared-panel outcomes were read. This is exploratory and was not preregistered.

For each sample we calculate class-specific signal-weighted mean-chain proxies, then weight the five classes equally. Below are old-minus-young differences on that shared panel, in carbon-proxy units. Intervals are individual 95% intervals; P values receive Holm adjustment within each unit's metric family. Intervals lack simultaneous coverage. All tissues, classes, four aggregate definitions, both panels and time strata remain available. 6 7 8 12

Lengthening does not disappear everywhere on a fixed panel, opposing the strong claim that class replacement creates everything. It is also not universal. Quadriceps has approximately −0.028 with a wide interval on the local complete panel, versus +0.063 on the shared one. The female complete-panel index is approximately +0.054 and uncertain, versus −0.008 on the shared panel. Included classes/species can alter direction and precision.

These proxies still depend on species responses, processed low signals, tissue composition and complete-panel selection; the common panel favors readily detected species. The negative WAT direction does not deny all age-related membrane changes, and human uncertainty does not prove no age effect. Different estimands should not be used to declare that another estimand failed replication.

Time strata further limit interpretation: shared-panel BAT has both positive and negative directions across six times. WAT point estimates are negative at all six, but each time has only four mice per age and none of these individual family-adjusted contrasts is conclusive. The balanced-day average does not certify each time or exclude cage/sampling-environment effects. 12

For four-day activity, shared-panel EXERCISE–SIT is approximately −0.011 (−0.032 to 0.011). Other contrasts and incomplete pairing are retained; no family-adjusted comparison of this index gives stable shortening. The paper also reports no global chain shift for this exercise experiment. This does not establish absence of other exercise benefits. 9 3

04Cross-species and intervention evidence has a defined scope

Eight new lipidomic CSVs contain upstream log2 changes, without per-animal abundance. We recalculated 479 class/condition entries using v1.0, retaining species, sex/age conditions and all BH results. Species counts are not animal n, and correlated classes are not independent animal replication. 18 19 20 21 22 23 24 25

Valid class correlations have medians approximately +0.321/+0.333 in the two fly batches, and −0.376 in rapamycin-treated HL-1 cells. In-vivo heart sex comparisons lack the same global shortening, whereas dietary-restricted hearts have more positive evidence. Cell treatment, systemic treatment and ages are different comparisons. Authors report two/three ten-fly pooled samples in the fly batches, five HL-1 samples per condition and six worm lipidomics samples per condition; aggregate CSVs do not recover that biological variance. 3 5

The authors already show that short-species loss can produce positive correlations, discussing the Lands cycle, beta-oxidation and remodeling. Relative enrichment of longer species does not directly measure elongation of existing chains or elongase flux. 3

05RNAi and supplementation do not yet isolate a unique length mediator

In the lifespan source table, both full-dose F36A2.9 RNAi experiments raise median survival from 23 to 27 days; elo-1 also has positive results. For elo-3, the displayed experiment changes 25 to 20 days, while another changes 23 to 25 with P approximately 0.147. A selected curve cannot define every elongase's fixed effect. 14

Combined RNAi, stress-pathway backgrounds and fatty-acid supplementation strengthen the mechanistic case. But C18:1,C20:1 and C22:1 all alter the RNAi survival pattern; supplementation also changes control survival. Identifying a unique length mediator requires same-batch worm composition/exposure, feeding, bacterial environment, other remodeling and formal interaction. Loss of significance against BSA alone is not a precise rescue estimate. 3 14

We read the complete lifespan summary and extracted 17 key summaries for arithmetic audits, without rebuilding Kaplan–Meier, censoring or log-rank tests. Shared controls are not new trials each time they reappear. The source prints 15.4% for medians of 22 and 26 days (18.18% by the printed medians), appends “less than 0.0001” to 0.000261, and has unclear RNAi/count formatting elsewhere. Conflicts stay unresolved. Identical medians with a significant log-rank result are not inherently inconsistent: log-rank compares survival distributions. 14

06Human genetics and membrane models contribute different support

MR exposure is whole-blood PLB1 expression, not measured chain length. Among six results, PLB1–frailty has b approximately 0.0195, SE 0.00675, source P 0.00389 and six-comparison Holm P 0.0233. Heart failure and grip strength lack equivalent evidence. This signal is worth preserving but is not a per-carbon harm or lifespan estimate. Many SNPs enter PCA; SNP weights/LD, colocalization and complete pleiotropy information were not obtained. We did not rerun MR or exclude other PLB1 functions. 13

With the same PC headgroup and one double bond per tail, all-atom model thickness rises from 32.09 to 51.57 angstrom as mean tail length rises 14 to 24. This supports a conditional physical effect. At 22 carbons, two monounsaturated tails give 46.84 angstrom, versus 37.34 for two tails with six double bonds each. Length alone is incomplete. 15

Mixed membranes contain multiple headgroups, and replacing tails also changes double-bond combinations. Their differences cannot all be assigned to length alone. Public tables summarize four replicas per system without trajectories. The P-value block does not clearly assign all-atom versus coarse-grained results; we do not assign it. Longer mixed membranes do not have monotonically greater thickness, while other properties can change. HeLa supplementation/toxicity shows molecular perturbations can have consequences. The 8.68%-to-15.16% odd-chain index uses carbon-weighted signals, rather than a direct molecular fraction or a complete uptake/oxidation flux measurement. 3 15 16 17

07Current judgment

Evidence supports relative within-tissue lipid remodeling, conditional membrane-physics effects and some worm lifespan benefits. It does not identify a stable chain-length age index across all tissues/classes or establish that shortening human lipid chains extends life.

A discriminating next experiment would keep species panels and within-class response calibration fixed, measuring remodeling, membrane properties, function and survival in the same animals/timeline. Supplementation should measure actual tissue exposure and separate length, saturation and other metabolic effects. Cardiolipin quantity, four-tail composition, mitochondrial abundance and muscle force require separate answers, as examined in our cardiolipin and muscle-function study.

Download the reproduction package. Thirteen aggregate/audit outputs, code and three original figures are traceable; third-party originals and human sample-level work stay private. The same Codex agent researched, calculated, self-reviewed and translated, without independent human/professional review. Full final text, original MS/death/MD records, clinical chain-length intervention and newly verified task backup remain unavailable or limited; tool checks do not certify science.