Results

Overview of statistical power and gene-level significance

The first and most important result is negative: no individual genes reached genome-wide FDR <0.05 for the age effect in control mice, the NMN effect at six months, or the age-by-treatment interaction term in any of the three tissues. This was true for skeletal muscle, liver, and WAT. Accordingly, the analysis does not provide definitive gene-level discoveries, and all subsequent candidate lists are hypothesis-generating under relaxed nominal criteria.

The relaxed rescue criteria produced sizable tissue-specific lists. Skeletal muscle had 421 nominal NMN-rescue candidates. Liver had 355. White adipose tissue had 397. These numbers should not be read as the number of statistically confirmed NMN-rescued genes. They are the number of genes that met the relaxed candidate definition: nominal age effect in control mice, nominal interaction effect, and opposite sign between aging and interaction coefficients. The distinction is central to the interpretation of the entire analysis.

The downstream stringency filters reduced these lists substantially. After applying the interaction effect-size threshold of 0.8, 180 genes remained in skeletal muscle, 171 in liver, and 135 in white adipose tissue. When the high-confidence rule was applied, using P_interaction <0.01 and an absolute interaction coefficient of at least 0.8, the counts dropped to 54 genes in skeletal muscle, 49 in liver, and 36 in white adipose tissue. These high-confidence genes remain nominally selected candidates rather than genome-wide FDR-significant findings, as summarized in Table 1.

Tissue | Age effect, FDR <0.05 | NMN at 6 months, FDR <0.05 | Interaction, FDR <0.05 | Relaxed rescue genes | Effect-filtered rescue genes | High-confidence genes | Selection mode
Skeletal muscle | 0 | 0 | 0 | 421 | 180 | 54 | Relaxed nominal P_age <0.05, P_interaction <0.05, opposite sign
Liver | 0 | 0 | 0 | 355 | 171 | 49 | Relaxed nominal P_age <0.05, P_interaction <0.05, opposite sign
WAT | 0 | 0 | 0 | 397 | 135 | 36 | Relaxed nominal P_age <0.05, P_interaction <0.05, opposite sign

Tissue-specific rescue patterns

Skeletal muscle showed the largest relaxed rescue list, with 421 genes. Of these, 178 rose with age in control mice and were lowered by NMN, whereas 243 fell with age and were raised by NMN. The skeletal-muscle pathway results were broad and noisy. Positive interaction-ranked GSEA terms included regulation of protein transport, multivesicular body organization, cytoplasmic transport-related themes, and some apoptotic and microRNA-associated terms. However, the FDR values for the top skeletal muscle terms were high, so these signals should not be treated as statistically strong pathway discoveries. They are better viewed as directional clues.

Liver showed 355 relaxed rescue candidates. Of these, 164 rose with age and were lowered by NMN, while 191 fell with age and were raised by NMN. Liver produced the clearest pathway-level signal in the entire reanalysis. NMN was associated with negative enrichment of fatty-acyl-coenzyme A (CoA) biosynthetic and metabolic processes. The strongest term was fatty-acyl-CoA biosynthetic process, with a normalized enrichment score (NES) of -1.973 and FDR 0.0109. Long-chain fatty-acyl-CoA metabolic process had NES -1.879 and FDR 0.0343, and fatty-acyl-CoA metabolic process had NES -1.858 and FDR 0.0423. KEGG fatty acid elongation also showed negative enrichment, with NES -1.774 and FDR 0.0567. This pattern suggests suppression of an age-associated hepatic lipid remodeling program. It is the most statistically credible pathway signal in the analysis.

White adipose tissue had 397 relaxed rescue candidates. Of these, 193 rose with age in control mice and were lowered by NMN, and 204 fell with age and were raised by NMN. The WAT pathway results pointed toward trafficking, ubiquitination, and metabolic signaling, but most were exploratory. The strongest negative terms included retrograde transport from endosome to Golgi, with NES -1.827 and FDR 0.0722, and protein K63-linked ubiquitination, with NES -1.808 and FDR 0.0593. Regulation of receptor-mediated endocytosis also appeared among the negative terms but with weaker FDR support. Positive terms included regulation of Ras protein signal transduction, regulation of target of rapamycin complex 1 (TORC1) signaling, and regulation of generation of precursor metabolites and energy, again with FDR values that call for caution. The directions of these rescue patterns are summarized in Table 2, and the pathway-level GSEA highlights are summarized in Table 3.

Tissue | Total relaxed rescue genes | Rose with age in controls; NMN lowered expression | Fell with age in controls; NMN raised expression
Skeletal muscle | 421 | 178 | 243
Liver | 355 | 164 | 191
WAT | 397 | 193 | 204

Tissue | Direction | Term | NES | P | FDR
Skeletal muscle | Positive | Extrinsic apoptotic signaling pathway via death domain receptors | +1.754 | 0.000 | 0.841
Skeletal muscle | Positive | Negative regulation of developmental growth | +1.721 | 0.000 | 0.792
Skeletal muscle | Positive | Regulation of protein transport | +1.719 | 0.000 | 0.550
Skeletal muscle | Positive | Negative regulation of neuron projection development | +1.718 | 0.000 | 0.423
Skeletal muscle | Positive | Production of miRNAs involved in gene silencing by miRNA | +1.706 | 0.004 | 0.420
Skeletal muscle | Positive | Negative regulation of phosphorylation | +1.692 | 0.000 | 0.453
Skeletal muscle | Positive | Cellular response to ionizing radiation | +1.686 | 0.000 | 0.432
Skeletal muscle | Positive | Phosphatidylethanolamine metabolic process | +1.679 | 0.002 | 0.431
Skeletal muscle | Positive | Multivesicular body organization | +1.676 | 0.002 | 0.401
Skeletal muscle | Positive | Regulation of vascular endothelial growth factor receptor signaling pathway | +1.672 | 0.000 | 0.389
Skeletal muscle | Negative | Diol biosynthetic process | -1.806 | 0.000 | 0.370
Skeletal muscle | Negative | Glycogen metabolic process | -1.775 | 0.000 | 0.366
Skeletal muscle | Negative | Oxoacid metabolic process | -1.764 | 0.000 | 0.303
Skeletal muscle | Negative | Aminoglycan biosynthetic process | -1.749 | 0.000 | 0.302
Skeletal muscle | Negative | Regulation of proteasomal protein catabolic process | -1.742 | 0.000 | 0.279
Skeletal muscle | Negative | Purine nucleotide metabolic process | -1.737 | 0.002 | 0.253
Skeletal muscle | Negative | Sphingosine metabolic process | -1.735 | 0.000 | 0.227
Skeletal muscle | Negative | Glycosaminoglycan biosynthetic process | -1.708 | 0.000 | 0.311
Skeletal muscle | Negative | Regulation of telomerase RNA localization to Cajal body | -1.700 | 0.000 | 0.315
Skeletal muscle | Negative | Positive regulation of telomerase RNA localization to Cajal body | -1.665 | 0.004 | 0.490
Liver | Positive | Endoplasmic reticulum organization | +1.859 | 0.000 | 0.156
Liver | Positive | Protein insertion into ER membrane | +1.836 | 0.000 | 0.116
Liver | Positive | Spliceosomal complex assembly | +1.819 | 0.000 | 0.114
Liver | Positive | Peptidyl-serine dephosphorylation | +1.814 | 0.000 | 0.097
Liver | Positive | Regulation of lipid catabolic process | +1.797 | 0.000 | 0.111
Liver | Positive | Cortical cytoskeleton organization | +1.790 | 0.000 | 0.105
Liver | Positive | Inner mitochondrial membrane organization | +1.779 | 0.002 | 0.113
Liver | Positive | Positive regulation of sphingolipid biosynthetic process | +1.753 | 0.000 | 0.145
Liver | Positive | Positive regulation of ceramide biosynthetic process | +1.753 | 0.000 | 0.145
Liver | Positive | Regulation of ceramide biosynthetic process | +1.749 | 0.000 | 0.140
Liver | Negative | Fatty-acyl-CoA biosynthetic process | -1.973 | 0.000 | 0.011
Liver | Negative | Global genome nucleotide-excision repair | -1.889 | 0.000 | 0.041
Liver | Negative | Long-chain fatty-acyl-CoA metabolic process | -1.879 | 0.000 | 0.034
Liver | Negative | Fatty-acyl-CoA metabolic process | -1.858 | 0.000 | 0.042
Liver | Negative | Acyl-CoA metabolic process | -1.788 | 0.002 | 0.143
Liver | Negative | KEGG fatty acid elongation | -1.774 | 0.002 | 0.057
Liver | Negative | Long-chain fatty-acyl-CoA biosynthetic process | -1.753 | 0.000 | 0.238
Liver | Negative | Positive regulation of transcription from RNA polymerase II promoter involved in cellular response to chemical stimulus | -1.736 | 0.000 | 0.274
Liver | Negative | Negative regulation of response to DNA damage stimulus | -1.730 | 0.002 | 0.269
Liver | Negative | Positive regulation of protein dephosphorylation | -1.714 | 0.000 | 0.320
WAT | Positive | Negative regulation of cell cycle G2/M phase transition | +1.765 | 0.000 | 0.186
WAT | Positive | Positive regulation of Ras protein signal transduction | +1.754 | 0.002 | 0.123
WAT | Positive | Negative regulation of G2/M transition of mitotic cell cycle | +1.753 | 0.000 | 0.083
WAT | Positive | Regulation of cellular amine metabolic process | +1.710 | 0.002 | 0.180
WAT | Positive | Regulation of TORC1 signaling | +1.699 | 0.000 | 0.184
WAT | Positive | KEGG choline metabolism in cancer | +1.693 | 0.000 | 0.086
WAT | Positive | Regulation of generation of precursor metabolites and energy | +1.681 | 0.000 | 0.233
WAT | Positive | Transcription by RNA polymerase III | +1.678 | 0.000 | 0.211
WAT | Positive | Amino acid import across plasma membrane | +1.662 | 0.002 | 0.259
WAT | Positive | Negative regulation of ubiquitin-dependent protein catabolic process | +1.648 | 0.004 | 0.303
WAT | Negative | Retrograde transport, endosome to Golgi | -1.827 | 0.000 | 0.072
WAT | Negative | Protein K63-linked ubiquitination | -1.808 | 0.000 | 0.059
WAT | Negative | Regulation of receptor-mediated endocytosis | -1.730 | 0.002 | 0.239
WAT | Negative | Regulation of cellular response to transforming growth factor beta stimulus | -1.680 | 0.002 | 0.498
WAT | Negative | Vacuolar acidification | -1.660 | 0.008 | 0.568
WAT | Negative | Response to organophosphorus | -1.660 | 0.000 | 0.477
WAT | Negative | Regulation of transmembrane receptor protein serine/threonine kinase signaling pathway | -1.649 | 0.004 | 0.494
WAT | Negative | Modulation by host of symbiont process | -1.604 | 0.009 | 0.944
WAT | Negative | Histone H3-K4 methylation | -1.603 | 0.004 | 0.855
WAT | Negative | Tetrapyrrole metabolic process | -1.601 | 0.016 | 0.790

Cross-tissue robust genes

The rescue programs were mostly tissue-specific. Pairwise overlap was modest: skeletal muscle and WAT shared 15 genes, liver and skeletal muscle shared 10, and liver and WAT shared 10. No gene was rescued in all three tissues. This finding argues against a universal NMN transcriptional rescue program across these metabolic organs, as summarized in Table 4.

Tissue set | Skeletal muscle | Liver | WAT
Skeletal muscle | 421 | 10 | 15
Liver | 10 | 355 | 10
WAT | 15 | 10 | 397

A total of 35 genes were rescued in at least two tissues. The two principal named candidates were Ras-related protein Rab-11A (RAB11A) and carnitine palmitoyltransferase 2 (CPT2); the remaining entries are retained as official mouse gene symbols. The robust set consisted of 0610009O03RIK, 0610012A05RIK, 1700024D23RIK, 2810405F18RIK, 2810441K11RIK, 6330581L23RIK, ACD, APBA2, BCHE, CMA2, CPT2, D430042O09RIK, DUSP11, EIF3S4, FMO2, GALR1, HS3ST6, LOC192950, MAP2K2, MOBKL2C, MYO15, MYOM2, OLFR508, OLFR627, PKNOX1, PSME3, RAB11A, RANGAP1, SLC13A5, SPG21, SYNGR2, TRP53INP2, V1RE1, WAP, and ZFAND2B. The cross-tissue robust genes are listed in Table 5.

Gene | Tissues | Coef. skeletal muscle | Coef. liver | Coef. WAT | Mean interaction | Direction consistent | High confidence | Mito- chondrial | GSEA lead terms | Best p | Priority
RAB11A | Skeletal muscle; WAT | +4.017 | NA | +0.464 | +2.241 | Yes | Yes | No | 39 | 2.01e-03 | 8
GALR1 | Liver; WAT | NA | +5.007 | -4.884 | +0.062 | No | Yes | No | 2 | 5.00e-12 | 7
CMA2 | Liver; WAT | NA | +5.386 | +6.181 | +5.783 | Yes | Yes | No | 0 | 5.99e-03 | 7
MYO15 | Skeletal muscle; WAT | +5.990 | NA | +6.091 | +6.040 | Yes | Yes | No | 0 | 6.74e-03 | 7
OLFR627 | Skeletal muscle; WAT | +3.915 | NA | +3.841 | +3.878 | Yes | Yes | No | 0 | 8.52e-03 | 7
D430042O09RIK | Liver; WAT | NA | +3.945 | +3.814 | +3.879 | Yes | Yes | No | 0 | 9.47e-03 | 7
CPT2 | Skeletal muscle; WAT | +0.259 | NA | +0.142 | +0.201 | Yes | No | Yes | 4 | 5.32e-03 | 7
OLFR508 | Skeletal muscle; WAT | +4.151 | NA | -5.906 | -0.878 | No | Yes | No | 0 | 6.39e-03 | 6
V1RE1 | Skeletal muscle; Liver | +6.116 | -3.794 | NA | +1.161 | No | Yes | No | 0 | 7.78e-03 | 6
MAP2K2 | Liver; WAT | NA | -0.072 | -0.104 | -0.088 | Yes | No | No | 4 | 4.61e-03 | 6
ACD | Liver; WAT | NA | -0.081 | -0.082 | -0.082 | Yes | No | No | 1 | 6.91e-03 | 6
WAP | Skeletal muscle; Liver | -3.722 | -5.066 | NA | -4.394 | Yes | No | No | 1 | 1.11e-02 | 6
ZFAND2B | Liver; WAT | NA | -0.065 | -0.065 | -0.065 | Yes | No | No | 2 | 1.23e-02 | 6
SLC13A5 | Liver; WAT | NA | +5.190 | +4.909 | +5.049 | Yes | No | No | 5 | 1.62e-02 | 6
APBA2 | Skeletal muscle; Liver | +3.449 | +5.047 | NA | +4.248 | Yes | No | No | 3 | 1.78e-02 | 6
SYNGR2 | Skeletal muscle; WAT | -0.209 | NA | -0.151 | -0.180 | Yes | No | No | 1 | 2.72e-02 | 6
RANGAP1 | Liver; WAT | NA | +0.137 | -0.131 | +0.003 | No | No | No | 5 | 4.96e-03 | 5
HS3ST6 | Skeletal muscle; WAT | -3.894 | NA | +4.690 | +0.398 | No | No | No | 3 | 1.11e-02 | 5
MYOM2 | Skeletal muscle; WAT | +3.636 | NA | -3.659 | -0.011 | No | No | No | 5 | 1.12e-02 | 5
1700024D23RIK | Skeletal muscle; Liver | +3.706 | +3.937 | NA | +3.821 | Yes | No | No | 0 | 1.20e-02 | 5
0610012A05RIK | Skeletal muscle; Liver | +3.795 | +5.151 | NA | +4.473 | Yes | No | No | 0 | 1.25e-02 | 5
6330581L23RIK | Skeletal muscle; WAT | +0.323 | NA | +0.311 | +0.317 | Yes | No | No | 0 | 1.41e-02 | 5
0610009O03RIK | Skeletal muscle; WAT | -0.112 | NA | -0.081 | -0.096 | Yes | No | No | 0 | 1.52e-02 | 5
SPG21 | Skeletal muscle; Liver | -0.153 | -0.067 | NA | -0.110 | Yes | No | No | 0 | 1.60e-02 | 5
MOBKL2C | Liver; WAT | NA | -0.106 | -0.163 | -0.134 | Yes | No | No | 0 | 1.60e-02 | 5
DUSP11 | Skeletal muscle; WAT | -0.070 | NA | -0.097 | -0.084 | Yes | No | No | 0 | 1.64e-02 | 5
2810405F18RIK | Skeletal muscle; Liver | +0.168 | +0.077 | NA | +0.123 | Yes | No | No | 0 | 1.99e-02 | 5
PKNOX1 | Skeletal muscle; WAT | -0.105 | NA | -0.116 | -0.110 | Yes | No | No | 0 | 2.17e-02 | 5
PSME3 | Skeletal muscle; Liver | +0.111 | +0.089 | NA | +0.100 | Yes | No | No | 0 | 2.21e-02 | 5
TRP53INP2 | Skeletal muscle; Liver | +0.147 | +0.143 | NA | +0.145 | Yes | No | No | 0 | 3.01e-02 | 5
EIF3S4 | Skeletal muscle; WAT | -0.060 | NA | -0.080 | -0.070 | Yes | No | No | 0 | 3.67e-02 | 5
FMO2 | Skeletal muscle; WAT | -0.224 | NA | +0.184 | -0.020 | No | No | No | 1 | 3.68e-02 | 5
2810441K11RIK | Skeletal muscle; Liver | -0.133 | -0.103 | NA | -0.118 | Yes | No | No | 0 | 4.13e-02 | 5
BCHE | Skeletal muscle; WAT | -0.191 | NA | +0.358 | +0.084 | No | No | No | 0 | 2.90e-03 | 4
LOC192950 | Liver; WAT | NA | +3.776 | -0.169 | +1.804 | No | No | No | 0 | 1.40e-02 | 4

Among these 35 robust genes, 26 were direction-consistent across tissues. Eight were also high-confidence by the stage 2 criteria. Only one robust gene was mitochondrial by the fallback mitochondrial annotation: CPT2. The mitochondrial overlap among robust genes was not statistically significant, with a hypergeometric P value of 0.109. At the tissue level, the fallback mitochondrial overlap identified CPT2, CS, and SUCLA2 in skeletal muscle; NDUFV1, POLG, and TFB2M in liver; and CPT2 in WAT.

The nonsignificant hypergeometric result does not support broad mitochondrial transcriptional restoration as a principal mechanism of NMN in this dataset. CPT2 is highlighted because it was the only mitochondrial gene among the robust cross-tissue candidates and because its biological function is directly relevant to fatty-acid substrate handling. It should not be interpreted as evidence of generalized mitochondrial biogenesis or broad mitochondrial gene-set activation, as summarized in Table 6.

Level | Tissue or gene set | Rescue genes | Mitochondrial hits | Mitochondrial genes | Hypergeometric p
Tissue | Skeletal muscle | 421 | 3 | CPT2, CS, SUCLA2 | 0.162087
Tissue | Liver | 355 | 3 | NDUFV1, POLG, TFB2M | 0.112503
Tissue | WAT | 397 | 1 | CPT2 | 0.734318
Robust set | Genes rescued in at least two tissues | 35 | 1 | CPT2 | 0.1093

RAB11A as the strongest cross-tissue candidate

RAB11A was the highest-priority robust gene in the final summary. It was rescued in skeletal muscle and WAT, with consistent positive interaction coefficients. In skeletal muscle, the control aging coefficient was -4.165, and the interaction coefficient was +4.017, with P_interaction = 0.00725. In WAT, the control aging coefficient was -0.408, and the interaction coefficient was +0.464, with P_interaction = 0.00201. Thus, in both tissues RAB11A fell with age in control mice and was shifted upward by NMN. It met high-confidence criteria in at least one tissue and appeared in 39 GSEA leading-edge terms, mostly in skeletal muscle.

The pathway context is important. RAB11A appeared in leading-edge terms related to multivesicular body organization and assembly, cytoplasmic microtubule organization, regulation of transport, protein localization to organelle, and plasma membrane-bounded cell projection assembly. These leading-edge appearances are exploratory and should be regarded as a prioritization heuristic rather than independent evidence that endocytic recycling changed functionally. The repeated appearance of RAB11A in transport-linked leading edges nevertheless makes the candidate biologically coherent.

Endocytic recycling controls the composition of the plasma membrane by returning internalized proteins and lipids back to the cell surface after sorting in endosomes. Grant et al. [14] emphasized that recycling pathways are spatially and temporally regulated and are important for processes such as cytokinesis, cell adhesion, morphogenesis, cell fusion, learning, and memory. More broadly, Rab guanosine triphosphatases coordinate membrane identity, vesicle budding, motility, and fusion [15]. In that context, RAB11A provides a plausible link between NMN response and cellular logistics.

The translational interpretation is that NMN may be associated with preservation of membrane trafficking or recycling capacity during aging in selected tissues. This could affect receptor availability, membrane protein quality control, and organelle-associated transport. It may also connect indirectly to insulin sensitivity, because cell-surface receptor trafficking is a key determinant of signaling competence. Rab11 has been associated with glucose transporter 4-containing vesicles and insulin-responsive trafficking, although that specific insulin-receptor link was not tested in this dataset [16]. RAB11A is therefore a testable candidate for follow-up, not a confirmed mediator of NMN action.

CPT2 as the main fatty-acid oxidation candidate

CPT2 was the only mitochondrial gene among the 35 robust cross-tissue rescue candidates. It was rescued in skeletal muscle and WAT, and the direction was consistent. In skeletal muscle, CPT2 had a control aging coefficient of -0.194 and an interaction coefficient of +0.259, with P_interaction = 0.0147. In WAT, it had a control aging coefficient of -0.088 and an interaction coefficient of +0.142, with P_interaction = 0.00532. The effect sizes were modest and did not meet the high-confidence effect threshold, but the consistency across two fat-oxidizing tissues makes CPT2 biologically compelling.

CPT2 is part of the carnitine-dependent system required for mitochondrial long-chain fatty-acid oxidation. Long-chain fatty acids must be converted and transported into mitochondria before β-oxidation can proceed efficiently. Longo et al. [17] described the carnitine system as essential for fatty-acid oxidation, and Zammit [18] framed carnitine palmitoyltransferase biology as central to metabolic function. The broader physiology of mitochondrial fatty-acid β-oxidation also supports interpreting CPT2 as a substrate-utilization candidate rather than as evidence of generalized mitochondrial activation [19]. The present data therefore support a narrow carnitine-shuttle and fatty-acid oxidation hypothesis, not a claim of broad mitochondrial transcriptional reprogramming.

CPT2 provides a direct bridge between the transcriptional results and the metabolic phenotypes reported in the original NMN study. The interpretation should still be measured. This analysis does not show that NMN increases CPT2 protein abundance, CPT2 enzyme activity, or fatty-acid oxidation flux. It only shows that CPT2 expression met the relaxed rescue definition in two tissues and was the only robust mitochondrial gene under the fallback annotation. That makes CPT2 a high-priority validation target, not a confirmed mechanism.

MYO15, CMA2, and other high-magnitude robust candidates

MYO15 and CMA2 also stood out because of their large, direction-consistent interaction effects. MYO15 was rescued in skeletal muscle and WAT. Its mean interaction coefficient across tissues was +6.040, with low cross-tissue variability, and it met high-confidence criteria. This pattern suggests a possible connection to cytoskeletal organization or motor-protein-related biology. Such a connection could matter in muscle and adipose tissue, where cellular structure, intracellular transport, and organelle positioning are important. However, the dataset does not establish a direct role for MYO15 in NMN response.

CMA2 was rescued in liver and WAT, with a mean interaction coefficient of +5.783 and direction consistency. It also met high-confidence criteria. CMA2 may point toward peptide processing or local signaling biology in metabolic tissues, but this interpretation is tentative. As with MYO15, the effect size is notable, but the relaxed statistical framework and microarray platform make independent validation essential.

Several other robust genes were prioritized by the downstream scoring, including GALR1, OLFR627, D430042O09RIK, OLFR508, V1RE1, MAP2K2, ACD, WAP, ZFAND2B, SLC13A5, APBA2, SYNGR2, and RANGAP1. Some of these genes are biologically interpretable, while others, especially olfactory receptor (Olfr), RIK, and LOC entries, may carry a higher risk of probe or annotation noise. High-magnitude Olfr, RIK, and LOC candidates should be considered the lowest priority for wet-laboratory follow-up until probe identity, hybridization specificity, and current gene annotation are independently verified. The presence of multiple such genes is one reason the gene-level results should not be over-read.

Direction-discordant genes

Nine robust genes showed direction discordance, meaning NMN moved expression in opposite directions across tissues. These were BCHE, FMO2, GALR1, HS3ST6, LOC192950, MYOM2, OLFR508, RANGAP1, and V1RE1. GALR1 is a useful example. It was rescued in liver and WAT, but with opposite interaction signs: positive in liver and negative in WAT. This does not necessarily make the result unimportant. It means the biology is tissue-dependent and should not be summarized as a single systemic effect.

Direction discordance is translationally relevant. If NAD+ precursor responses differ across tissues, then biomarkers from one tissue may not generalize to another. It also suggests that safety and efficacy should be assessed in a tissue-aware way. A gene that appears "rescued" in two tissues may still reflect different mechanisms in each tissue.