Restoring information in aged gene regulatory networks by single knock-ins

Abstract: A hallmark of aging is loss of information in gene regulatory networks. These networks are tightly connected, raising the question of whether information could be restored by perturbing single genes. We develop a simple theoretical framework for information transmission in gene regulatory networks that describes the information gained or lost when a gene is "knocked in" (exogenously expressed). Applying the framework to gene expression data from muscle cells in young and old mice, we find that single knock-ins can restore network information by up to 10%. Our work advances the study of information flow in networks and identifies potential gene targets for rejuvenation.

• The paper demonstrates that single gene knock-ins can partially restore mutual information lost in aged gene regulatory networks.
• A minimalist binary model validated with single-cell RNA-seq data quantifies the impact of genetic interventions on network interactions.
• The study identifies key restorative genes, notably linked to mitochondrial function, underscoring potential avenues for targeted gene therapies.

Restoring Information in Aged Gene Regulatory Networks via Single Knock-ins

Abstract and Motivation

This study addresses the information loss observed in gene regulatory networks (GRNs) during aging, with a focus on restoration through targeted single gene perturbations. Aging is typified by functional dysregulation at the molecular level including impaired gene expression coordination. Metrics from information theory, notably mutual information (MI), have revealed this gradual loss in GRNs. The pivotal question investigated here is whether precise genetic interventions, specifically single gene knock-ins (exogenous upregulation), can remediate the global information deficit characteristic of aged networks.

Theoretical Framework and Model

A minimalist theoretical model is constructed for information transmission in binary gene regulatory motifs. Each gene (transcription factor, TF; target gene, TG) is described as being in either an 'on' or 'off' state, with endogenous switching rates $\alpha$ (TF activation), $\beta$ (TG activation when TF is off), and $\gamma$ (TG activation when TF is on). The model allows exact inference of these rates from processed single-cell RNA-seq data—no parameter fitting is required. Knock-ins are modeled by augmenting the endogenous activation rates with an exogenous term $k$, which directly alters steady-state probabilities and consequently the MI within TF-TG pairs.

Joint probabilities $p_{ij}$ for gene activation states ($i$ for TF, $j$ for TG) are empirically derived and used to quantify MI:

$I = \sum_{ij} p_{ij} \log \frac{p_{ij}}{q_i r_j}$

where $q_i$ and $r_j$ are the marginals. This model permits efficient quantification of information transmission and its change post-knock-in at all network distances.

Analysis of Gene Expression and Information Loss

Empirical investigation leverages SMART-seq2 single-cell RNA-seq data from young (3 months) and old (24 months) murine limb muscle cells (Tabula Muris Senis) and a curated mouse GRN (TRRUST v2). Analysis confirms a substantial shift towards lower expression levels in aged cells. MI computed across 6,253 TF-TG pairs also shows a statistically significant decrease with age, with average MI per pair dropping by an order exceeding the associated standard error.

Predictive Results of Single Knock-ins

Application of the framework predicts the restoration of network information via single gene knock-ins. Genes are perturbed individually, with MI and expression profiles recalculated for all network interactions:

• Direct effects (distance 1): The maximal restored information for directly affected pairs (i.e., those where the knocked-in gene is a TF or TG) reaches up to 0.02 nats per pair. However, since any single gene often regulates a minor subset (<1%) of the network, the average impact across all pairs is limited to ~10% of the age-induced deficit.
• Network propagation: Effects are propagated recursively through the network, accounting for feedback and multi-input regulation via randomized selection and averaging. The signal decay is pronounced, but knock-ins affecting highly connected hubs propagate benefits modestly further, especially for genes with extensive downstream regulation.

Interestingly, optimal restoration for many high-impact genes is achieved at maximal expression levels (on-probability near unity). This simplifies experimental design, making full knock-in more advantageous than fine-tuned partial expression.

Implications and Biological Target Identification

The study highlights five genes with the highest restorative effects: Ppara, Phox2b, Esrra, Med23, and Ppargc1b. Many are notably associated with mitochondrial function (Ppara, Esrra, Ppargc1b), aligning with established links between mitochondrial decline and aging—particularly in skeletal muscle. The identification of Phox2b and Med23, with no prior explicit association to aging hallmarks, underscores the model’s utility for uncovering novel regulatory targets.

The additive nature of information restoration from multiple knock-ins—due to topological separation in the network—suggests a compounded benefit if multiple gene interventions are applied. Model limitations relate to the binary representation and simplified feedback handling; further refinement could improve biological granularity and predictive scope.

Future Prospects in Network Rejuvenation

Practical implications of these findings include the potential rational design of gene therapies for aging mitigation. The theoretical framework is directly extensible to other tissues or multi-omics data sets. Algorithmic advances, such as multi-state gene representation or full feedback incorporation, could enable more precise mapping of rejuvenation landscapes and facilitate ex vivo or in vivo experimental validation.

Conclusion

This work delivers a robust, computationally tractable approach for quantifying and restoring lost information in aged GRNs using single-gene knock-in perturbations (2601.04016). The model provides exact inference from expression data and accurately predicts restorative targets and the network-wide consequences of genetic interventions. The implications span both theoretical understanding of aging dynamics and practical rejuvenation strategies, with the possibility for significant enhancement upon combinatorial perturbations and algorithmic upgrades.