Most people know creatine as the supplement that helps you lift a little more and recover a little faster. Far fewer know what actually happens inside a muscle or brain cell once creatine arrives there. The standard explanation, that creatine tops up phosphocreatine and helps regenerate ATP, is correct, but it is only the first chapter. Professor Diego Bonilla of DBSS International argues that creatine also acts as a signal, nudging genes, kinases and transport proteins in ways that a single experiment can never fully reveal. In a conference talk on his 2021 paper in the journal Nutrients, he described how his group used bioinformatics to map those signals. This article walks through what he said, what the underlying paper actually found, and how well those findings hold up against the wider research.
Beyond the Energy Story
Bonilla begins by placing his work in the history of sports nutrition. For decades, he explains, researchers evaluated a supplement or an exercise programme by measuring a handful of physiological or metabolic outcomes, one at a time. That approach produced the solid evidence base creatine now enjoys. The International Society of Sports Nutrition’s position stand, for example, summarises hundreds of trials and describes creatine monohydrate as the most studied and effective ergogenic supplement available.
What that approach cannot do, according to Bonilla, is explain how a cell integrates the many changes creatine sets in motion at the same time. That is where he sees the newer “omics” technologies and bioinformatics tools earning their place. Instead of asking whether one protein went up or down, transcriptomics can measure thousands of gene transcripts at once, proteomics can do the same for proteins, and metabolomics for small molecules. He notes that the preceding talk at the same conference, by Dr Ralf Jäger, had covered metabolomics, so he framed his own work as the gene expression side of the same picture.
The paper itself makes the same case in more formal language. Its authors write that individual experiments can only identify targeted regulators chosen in advance, and that a systems biology approach is needed to see the molecular signatures of cellular processes at scale. They describe the creatine kinase and phosphocreatine system not just as an energy buffer but as a dynamic biosensor of the cell’s energetic and mechanical state.
Why the Central Dogma Is Not Enough
To explain why gene expression matters here, Bonilla revisits the central dogma of molecular biology: DNA is transcribed into RNA, and RNA is translated into protein. He points out that this one way street is an oversimplification. Proteins loop back and regulate DNA, most obviously as transcription factors that switch genes on and off. Chemical tags on DNA and on the histone proteins that package it, such as methylation and acetylation, change how accessible a gene is without changing its sequence.
Bonilla adds a detail that shows how fast the field moves: lactylation, a histone modification derived from lactate, was only described a few years before his talk. That claim checks out. In 2019, Zhang and colleagues reported in Nature that lactate can be attached to lysine residues on histones and that this modification directly stimulates gene transcription. They identified 28 lactylation sites on core histones in human and mouse cells. For anyone interested in creatine, this is more than trivia, because creatine supplementation is known to reduce lactate accumulation during hard exercise, and lactate turns out to be a signalling molecule as well as a waste product.
He also insists on not skipping microRNAs, short pieces of RNA that do not code for protein but instead bind to messenger RNA and suppress it. He calls them very important regulators of gene expression, and the mainstream view agrees.
Humans, Mice and Shared Wiring
One of the more surprising parts of the talk is Bonilla’s justification for mixing human and mouse data. He reminds the audience that any two humans are about 99.9 percent identical at the DNA level, and that what makes us different lives in a tiny fraction of the genome. The National Human Genome Research Institute states exactly this: regardless of ethnic or racial self identity, two people are 99.9 percent the same genetically, and small sequence changes account for individual differences in disease risk and drug response.
Bonilla then extends the logic across species. Humans, mice and frogs, he says, share patterns and sections of DNA that evolutionary biology and phylogenetics have mapped in detail. He also mentions that the most distant species can share a surprising fraction of sequence with us, but he quotes a specific figure that I could not verify in any independent source, so it is best treated as an illustration rather than a fact.
The conserved wiring is what makes his method, convergent functional genomics, possible. The published paper explains the trade off clearly: human data increase clinical relevance, animal data increase the ability to detect a signal, and combining them helps separate signal from noise even when the datasets are small. The technique has been used before to study candidate genes in psychiatric disorders, epilepsy and chronic fatigue syndrome, which gives some confidence that it is a recognised approach rather than an improvised one.
Emergent Properties and the Data Behind Them
Before describing the results, Bonilla makes a conceptual point that is worth pausing on. Every complex system, whether a cell, a tissue or a whole body, has what he calls emergent properties: characteristics that arise from the interactions between its components and that are not present in any single component on its own. This is the standard definition used in systems biology, and it explains why he thinks studying one protein in isolation can miss the point.
To study interactions, though, you need large amounts of data, and Bonilla is candid that his group did not generate it themselves. Instead they used public repositories. He names the databases maintained by the United States National Library of Medicine, part of the NIH, and by the European Bioinformatics Institute, and praises the manual curation by biology experts that makes the data reusable. The paper confirms that the team searched the NCBI Gene Expression Omnibus and the ArrayExpress Archive in January 2021, following systematic review guidelines, and even emailed corresponding authors to try to obtain data that had not been deposited.
He is also candid about how little there was to find. After screening, only three datasets met the criteria: one from a human clinical trial and two from mice. Bonilla attributes the human data to Safdar and colleagues, which is correct. That 2008 study in Physiological Genomics gave 12 healthy young men either creatine (a 3 day loading phase at 20 grams per day, then 5 grams per day for a week) or placebo in a crossover design, then profiled gene expression in thigh muscle biopsies. Of the two mouse datasets, one came from brains of animals fed creatine for six months, and the other from fibroblast cells engineered to overexpress the creatine transporter and then exposed to creatine in a dish. Bonilla describes these as one in vivo tissue study and one in vitro study, which is accurate.
The scarcity is itself a finding. The paper’s authors list it as their first limitation: only one human transcriptomic dataset existed in public repositories, and they call for more studies across genomics, transcriptomics, proteomics and metabolomics.
What the Gene Networks Point To
With the datasets in hand, Bonilla explains, the team reran the differential expression analysis to find genes whose activity changed after creatine, then looked for overlap between species. According to the paper, only one gene converged between the human and mouse data, a scaffold protein called PQBP1 involved in transcription, RNA splicing, innate immunity and neuron projection development. The final list for analysis contained 35 genes, almost all from the human muscle study.
The next step was enrichment analysis, which Bonilla describes as aligning the gene list against curated libraries from the literature and ranking the results by statistical significance. He gives credit to the Gene Ontology, an effort to standardise how gene functions are described, and the Gene Ontology Consortium does exactly that: founded in 1998, it maintains a structured, computable vocabulary of gene functions that has become the standard for this kind of analysis.
Bonilla summarises the results as pointing to neural differentiation, regulation of angiogenesis, mammary gland development and DNA regulation. The paper’s table bears him out, with some fine print. The strongest hits among biological processes were mammary gland development, negative regulation of programmed cell death and regulation of angiogenesis, all with adjusted p values below 0.05. Dopaminergic neuron differentiation came next, but at an adjusted p value of 0.055, just outside the usual threshold. The molecular function terms he groups under DNA regulation included microRNA binding and pre messenger RNA binding, again with adjusted values of around 0.05. In plain terms, the analysis suggests creatine affects genes involved in cell survival, differentiation and blood vessel regulation, but the signal is modest and rests on a small gene list.
The upstream analysis is where the picture becomes more familiar to exercise physiologists. Bonilla describes a three layer reconstruction: transcription factors that likely regulate the network, intermediate proteins that connect them, and finally the kinases, the secondary messengers that carry signals from receptors at the cell surface. The paper names the transcription factors, including FOS, SP1, MYC and the EZH2 and SUZ12 complex, all involved in cell survival, proliferation and differentiation. Among the kinases it lists p38 (also called MAPK14), cyclin dependent kinases, casein kinase II, ERK, Akt/PKB and JNK.
Kinases, Myostatin and the Evidence Trail
Bonilla’s central argument is that these computational predictions match what earlier, targeted experiments had already found. He offers three examples.
First, creatine supplementation raises phosphocreatine, and there is experimental evidence that this creatine enriched environment activates Akt, also known as protein kinase B. The best known demonstration is a 2007 study by Deldicque and colleagues in mouse muscle cells. Adding 5 millimolar creatine increased Akt/PKB phosphorylation by about 60 percent, increased phosphorylation of downstream targets involved in protein synthesis, and boosted the fusion of muscle precursor cells and their expression of muscle proteins. Importantly, the researchers showed that mannitol, taurine and beta alanine did not reproduce the effect, which rules out simple cell swelling as the explanation. The paper Bonilla presents also cites human data from Safdar’s trial and other studies showing higher Akt activity after creatine.
Second, Bonilla mentions reports of myostatin reduction in humans doing resistance training with creatine. Myostatin is a protein that limits muscle growth, so lower levels are, in principle, favourable. The evidence here comes mainly from one study. In 2010, Saremi and colleagues assigned 27 healthy young men to a control group, resistance training with placebo, or resistance training with creatine for eight weeks. Training lowered serum myostatin in both training groups, and the creatine group saw a greater decrease. Creatine did not further change GASP-1, a related regulatory protein. It is a small, single trial, and Bonilla is accurate in describing it as a report rather than an established mechanism.
Third, he points to activation of the p38 pathway and other MAP kinases, which regulate cell differentiation. The same Deldicque study found that creatine changed p38 phosphorylation and the nuclear content of its downstream targets, and that blocking p38 prevented differentiation. Bonilla’s paper adds that MAPK activation could be triggered by mechanosensing and osmosensing, that is, by the cell detecting the mechanical and osmotic changes that accompany creatine loading, and notes that Safdar’s human study found increased expression of an osmosensing gene and several MAP kinases after ten days of creatine.
Bonilla’s broader claim is that this agreement between prediction and experiment validates the bioinformatics approach and saves researchers time when they decide which protein to investigate next. That is a reasonable reading, with one caution: the kinase predictions were partly derived from a human dataset that had itself already measured some of these kinases, so the agreement is not fully independent.
Klotho and the Creatine Transporter
The talk then turns to how creatine gets into cells in the first place, through a transporter protein called SLC6A8 or CRT. Bonilla mentions that Sergej Ostojic had recently published on Klotho, a protein linked to ageing and lifespan, and that experiments in oocytes using electrical recording techniques showed Klotho increases and stabilises the creatine transporter.
The science is correct, but the credit needs adjusting. The oocyte experiments were performed by Florian Lang’s group at the University of Tübingen and published in 2014. They injected frog oocytes with the genetic instructions for the creatine transporter, with or without Klotho, and measured creatine induced current using two electrode voltage clamp. Coexpressing Klotho significantly increased the current, delayed its decline when new transporter insertion was blocked, and the effect was reversed by an inhibitor of Klotho’s beta glucuronidase activity. Their conclusion was that Klotho upregulates the transporter by stabilising it in the cell membrane. Ostojic’s 2021 papers in Frontiers in Nutrition are opinion pieces that summarise this work and propose that boosting Klotho, possibly through diet, might improve creatine uptake into the brain. Bonilla’s own paper cites the Tübingen study correctly and notes that Klotho has also been linked to inhibition of glycolysis and anticancer activity, which it says deserves more research.
Why does this matter? Muscle takes up supplemental creatine readily, but the brain does not, and creatine transporter deficiency causes intellectual disability and seizures in the people born with it. Any factor that stabilises the transporter is therefore of real interest, even if the current evidence comes from frog eggs rather than people.
Why Creatine Uptake Switches Itself Off
Bonilla describes a second layer of transporter regulation that runs in the opposite direction. Creatine accumulation, he explains, is concentration dependent: once cellular creatine climbs to a ceiling, roughly 150 to 160 millimoles per kilogram of dry muscle mass, uptake slows. That figure agrees with the ISSN position stand, which puts the average total creatine pool at about 120 millimoles per kilogram of dry muscle and the upper storage limit at around 160.
He then identifies a mechanism that could enforce the ceiling. A gene called TXNIP, which encodes thioredoxin interacting protein, was upregulated in one of the analysed models, and it sits on a pathway that can pull the creatine transporter off the cell surface by endocytosis, effectively reducing further uptake. The original evidence comes from Zervou and colleagues at Oxford in 2013. In cells overexpressing the transporter, creatine uptake plateaued after three hours of exposure to 5 millimolar creatine. A genome wide screen found TXNIP to be the only gene significantly upregulated, by 46 percent, and the response was specific to creatine rather than to osmotic stress, since taurine did not trigger it. Silencing TXNIP kept uptake at normal levels and prevented the transporter’s messenger RNA from falling. The same team found that mouse hearts overexpressing the transporter had higher TXNIP, while creatine deficient hearts had about 40 percent less.
Bonilla describes this as an animal model, which is a slight simplification: the core experiments were in a mouse cell line, with supporting data from mouse hearts. His paper builds on it to propose two complementary networks: a kinase driven mechanism that moves transporters to the membrane when creatine first rises, and a ubiquitin driven mechanism, involving TXNIP and the ubiquitin ligase Nedd4-2, that internalises transporters when creatine becomes excessive. The authors label this a cautious suggestion, and Bonilla is honest in the talk that several arrows in the pathway figure carry question marks. This is a hypothesis assembled from real pieces of evidence, not a mechanism anyone has yet watched operate in a supplemented human.
The MicroRNA Question
The final substantive part of the talk concerns microRNAs. Bonilla reports that the team asked whether any microRNAs were predicted to regulate the gene network, and that the ones that surfaced have been linked in other research to cancer regulation, immune responses and neuroprotection. He suggests that hidden and intricate molecules, especially microRNAs, might turn out to regulate the physiological effects attributed to creatine.
This is the most speculative claim in the talk, and the paper treats it that way. The microRNA predictions came from a database of experimentally validated microRNA to messenger RNA interactions, but the adjusted p values for the top hits ranged from about 0.09 to 0.25, so none met conventional significance. The authors say the analysis revealed for the first time that creatine might impact certain microRNAs, and immediately add that more experimental evidence is needed. Bonilla’s phrasing, that there are perhaps hidden regulators, is appropriately hedged.
The cancer connection deserves its own honest treatment, because the wider evidence is contested. On one side, a 2021 study in Cell Metabolism reported that creatine, whether from diet or synthesised in the body, promoted colorectal and breast cancer metastasis in mouse models and shortened survival, acting through a signalling protein called Smad2/3. Its authors called for caution about dietary creatine in cancer patients. On the other side, a 2019 study in the Journal of Experimental Medicine found that tumour fighting CD8 T cells depend on creatine uptake for energy, that mice lacking the transporter mounted weaker antitumour responses, and that creatine supplementation suppressed tumour growth and worked synergistically with checkpoint inhibitor immunotherapy. Bonilla’s paper cites both lines of evidence and notes that low creatine and reduced creatine kinase are associated with progression of some sarcomas. Both studies were in mice. Nobody yet knows which effect, if either, dominates in people, and a commentary by leading creatine researchers disputed whether the metastasis paper justified warnings about human health. The responsible summary is that the microRNA hints are intriguing and the cancer question is unresolved.
The neuroprotection thread is similarly mixed. A 2024 meta analysis of 16 randomised trials found small benefits of creatine on memory, attention time and processing speed, but no effect on overall cognitive function or executive function, and it has been criticised for counting multiple tests from the same participants as independent results. In November 2024, the European Food Safety Authority reviewed 21 human trials for a proposed health claim and concluded that a cause and effect relationship between creatine and improved cognition had not been established, noting that acute effects on working memory appeared only at 20 grams per day and not at lower or sustained doses. The European Commission subsequently refused the claim. Creatine may well support brain energy metabolism, and Bonilla’s gene network points that way, but the human outcome data do not yet justify confident claims.
What This Means for Anyone Taking Creatine
Bonilla closes by looking forward, suggesting that science is evolving toward models, including artificial intelligence, that could predict how biological systems behave. He illustrates the idea with fractals, showing how simple repeated rules generate complex structures, and compares that to how systems biology views cells: complex behaviour emerging from many simple, repeated interactions. This is a philosophical framing rather than a testable claim, and he presents it as such.
For a reader deciding whether any of this changes how they use creatine, the practical takeaways are modest and honest. The strong, well replicated evidence remains what it was: creatine monohydrate raises muscle creatine stores, improves performance in high intensity exercise and, combined with resistance training, supports gains in strength and lean mass. Bonilla’s work does not add a new benefit. What it adds is a plausible molecular map of how those benefits might arise, connecting the energy story to signalling through Akt, p38 and related kinases, and explaining why the body limits creatine uptake once stores are full.
The map is early stage. It rests on three datasets, one of them human, and its most interesting proposals, the two stage transporter regulation and the microRNA connections, remain computational predictions awaiting validation. Bonilla says so himself, and the published paper lists the limitations plainly. That transparency is a good sign. It means the next decade of creatine research has a clear set of hypotheses to test, and readers have a clear sense of which claims are settled and which are still being worked out.