Housekeeping genes (HKGs) serve as foundational internal standards for normalizing gene expression profiles in quantitative molecular profiling assays, most notably reverse transcription quantitative real-time polymerase chain reaction (RT-qPCR) and RNA sequencing (RNA-seq)1,2. Accurate transcript normalization remains indispensable for mitigating confounding technical variability introduced by fluctuations in initial RNA yield, RNA integrity, enzymatic reverse transcription efficiency, and exponential PCR amplification kinetics2,3. Historically, classical reference transcripts—such as glyceraldehyde-3-phosphate dehydrogenase (GAPDH), beta-actin (ACTB), 18S ribosomal RNA (18S rRNA), and hypoxanthine phosphoribosyltransferase 1 (HPRT1)—were widely assumed to be constitutively and uniformly expressed across distinct cell lineages, leading to their routine application without prior experimental validation1,4.

Mesenchymal stem/stromal cells (MSCs) have emerged as an essential platform in regenerative medicine, immunomodulatory therapy, and tissue engineering, underpinning extensive preclinical investigations and clinical trials5. MSCs can be isolated from a diverse spectrum of anatomical sources, including bone marrow, adipose tissue, umbilical cord matrix, placenta, and other perinatal tissues, with each compartment imparting distinct biological, phenotypic, and transcriptomic properties5,6,7,8,9. This pronounced innate heterogeneity, compounded by the extensive in vitro ex vivo expansion required to generate clinically meaningful cell dosages, poses substantial hurdles for standard transcript normalization strategies10,11.

A burgeoning body of empirical evidence underscores that the transcript abundance of conventional HKGs in MSCs is exquisitely sensitive to culture context. Factors including anatomical tissue derivation, inter-donor variability, cumulative passage number, cellular seeding density, multilineage differentiation commitment, ambient hypoxia, inflammatory activation, and mechanical loading have all been shown to disrupt HKG expression stability12,13. Crucially, classical reference genes such as GAPDH and ACTB actively participate in core glycolytic and cytoskeletal networks, rendering their transcription highly vulnerable to common experimental perturbations14. Consequently, the uncritical adoption of single unvalidated HKGs in MSC research has frequently precipitated flawed normalization, compromised data reproducibility, and yielded contradictory biological interpretations15 (Figure 1).

The ramifications of reference gene instability are particularly acute in translational and clinical manufacturing settings. Clinical MSC therapies demand extensive ex vivo expansion to attain targeted therapeutic doses, an intervention inextricably linked to replicative senescence, profound cytoskeletal remodeling, and genome-wide transcriptomic rewiring16. Furthermore, the regulatory transition from fetal bovine serum (FBS)-supplemented culture media toward chemically defined serum-free (SFM) or xeno-free (XFM) formulations—mandated by Good Manufacturing Practice (GMP) compliance—imposes additional transcriptional shifts that profoundly alter baseline HKG stability13. In these bioprocessing environments, selecting an inappropriate reference gene can introduce false readouts in quantitative potency assays, distorting the perceived efficacy and release criteria of critical immunomodulatory and differentiation pathways13,16.

To address these limitations, a variety of statistical and computational validation frameworks have been established, including geNorm, NormFinder, and BestKeeper, alongside integrated consensus tools such as RefFinder17,18. Because reference gene stability is inherently context-dependent, no single universal HKG can perform reliably across all MSC experimental conditions13. Instead, genes participating in basal transcription initiation or proteostatic turnover—such as TATA-box binding protein (TBP), ribosomal protein L13a (RPL13A), peptidylprolyl isomerase A (PPIA), and tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein zeta (YWHAZ)—frequently demonstrate superior stability within tailored biological contexts19,20,21. In recent years, these accumulated insights have catalyzed a fundamental paradigm shift toward multi-gene normalization strategies and rigorous pre-validation, principles formally codified in the updated Minimum Information for Publication of Quantitative Real-Time PCR Experiments (MIQE 2.0) guidelines22.

This narrative review synthesizes contemporary experimental and computational findings examining HKG expression variability in MSCs across diverse biological and technical landscapes, including multilineage differentiation, hypoxia, inflammatory priming, mechanical stimulation, and GMP-compliant biomanufacturing. We critically evaluate established validation algorithms, explore emerging frontiers—such as RNA-seq–assisted candidate preselection, small non-coding RNA normalization, and machine learning–driven discovery models15—and provide a structured series of practical recommendations to maximize experimental validity, methodological reproducibility, and translational reliability in MSC gene expression profiling.

This narrative review was informed by a systematic literature search conducted across PubMed, Web of Science, and Google Scholar, capturing peer-reviewed studies published up to 2025. Search strings incorporated combinations of the following terms: "mesenchymal stem cells", "mesenchymal stromal cells", "housekeeping genes", "reference genes", "gene expression normalization", "RT-qPCR", "RNA-seq", "hypoxia", "multilineage differentiation", "inflammatory priming", "serum-free media", "xeno-free culture", and "GMP biomanufacturing". Articles evaluating reference gene stability in MSCs under defined biological or physical perturbations were critically appraised. Given the broad methodological diversity across MSC sources and experimental systems, findings were qualitatively synthesized and structured thematically by experimental context. Representative empirical studies were prioritized to illustrate recurrent patterns of transcript stability and instability rather than to provide an exhaustive enumeration.

Classical molecular biology traditionally defines housekeeping genes as an essential cohort of constitutively expressed genes required for basal cellular maintenance, encompassing central metabolic processing, cytoskeletal integrity, ribosome biogenesis, and basic transcriptional upkeep23. Historically, these genes were assumed to exhibit uniform transcription across diverse cell lineages and physiological states, independent of external conditions23. However, modern high-throughput transcriptomic profiling and rigorous empirical validation have revealed this paradigm to be fundamentally oversimplified. Even under baseline homeostatic conditions, conventional candidate HKGs display notable transcriptional fluctuations across distinct cell types, developmental stages, and microenvironmental niches, undermining their validity as invariant internal standards23. Consequently, candidate reference genes must no longer be viewed as absolute constants, but rather as empirical candidates whose expression stability must be experimentally verified in each specific biological system10.

This conceptual limitation is uniquely pronounced in MSCs. Beyond their characteristic trilineage differentiation capacity, MSCs possess exceptional phenotypic plasticity and environmental sensitivity. During osteogenic, adipogenic, and chondrogenic differentiation, MSCs undergo sweeping transcriptional reprogramming that reconfigures wide functional networks rather than merely altering isolated lineage markers24,25. These lineage specifications are intrinsically coupled to extensive metabolic transitions, such as switching from aerobic glycolysis to mitochondrial oxidative phosphorylation (or vice versa), which causally drives the upregulation or downregulation of metabolic genes routinely employed as reference standards24,25. Furthermore, culturing MSCs under physiological or pathophysiological hypoxia—frequently utilized to sustain stemness, promote expansion, or mimic the native in vivo niche—stabilizes hypoxia-inducible factor 1-alpha (HIF-1α) and transactivates glycolytic target genes, triggering substantial expression shifts in metabolic HKGs such as GAPDH and PGK126.

In addition to metabolic rewiring, MSCs mounting immunomodulatory responses exhibit pronounced transcriptional plasticity. Exposure to proinflammatory cytokines, such as interferon-gamma (IFN-γ) or tumor necrosis factor-alpha (TNF-α), initiates broad stress-responsive and immune-regulatory cascades27,28. Concurrently, biophysical cues—such as substrate stiffness, cyclic tensile stretch, and three-dimensional (3D) scaffold architecture—activate cytoplasmic mechanotransduction pathways and induce extensive cytoskeletal remodeling28,29. Because structural genes such as ACTB and TUBA1B are central components of the actin and microtubule machinery, their transcription fluctuates dynamically during physical perturbation. Moreover, baseline transcriptomic heterogeneity stemming from donor demographics, tissue origin, and isolation procedures introduces substantial inter-donor variability even prior to experimental stimulation27.

Collectively, these unique attributes position MSCs as a biologically sensitive system in which subtle experimental perturbations elicit pronounced transcriptional responses. Secondary variables that exert minimal impact on immortalized cell lines—such as subculture passage number, cell confluency, pericellular oxygen tension, and basal medium formulation—profoundly modulate the MSC transcriptome10,27. From a conceptual standpoint, reference gene stability in MSC models cannot be assumed a priori but must be rigorously demonstrated within each designated experimental framework.

Mechanistic insight into reference gene instability can be obtained by examining the functional roles of candidate genes within MSC biology. Instability is rarely a stochastic phenomenon; rather, it directly mirrors the functional involvement of candidate genes in the cellular pathways undergoing experimental remodeling26,30. Cytoskeletal genes (e.g., ACTB, TUBA1B) fluctuate extensively because actin dynamics and cell morphology are tightly linked to lineage commitment and mechanotransduction28,29. Ribosomal protein genes (e.g., RPL13A, RPLP0) frequently exhibit robust stability under moderate conditions but can drift significantly in response to altered proliferation rates, fluctuating translational demand, or metabolic stress31. Transcripts involved in basal transcriptional initiation (e.g., TBP, POLR2A) generally display superior stability because they regulate global RNA polymerase II assembly rather than stimulus-specific responses4,30. Similarly, molecular chaperones and proteostatic regulators (e.g., PPIA, HSP90AA1, UBC) often maintain stable expression during standard expansion and mild physical stimulation, although they may fluctuate under severe cellular stress, proteasomal saturation, or replicative senescence32,33.

This biological interconnectedness introduces a major hazard in reference gene selection: functional co-regulation. Evaluating multiple candidate genes belonging to identical metabolic or structural pathways can yield artificially low pairwise variation scores, generating misleadingly stable normalization factors that merely mirror shared biological regulation. As detailed in Table 1, candidate panels must deliberately integrate genes from disparate, non-overlapping functional categories. Given the intrinsic plasticity of MSCs, an ideal reference gene represents an operational definition rather than a universal transcript. Endogenous reference controls should possess low intra- and inter-group transcriptional variance, moderate baseline expression (avoiding extreme saturation or low-abundance dropout), consistent amplification efficiency (90–110%), and no mechanistic involvement in the experimental conditions under study. Because no single gene completely fulfills these criteria across all MSC platforms, contemporary consensus strongly advocates a multi-gene normalization paradigm: screening a diverse set of candidate genes and applying the geometric mean of two or three validated, stably expressed transcripts (e.g., TBP, RPL13A, PPIA, YWHAZ) (Table 1).

To quantify reference gene stability rigorously, researchers employ an array of statistical algorithms alongside high-throughput transcriptomic profiling tools (RT-qPCR, RNA-seq, and microarrays). For instance, Ong et al. (2023) established an RNA-seq-guided selection pipeline in human AD-MSCs: candidate genes from literature databases were cross-referenced against whole-transcriptome RNA-seq profiles of normoxic and hypoxic cells. Candidates exhibiting minimal biological variance (low CV and log₂ fold-change near zero) were shortlisted, validated by RT-qPCR, and evaluated using geNorm, NormFinder, and BestKeeper to pinpoint the most resilient reference set43.

Algorithmic platforms such as geNorm, NormFinder, and BestKeeper assess relative stability by evaluating candidate gene expression across experimental cohorts (Table 2). To streamline multi-algorithm comparisons, RefFinder integrates geNorm, NormFinder, BestKeeper, and the comparative ΔC_t method into a unified web-accessible platform17. RefFinder assigns an overall stability ranking based on the geometric mean of the individual algorithm ranks, thereby minimizing single-method bias. Nevertheless, RefFinder possesses intrinsic constraints: it lacks customizable weighting parameters, cannot perform advanced multivariate modeling, handles large-scale genomic datasets slowly, and its consensus metric may obscure nuanced, context-dependent variances.

An enhanced geNorm module has been incorporated into Biogazelle's qbase+ analytical software suite54. Compared to the original macro-based geNorm program, qbase+ offers accelerated processing speeds, automated QC reporting, robust handling of missing values, identification of the single most stable reference transcript (rather than only the final stable gene pair), and cross-platform compatibility across Windows, macOS, and Linux systems.

Despite their widespread adoption, geNorm and BestKeeper rely predominantly on pairwise correlation and similarity metrics. As a consequence, functionally co-regulated genes may be falsely categorized as highly stable if they exhibit parallel expression shifts in response to a common perturbation. For example, transcripts involved in shared metabolic pathways may display strong mutual correlation despite both being modulated by experimental treatments. NormFinder partially circumvents this limitation through an ANOVA-based model that separates overall expression variance into intra-group and inter-group components. Nonetheless, if a gene displays uniform, low-variance changes across all experimental groups, NormFinder may still classify it as stable despite biologically meaningful modulation. Therefore, mathematical stability rankings must always be cross-referenced against biological plausibility, ensuring that candidate panels are populated by genes originating from distinct, non-overlapping functional categories to preclude co-regulation artifacts.

Reference gene stability directly dictates the precision, validity, and biological interpretability of gene expression profiling. Normalizing datasets against unverified, unstable internal controls introduces severe mathematical bias and systematic error (Figure 2)59. The updated MIQE 2.0 guidelines explicitly assert that qPCR datasets normalized against unvalidated reference genes cannot be considered analytically valid22.

Across diverse MSC sources (adipose tissue, bone marrow, umbilical cord matrix), conventional HKGs such as GAPDH, ACTB, and PGK1 frequently display pronounced expression instability, whereas non-traditional candidates such as PPIA, YWHAZ, TBP, RPL13A, RPLP0, 18S rRNA, or RRP1 provide superior stability in specific biological contexts (Table 1). External factors—including tissue harvest site12, donor characteristics10, pericellular hypoxia43, culture media shifts (FBS to SFM/XFM)13, inflammatory activation, serum starvation48, and 3D scaffold encapsulation50—profoundly alter basal HKG transcription. Choosing an inappropriate reference gene can artificially amplify or dampen apparent expression differences in target genes. In MSC differentiation assays, sub-optimal reference gene selection can entirely mask lineage-specific induction markers or, conversely, create artifactual differences in differentiation efficiency60. In severe instances, improper normalization can invert biological conclusions. For example, Casas et al. (2022) demonstrated that standardizing stroke-related NOX4 expression against different HKGs generated contradictory conclusions regarding whether the target transcript was significantly upregulated or downregulated59. In cytokine-primed MSCs, normalizing against ACTB completely obscured the upregulation of the key immunosuppressive enzyme indoleamine 2,3-dioxygenase (IDO), whereas stable reference genes revealed robust induction. Such discrepancies are especially detrimental in quantitative potency assays, where misquantifying immunomodulatory transcripts can compromise therapeutic efficacy and product characterization.

At the bioprocessing and translational scale, reference gene instability increases experimental variance, generating false-positive or false-negative readouts that directly impact cell therapy release criteria13. Ragni et al. (2024) demonstrated that when comparing AD-MSCs expanded in SFM versus FBS, unstable ACTB expression falsely indicated an upregulation of the proinflammatory chemokine CCL5 in SFM products; conversely, normalization against the stable reference gene TBP demonstrated that CCL5 transcription was unchanged or slightly downregulated13. Similarly, during immunological evaluations of MSCs encapsulated in 3D biomaterials or subjected to inflammatory challenge, ACTB normalization severely skewed perceived immune-modulating potency50,61.

In commercial cell manufacturing and clinical translation, MSC products must adhere strictly to GMP regulations. Regulatory agencies, including the US Food and Drug Administration (FDA) and the European Medicines Agency (EMA), mandate validated quantitative potency testing prior to lot release. Within these release assays, reference genes represent an indispensable component of the analytical validation framework, ensuring assay accuracy, linearity, and precision62. Normalizing against unvalidated reference genes can cause release metrics to fall artificially outside predefined acceptance ranges. This introduces substantial operational risks: high-quality cell batches may be erroneously rejected due to normalization artifacts—resulting in substantial economic loss—or compromised cell batches may be mistakenly released, posing direct clinical risks to patients.

The selection of housekeeping genes in mesenchymal stem cell research cannot be addressed through simple, one-size-fits-all assumptions. MSCs display remarkable biological plasticity, undergoing profound transcriptomic rewiring during multilineage differentiation, metabolic reprogramming, hypoxic adaptation, inflammatory licensing, mechanical stimulation, and extended ex vivo expansion. This intrinsic responsiveness systematically disrupts the transcriptional stability of traditional reference genes such as GAPDH and ACTB. Empirical evidence firmly establishes that no universal reference gene exists across all MSC experimental conditions. Accurate, reproducible normalization requires a context-specific validation strategy that merges rigorous mathematical algorithms with sound biological rationale. Failing to validate reference genes risks generating distorted expression metrics, false-positive or false-negative findings, and contradictory conclusions in potency and translational assays. Integrating multi-gene normalization panels, adhering strictly to MIQE 2.0 reporting standards, and leveraging RNA-seq preselection and machine learning models provide a robust methodological foundation to ensure experimental rigor, reproducibility, and therapeutic translatability in MSC biology.

18S rRNA: 18S ribosomal RNA; ACTB: Beta-actin; AD-MSCs / ASCs: Adipose-derived mesenchymal stem/stromal cells; AI: Artificial intelligence; B2M: Beta-2-microglobulin; BM-MSCs / BMSCs: Bone marrow–derived mesenchymal stem/stromal cells; C_q / C_t: Quantification cycle / Cycle threshold; CV: Coefficient of variation; ΔC_t: Delta cycle threshold; ECM: Extracellular matrix; EF1A: Eukaryotic translation elongation factor 1 alpha 1; EMA: European Medicines Agency; EVs: Extracellular vesicles; FBS: Fetal bovine serum; FDA: Food and Drug Administration; GAPDH: Glyceraldehyde-3-phosphate dehydrogenase; GMP: Good Manufacturing Practice; hAM-MSCs / hAMSCs: Human amniotic membrane–derived mesenchymal stem/stromal cells; HIF-1α: Hypoxia-inducible factor 1 alpha; HKGs: Housekeeping genes; hPL: Human platelet lysate; HPRT1: Hypoxanthine phosphoribosyltransferase 1; HRE: Hypoxia response element; HSP90AA1: Heat shock protein 90 alpha family class A member 1; IDO: Indoleamine 2,3-dioxygenase; IFN-γ: Interferon gamma; IL-1β: Interleukin 1 beta; ISCT: International Society for Cell & Gene Therapy; miRNA: MicroRNA; MIQE: Minimum Information for Publication of Quantitative Real-Time PCR Experiments; ML: Machine learning; MSCs: Mesenchymal stem/stromal cells; NRTC: No-reverse-transcription control; NTC: No-template control; PGK1: Phosphoglycerate kinase 1; POLR2A: RNA polymerase II subunit A; PPIA: Peptidylprolyl isomerase A (Cyclophilin A); QC: Quality control; RIN: RNA integrity number; RNA-seq: RNA sequencing; RPL13A: Ribosomal protein L13a; RPLP0: Ribosomal protein lateral stalk subunit P0; RPS13: Ribosomal protein S13; RPS17: Ribosomal protein S17; RRP1: Ribosomal RNA processing 1; RT-qPCR: Reverse transcription quantitative real-time polymerase chain reaction; SASPs: Senescence-associated secretory phenotypes; SD: Standard deviation; SFM: Serum-free medium; SOP: Standard operating procedure; SV: Stability value; TBP: TATA-box binding protein; TNF-α: Tumor necrosis factor alpha; TUBA1B: Tubulin alpha 1b; UBB: Ubiquitin B; UBC: Ubiquitin C; UC-MSCs: Umbilical cord–derived mesenchymal stem/stromal cells; XFM: Xeno-free medium; YWHAZ: Tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein zeta.

None.

T.K.N. Nguyen, N.T.T. Nguyen, and A.M. Nguyen conducted the literature investigation, performed data curation, and prepared the original manuscript draft. K.V.T. Nguyen and K.D. Bui supported formal comparative analyses and contributed to critical manuscript editing. P.D. Huynh conceptualized and supervised the study, oversaw methodology and project administration, and critically revised and finalized the manuscript. All authors read and approved the final manuscript.

This research was funded by Vietnam National University Ho Chi Minh City (VNU-HCM) under grant number C2026-18-27.

None.

None.

None.

During the preparation of this manuscript, the authors utilized ChatGPT and Grammarly solely for English language editing and stylistic refinement. Following the use of these tools, the authors thoroughly reviewed, validated, and edited the text to ensure factual accuracy and scientific rigor. The authors take full institutional and intellectual responsibility for the contents of this publication, and all final interpretations and conclusions were determined entirely by the authors.

The authors declare that they have no competing interests.