Non-coding small RNAs buffer protein interactions to prevent oncogenic aggregation: structural dampening of aberrant PPIs by RNA

ABSTRACT

Non-coding RNAs (ncRNAs) modulate protein–protein interactions (PPIs) by shaping the structural context in which binding occurs, rather than acting as direct inhibitors or enhancers. Using an integrative framework combining catRAPID RNA–protein interaction prediction and AlphaFold3-based structural modelling, we analysed RNA-dependent modulation of interaction states across physiological and oncogenic protein complexes. At the network level, physiological PPIs exhibit high shared ncRNA buffering capacity, whereas oncogenic interactions are characterized by reduced or absent RNA overlap. AlphaFold3 modelling of mutant IDH1/2 complexes illustrates how loss of RNA buffering permits excessive stabilization of enzyme-associated interfaces, reflected by directional changes in buried surface area (ΔBSA) and contact heterogeneity.

HIGHLIGHTS

• Non-coding RNAs shape protein–protein interaction states by acting as contextual structural buffers rather than direct inhibitors.

• AlphaFold3 modelling reveals RNA-dependent modulation of interface geometry, including buried surface area (ΔBSA) and contact heterogeneity.

• Loss or exclusion of RNA buffering permits excessive stabilization and pathological fixation of oncogenic protein complexes.

• Integrative AlphaFold3 and catRAPID analyses support an RNA-mediated framework that delineates physiological reversibility from oncogenic aggregation.

Introduction

Protein–protein interactions (PPIs) underpin virtually every biological pathway, orchestrating cellular signalling, structural integrity, and metabolic regulation [1]. The modulation of their strength is essential for transient signalling and homoeostasis, where finely tuned interaction kinetics prevent both constitutive activation and aggregation. Traditionally, binding affinity has been quantified using dissociation constants (Kd); however, in vivo environments introduce complex effects of molecular crowding, hydration shells, and allosteric interference that static biochemical assays fail to capture. Such contextual modulation suggests that additional molecular layers – beyond direct protein interactions – may act as dynamic regulators of binding equilibria [2,3]. Small non-coding RNAs (sncRNAs), particularly those incorporated into RNA–protein complexes such as ribonucleoproteins and snoRNPs [4], have emerged as multifaceted modulators of cellular function and disease. While their canonical role involves post-transcriptional regulation, increasing evidence links sncRNAs to the structural organization of the proteome. Interestingly, our earlier work [5] identified physical associations between the HPV16 E7 oncoprotein and specific small RNAs – a phenomenon rarely observed in low-risk viral strains. This finding raises the possibility that small RNAs not only regulate gene expression but also modulate the physical interaction landscape of proteins. From a biophysical standpoint, homologous or repetitive RNA sequences exhibit strong self-association via base pairing and π-stacking, forming viscous or networked molecular buffers [6]. We hypothesize that such RNAs may influence PPI dynamics by acting as structural dampeners – interfering with aberrant or overly stable complexes through multiple mechanisms. These include (i) steric hindrance at the binding interface, (ii) local charge redistribution that perturbs electrostatic complementarity, and (iii) allosteric alteration of domain flexibility or solvent exposure. In structural terms, RNA binding could transiently reduce the buried surface area (BSA) of critical protein interfaces, thereby attenuating pathological aggregation while maintaining reversible physiological contacts [7]. In this study, we integrate AlphaFold3 structural modelling [8] and catRAPID RNA–protein interaction predictions [9] to explore this concept of RNA-mediated structural buffering. By integrating network-level ncRNA commonality analyses across oncogenic complexes, including Rb1–E7 [10], Kras–Raf1 [11] and by quantifying ΔBSA and interface contact variance in a representative structural case study, we investigate how ncRNAs may modulate binding strength and specificity. This structural and RNA-interference-based stratification parallels the transcriptome-based molecular subclasses described by Hoshida et al. [12], suggesting that physical interaction metrics may reflect similar biological divergence. We further propose that pathogenic mutations disrupt shared ncRNA interaction motifs, weakening this buffering capacity and promoting oncogenic aggregation. We further propose that pathogenic mutations disrupt shared ncRNA interaction motifs, weakening this buffering capacity and promoting oncogenic aggregation. This framework extends the traditional view of protein regulation by introducing an RNA-driven layer of structural homeostasis within cellular networks. Here, IDH1/2 is used as an orthogonal case study to demonstrate substrate-linked state bias under RNA buffering, complementing the PPI-centred analyses presented below.

Methods

Prediction of ncRNA–protein interactions

Candidate non-coding small RNAs (ncRNAs) that interact with proteins were predicted using catRAPID v2.1 (Istituto Italiano di Tecnologia) [9]. Amino acids of protein full sequences were retrieved from UniProt and partial sequences used are followed.

Wild-type

IDH1: GTVFREAIICKNIPRLVSGWVKPIIIGRHAYGDQYRATDFVVPGPGKVEMTY

Mutant IDH1 (R132H):

GTVFREAIICKNIPRLVSGWVKPIIIGHHAYGDQYRATDFVVPGPGKVEMTY

Wild-type IDH2:

REPIICKNIPRLVPGWTKPITIGRHAHGDQYKATDFVADRAGTFKMVFTPKD

Mutant IDH2 (R172K):

REPIICKNIPRLVPGWTKPITIGKHAHGDQYKATDFVADRAGTFKMVFTPKD

Kras:

YKLVVVGAGGVGKSALTIQLIQNHFVDEYDPTIEDSYRKQVVIDGETCLLD

kras(G12C):

YKLVVVGAGCGVGKSALTIQLIQNHFVDEYDPTIEDSYRKQVVIDGETCLLD

raf1:

TDPSKTSNTIRVFLPNKQRTVVNVRNGMSLHDCLMKALKVRGLQPECCAVFRLLHEHKGKKARLDWNTDAASLIGE

Free-energy modeling of post-transcriptional interferenc

Predicted hybridization energies (ΔG) and minimum free-energy configurations were used to infer probable binding sites that might contribute to post-transcriptional interference.

AlphaFold 3 modelling and structural prediction

The three-dimensional structures of wild-type and oncogenic mutant IDH1 and IDH2 dimers were predicted using AlphaFold 3 (6, DeepMind, 2024). The models were analysed as a structural case study to assess mutation-associated changes in interface geometry. Because AlphaFold3 generates ranked prediction instances rather than independent experimental replicates, structural metrics are interpreted descriptively, emphasizing directionally consistent trends rather than atomic-level accuracy.

Interface and buried-surface-area (BSA) analysis

AlphaFold3-predicted IDH1 and IDH2 complexes were analysed as a structural case study using UCSF ChimeraX v1.6.2. Solvent-excluded surfaces were computed with a 1.4 Å probe radius to obtain interface buried surface area (BSA) and non-contact surface ratio (NCSR). Because AlphaFold3 generates ranked prediction instances rather than independent experimental replicates, interface metrics are interpreted descriptively, emphasizing directionally consistent trends across models rather than inferential statistics.

Measurement of antigen–antibody interactions

To evaluate the influence of RNA composition on antigen–antibody (Ag–Ab) binding, we compared binding efficiency under two conditions: one enriched for a single microRNA species (miR-6089), and another containing heterogeneous endogenous small RNAs. HEK 293 cells (2 × 10⁵) were transfected with either a plasmid encoding human miR-6089 or a control empty vector, using a standard lipid-based transfection protocol (Lipofectamine; Thermo Fisher). Cells were cultured overnight post-transfection, and small RNAs were extracted using the E.Z.N.A. miRNA Isolation Kit (Omega Bio-Tek). The miR-6089 expression cassette was driven by a human U6 promoter, a canonical RNA polymerase III promoter commonly used for shRNA and synthetic miRNA expression. The mature miR-6089 sequence was placed immediately downstream of the U6 promoter, and successful expression was confirmed by gel electrophoresis. For the antigen–antibody binding assay, 0.1 μg of purified small RNA was added to wells of a 96-well plate pre-coated with human IgG. FITC-conjugated anti-human IgG was then added to each well. After incubation, wells were washed three times with PBS to remove unbound material. Fluorescence intensity was measured using a plate reader (values of each five samples were averaged) with excitation at 489 nm and emission at 520 nm.

Results

Venn diagram of representative physiological and oncogenic PPIs

We first assessed ncRNA overlap by predicted small RNAs using CatRAPID among physiological protein–protein interactions (PPIs), including Rb1-E2F1, p53-MDM2, CDK2-CycE (Figure 1(A)). They show high ratio of overlap and also close clustalw dnd as shown Figure 2. They were performed using full sequence. To more precisely delineate the RNA-binding specificity associated with the Rb–E7 interface, we transitioned from full-length protein sequences to focused domain regions. Initial catRAPID analysis using the entire Rb and HPV16-E7 sequences revealed substantial overlap in predicted RNA partners (460 shared targets), consistent with a broadly similar physicochemical RNA-binding profile.

Figure 1.

Qualitative illustration of ncRNA commonality across physiological and oncogenic protein–protein interactions.

(A) Venn diagrams illustrate representative patterns of small non-coding RNA (ncRNA) commonality for direct physiological protein–protein interactions (PPIs), including Rb1–E2F, p53–MDM2, and CDK2–Cyclin E. Numbers indicate the counts of predicted shared and unique ncRNAs based on catRAPID analysis using full-length protein sequences. These physiological PPIs exhibit substantial ncRNA overlap, consistent with broadly shared RNA-buffering environments associated with reversible and regulated interactions. (B) Venn diagrams for the oncogenic Rb–HPV16 E7 interaction illustrate the dependence of ncRNA commonality on sequence scope. When full-length sequences are analysed (left), substantial overlap is observed, reflecting network-level RNA association. In contrast, restriction to interface-proximal sequence regions (Rb residues 720–770 and E7 residues 20–70; right) results in minimal or absent ncRNA overlap, indicating localized exclusion of RNA buffering at a compact, high-affinity oncogenic interface. These panels are presented as qualitative indicators rather than quantitative measures. (C) A representative example of another oncogenic interaction network involving KRAS and RAF1 is shown to illustrate a similar trend. While whole-sequence analysis reveals shared ncRNA associations among wild-type components, restriction to mutation-associated or interface-focused regions (e.g. KRAS mutant versus wild-type contexts) markedly reduces ncRNA commonality. This pattern supports the generality of interface optimization-associated RNA exclusion across oncogenic PPIs.

Across all panels, ncRNA commonality values are used to illustrate context-dependent RNA-buffering tendencies, recognizing that absolute overlap values depend on sequence scope and computational thresholds. These Venn diagrams therefore complement, rather than replace, structural case studies based on interface geometry and non-contact surface ratio (NCSR).

Figure 2.

Interaction-focused sequence similarity highlights functional convergence rather than evolutionary relatedness.

This figure shows a distance tree generated from interaction-relevant sequence features of selected host and viral proteins, including p53, MDM2, Rb1, CDK2, Cyclin E, adenoviral E1A, and HPV16 E7. Distances reflect similarity within sequence regions implicated in protein–protein interactions rather than full-length evolutionary relationships. Viral oncoproteins (HPV16 E7 and adenoviral E1A) cluster proximal to host interaction partners despite clear evolutionary distance, consistent with functional convergence and molecular mimicry rather than shared ancestry. This analysis illustrates how interaction-optimized sequence features can converge across unrelated proteins, providing a framework for understanding how viral proteins access host regulatory interfaces without implying genetic relatedness.

Qualitative illustration of ncRNA commonality depends on sequence scope

When restricting the analysis to the known interaction domains—Rb residues 720–770 and E7 residues 20–70, which encompass the LxCxE-binding motif – this overlap was completely lost, with no shared RNA targets identified (Figure 1(B)). This transition from global to local sequence analysis underscores that the RNA interactome of the Rb–E7 axis is mediated by a conserved domain-level RNA-binding profile. ncRNA commonality is highly sensitive to sequence scope and analytical boundaries. Accordingly, these Venn diagrams are presented as qualitative indicators of RNA-buffering tendencies, rather than as absolute quantitative measures. This context-dependent loss of shared ncRNA association is consistent with localized RNA-buffer exclusion at compact, high-affinity interfaces and complements the structural case studies based on interface geometry. This asymmetric loss of shared RNA buffering suggests that RNA availability constrains interaction reversibility and that oncogenic PPIs preferentially emerge in RNA-poor interface contexts.

Effect of RNA composition on antigen–antibody interaction efficiency

This result provides a proof-of-concept indication that RNA composition may influence interaction efficiency, without implying a direct or specific RNA–protein binding mechanism (Figure 3).

Figure 3.

RNA composition modulates antigen–antibody interaction efficiency in a proof-of-concept assay.

This Figure illustrates the effect of RNA composition on antigen–antibody (Ag–Ab) interaction efficiency. Fluorescence intensity at 520 nm is shown for reactions conducted under three RNA conditions (1): baseline control (2), heterogeneous ensemble RNA environment, and (3) homogeneous microRNA-enriched environment. Compared with ensemble RNA conditions, reactions in the presence of homogeneous small RNAs exhibit reduced fluorescence intensity, indicating lower Ag–Ab interaction efficiency. This assay is presented as a proof-of-concept illustration demonstrating that RNA composition can modulate protein–protein interaction efficiency, without implying direct RNA–protein binding or oncogenic specificity.

Structural case study: mutation-associated interface state bias in IDH1/2

To complement the PPI-centred analyses above, we examined the IDH1/2 system as an orthogonal structural case study to illustrate mutation-associated state bias at enzyme interfaces (Table 1). AlphaFold3 was used to predict structures of wild-type IDH1 and IDH2 homodimers and their corresponding oncogenic mutants (IDH1 R132H and IDH2 R172K). Comparison of wild-type and mutant complexes revealed that mutation-adjacent interface regions in the oncogenic variants exhibited increased buried surface area accompanied by a modest reduction in mean non-contact surface ratio (NCSR), indicating a shift towards more compact and stabilized interface states.

Table 1.

Structural case study of mutation-associated interface state bias in IDH1/2.

Complex	Interface_BSA (Å)	NCSR_mean (Å)
IDH1 hom	2976.6	0.4955
IDH1 het	3109.9 ↑	0.4745 ↓
IDH2 hom	3080.7	0.468
IDH2 het	3171.2 ↑	0.4655 ↓

Interface buried surface area (Interface BSA) and mean non-contact surface ratio (NCSR_mean) were calculated for wild-type (hom) and oncogenic mutant (het) IDH1 and IDH2 dimers using AlphaFold3-predicted structures. Across both enzymes, oncogenic mutations are associated with an increase in interface buried surface area accompanied by a reduction in NCSR_mean, indicating a shift towards more compact and stabilized interface states. Arrows denote the direction of change relative to the corresponding wild-type complex, emphasizing directionally consistent trends rather than quantitative inference. Because AlphaFold3 prediction instances do not represent independent experimental replicates, these values are interpreted descriptively as a structural case study illustrating mutation-associated interface state bias.

Although AlphaFold3 prediction instances do not represent independent experimental replicates, the directionally consistent changes observed across models support a trend-level interpretation in which oncogenic mutations bias enzyme-associated interfaces towards reduced reversibility. This structural case study provides a quantitative bridge between network-level RNA buffering tendencies and interface-level stabilization, without requiring explicit RNA–protein complex modelling.

Mutations disrupt ncRNA overlap and enhance binding

Mutant forms of key cancer genes (e.g Kras G12C) were analysed for changes in NCSR overlap. Compared to wild type, these mutants showed either reduced commonality or increased heterogeneity, both indicative of buffer loss and tighter binding affinity (Figure 1(C)). This finding reinforces that ncRNA profiles are not passive artefacts but active determinants of interaction strength and stability.

Evolutionary convergence and motif-level mimicry

As shown in Figure 2, evolutionary clustering of viral oncoproteins with Rb1 supports structural mimicry as a mechanism of functional interference. When minimal shared motifs are mapped onto a phylogenetic tree including viral oncoproteins and host factors (p53, Rb1, MDM2, CDK2, Cyclin E), the viral proteins that cluster closest to Rb1 (HPV16 E7, Ad E1A) are precisely those with the fewest shared ncRNA motifs with their hosts. This suggests that structural mimicry depends not on global sequence similarity but on conservation of short ncRNA-binding or motif-interaction patches.

From an RNA-commonality perspective, viral proteins appear to converge on interaction-relevant sequence features that overlap with regions involved in host protein regulation. Accordingly, the minimal overlap emphasized in Figure 1(B) is interpreted here as a qualitative indicator of the functional substrate on which convergent interference can occur, rather than as evidence of direct evolutionary relatedness.

Integration of AF3 and RNA binding profiles

By integrating catRAPID-predicted RNA binding tendencies with AlphaFold3-derived interface metrics, we observed a consistent directional relationship in which proteins exhibiting lower ncRNA commonality tended to display larger increases in buried surface area (ΔBSA) and higher aggregation-associated interface scores. Because AlphaFold3 prediction instances do not constitute independent experimental replicates, this relationship is interpreted descriptively as a trend-level association rather than a statistically definitive correlation. Together, these observations support a model in which ncRNA binding modulates the structural context of protein complexes, contributing to proteostatic balance in oncogenic settings.

Discussion

Theoretical implications

This model aligns with emerging views of RNA as a structural regulator in liquid–liquid phase separation (LLPS), nuclear body assembly, and even cytoskeletal organization.

However, unlike these macroscopic condensates, our hypothesis focuses on microscale protein–protein interactions (PPIs) – subtle yet cumulative shifts in binding affinity that can ultimately dictate cell fate. At this fine structural level, RNA acts not as a scaffold but as a molecular rheostat, tuning the reversibility of transient interactions by modulating surface complementarity and electrostatic charge distribution. From an evolutionary perspective, viral oncoproteins such as HPV16 E7 and adenoviral E1A may have evolved to exploit unbuffered regions of the host proteome, targeting tumour suppressors like Rb1 through low-redundancy ncRNA interfaces. This strategy minimizes competition from host RNAs, thereby maximizing oncogenic efficiency. The convergence of these viral proteins towards Rb1 in both motif space and structural distance (Figure 2) suggests that viral mimicry operates by preserving small RNA-binding motifs rather than global sequence similarity – a principle that could redefine our understanding of viral hijacking mechanisms.

At the structural level, AlphaFold 3 modelling and ΔBSA analysis demonstrate that ncRNA binding can reduce buried-surface area and contact variance, thereby acting as a structural dampener that stabilizes native complexes and prevents pathological aggregation.

This offers a mechanistic explanation for how ncRNA diversity contributes to proteome stability, extending the role of RNA beyond transcriptional control to direct biophysical regulation of protein homoeostasis.

Predictions and testable consequences

This model generates several experimentally testable predictions:

1. RNA supplementation and aptamer design:

   Increasing homologous ncRNAs – via miRNA mimics, aptamers, or synthetic scaffolds – should decrease binding affinity in otherwise tight oncogenic complexes such as Rb1–E7 or Kras–Raf1. Such effects can be validated by biolayer interferometry, microscale thermophoresis, or AlphaFold-guided mutagenesis of RNA-binding surfaces.

2. Loss of RNA-processing proteins:

   Depletion or mutation of RNA-binding proteins (e.g. hnRNPs, Dicer) is predicted to reduce global ncRNA buffering capacity, shifting PPI networks towards tighter, aggregation-prone states that favour transformation. This provides a plausible structural rationale for the frequent co-occurrence of Dicer or AGO2 loss in cancer.

3. Restorative RNA therapeutics:

   Rationally designed RNA aptamers [13] with high complementarity to interface-proximal sequences could restore the buffering function lost through mutation or viral interference.

   These molecules would not inhibit binding entirely, but rather restore reversibility, thereby normalizing proteostatic dynamics.

Therapeutic and conceptual outlook

If validated, ncRNA-mediated buffering would represent a new therapeutic dimension in oncology-one that targets interaction context rather than binding sites.

RNA-based molecules could be used to mimic or restore buffering RNAs, stabilizing critical complexes such as p53–MDM2 or Rb1–E2F under stress conditions.

Conversely, oncogenic interactions characterized by low ncRNA overlap (e.g. Rb1–E7, Kras–Raf1) could be disrupted or re-balanced by ncRNA analogs that re-introduce structural interference.

This concept reframes RNA from a passive messenger to an active structural modulator of protein networks. It also suggests that RNA dysregulation – whether through mutation, viral competition, or miRNA depletion – may be a first-order driver of oncogenic proteome remodelling.

In a broader sense, the ncRNA-buffering model bridges molecular evolution, structural biology, and therapeutic design, providing a unifying principle for how subtle RNA–protein couplings can determine the boundary between physiological reversibility and pathological fixation.

Conclusion

Non-coding RNAs may act as structural moderators of protein–protein interactions, determining whether a binding event remains reversible or becomes pathologically stable.

By integrating computational modelling (AlphaFold 3 ΔBSA, catRAPID), evolutionary clustering, and experimental validation, this study introduces a biophysical framework for understanding how ncRNAs buffer the proteomic landscape.

Future investigations using single-molecule or cryo-EM methods could verify these predicted RNA-induced conformational constraints and pave the way towards RNA-guided correction of oncogenic misinteractions.

Note

‘The AlphaFold 3 ΔBSA and interface datasets described in this study will be made publicly available in conjunction with forthcoming publications’.

Funding Statement

This research received no external funding and was supported by the internal resources of the BFSR Institute.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

All computational data generated in this study, including catRAPID RNA–protein interaction scores, ncRNA overlap tables, AlphaFold3 structural models of IDH1/2 complexes, ΔBSA calculations, and interface contact metrics, are available from the corresponding author upon reasonable request. Structural datasets will be deposited in a public repository upon acceptance of the manuscript.