Integrated Gene Regulatory Network Analysis Reveals Coordinated Transcriptional Reprogramming in the Arabidopsis thaliana–Trichoderma atroviride Interaction

Abstract

Mutualistic interactions between plants and beneficial fungi rely on extensive transcriptional reprogramming in both partners, yet the underlying regulatory mechanisms coordinating these responses remain incompletely understood. Here, we combined a transcriptomics analysis with a gene regulatory network (GRN) inference to dissect the interaction between Arabidopsis thaliana and the growth-promoting fungus Trichoderma atroviride. At an early but established stage of colonization (72 h post-inoculation), we identified widespread transcriptional changes in both of the organisms, including host activation of hypoxia, stress and root development-related pathways in Arabidopsis, and fungal reprogramming of membrane transport and primary metabolism. Using DNA-binding motifs and GENIE3-based regulatory inference, we reconstructed interaction-specific GRNs for each species. The subnetworks focused on the main differentially expressed biological processes and uncovered ERF-, WRKY-, NAC- and DOF-centered modules linking hypoxia responses with developmental remodeling in the plant, whereas the putative orthologs of TFs involved in developmental and stress-related TFs in fungi, such as CrzA, RME1, NsdC, PacC and RPN4, formed a regulatory core coordinating fungal transport and metabolic adjustment. In parallel, we uncovered contrasting sRNA dynamics between the partners. While the Arabidopsis sRNA changes were limited, T. atroviride exhibited a strong induction of 20–22 nt sRNAs, including a small set of high-confidence sRNA–mRNA interactions targeting host genes involved in root function and immunity. Together, our results extend previous pathway-based descriptions of the Arabidopsis–Trichoderma mutualism and provide a systems-level, testable framework for how coordinated regulatory programs in both of the partners support the interaction.

1. Introduction

Fungi play fundamental roles in terrestrial ecosystems as primary decomposers, recycling organic matter and releasing essential nutrients for plants [1,2]. Their early symbiotic associations with ancestral land plants established lasting relationships between specific fungal lineages and their hosts [3,4]. These associations have diversified over evolutionary time, giving rise to a wide spectrum of lifestyles, such as saprotrophic, pathogenic and mutualistic, each shaped by host signals and environmental conditions [5,6]. Most of the research on plant–fungi interactions has been focused on harmful fungal pathogens such as Magnaporthe oryzae and Botrytis cinerea due to their devastating impact on crops and native plants [2,7,8]. More recently, attention has turned toward beneficial rhizospheric fungi that enhance plant fitness. These organisms, collectively referred to as plant-growth-promoting fungi (PGPF) [9,10,11], improve plant stress tolerance, stimulate growth, and suppress pathogens, thereby contributing to sustainable crop resilience [2,12]. Once established in plant roots, PGPF can enhance nutrient acquisition, strengthen immunity, increase stress resistance, and promote overall plant growth and development [2,4,5,10,13,14].

Trichoderma species are widespread soil fungi, and some of them are PGPF. They combine a saprophytic lifestyle with the ability to antagonize phytopathogens and colonize plant roots as avirulent endophytes [9,15]. Successful fungal colonization relies on multiple factors, including fungal redox signaling, root adhesion mechanisms, and secreted effector-like proteins [16,17,18,19]. Upon root colonization, Trichoderma activates both systemic acquired resistance (SAR) and induced systemic resistance (ISR) in the plant, priming the plant immune system against diverse pathogens [20,21,22]. These beneficial effects are mediated by secreted fungal enzymes, metabolites and small molecules that modulate plant hormonal signaling, reprogram resource allocation, and reshape root architecture to optimize nutrient uptake [10,12,15,16,17,23,24].

As a model for endophytic interactions, the Arabidopsis thaliana–Trichoderma system has been instrumental to dissect these processes. In A. thaliana, Trichoderma enhances biomass accumulation and lateral root development through auxin-dependent pathways improves salt tolerance and modulates sucrose transport to promote root branching [25,26,27,28]. The outcome of this symbiosis is highly context-dependent, varying with the fungal species and strain, the growth stage of the plant, or the environmental cues [16,29,30]. At the root–fungus interface, Trichoderma influences plant growth and defense by modulating reactive oxygen species (ROS) dynamics and secondary metabolite production in A. thaliana, processes regulated by fungal nutrient status and NADPH oxidase activity [29,31,32,33]. These physiological and molecular adjustments provide a mechanistic framework to explain the variable outcomes of the association under different conditions.

Recent transcriptomic studies have profiled global changes in gene expression during plant–Trichoderma interactions [19,34,35,36]. RNA-seq analyses in crops such as rice, tomato, wheat, and maize revealed that Trichoderma exposure alters the expression of genes related to defense, photosynthesis, nutrient transport, secondary metabolism, and root development [37,38,39,40,41,42,43]. On the fungal side, the plant modulates the expression of genes involved in carbohydrate degradation, antioxidant transport, and detoxification [19,44,45]. However, most studies have analyzed each organism in isolation, obscuring the reciprocal regulatory dynamics that underlie the interaction.

More integrative approaches, such as dual RNA-seq and gene co-expression network analyses, have begun to bridge this gap [46,47,48,49]. Dual RNA-seq enables the simultaneous profiling of the transcriptomes from both of the interacting partners by mapping the reads to each genome to resolve coordinated, cross-organism expression changes [50,51]. On the other hand, gene co-expression networks have been used to study the transcriptional response of Arabidopsis thaliana during its interaction with two Trichoderma species, T. atroviride and T. virens [52]. These analyses identified groups of co-expressed genes related to chorismate metabolism, defense responses and the aerobic electron transport chain. Connectivity analyses further revealed a highly connected transcription factor (TF), ERF71 (AT2G47520), required for hypoxia tolerance, suggesting a link between hypoxia, transcriptional reprogramming and plant defense mechanisms [52]. This work underscores the value of network biology approaches for uncovering novel mechanisms of plant–fungal interactions. However, co-expression networks remain correlative and lack causality, limiting their ability to identify the central controllers that drive transcriptome-level changes in both of the interacting partners.

Gene regulatory networks (GRNs) provide a causal, mechanistical basis to overcome this limitation. GRNs represent how genes are regulated and coordinated within a cell, describing interactions between regulatory factors (e.g., TFs, non-coding RNAs) and their targets, thereby determining when, where, and to what extent the genes are expressed [53]. Although GRN approaches have successfully dissected regulatory interactions in plant pathosystems such as tomato–Alternaria, potato–PSTVd, and wheat–Fusarium [54,55,56], their application to model beneficial symbioses remains scarce. Consequently, the regulatory architecture that underpins mutualistic plant–fungus communication is still poorly understood.

Here, we address this gap by constructing the first GRN models for the A. thaliana–T. atroviride symbiosis, integrating transcriptomic data and TF–target inference to define the core transcriptional regulatory layers in both of the partners. In addition, we incorporate small RNA (sRNA) profiling to explore whether post-transcriptional regulation contributes to plant–fungus coordination during early symbiosis. By moving beyond correlative analyses, this framework identifies candidate regulatory TFs, combinatorial control modules, and putative sRNA-mediated interactions that may underlie the developmental reprogramming, stress priming, and metabolic adjustment associated with mutualism. Together, this systems-level perspective refines the mechanistic understanding of Trichoderma-induced reprogramming in plants and provides a transferable basis for predictive regulatory models with potential relevance for improving crop resilience and supporting more sustainable agricultural practices.

2. Results

2.1. Global Transcriptional Reprogramming in Arabidopsis thaliana and Trichoderma atroviride Reveals Coordinated Metabolic and Signaling Adjustments During Interaction

To obtain a comprehensive view of the molecular events underlying the interaction of Arabidopsis thaliana and Trichoderma atroviride, we simultaneously profiled transcriptome-wide gene expression changes in Arabidopsis roots and in T. atroviride mycelia after 72 h of co-culture to identify differentially expressed genes (DEGs) (Figure S1). At this stage, root colonization is established and sufficient fungal material can be recovered for downstream analyses while avoiding extensive fungal overgrowth of the root system. Importantly, this time point captures an early regulatory phase of the Arabidopsis–T. atroviride interaction, in which transcriptional reprogramming and signaling events are already active, whereas macroscopic growth phenotypes remain subtle or absent [9,24,57,58,59]. Our RNA-seq analysis revealed extensive transcriptional reprogramming in both of the partners. In A. thaliana, 2979 genes (10.81% of total protein coding genes (PCGs)) were differentially expressed, with 1855 of them (62.27% of A. thaliana DEGs) being up-regulated and 1124 (37.73% of DEGs) being down-regulated. Within this DEG set, we identified 216 transcription factors (TFs), representing 12.58% of the annotated TF loci in A. thaliana and 7.25% of all DEGs. Of these differentially expressed TFs, 154 were up-regulated (71.30%) and 62 were down-regulated (28.70%) (Table 1 and Table S1). This predominance of induced TFs likely reflects a dynamic reprogramming to adapt to colonization. Among the TFs, one of the families that presented a relevant response to the T. atroviride interaction was the WRKY family, known for playing roles in the regulation of plant–microbe responses. From the 72 WRKY loci annotated in The Arabidopsis Information Resource (TAIR, www.arabidopsis.org), 17 were differentially expressed, and all were up-regulated at 72 h post-inoculation (Table S2). The most strongly induced WRKYs included WRKY22, WRKY8, WRKY18 and WRKY29 (log₂ fold-change > 3). The functional annotations compiled from TAIR and the BAR/ePlant Plant Connectome resource (bar.utoronto.ca/eplant/, accessed on 27 December 2025) indicate that these WRKYs collectively span immune- and hormone-associated processes including defense-related programs annotated as hormone-responsive (including SA- and JA-associated terms) and stress-related programs annotated as responsive to ABA, as well as developmental and nutrient-associated functions (Table S2). This suggests that WRKY induction accompanies not only defense priming but also broader reprogramming of stress adaptation and resource/developmental pathways during root colonization.

Table 1.

Differential expression analysis of mRNA-seq libraries from Arabidopsis thaliana roots and Trichoderma atroviride mycelium during the interaction. Summary of transcriptomic changes detected after 72 h of co-culture. The Table reports the number of differentially expressed (DE) genes (DEGs) and their percentage relative to all protein coding genes (PCGs), to DEGs, to annotated TFs in the corresponding organism, or to DE TFs. Up- and down-regulated genes were defined relative to non-inoculated control roots (Arabidopsis thaliana) or to mycelia grown in the absence of plants (Trichoderma atroviride). Differential expression was determined using a Wald test with Benjamini–Hochberg correction (adjusted p-value < 0.01 and |log₂ fold-change| > 1).

Gene Category	Arabidopsis thaliana		Trichoderma atroviride
	Number	% (Gene Set)	Number	% (Gene Set)
DEGs	2979	10.81 (of PCGs)	1235	10.41 (of PCGs)
Up-regulated DEGs	1855	62.27 (of DEGs)	484	39.19 (of DEGs)
Down-regulated DEGs	1124	37.73 (of DEGs)	751	60.80 (of DEGs)
DE TFs	216	12.58 (of TFs)	60	11 (of TFs)
Up-regulated TFs	154	71.30 (of DE TFs)	20	33.3 (of DE TFs)
Down-regulated TFs	62	28.70 (of DE TFs)	40	66.7 (of DE TFs)

In T. atroviride, 1235 DEGs were identified, representing 10.41% of its PCGs, with 484 (39.19%) being up-regulated and 751 (60.81%) being down-regulated. Among these DEGs, 60 TFs were detected, corresponding to 11% of the annotated TF repertoire of T. atroviride and 4.86% of all DEGs. Within this set of differentially expressed TFs, 20 showed increased expression (33.3%), whereas 40 exhibited reduced expression (66.7%) (Table 1; Table S1), indicating a possible transition from a transcriptionally active, saprotrophic state toward a symbiotic mode.

To functionally contextualize the transcriptomic responses of each interacting partner, the DEGs were mapped to KEGG Ortholog (KO) identifiers for the major KEGG pathway categories (Table S3) and supplemented with Gene Ontology (GO) term enrichment (Table S4). In Arabidopsis thaliana, the KEGG mapping analysis showed that the DEGs were primarily associated with metabolism-related categories, including carbohydrate, amino acid, energy, and lipid metabolism. Pathways classified under environmental information processing, particularly signal transduction, were also mapped to the DEGs, as well as those involved in metabolism and signaling (Figure S2A and Table S3), consistent with cellular adaptation to fungal colonization. A separate analysis of up- and down-regulated genes revealed that the up-regulated DEGs were predominantly involved in plant–pathogen interactions, photosynthesis, MAPK signaling, endocytosis, protein processing and carbon metabolism (Figure S2B and Table S3). The KEGG map “Plant hormone signal transduction” (map04075) was present in both the up-regulated and the down-regulated genes (Figure S2B,C), but it was represented by different hormone branches: the up-regulated genes mapped mainly to abscisic acid (ABA), jasmonate (JA), cytokinin (CK) and salicylate (SA)-related genes (e.g., PYL/SnRK2, JAZ, ARR-B/AHK, NPR1), whereas the down-regulated genes mapped predominantly to auxin-related components (e.g., TAA1/YUCCA, AUX1/LAX, GH3/SAUR), ABA negative regulatory components (PP2C), as well as gibberellin (GA), brassinosteroid (BR) and ethylene (ET) genes (DELLA, BZR1/2, BES1, EBF1/2) (Table S3). The down-regulated genes were additionally associated with cysteine/methionine and starch/sucrose metabolism, ribosome biogenesis, cell cycle and DNA replication (Figure S2C and Table S3). KEGG Mapper further reconstructed the complete metabolic modules, including an up-regulated ethylene biosynthesis pathway (M00368) and a down-regulated terminal gibberellin biosynthesis module (GA12/GA53→GA4/GA1; M00928) (Figure S3A,B). Our GO term enrichment analysis supported these trends, showing that the up-regulated genes were associated with hypoxia responses, defense responses and ABA and JA signaling, whereas the down-regulated genes were enriched in developmental and structural processes, including cell wall organization and modification, microtubule organization, and DNA replication (Figure S2D,E and Table S4). These findings reveal a stress-primed transcriptional state characterized by the activation of defense and hormone signaling pathways and the down-regulation of developmental and metabolic programs, reflecting a shift toward adaptive root remodeling during the interaction with T. atroviride.

In T. atroviride, the KEGG mapping mirrored the host’s broad metabolic adjustments while highlighting fungal-specific reprogramming (Figure S4A and Table S5). The DEGs were mainly associated with carbohydrate, amino acid, lipid, glycan and energy metabolism, as well as signal transduction, transport and catabolism (Figure S4A). As in A. thaliana, separating the up- and the down-regulated genes allowed a clearer distinction between activated and repressed processes. The up-regulated genes were related to amino acid, glyoxylate, pyruvate and nitrogen metabolism (Figure S4B). Consistently, KEGG Mapper identified two complete, up-regulated nitrogen-assimilation modules, consisting of assimilatory nitrate reduction (M00531) and nitrate assimilation (M00615) (Figure S3C,D). The down-regulated genes were associated with MAPK signaling, amino- and nucleotide-sugar metabolism, ER protein processing, inositol phosphate metabolism, protein export and cell cycle (Figure S4C). The GO enrichment analysis for all of the T. atroviride DEGs was consistent with these findings, highlighting changes in membrane transport and transport terms (Figure S4D and Table S6). These patterns suggest a metabolic shift toward nutrient recycling and energy optimization, reflecting fungal adaptation to the symbiotic lifestyle, as previously reported [60].

To further explore the fungal genes potentially mediating communication with the host, we analyzed the expression of the predicted T. atroviride elicitors, effector-like proteins and carbohydrate-active enzymes (CAZymes). Among the 964 annotated CAZymes, 132 were differentially expressed (44 up- and 88 down-regulated). Similarly, of the 94 predicted effectors, 12 were differentially expressed (1 up- and 11 down-regulated). A broader set of 27 putative elicitor-like proteins (e.g., cerato-platanins, xylanases, chitinases, cutinases, and pectate lyases) lacking transmembrane domains included 10 induced and 17 repressed members (Table S7). Notably, the well-characterized cerato-platanin Epl1, previously implicated in growth promotion and stress priming in plants [61,62], was down-regulated at 72 h of interaction compared with the fungus grown in hydroponic MS medium without plants. This repression, together with the general down-regulation of secreted and CAZyme-related genes, supports a shift from a saprotrophic, exploratory mode to a low-immunogenic, host-adapted colonization state at this stage of the symbiosis.

Together, these analyses reveal coordinated metabolic and signaling reprogramming in both of the partners, reflecting complementary adjustments that enable the mutualistic association.

2.2. Construction of GRN Models for Arabidopsis thaliana and Trichoderma atroviride During the Interaction

To interpret the transcriptional reprogramming underlying the Arabidopsis–Trichoderma interaction, we first defined the reference GRNs that connect the TFs to their putative targets. For T. atroviride, we used a previously published reference GRN, which is based on the presence of cis-binding motifs in promoter regions and TF–target inferences derived from transcriptomic data using the GENIE3 algorithm [63]. For Arabidopsis, we constructed a reference GRN using the same pipeline. We retrieved a catalog of 1717 TFs from PlantTFDB [64] and assigned position weight matrices (PWMs) to these TFs using information from the CisBP database [65]. We were able to assign experimentally determined (direct) binding motifs for 762 TFs and inferred motifs (indirect) for an additional set of 391 TFs (Table S8). The promoter regions (2 Kb upstream of the transcription start sites) of all of the annotated genes were scanned using FIMO [66], generating TF–target links for 993 TFs. To refine these motif-based predictions, we integrated regulatory inferences from GENIE3, trained on a compendium of 3493 curated Arabidopsis Illumina RNA-seq libraries (Table S9). We generated three GENIE3 models, retaining the top 10%, 20%, and 30% of the ranked edges (Table S10); we then calculated the area under the receiver operating characteristic (AUROC) and the precision-recall (AUPR) curves for each, and we benchmarked them against a gold standard experimental TF–target network from ConnecTF (https://connectf.org/, accessed on 20 March 2024) [67]. The network filtered at the top 20% of GENIE3 scores achieved the best balance between precision and coverage (AUPR = 0.436; AUROC = 0.563) (Tables S11 and S12). To further assess the predictive performance of our Arabidopsis GRN, we benchmarked the network against an independent gold standard curve derived from DAP-seq TF–target interactions obtained from the Plant Cistrome Database (http://neomorph.salk.edu/dap_web/pages/, accessed on 27 December 2025). Our comparison with the DAP-seq dataset yielded an AUROC of 0.520 and an AUPR of 0.367, values comparable to those obtained using ConnecTF and consistent with the partial concordance expected between in vitro binding data and context-specific regulatory interactions inferred from transcriptomics. To evaluate whether this performance exceeded random expectation, we generated 10,000 randomized networks preserving the number of nodes and edges. None of the randomized networks achieved an AUPR higher than that of the inferred GRN (mean AUPR = 0.325, median AUPR = 0.329) (Figure S5), indicating that the observed precision-recall performance is significantly better than chance (p < 0.001) and supporting the reliability of the network for predicting regulatory interactions involving TFs lacking experimental validation.

Topologically, the selected network forms a single connected component with a mean shortest path of 3.12, non-trivial clustering (global coefficient = 0.0062 and local coefficient = 0.17), and several high-degree hubs (≥500 edges) (Table S13), consistent with the small-world, hub-and-module organization characteristic of biological GRNs [68,69,70,71].

To investigate transcriptional regulation in the context of the Arabidopsis–Trichoderma interaction, the T. atroviride [63] and A. thaliana reference GRNs were intersected with the respective set of DEGs identified in the co-culture experiment to extract context-specific interaction networks for both of the partners. The resulting A. thaliana interaction GRN (AiGRN) comprised 2593 nodes, including 206 TFs, connected by 11,646 TF–target edges, encompassing 87% of all plant DEGs (Table S14). The T. atroviride interaction network (TiGRN) consisted of 1005 DE nodes, including 46 TFs and 3260 TF–target edges, covering 81.4% of fungal DEGs (Table S14).

2.3. Modular Gene Regulatory Networks Coordinate Hypoxia Signaling and Root Development During Trichoderma Interaction

To explore the transcriptional regulatory programs underlying the Arabidopsis–Trichoderma interaction, we focused on two biological processes whose associated terms were consistently enriched among the Arabidopsis DEGs: response to hypoxia and root development (Figure S2, Tables S3 and S4). Hypoxia-related responses have previously been linked to developmental and defense programs activated during Trichoderma colonization [33,52,72,73], while extensive remodeling of the root system architecture in response to Trichoderma has been widely reported [24,28,29,74]. However, the TFs orchestrating these responses and their regulatory organization remain poorly characterized.

To generate the process-specific regulatory subnetworks, we first identified all of the DE genes annotated to the GO biological processes “cellular response to hypoxia” (GO:0071456), “lateral root development” (GO:0048527), “root hair cell tip growth” (GO:0048768), “regulation of lateral root development” (GO:2000023), and “regulation of root meristem growth” (GO:0010082). Using the AiGRN, we then extracted all of the regulatory relationships linking these hypoxia- and root development-related genes to their upstream TFs, retaining regulatory TFs with an outdegree greater than 1. This approach resulted in a hypoxia subnetwork comprising 87 nodes (27 TFs and 60 other genes) and 321 edges (Table S15), and a root development subnetwork consisting of 80 nodes (63 TFs and 17 other genes) connected by 334 edges (Table S16). Importantly, we independently validated the regulatory edges in both of the subnetworks using the TF–target relationships derived from the DAP-seq experiments available in the Plant Cistrome Database. This analysis confirmed regulatory interactions for 16 TFs in the hypoxia subnetwork and 25 TFs in the root subnetwork with available DAP-seq data, corresponding to 40.5% and 42.9% of the regulatory interactions for these TFs, respectively (Tables S15 and S16). This independent support strengthens our confidence in the inferred regulatory structure and indicates that these subnetworks capture biologically meaningful transcriptional control relationships.

For the hypoxia subnetwork (Figure 1A), the hypoxia-related targets encompassed a diverse but functionally coherent set of processes. Metabolic reprogramming was represented by the genes involved in amino acid and nitrogen metabolism, such as the alanine aminotransferase AlaAT1. Redox balance and detoxification were prominent features, with glutathione S-transferases (e.g., GSTF8, At4g19880), peroxidases (PER4, At1g14550) and oxidoreductases such as OPR1. Cell wall remodeling and polysaccharide turnover were also represented through genes encoding a chitinase (At1g02360), an endoglucanase (At4g30380) and the xyloglucan hydrolase XTH18. In addition, the network captured stress signaling and protein homeostasis components, including calcium-binding proteins (CML38, TCH3), the lectin-related GAL1 and multiple kinase-associated regulators (MAPKKK13, S6K2, STY46). Defense-related genes were also present, notably, TIR-NBS-LRR genes (TN3, DM10, TN7; At1g57630; At4g19520) and the membrane-associated flotillin FLOT1, linking hypoxia-associated signaling to immune-related pathways (Figure 1A and Table S15).

Figure 1.

Hypoxia-related gene regulatory subnetwork. (A) Each node represents a differentially expressed gene in Arabidopsis roots identified during co-culture with T. atroviride. Transcription factors (TFs) are represented as triangles and non-TF genes are represented as circles. The size of the nodes represents the node outdegree. Genes annotated with the GO Biological Process “response to hypoxia” (GO: 0071456) are colored blue, and other nodes are colored orange. The node border color represents the Cluster the node belongs to (yellow: Cluster 1; red: Cluster 2; black: Cluster 3; green: Cluster 4). Edge colors represent whether the regulatory interaction was only found by FIMO-GENIE3 analysis (grey) or if the edge was validated with DAP-seq data from the Plant Cistrome Database (brown). (B–D) Examples of feed-forward loop (B), single-input module (C) and bifan (D) in the subnetwork.

To dissect the regulatory logic of the subnetwork, we performed community detection using the GLay clustering algorithm [75], which organized the hypoxia network into four discrete but highly interconnected clusters each defined by a distinct TF composition (Figure 1A). Cluster 1 was dominated by members of the ERF family, composed mainly of members of the ERF family, including ERF-1, ERF112, ERF114, ERF19, ERF7/WIND4, ERF6, ERF73/HRE1, ABR1 and DREB2B. These TFs are well-known regulators of stress-responsive programs, integrating signals related to abscisic acid (ABA), reactive oxygen species (ROS), and osmotic stress signaling (Table S15). Notably, ERF73/HRE1 is a central regulator of low-oxygen acclimation and a key component of hypoxia signaling in Arabidopsis [76,77], positioning this cluster as a core hypoxia-responsive module. This cluster also included bZIP53, involved in amino acid metabolism [78,79], and NTM1, a NAC TF that emerged as the most highly connected regulator within the module. NTM1 has been implicated in cytokinin-associated signaling and the control of cell division [80], suggesting the integration of hypoxia and stress signaling with growth-related regulatory inputs. Cluster 2 was enriched in WRKY TFs, including WRKY11, 15, 22, 25, 33 and 40. These TFs are established regulators of biotic and abiotic stress responses and participate in SA, JA and ABA-associated signaling pathways (Table S15). Their co-occurrence suggests a coordinated WRKY-driven regulatory program integrating immune signaling with stress adaptation under hypoxia-associated conditions. This cluster also included DOF1/DOF1.7, a TF implicated in N responses [81], which emerged as the most highly connected regulator in the hypoxia network, positioning it as a key integrator linking WRKY-mediated defense signaling with broader metabolic and developmental reprogramming. Clusters 3 and 4 comprised members from multiple families including ERFs (CRF4, DREB2C, ERF8, ERF013, ERF025), Dof (OBP3 and CDF5), the G2-like TF HHO3/NIGT1.1 and MYB15.

These TFs are involved in defense, developmental and growth responses (Table S15). For example, HHO3/NIGT1.1 integrates N and P signaling [82]; OBP3 participates in iron homeostasis, SA responses and root, hypocotyl and cotyledon development [83,84,85]; and MYB15 regulates lignification during pathogen attack [86].

To further delve into the regulatory logic of this subnetwork, we performed a motif analysis (Figure 1B–D and Table S17). In total, we identified 50 bifan motifs, 100 feed-forward loops (FFLs), 7 feed-back loops (FBLs) and 2 single-input modules (SIMs). The high number of FFLs suggest a regulatory strategy that filters transient signals, enabling the stable activation of hypoxia responses only under sustained low-oxygen conditions. The bifan motifs further support the integration of multiple regulatory inputs, consistent with hypoxia signaling intersecting with hormonal, stress and developmental pathways during root colonization. The FFLs, although less abundant, likely contribute to the stabilization and fine-tuning of transcriptional states once activated. In contrast, the scarcity of the SIMs highlights a distributed regulatory architecture rather than a hierarchical one.

For the root development subnetwork, we found a high density of transcriptional regulators (Figure 2A and Table S16). Of the 80 nodes, 63 (78.8%) corresponded to TFs, indicating strong combinatorial control over root development-related gene expression. Moreover, 25 of these TFs have previously been annotated as root development-related (Figure 2A), underscoring the central role of transcriptional regulation in shaping root system architecture in response to the Trichoderma atroviride interaction. The target genes within this subnetwork included the key components of peptide growth factor signaling (RGF5, RGF7, RGI1, RGI4 and RGI5) that sustain meristem activity [87]; the nitrate transporter/sensor NRT1.1/NPF6.3, linking nitrogen status to lateral root formation [88]; the phosphatidylinositol-transfer protein COW1, essential for root-hair elongation [89]; the Casparian strip-associated peptide CIF1 [90] and the CLE6 peptide, involved in root growth and systemic signaling [91,92]; the extracellular signaling peptide CEP3 that controls primary and lateral root growth [93]; and the auxin-induced protease AIR3, which promotes lateral root emergence [94]. Together, these targets support meristem maintenance, lateral root initiation, and root hair differentiation, core processes shaping root system architecture during the interaction with T. atroviride.

Figure 2.

Root development-related gene regulatory subnetwork. (A) Each node represents a differentially expressed gene in Arabidopsis roots identified during co-culture with T. atroviride. Transcription factors (TFs) are represented as triangles and non-TF genes are represented as circles. The size of the nodes represents the node outdegree. Genes annotated with the GO Biological Processes “cellular response to hypoxia” (GO:0071456), “lateral root development” (GO:0048527), “root hair cell tip growth” (GO:0048768), “regulation of lateral root development” (GO:2000023), and “regulation of root meristem growth” (GO:0010082) are colored green, and other nodes are colored orange. The node border color represents the Cluster the node belongs to (yellow: Cluster 1; red: Cluster 2; black: Cluster 3; green: Cluster 4). Edge color represents whether the regulatory interaction was only found by FIMO-GENIE3 analysis (grey) or if the edge was validated with DAP-seq data from the Plant Cistrome Database (brown). (B–E) Examples of feed-forward loop (B), feedback loop (C), single-input module (D) and bifan (E) in the subnetwork.

The community detection again resolved the root development subnetwork into four regulatory clusters (Figure 2A and Table S16). Cluster 1 included TFs from multiple families, with the Dof factor OBP3 as the most connected node, along with the NAC TFs involved in growth and development (NTM1, NAC038, NAC011, ANAC087), the MYBs (MYB30, MYB108) linked to development and immunity, the G2-like TF HRS1, integrating in N and P signaling [82,95,96], GATA TFs (GATA3, GATA16), ERFs (ERF73, ERF9), the MADS-box TF AGL44/ANR1, involved in LR elongation in response to nitrate [97], and the heat shock factors HSFA7A and HSFB2A. Cluster 2 was enriched in the WRKY TFs associated with immune responses and development (WRKY11, WRKY15, WRKY18, WRKY25, WRKY29, WRKY36, WRKY40 and WRKY6), along with the GATA TFs involved in development and gibberellin-related pathways (GATA12 and GATA19), MYB27, and members of the NAC, ERF, C2H2 and bHLH families. Cluster 3 comprised the ERFs involved in stress responses (ABR1, DREB26, DREB2B, ERF-1, ERF013, ERF19, ERF6, TDR1), the MYBs involved in development (MYB15, MYB3R-5, MYB51, MYB68), and DOF1 and HHO3/NIGT1.1, linking nutrient signaling with developmental regulation. Cluster 4 included ERFs (DREB2C, ERF112, CRF4, DEAR5), bZIPs (bZIP25, GBF1, BZO2H3), NAC069, the MYB TF RVE8, WRKY75, the GRAS TF HAM3, and bHLH129, highlighting the further integration of hormonal and stress-responsive regulatory inputs (Figure 2A and Table S16). Our network motif analysis of the root development-related Arabidopsis GRN revealed a strong enrichment of bifan (244) and FFL (267) motifs, together with fewer FBLs (22) and only two SIMs (Figure 2B–E and Table S18). The high prevalence of bifan motifs supports extensive co-regulation by TF pairs, consistent with the need to integrate multiple inputs (developmental state, nutrient cues, and stress/hormone signaling) when remodeling root architecture during colonization. The large number of FFLs further suggests strong temporal structuring and signal filtering, in which downstream developmental targets are activated only after upstream regulatory signals persist, enabling ordered transitions from signaling to developmental programs (e.g., meristem maintenance, lateral root initiation, root hair growth). Although less frequent, FBLs likely contribute to stabilizing and fine-tuning transcriptional states once developmental reprogramming is engaged. Finally, the scarcity of SIMs indicates that control is distributed, with root remodeling emerging from coordinated TF interactions rather than from single hierarchical regulators. Together, the motif profile supports a regulatory architecture optimized for robustness, integration, and staged activation, consistent with root development being executed through tightly coordinated modules during the Arabidopsis–T. atroviride interaction.

2.4. Integration of Membrane Transport and Primary Metabolism Through a Common Regulatory Core in T. atroviride

The GO and KEGG analyses revealed that membrane transport and primary metabolism were among the most relevant processes in T. atroviride during its interaction with Arabidopsis thaliana (Figure S5, Tables S5 and S6). To dissect the transcriptional regulation underlying these functions, we applied the same network-based strategy used for Arabidopsis to identify the central regulators in T. atroviride. The genes annotated as “transmembrane transport” (GO:0055085) were used to construct the transport subnetwork (Table S19), while those associated with “primary metabolic process” (GO:0044238) defined the primary metabolism subnetwork (Table S20).

The transmembrane transport subnetwork comprised 80 nodes and 231 edges, including 24 TFs and 56 non-TF genes (Figure 3A). The target genes in this network were dominated by putative transport-related proteins, most notably 25 predicted members of the major facilitator superfamily (MFS), a group broadly associated with sugar and multi-ion transport including Zn, Ca, Fe, phosphate and urea [98,99]. Additional components included putative orthologs of ABC transporters, carbohydrate transporters and an auxin efflux carrier (Table S19). Together, these transport-associated components suggest that T. atroviride establishes a nutrient- and signal-responsive interface with the plant, simultaneously acquiring host-derived metabolites and modulating auxin fluxes to coordinate plant growth.

Figure 3.

Transmembrane transport-related gene regulatory subnetwork. (A) Each node represents a differentially expressed gene in T.atroviride identified during co-culture with A. thaliana. Transcription factors (TFs) are represented as triangles and non-TF genes are represented as circles. The size of the nodes represents the node outdegree. Genes annotated with the GO Biological Process “transmembrane transport” (GO:0055085) are colored purple, and other nodes are colored orange. The node border color represents the Cluster the node belongs to (yellow: Cluster 1; red: Cluster 2; black: Cluster 3; green: Cluster 4). Grey edges represent whether the regulatory interaction was found by FIMO-GENIE3 analysis. (B–D) Examples of feed-forward loop (B), single-input module (C) and bifan (D) in the subnetwork.

Our community detection separated the genes into four major clusters (Figure 3A). These clusters contained several TF nodes from different families. Although the functional characterization of most Trichoderma TFs remains limited, eight regulators could be confidently assigned putative homologs in other fungi based on their sequence identity and conserved domain architecture (Figure S6). These included putative homologs of RPN4, VSD-6 and TAH-3 in Cluster 1, of VAD-11 in Cluster 2, of RME1 and NsdC in Cluster 3, and of CrzA and PacC in Cluster 4. Notably, the homologs of these TFs in other fungal species are established regulators of key physiological and developmental processes. RPN4 controls proteasome homeostasis, cell cycle progression, differentiation and stress responses [100,101], while TAH-3 has been implicated in endoplasmic reticulum stress resistance through the maintenance of intracellular sterol homeostasis in N. crassa [102]. RME1 regulates cell cycle and growth via control of cellulolytic gene expression in T. reesei [103], and NsdC governs vegetative growth and conidiation in A. nidulans [104]. In addition, CrzA mediates calcium-dependent stress responses and growth regulation [105], and PacC is a conserved regulator of ambient pH signaling in filamentous fungi, including N. crassa, where it controls the expression of pH-responsive genes [106]. The motif analysis of the transmembrane transport subnetwork revealed a highly structured regulatory architecture dominated by combinatorial control (Table S20 and Figure 3B–D). The network contained 107 bifan motifs, indicating an extensive shared regulation of transport-related targets by pairs of transcription factors, consistent with the coordinated control of membrane transport functions. In addition, 51 FFLs were identified, suggesting the stabilization of transcriptional responses, signal filtering and the temporal ordering of gene activation. By contrast, only four SIMs were detected, suggesting that the regulation of transport genes is rarely driven by a single dominant regulator but instead relies on multi-TF integration. Together, this motif composition supports a regulatory organization in which transmembrane transport processes are tightly co-regulated, robust to fluctuations, and integrated with broader signaling and metabolic programs during the T. atroviride interaction with the plant.

The primary metabolism subnetwork comprised 94 nodes and 135 edges, including 13 TFs and 38 non-TF nodes (Figure 4A and Table S21). As for the transmembrane transport subnetwork, the community detection analysis also separated the network into four highly interconnected clusters. The non-TF genes in the network include putative enzymes involved in energy balance and growth, such as glycoside hydrolases, aminotransferases, dehydrogenases, phosphodiesterases, and reductases (Table S21). Interestingly, regulatory control greatly overlapped with that of the transmembrane transport subnetwork, presenting 11 common TFs. Among these, we found the same most highly connected nodes, including CrzA, RME1 and NsdC in Cluster 1, RPN4 in Cluster 2, TAH-3 in Cluster 3 and PacC in Cluster 4. The motif analysis revealed clear differences in the regulatory organization of the primary metabolism and transmembrane transport subnetworks. While both of the networks share a highly overlapping set of upstream regulators and a similar cluster structure, their motif compositions diverged markedly. The primary metabolism subnetwork contained 42 bifan motifs and only two SIMs, and it lacked FFLs or FBLs (Table S22 and Figure 4B,C). The reduced motif complexity in the primary metabolism network suggests a regulatory strategy favoring coordinated but comparatively direct control, consistent with the need to maintain stable core metabolic functions, in contrast to the enriched motif architecture of the transport network supporting dynamic signal integration and fine-tuned responsiveness to environmental and host-derived cues.

Figure 4.

Primary metabolism-related gene regulatory subnetwork. (A) Each node represents a differentially expressed gene in T. atroviride identified during co-culture with A. thaliana. Transcription factors (TFs) are represented as triangles and non-TF genes are represented as circles. The size of the nodes represents the node outdegree. Genes annotated with the GO Biological Process “primary metabolic process” (GO:0044238) are colored yellow, and other nodes are colored orange. The node border color represents the Cluster the node belongs to (yellow: Cluster 1; red: Cluster 2; black: Cluster 3; green: Cluster 4). Grey edges represent whether the regulatory interaction was found by FIMO-GENIE3 analysis. Examples of single-input module (B) and bifan (C) motifs.

2.5. Small RNAs Provide an Additional Regulatory Layer During Arabidopsis–Trichoderma Symbiosis

To evaluate whether small RNAs (sRNAs) contribute to regulatory coordination during the Arabidopsis–T. atroviride interaction, we performed sRNA sequencing on the same total RNA samples used for the mRNA profiling. The reads were aligned to a merged Arabidopsis–Trichoderma genome to assign the organismal origin of the reads with high confidence (Figure S7A, B). The size distributions were consistent with canonical plant sRNAs (20–24 nt), while the Trichoderma sRNAs showed a narrower peak centered around 21 nt (Figure S7B), suggesting distinct sRNA biogenesis profiles in the two organisms. A de novo sRNA annotation under control conditions identified 6683 loci in Arabidopsis roots and 471 loci in T. atroviride (Figure 5A and Table S23). As expected, the Arabidopsis sRNA loci were dominated by 23–24 nt species, consistent with small interfering (si)RNA-rich populations, whereas the T. atroviride loci were enriched for 20–22 nt sRNAs (Table S23). A subset of loci classified as “N” showed heterogeneous size profiles and likely represent mRNA degradation products rather than functional sRNAs.

Figure 5.

Small RNA function in symbiosis. (A) Counts of de novo sRNA annotations found in Arabidopsis (olive green bars) and Trichoderma atroviride (goldenrod bars). Size describes loci which are majority for a specific sRNA-length, with “N” loci non-specific. (B) Differential expression analysis of sRNA loci, highlighting those which are not DE (black), DE 20–22 sized (red), DE 23–24 sized (gold), and DE canonical miRNAs (blue). Y-axis shows log₂ fold-change, oriented as interaction/non-interaction. (C) Allen et al. score profiles for predicted targets of DE sRNAs, separated by their source organism, target organism, and those that are miRNAs. (D) Heatmap showing the expression profile of putative target transcripts and a targeting sRNA. Expression is normalized to the mean expression within sRNAs and mRNAs separately, shown as log₂ fold-change. Source (sRNA) and target (transcript) are colored by species (Trichoderma: goldenrod, Arabidopsis: olive green). Gene symbols/aliases and miRNA identifiers are shown where available or applicable.

The differential expression analysis revealed contrasting sRNA dynamics between partners. In Arabidopsis roots, most of the differentially expressed sRNA loci were down-regulated upon fungal colonization (74 down, 16 up; Figure 5B and Table S24), with relatively few changes affecting the 20–22 nt size class typically associated with trans-acting regulation. Notably, several conserved miRNA families (including miR156, miR167, miR168, and miR403) were among the repressed loci. In contrast, Trichoderma exhibited a strong induction of sRNA loci upon contact with the plant (119 up, 6 down), almost exclusively within the 20–22 nt range, consistent with the activation of a responsive sRNA program during symbiosis.

To assess whether these sRNAs could contribute to gene regulation within or across species, we performed a target prediction against both Arabidopsis and T. atroviride transcriptomes, testing for putative cis- and trans-kingdom interactions. The predictions were evaluated using the Allen et al. scoring metric [107], where scores ≤3 indicate high-confidence miRNA-like targeting. While most sRNA–mRNA pairs did not meet this stringent threshold, a small subset did (Figure 5C and Table S25). As expected, the Arabidopsis miRNA cis-target predictions showed the strongest scores, serving as an internal positive control.

Applying conservative filters (Allen score ≤3 and differentially expressed targets), we identified 19 high-confidence sRNA–mRNA pairs (Figure 5D). Importantly, 14 of these pairs displayed inverse expression patterns, consistent with functional repression, representing a significant enrichment over random expectation (one-sided Fisher’s exact test, p = 0.049). All of the high-confidence targets corresponded to Arabidopsis genes, approximately half of which were predicted targets of the Trichoderma-derived sRNAs. The cis-target interactions in Arabidopsis primarily involved well-characterized miRNAs (miR156 and miR403) and their established SPL and AGO family targets, respectively [107,108]. The reduced expression of these miRNAs upon T. atroviride colonization coincided with the increased expression of their target mRNAs, consistent with canonical miRNA-mediated regulation. Strikingly, several predicted Trichoderma-to-Arabidopsis trans-targets encode proteins involved in root function and immune responses, including PRL [109], YSL8 [110], and RGI5 [111], suggesting a potential role for fungal sRNAs in modulating host developmental and defense pathways. A minority of the predicted interactions showed concordant expression changes (up–up) (Figure 5D), which may reflect the false positives inherent to a sequence-based prediction. However, in the context of trans-species regulation, such patterns could also indicate the fine-tuning of the induced host genes rather than strict repression.

Collectively, these results indicate that while sRNA-mediated regulation does not globally dominate the transcriptional response during early symbiosis, a limited and highly specific set of sRNA–mRNA interactions, particularly involving fungal sRNAs targeting host genes, may contribute to shaping the key aspects of root development and immune signaling. These findings position sRNAs as a complementary regulatory layer that operate alongside transcriptional and GRN-level control during the Arabidopsis–Trichoderma atroviride interaction and provide testable hypotheses for future functional validation.

3. Discussion

3.1. Regulatory and Metabolic Coordination Underlying the Arabidopsis–Trichoderma atroviride Symbiosis

Numerous transcriptomic studies of plant–Trichoderma interactions have focused on a single partner, most often the plant. On the plant side, these works have linked growth promotion and enhanced resilience to the transcriptional reprograming of defense responses, metabolic reconfiguration, hormone signaling, and the activity of transcription factor families such as ERFs, WRKYs, NACs and MYBs [34,35,37,40,112]. In the specific case of the Arabidopsis thaliana–Trichoderma atroviride association, the transcriptomic literature is more limited, but the available studies report a characteristic activation of hypoxia-related defense programs and ROS-detoxification genes in A. thaliana, together with shifts in T. atroviride carbon metabolism, that favors the utilization of simple carbon sources [32,52]. Despite the differences in experimental design and only a modest overlap in the individual DEGs across these assays, a coherent picture has emerged on the plant side: Trichoderma colonization typically enhances immune and stress signaling while modulating growth-related pathways. Consistent with this, WRKYs represented one of the most uniformly induced TF families in our dataset, contrasting with the mixed or transient expression patterns reported at earlier interaction stages by [113]. This likely reflects the progression from the early perception and signaling phases to a stabilized colonization state, where WRKY-dependent transcription integrates hormonal and developmental pathways in defense priming.

Our RNA-seq data are consistent with, and extend, the previously reported gene expression changes. After 72 h of co-culture, approximately 10–11% of the protein coding genes in both A. thaliana and T. atroviride are differentially expressed, indicating extensive transcriptional reprogramming in each partner. A limitation of this study is that we profile a single early colonization time point. While time-series sampling would be valuable to resolve early recognition (<24 h) and subsequent regulatory transitions, such sampling is constrained in our hydroponic dual-transcriptome design: at early times, the fungal biomass is insufficient for reliable fungal transcript quantification, whereas at later times, hyphal overgrowth increasingly complicates partner separation and introduces strong secondary effects. Accordingly, we selected 72 h as a practical early-stage window, consistent with published studies (24–72 h classified as early interaction) [113], enabling robust bidirectional transcriptional profiling under established but non-overgrown colonization.

In Arabidopsis, KEGG identification and GO enrichments for defense programs, hypoxia/low-oxygen responses, hormone signaling, and ROS-mitigating processes, together with repression of the cell cycle, complex carbon metabolism, organ/root system development and cell wall organization, are closely consistent with previous reports of a trade-off between growth and defense during beneficial Trichoderma interactions [2,9,10,43,114,115,116]. As in previous Arabidopsis–Trichoderma studies [32,52], we found a robust induction of hypoxia-associated programs together with immune and hormone pathways, supporting the view that T. atroviride imposes a mild stress state that triggers controlled developmental reprogramming rather than simple growth inhibition [116]. This transcriptional profile is consistent with phenotypic reports obtained under in vitro Arabidopsis–Trichoderma interaction systems, which describe the inhibition of primary root elongation accompanied by a stimulation of lateral roots and root hairs [24,25,30]. Importantly, these developmental defects have been shown to depend on the experimental context. González-Pérez et al. (2018) [29] demonstrated that primary root inhibition occurs under in vitro conditions and is likely mediated by the accumulation of volatile fungal compounds, such as 6-pentyl-2H-pyran-2-one (6PP). Notably, this inhibitory phenotype was not observed in distance (split-plate) assays or in soil-grown plants, where secreted or volatile metabolites are diluted or dispersed. Accordingly, the transcriptional and regulatory responses discussed here should be interpreted in the context of controlled in vitro colonization, which captures early molecular reprogramming events but may not fully recapitulate developmental outcomes under soil or field conditions. Rather than reflecting generalized growth suppression, our data support a redirection of developmental programs under in vitro interaction settings toward a remodeled root system architecture that optimizes nutrient acquisition while enhancing stress resilience.

On the fungal side, most transcriptomic studies of Trichoderma–plant interactions have focused on how the fungus attenuates a saprotrophic lifestyle and adapts to the rhizosphere. Work on different Trichoderma species and hosts has consistently reported the repression of plant cell wall-degrading enzymes and CAZymes, together with the reduced expression of genes involved in complex polysaccharide degradation and broad-spectrum secretory activity [19,32,38,44,117]. Consistent with these transcriptional shifts, T. atroviride showed a broad down-regulation of secreted proteins, CAZymes, and elicitor-like genes, including the well-known Epl1 [61,62]. While Epl1 homologs have been shown to act as early-stage elicitors of plant immunity (e.g., Sm1 in T. virens [118]), its repression at 72 h in our system suggests a transition from early activation to a stable colonization phase in which T. atroviride minimizes immune stimulation.

The KEGG and GO analyses highlight a pronounced reconfiguration of carbohydrate, amino acid, lipid, glycan and energy metabolism, together with multiple membrane transport categories, rather than a generic stress response. In particular, the induction of nitrate-assimilation and amino acid-related pathways, coupled with the reduced expression of MAPK signaling, ER protein processing and cell-cycle functions, is consistent with a shift from a predominantly saprotrophic, secretory program toward a nutrient-recycling, energy-efficient physiology adapted to the root environment. These results agree with previous transcriptomic and metabolomic studies of Trichoderma–plant interactions, which likewise infer a transition toward simple carbon utilization and a restrained, plant-compatible lifestyle [32,38,60,119]. Most of these observations were made under experimental conditions in which Trichoderma was grown in MS-based media, similar to the setup used here, underscoring that the observed regulatory shifts are driven by host-derived cues rather than by a transition from complex to simple carbon substrates. In this context, fungal growth in MS medium without plants represents a controlled, host-free baseline rather than a fully saprotrophic state, allowing plant-induced regulatory reprogramming to be isolated without the confounding effects of carbon source complexity. Thus, the repression of saprotrophy-associated pathways observed here should be interpreted as a relative shift toward a plant-adapted, endophytic-like transcriptional program, rather than as a comparison to classical saprotrophic growth on lignocellulosic substrates. Future studies incorporating complex polymers or soil-derived matrices will be valuable to further dissect how environmental carbon availability modulates fungal regulatory strategies during symbiosis.

3.2. A Gene Regulatory Network Resource for the Arabidopsis–Trichoderma atroviride Interaction

By constructing GRNs for A. thaliana and T. atroviride with the same motif-scanning plus GENIE3 pipeline, we obtained a mechanistically interpretable and predictive model that links transcriptional changes to candidate regulators. Several Arabidopsis GRNs based on ChIP-/DAP-seq and perturbation data are available [67,120,121,122,123], but they differ in coverage, tissues and conditions. Reconstructing both networks de novo with the same strategy therefore allowed a comparable analysis of the regulatory architecture in the two interacting organisms. The AUROC/AUPR values obtained using the ConnecTF gold standard [67] or a gold standard consisting of TF–target interactions derived from DAP-seq data [65], indicate that the chosen GENIE3 threshold recovers the known Arabidopsis interactions from previous genome-scale models while extending the coverage to TFs without a direct experimental characterization with an above-chance performance [124,125]. Moreover, moderate AUPR values similar to the ones found in this work have been reported for other Arabidopsis GRNs and have proven effective for identifying biologically meaningful regulators and modules [120,126]. Topologically, the Arabidopsis GRN model shows a short characteristic path length (3.12) and a small diameter (8) despite a very low edge density (0.0062), together with a non-trivial average clustering coefficient (0.17), a combination consistent with the Watts–Strogatz-type small-world organization described for other biological regulatory networks [70,71,127,128].

Intersecting the RNA-seq data onto these reference GRNs yielded interaction-specific networks that capture most of the transcriptional response (AiGRN: 87% of plant DEGs; TiGRN: 81.4% of fungal DEGs), extending the published T. atroviride network into the context of root colonization [63]. It is important to note that root colonization by T. atroviride is spatially heterogenous, involving distinct integration dynamics at root tips, elongation zones and differentiated tissues. While bulk root RNA-seq does not resolve cell type- or zone-specific transcriptional programs, our goal here was to establish a systems-level, bidirectional regulatory model capturing the dominant transcriptional and regulatory coordination between Arabidopsis thaliana and Trichoderma atroviride. Bulk transcriptomics remains a widely used and tractable approach for integrating gene expression data with gene regulatory network inference across interacting organisms, enabling the identification of central regulators and regulatory modules that operate at the whole-root level. Future work combining GRN inference and spatially resolved or single-cell transcriptomic approaches would provide valuable additional resolution to dissect zone- or cell type-specific regulatory programs during colonization. These context-specific GRNs thus provide a compact systems-level scaffold for interpreting DEG patterns in terms of regulatory topology, but all of the edges remain computational predictions rather than confirmed binding or causal interactions. Consequently, the networks should be viewed as structured hypotheses that prioritize regulatory scenarios for future functional validation, rather than definitive maps of transcriptional control.

3.3. A Hypoxia-Centered Module Links Stress Cues with Developmental Control

The hypoxia-associated regulatory subnetwork uncovered in this work supports the view that low-oxygen signaling is not an isolated stress response during Trichoderma colonization, but rather a central organizing axis that coordinates metabolic adjustment, immune response, and developmental control in Arabidopsis roots. The co-occurrence of genes involved in nitrogen and amino acid metabolism, redox homeostasis, and cell wall remodeling indicates a transition toward metabolic states compatible with limited oxygen availability and altered resource allocation, consistent with previous reports linking hypoxia signaling to root developmental plasticity and stress adaptation [72,129]. Importantly, the presence of immune-related components, including TIR-NBS-LRR genes and membrane-associated signaling factors such as FLOT1, suggests that hypoxia-responsive transcriptional programs intersect directly with defense pathways, reinforcing a model in which hypoxia acts as a modulatory signal for immunity rather than a purely metabolic constraint.

At the regulatory level, the partitioning of the hypoxia subnetwork into multiple interconnected TF clusters reveals a distributed control architecture. ERF-dominated modules centered on ERF73/HRE1 provide a canonical hypoxia-responsive backbone [76], consistent with what has been previously described for ERF71/HRE2 using a co-expression network analysis of the A. thaliana–T. atroviride interaction [52]. Moreover, WRKY-enriched clusters integrate immune and hormone-associated signaling, and DOF- and G2-like-related regulators link nutrient sensing to transcriptional reprogramming. The identification of DOF1 as a highly connected hub highlights its potential role as a key integrator of nitrogen status, stress signaling, and hypoxia-responsive gene expression during root colonization. Similarly, the prominence of NTM1 suggests that cytokinin-related growth control interacts with hypoxia-associated regulatory circuits, supporting a model in which growth restraint and adaptation are actively coordinated rather than passively imposed under low-oxygen conditions.

The network motif analysis further refines this interpretation by revealing a regulatory logic dominated by feed-forward loops and bifans. The abundance of feed-forward loops is consistent with a system designed to buffer transient or noisy signals, ensuring that hypoxia-responsive transcriptional states are activated only under sustained conditions associated with fungal colonization. The bifan motifs, in turn, point to extensive combinatorial regulation, allowing multiple TFs from distinct signaling pathways to converge on shared targets. Together, these features indicate that hypoxia signaling during the Arabidopsis–T. atroviride interaction is implemented through a robust, multilayered regulatory architecture that integrates metabolic, hormonal, immune, and developmental cues. Rather than acting as a simple stress response, hypoxia emerges as a systems-level signal that helps coordinate root adaptation to microbial colonization.

Compared with previous analyses of Trichoderma-induced hypoxia responses, which were based on pathway enrichment [32,52], our GRN-based approach provides a mechanistic view of how specific TF families and hubs may coordinate hypoxia, ROS detoxification, and developmental adjustments during colonization. However, we agree that a definitive demonstration of causality between reduced oxygen availability and an enhanced immune or developmental response would require complementary physiological and biochemical measurements, such as root oxygen profiling, the study of the stability of hypoxia-responsive TFs, or the direct assessment of anaerobic metabolic fluxes.

3.4. Root Developmental Regulatory Network Integrates Environmental and Hormonal Signals During Arabidopsis–Trichoderma Interaction

The root development-associated GRN reveals a highly TF-dense and combinatorial regulatory architecture, indicating that Trichoderma-induced root remodeling is implemented through coordinated transcriptional control rather than through a small number of dominant regulators. The strong representation of TFs previously linked to root development, together with targets involved in meristem maintenance, lateral root initiation, root hair growth and nutrient sensing (e.g., RGF–RGI signaling [130,131], NRT1.1 and ANR1 [88,132], COW1, CEP and CLE peptides), supports the idea that T. atroviride reconfigures endogenous developmental programs to reshape the root system architecture.

Our community detection uncovered four interconnected regulatory modules integrating developmental, nutritional, immune and hormonal signals. Central hubs such as OBP3, DOF1, NTM1 and HHO3/NIGT1.1 link nutrient sensing and hormonal regulation with developmental outputs, while WRKY-, ERF-, NAC- and GATA-containing clusters highlight extensive cross-talk between stress, immune and growth pathways. Rather than acting solely as defense regulators, WRKY TFs are embedded in mixed modules that couple immunity with developmental regulation, consistent with a growth–defense coordination model at the root level. The network motif analysis further supports this interpretation. The strong enrichment of bifan motifs indicates extensive co-regulation by TF pairs, enabling the integration of multiple inputs such as nutrient status, microbial cues and hormonal signals. The high number of feed-forward loops suggests temporal filtering, ensuring that developmental programs are activated only after a sustained regulatory input, while the relative scarcity of single-input modules points to a distributed rather than a hierarchical control. Together with the hypoxia subnetwork, these results define a shared regulatory logic in which ERF-centered stress and metabolic signaling interfaces with nutrient-responsive and developmental TF modules. This integrated architecture provides a mechanistic basis for how Arabidopsis coordinates metabolic adjustment and root architectural remodeling during Trichoderma colonization, allowing flexible yet robust developmental responses to a beneficial fungal partner.

3.5. Regulatory Logic of Fungal Metabolic Adaptation During Symbiosis

Our network analyses indicate that T. atroviride deploys a tightly coordinated regulatory program to integrate membrane transport and primary metabolism during its interaction with Arabidopsis thaliana. Both of the processes emerge as central components of the fungal response and are controlled by a largely overlapping set of transcription factors, suggesting that nutrient acquisition, signaling at the host–fungus interface, and internal metabolic adjustment are not regulated independently but instead form an integrated regulatory module.

The transmembrane transport network highlights the importance of transporter-mediated exchanges during colonization. The strong representation of MFS transporters, ABC transporters and ion carriers supports a model in which T. atroviride dynamically modulates the uptake of host-derived nutrients and ions while adjusting intracellular homeostasis. The enrichment of these transporters is also consistent with comparative genomic analyses showing the expansion of these families in Trichoderma and other mycoparasitic or root-associated fungi, where they support nutrient uptake, ion homeostasis, detoxification and resistance to antifungals [133,134,135]. The presence of an auxin efflux carrier further suggests that transport processes may also contribute to interspecies signaling, potentially influencing plant developmental responses. Importantly, the regulatory hubs identified in this network, CrzA, RME1, NsdC, RPN4, TAH-3 and PacC, are conserved fungal regulators known to integrate stress sensing, nutrient status, pH signaling and growth control in other species, consistent with a coordinated physiological adaptation to the host environment. Of note, these functional roles are inferred based on a conserved domain architecture and prior studies in other fungi, thus their precise regulatory functions in T. atroviride should be viewed as hypotheses supported by network context rather than direct functional validation. Future genetic or biochemical studies will be required to experimentally resolve the species-specific regulatory mechanisms. Furthermore, the integration of metabolomics and secretome analyses will be key to support our conclusions and further understand the communication established between the fungus and the plant.

Our motif analysis revealed that this transport network is governed by a highly combinatorial regulatory architecture, characterized by abundant bifan and feed-forward loop motifs. This structure is well suited for integrating multiple environmental and host-derived inputs, filtering transient signals, and stabilizing transcriptional outputs. Such an organization likely confers robustness and flexibility, enabling T. atroviride to fine-tune transporter expression in response to fluctuating nutrient availability and signaling cues at the root interface. In contrast, the primary metabolism subnetwork exhibited a markedly simpler motif architecture despite sharing most of its upstream regulators with the transport network. The reduced number of motifs and the absence of feed-forward or feed-back loops suggest a more direct and conservative regulatory strategy, consistent with the need to maintain stable core metabolic functions during symbiosis. Rather than dynamic reprogramming, primary metabolism appears to be adjusted through a coordinated but comparatively straightforward control by a shared regulatory core, ensuring metabolic continuity while resources are redirected toward the host interaction.

Together, these findings support a model in which T. atroviride employs a hierarchical division of regulatory labor: dynamic, highly integrated control for transport processes that interface directly with the host, and more stable regulation of internal metabolic pathways. Notably, this fungal regulatory logic mirrors the key features observed in the Arabidopsis GRNs described here, where hypoxia, nutrient sensing and developmental responses are likewise governed by combinatorial, ERF- and WRKY-centered modules enriched in bifans and feed-forward loops. Thus, both of the partners appear to rely on distributed, multi-input regulatory architectures to coordinate metabolic adaptation and developmental outcomes during symbiosis.

3.6. sRNAs as a Complementary Regulatory Layer in Arabidopsis–Trichoderma Interaction

Our sRNA profiling revealed that sRNAs contribute an additional, more targeted regulatory layer during the Arabidopsis thaliana–Trichoderma atroviride interaction, operating alongside the transcriptional and network-level regulatory programs described above. In contrast to the extensive rewiring observed at the mRNA and GRN levels, sRNA-mediated regulation appears selective rather than global, consistent with a role in fine-tuning specific developmental and immune outputs rather than driving broad transcriptional reprogramming.

The two partners displayed markedly asymmetric sRNA responses. The Arabidopsis roots exhibited a net reduction in sRNA abundance upon colonization, including the repression of conserved miRNA families such as miR156 and miR403. These miRNAs regulate key transcriptional regulators and RNA silencing components, and their down-regulation is consistent with the up-regulation of developmental and stress-related gene expression programs that accompany root remodeling and immune priming. In contrast, T. atroviride showed a strong induction of 20–22 nt sRNAs upon contact with the plant, suggesting the activation of a responsive sRNA program during symbiosis.

Although the high-confidence sRNA–mRNA interactions were limited in number, the enrichment of inverse expression patterns among the predicted targets supports their functional relevance. Notably, approximately half of the strongest predicted interactions involved fungal sRNAs targeting Arabidopsis genes, including regulators of root development and defense. This finding aligns with the growing evidence that beneficial fungi, similar to pathogens, may deploy sRNAs to modulate host gene expression, albeit in a more restrained and context-dependent manner [136,137]. In this scenario, fungal sRNAs may contribute to shaping host developmental and immune states that favor stable colonization rather than the deployment of defense mechanisms.

Importantly, the limited scope of the sRNA-mediated regulation observed here contrasts with the extensive combinatorial control uncovered in Arabidopsis GRNs governing hypoxia responses and root development. This suggests a hierarchical regulatory organization in which transcriptional networks establish the dominant physiological and developmental states while sRNAs provide an additional layer of specificity and robustness, potentially buffering or refining key regulatory nodes. In T. atroviride, the induction of sRNAs parallels the motif-rich architecture of transport and signaling networks, further supporting a role for sRNAs in modulating adaptive responses at the host–fungus interface.

Together, these results position sRNAs as modulators rather than primary drivers of symbiotic reprogramming, complementing transcription factor-centered regulatory networks in both of the partners. While experimental validation will be required to establish the mechanistic roles of the individual trans-species sRNAs, our analysis identifies a restricted set of high-confidence candidates and provides a framework for integrating post-transcriptional regulation into systems-level models of plant–fungus symbiosis.

4. Materials and Methods

4.1. Plant and Fungal Material and Growth Conditions

Arabidopsis thaliana (ecotype Columbia-0, Col-0) and Trichoderma atroviride (IMI206040) were used in this study. A. thaliana seeds were surfaced-sterilized and sown over an inert nylon mesh (60–250 ASTM Nytex mesh, SEFAR, Heiden, Switzerland) supported by a plastic support inside a sterile plastic tray (PhytatrayTM II, Sigma-Aldrich, St. Louis, MI, USA, catalog number P5929). The trays contained 100 mL of basal 0.5X Murashige and Skoog (MS) plant media (Phytotechnology Laboratories, Lenexa, KA, USA, catalog number M519) supplemented with 0.05% (w/v) sucrose and 2.5 mM MES, pH 5.7. This configuration ensured that plants were not flooded and allowed normal gas exchange for the aerial parts of the plants, preventing flooding-associated stress responses. Plants were grown for 11 days at 22 °C with a regime of 16 h of light (100 µE m⁻² s⁻¹ light intensity) and 8 h of darkness. T. atroviride conidiophores were grown on potato dextrose agar (PDA) medium (Becton Dickinson and Co., Franklin Lakes, NJ, USA, catalog number 213400) in axenic conditions at 22 °C under the same light regime as the plants. Conidiophores were harvested from 7-day cultures, once the T. atroviride mycelia developed a green coloration [138].

4.2. Arabidopsis–Trichoderma Interaction Assay

Approximately 50,000 freshly collected T. atroviride conidiophores were inoculated in 100 mL of 0.5X MS medium supplemented with 0.05% (w/v) sucrose and 2.5 mM MES (pH 5.7). Plastic supports containing Arabidopsis seedlings (11 days post-sowing) were transferred to trays containing the fungal inoculum and were co-cultivated for 72 h under long-day conditions. This time was selected to capture an established yet balanced stage of colonization, at which sufficient fungal material can be recovered for analysis without extensive fungal overgrowth of the root system. Control treatments included seedlings transferred to identical media without T. atroviride and T. atroviride cultures grown in 0.5X MS medium without plants. After 72 h of co-culture, root and mycelium samples were collected separately for both the control and interaction conditions. All assays were performed in three independent biological replicates.

4.3. RNA Extraction, Library Preparation, and Sequencing

Frozen root and mycelia samples were ground using a Tissuelyser II (QIAGEN, Venlo, The Netherlands). Total RNA was extracted using the mirVana™ miRNA Isolation Kit (Thermo Fisher Scientific Inc., Waltham, MA, USA, catalog number AM1560) following the manufacturer’s instructions. RNA quantity and purity were determined using a NanoQuant Infinite 200 PRO (Tecan Trading AG, Männedorf, Switzerland) and a Qubit 3 fluorometer (Thermo Fisher Scientific Inc., Waltham, MA, USA). RNA integrity was assessed using an Agilent BioAnalyzer 2100 instrument (Agilent Technologies, Inc., Santa Clara, CA, USA), and RNA samples with RIN ≥ 7.5 were selected for further processing. mRNA libraries were prepared using the TruSeq^® Stranded RNA Library Preparation Kit (Illumina Inc., San Diego, CA, USA, catalog number 20020594), and sRNA libraries were prepared using the TruSeq^® Small RNA Library Preparation Kit (Illumina Inc., San Diego, CA, USA, catalog number RS-200-0012). Sequencing was performed on an Illumina^® HiSeq2000 sequencer using 150 bp paired-end mode for mRNA libraries and 50 bp single-end mode for sRNA libraries.

4.4. Bioinformatic Processing of the Libraries

Read quality was assessed with FastQC [139] version 0.12.0. For mRNA libraries, low-quality reads and adapter sequences were filtered out using BBDuk [140] version 38.90 with parameters trimq = 25, qtrim = rl, minavgquality = 25, hdist = 2 and k = 21 tpe tbo, as previously described [63]. The A. thaliana and T. atroviride filtered reads were pseudo-aligned to the Araport11 [141] or Trichoderma atroviride v2.0 [142] transcriptomes, respectively, using Kallisto (version 0.44.0) [143] under default settings. Transcript abundances were summarized to gene-level counts using the tximport package (version 1.1.2) [144], which aggregates reads from all annotated transcript isoforms corresponding to the same gene locus. Differential expression analyses were performed on gene-level abundance matrices using DESeq2 [145] version 1.48.1 applying a Wald test and Benjamini-Hochberg correction [146] (adjusted p-value < 0.01 and |log₂ fold-change| > 1).

sRNA libraries were processed and annotated using the YASMA pipeline (version 1.1.0) [147]. Adapters were automatically detected with YASMA and reads were trimmed using Cutadapt (version 5.2) [148] with the following command: ‘cutadapt -a [adapter] --minimum-length 15 --maximum-length 50 -O 4 --max-n 0 --trimmed-only’. Reads were aligned to a merged genome of Arabidopsis and T. atroviride genomes. For this, we used ‘YASMA align’ with default settings, which utilizes bowtie1 [149] as the alignment engine while following the weighted multi-mapper placement described in [150]. Merged alignments were then split to separate the organisms. Arabidopsis and T. atroviride alignments were then annotated separately using ‘YASMA tradeoff’ with default settings, using control conditions (ath-root, tat-mycelium) to identify sRNA loci. Counts tables were generated using ‘YASMA count’.

Differential expression analysis of sRNA-loci was performed using DEseq2 [145] version 1.48.1, using raw counts from the annotations with no locus filtering. Comparisons always compared the change in interaction conditions (Arabidopsis and Trichoderma) to conditions with an organism alone (Arabidopsis or Trichoderma). Differentially expressed loci were identified based on an adjusted p-value threshold of 0.1 and |log₂ fold-change| > 1.

4.5. Target Prediction

Prediction of sRNA-transcript targeting relationships was performed using GSTAr.pl (version 1.0) (https://github.com/MikeAxtell/GSTAr, accessed on 5 January 2026), which is based on RNA–RNA duplex models from the RNAplex (version 2.4.14) module of the ViennaRNA modeling package [151]. Targets were predicted only for sRNAs which were identified as DE, with a merged Arabidopsis–Trichoderma atroviride transcriptome as the subject. Prediction quality was assessed with Allen et al. scores [107], where lower scores are better. Global analysis of targets utilized the best target relationship for each sRNA (minimum score). Predictions were further filtered for those with an Allen et al. score of 3 or less and for those where the sRNAs and targets are both DE.

4.6. Functional Enrichment Analysis

Functional annotation for T. atroviride genes was obtained from [63], FungiFun2 and FungiDB (release 68) [152,153]. A. thaliana annotations were retrieved from The Arabidopsis Information Resource (TAIR, www.arabidopsis.org, accessed on 20 March 2024) [154]. Gene Ontology (GO) enrichment was performed using BiNGO (version 3.0.3) [155], and KEGG pathway reconstruction was conducted using BlastKOALA [156] version 3.1 and KEGG mapper [157] version 5. Gene Set Enrichment Analysis (GSEA) [158] was applied to identify overrepresented GO biological processes using a hypergeometric test with Benjamini and Hochberg false discovery rate (FDR) correction (adjusted p-value < 0.05).

4.7. CAZyme, Elicitors and Effectors Prediction Pipeline

Putative genes encoding CAZymes (including GH, GT, PL, CE, AA, and CBM families) in T. atroviride were identified using the local implementation of dbCAN version 3.0, integrating the dbCAN CAZyme domain (HMMER), short conserved motif (Hotpep), and CAZy (DIAMOND) databases [159]. Independently, the secretome was predicted using SignalP 5.0 [160], excluding transmembrane proteins with TMHMM 2.0 [161], and subsequently screened for candidate effectors using EffectorP 3.0 [162]. To ensure that the functional categories were mutually exclusive and to avoid redundancy in the downstream analysis, the lists of predicted effectors and putative elicitors were cross-referenced against the annotated CAZymes. This step confirmed that no single protein was assigned to multiple functional groups in the final dataset.

4.8. Inference of TF Binding Sites

A list of 1717 A. thaliana TFs was obtained from the Plant Transcription Factor Database (PlantTFDB, https://planttfdb.gao-lab.org/, accessed on 20 March 2024) [64]. Position Weight Matrices (PWMs) for these TFs were obtained from the Cis-BP version 2.0 database [65] including both experimentally determined and inferred PWMs. Promoter regions (2 kb upstream of the transcription start site (TSS)) were extracted from the Araport11 genome annotation using BEDTools [163] version 2.31.0. Each PWM was subsequently scanned across this set of promoter sequences with the Find Individual Motif Occurrences (FIMO) tool of the MEME suite version 4.11.2 [66] using default settings and applying a p-value < 1 × 10⁻⁴, as previously reported [63].

4.9. Regulatory Inference Using GENIE3

A compendium of A. thaliana RNA-seq libraries was assembled from the NCBI Sequence Read Archive (SRA). A total of 280 Illumina-based BioProjects were downloaded, yielding 5220 libraries in total. Raw reads were trimmed and filtered with BBDuk version 38.90 to remove adapter sequences and low-quality reads (average quality q < 20 and length < 35 bases). Libraries containing fewer than 5 million reads after filtering were discarded to reduce noise in downstream analyses. The remaining libraries were pseudoaligned to the Arabidopsis reference transcriptome (Araport11) using Kallisto (version 0.44.0), and transcript-level estimates were summarized to gene-level counts with the Bioconductor package tximport (version 1.1.2). Libraries with less than 25% of the coding genes detected were further excluded, resulting in a final dataset of 3493 libraries. For each retained library, gene expression levels were normalized as transcripts per million (TPM).

The normalized gene expression matrix, together with the list of 1717 Arabidopsis TFs, were provided as input for Gene Network Inference with Ensemble of Trees (GENIE3) [124]. Regulatory edges were inferred using a fixed random seed (123) to ensure reproducibility, the default setting K = sqrt, and 1000 decision trees to reduce stochastic variability and obtain stable edge-weight estimates. The resulting ranked TF–target edges were used to construct subnetworks corresponding to the top 10%, 20% and 30% of interactions, in line with previously published GENIE3-based networks [63,164,165]. Each of these three subnetworks was then intersected with the TF–target predictions from FIMO, retaining only regulatory edges supported by both approaches for subsequent evaluation.

4.10. Evaluation of Network Performance and Generation of an Arabidopsis GRN Model

Experimentally validated TF–target interactions were retrieved from ConnecTF [67], which compiles A. thaliana data generated by ChIP-seq, DAP-seq and TARGET assays. These curated interactions were assembled into a gold standard GRN that served as a benchmark. Each candidate network was evaluated against this benchmark by calculating the area under the receiver operating characteristic curve (AUROC) and the area under the precision recall curve (AUPR), as described in [126]. The model achieving the highest AUROC and AUPR values was selected as the reference GRN, while the other networks were excluded from further analyses. This network was further benchmarked against a gold standard GRN containing experimentally derived TF–target interactions obtained from the Plant Cistrome Database, following the same procedure.

4.11. Network Visualization and Topological Analysis

Networks were visualized in Cytoscape version 3.8 [166], and network topology analyses were conducted using the Cytoscape NetworkAnalyzer tool. Cluster analysis was performed using ClusterMaker2 [167] using the Community Clustering (GLay) algorithm [75].

4.12. Network Topology and Regulatory Motif Analysis

The topological structure of the inferred Gene Regulatory Network was analyzed to identify key local regulatory patterns. Motif detection was performed using custom scripts written in Python (version 3.14.0) utilizing the NetworkX library [168] version 3.6.1. The network was systematically screened for four specific structural motifs feed-forward loops (FFLs), feed-back loops restricted to short cycles of 2 or 3 nodes, bifans, and single-input modules (SIMs). Only directed interactions were considered for motif detection.

4.13. Data Availability

Publicly available RNA-seq datasets used for GENIE3 network inference are listed in Table S1. Newly generated RNA-seq and sRNA-seq data was deposited in the SRA under BioProject accession PRJNA1090539.

5. Conclusions

By integrating RNA-seq with gene regulatory network modeling, this study moves beyond pathway-level descriptions and provides a predictive, systems-level view of regulatory programs underlying the Arabidopsis thaliana–Trichoderma atroviride interaction. Consistent with previous work on beneficial plant–fungal associations, we identify recurring hallmarks of the interaction, including host hypoxia- and defense-related transcriptional activation, extensive root developmental reprogramming, and fungal shifts in membrane transport and primary metabolism. Within this framework, the Arabidopsis process-focused subnetworks highlight how hypoxia, stress and developmental responses converge on shared transcription factor hubs, while the fungal networks reveal a conserved, orthology-supported regulatory core coordinating transport and metabolic functions.

At the same time, our conclusions are subject to important limitations. The transcription factor identities and the functional annotations in T. atroviride are inferred from homology, and the regulatory edges in both of the organisms are predicted from motif occurrence and expression co-variation rather than direct binding or perturbation assays. Accordingly, the networks and the regulatory hubs identified here should be interpreted as testable hypotheses rather than definitive causal models. Nevertheless, the fact that these networks are centered on deeply conserved transcription factor families and recurrent regulatory motifs suggests that they capture a core regulatory logic that is likely shared across Arabidopsis accessions, Trichoderma species, and interaction contexts, even if the magnitude and the timing of individual responses vary. Future work combining targeted genetic perturbations, chromatin-based TF binding assays, and functional interaction experiments across diverse genotypes and environments will be essential to validate the individual regulators, refine the proposed regulatory edges, and establish the broader applicability of the networks proposed in this work.

Despite these constraints, the GRN models presented here provide a foundation for hypothesis-driven experimentation and offer candidate regulatory nodes of potential translational relevance. In plants, ERF-, WRKY-, NAC- and DOF-centered modules may represent key points for modulating root architecture, nutrient use and stress resilience, while conserved fungal regulators such as CrzA, PacC, RME1 and NsdC define targets for future efforts to optimize Trichoderma strains for beneficial interactions. More broadly, this work illustrates how GRN-based approaches can inform rational strategies to improve plant–microbe interactions while remaining grounded in experimentally testable predictions.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/plants15040533/s1.

Author Contributions

Conceptualization, E.S. and E.A.V.; methodology, E.S., L.M., N.R.J., T.C.M., C.O.-Y. and J.C.; formal analysis, E.S., L.M. and T.C.M.; investigation, E.S., J.C., N.R.J., T.C.M., J.D.F., L.M. and D.P.-S.; data curation, E.S., C.O.-Y., L.M. and T.C.M.; writing—original draft preparation, E.S. and E.A.V.; writing—review and editing, E.S., N.R.J., C.O.-Y., J.D.F., T.C.M., J.C., D.P.-S., L.M. and S.R.-C.; visualization, E.S., J.D.F. and L.M.; supervision, E.A.V., N.R.J., C.O.-Y. and S.R.-C. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

The mRNA-seq and sRNA-seq datasets generated and analyzed during the current study are available in the NCBI Sequence Read Archive (SRA) repository, accession PRJNA1090539, available at https://www.ncbi.nlm.nih.gov/sra/PRJNA1090539 (accessed on 10 January 2026). All the other data generated or used during this study are included in this published article and its Supplementary Information Files.

Conflicts of Interest

The authors declare no conflicts of interest.

Funding Statement

This work was supported by the Agencia Nacional de Investigación y Desarrollo (ANID)-Millennium Science Initiative Program (Millennium Institute for Integrative Biology iBio) ICN17_022 to E.A.V., N.R.J., and C.O-Y.; the ANID-Millennium Nucleus in Data Science for Plant Resilience NCN2024_047 to E.A.V.; the ANID-Fondo de Desarrollo Científico y Tecnológico (FONDECYT) 11240968 to C.O-Y. and 3250452 to D.P-S.; and the Beca Doctoral Universidad Mayor to E.S. and L.M.