Stress Knowledge Map: A knowledge graph resource for systems biology analysis of plant stress responses

Stress Knowledge Map (SKM; https://skm.nib.si) is a publicly available resource containing two complementary knowledge graphs that describe the current knowledge of biochemical, signaling, and regulatory molecular interactions in plants: a highly curated model of plant stress signaling (PSS; 543 reactions) and a large comprehensive knowledge network (488 390 interactions). Both were constructed by domain experts through systematic curation of diverse literature and database resources. SKM provides a single entry point for investigations of plant stress response and related growth trade-offs, as well as interactive explorations of current knowledge. PSS is also formulated as a qualitative and quantitative model for systems biology and thus represents a starting point for a plant digital twin. Here, we describe the features of SKM and show, through two case studies, how it can be used for complex analyses, including systematic hypothesis generation and design of validation experiments, or to gain new insights into experimental observations in plant biology.


INTRODUCTION
The already apparent effects of climate change on agriculture (Shukla et al., 2022), the spread of pests into new regions (Garrett, 2013;IPPC Secretariat, 2021), and rapid population growth (United Nations Department of Economic and Social Affairs Population Division, 2022) present immediate challenges to global food security (Steinwand and Ronald, 2020).Projections show that an increase of up to 75% in crop production is required to meet the 2050 demand (Hunter et al., 2017).This can be achieved with yield improvements through development of stress-resilient crops, a process that requires a holistic understanding of the effects of stressors on plants.The rapid development of modern ''omics'' technologies enables the generation of large and complex datasets characterizing system-wide responses.To understand the biological meaning of these large-scale datasets and generate meaningful hypotheses, contextualization within current knowledge is needed.We have assembled an integrated resource for plant signaling, Stress Knowledge Map (SKM; https://skm.nib.si), which provides a single, up-to-date entry point for plant-response investigations.
SKM integrates knowledge on plant molecular interactions and stress-specific responses from a wide diversity of sources, combining recent discoveries from journal articles with knowledge present in established resources such as KEGG (Kanehisa et al., 2016), STRING (Szklarczyk et al., 2023), MetaCyc (Caspi et al., 2016), and AraCyc (Mueller et al., 2003).SKM extends other aggregated resources (listed in Supplemental Table 1), including the heterogeneous knowledge graphs of KnetMiner (Hassani-Pak et al., 2021), Biomine Explorer (Podpe can et al., 2019), and ConsensusPathDB (Herwig et al., 2016), in that it enables conversion of biochemical knowledge to diverse mathematical modeling formalisms and integration with multi-omics experiments, in addition to enabling interactive exploration of current knowledge that is constantly reproducibly updated.SKM is a versatile resource that assists diverse users, from plant researchers to crop breeders, in investigating current knowledge and contextualizing new datasets in existing plant research.A number of tools have been developed within the SKM environment to support this aim and enable efficient linking to complementary tools.

RESULTS
SKM is a resource that combines two knowledge graphs resulting from the integration of dispersed published information on current biochemical knowledge: the Plant Stress Signaling model (PSS) and the Comprehensive Knowledge Network (CKN) of plant molecular interactions.SKM enables interactive exploration of its contents and represents a basis for diverse systems biology modeling approaches, from network analysis to dynamical modeling.

The Plant Stress Signaling model
PSS is an ongoing endeavor to assemble an accurate and detailed mechanistic model of plant stress signaling by extracting validated molecular interactions from published resources (Miljkovic et al., 2012;Ram sak et al., 2018).Currently, PSS covers the complete stress response cascade within the plant cell (Figure 1), initiating with abiotic (heat, drought, and waterlogging) and biotic stressors (extracellular pathogens, intracellular pathogens, and necrotrophs; Layer 1).Perception of these stressors through diverse receptors (Layer 2) initiates Ca 2+ , reactive oxygen species (ROS), and MAPK signaling cascades, as well as phytohormone biosynthesis and signaling pathways (abscisic acid [ABA], jasmonic acid [JA], salicylic acid [SA], ethylene, auxin, gibberellins, and cytokinins; Layer 3).These translate perception into a cellular response, resulting in activation of processes that execute protection against stress (Layer 4).Within and across these layers, relevant transcriptional (transcription factors known to act downstream of phytohormones) and posttranscriptional (e.g., small-RNA-transcript regulation known to participate in stress signaling) regulation is included.To capture the relationships between stress responses and growth and development, PSS also contains the major known regulators of growth (target of rapamycin signaling and the above-mentioned hormones) and major primary metabolism processes.Finally, tuberization signaling from potato is included as an example for evaluating potential effects on crop yields.
PSS is based primarily on the model plant Arabidopsis thaliana and also contains pertinent information from several crop species, predominantly potato (Solanum tuberosum).It currently includes 1425 entities and 543 reactions, a substantial update from the preceding model with 212 entities and 112 reactions (Ram sak et al., 2018).PSS entities include genes and gene products (proteins, transcripts, small RNAs), complexes, metabolites, and triggers of plant stress.Genetic redundancy (Cusack et al., 2021) is incorporated using the concept of functional clusters-groups of genes (possibly across species) that are known to mediate the same functions.Functional clusters can be used to obtain a list of candidate genes linked to a particular use case.For further analysis, individual genes within the functional clusters can be prioritized on the basis of context-specific experimental data (e.g., results of transcriptomics or proteomics analysis).Interactions between these entities include protein-DNA (e.g., transcriptional regulation), non-coding RNA-transcript, and protein-protein interactions, as well as enzymatic catalysis and transport reactions.The majority of these interactions were compiled from peerreviewed articles with targeted experimental methodology, giving them a high degree of confidence.PSS also contains relevant signaling-associated pathways from KEGG (Kanehisa et al., 2016) and AraCyc (Mueller et al., 2003).

The Comprehensive Knowledge Network
Complementary to PSS, CKN is a large-scale condition-agnostic assembly of current knowledge, offering broader insights into not only stress signaling but also any other plant process.CKN is a network of experimentally observed physical interactions between molecular entities, encompassing protein-DNA interactions, interactions of non-coding RNA with transcripts, posttranslational modifications, and protein-protein interactions (Table 1) in A. thaliana.Here we present an update of the previous version, which involved 20 012 entities and 70 091 interactions (Ram sak et al., 2018), to the current version, which provides 30% more entities (26 234 entities) and an almost seven-fold increase in the number of molecular interactions (488 390 unique interactions, Table 1).Entities in CKN include 24 829 of 38 202 genes registered in Araport11 (Cheng et al., 2017).
During the update, only STRING was found to have been altered since 2018 (updated to v.11.5 in 2021) and was thus reintegrated.In addition, nine novel sources of information were added, bringing the total number of sources integrated into CKN to 25 (Supplemental Table 2).Interactions are annotated with the interaction type and whether the interaction has directionality (e.g., undirected binding vs. transcription factor regulation).A ranking system for the interaction reliability (Table 1 legend) enables researchers to evaluate the biological credibility and relevance of individual interactions.CKN includes all relevant reactions from PSS to enable direct comparison of results obtained through both networks.

SKM environment and features
To enable accessibility and exploitation of the resources within SKM, we have developed an encompassing environment (Figure 2).The main features include content exploration and visualization, access to various export formats, and the ability to contribute improvements based on novel biological knowledge.The SKM webpage is publicly available at https:// skm.nib.si/.

Stress Knowledge Map
Exploration SKM provides a number of options for the exploration of its contents, including interactive network visualizations of both PSS (PSS Explorer, Figure 2C) and CKN (CKN Explorer, Figure 2F), offering neighborhood extraction of selected entities, shortestpath detection between multiple entities of interest, and on-thefly exports.Both explorers provide direct references to the object provenance, as well as links for the corresponding A. From top to bottom: stressors (Layer 1) acting on the plant are first perceived (Layer 2), resulting in a signaling (Layer 3) cascade that leads to plant defense and/or adaptive changes in the form of executor molecules and processes  2D).A separate search interface using internal and external database identifiers (e.g., DOI, KEGG) is also available for PSS.

Photosynthesis
Modeling and analysis support PSS is available for download in a number of domain-standard formats (Figure 2H; summarized in Table 2) enabling further visualizations, analysis, and dynamical modeling.A suite of tools implemented in Python (SKM-tools, Figure 2I) has been developed to support additional network analysis of CKN and PSS (described in Table 3).

Extending and improving SKM
The contribution interface of PSS enables constant updates based on novel discoveries (Figure 2B).2G).

Case studies
To showcase the benefits of SKM, we present two case studies demonstrating the use of SKM for contextualization of experi-mental results within prior knowledge networks.The first case study concerns jasmonates (JA) and SA interference with ABAmediated activation of RESPONSIVE TO DESICCATION 29 (RD29) transcription, and the second, a proteomics analysis of Ca 2+ -dependent redox responses.
Case study 1: Interaction of ABA, JA, and SA in the activation of RD29 transcription In A. thaliana, the RESPONSIVE TO DESICCATION 29 A gene (AtRD29A) plays a pivotal role in stress acclimation (Baker et al., 1994) and is transcriptionally regulated via several promoter elements, including the ABA-responsive binding motif ABRE (ACGTG), located close to the transcription initiation site.The 1-kb upstream region of the potato StRD29 transcription initiation site also contains ABRE and several other abiotic-stressresponsive binding elements (Supplemental Figure 1).
ABA treatment of leaf discs from tobacco plants transiently transformed with pStRD29::fluc and from transgenic potato plants (cv.De ´sire ´e) carrying the pStRD29::mScarlet-I (Supplemental Figure 2) construct strongly induced pStRD29 activity, which reached its highest amplitude after approximately 4 h in the ABA solution (Figure 3A).Treatments with either jasmonates (JA/MeJA) or SA alone did not lead to an increase in pStRD29 activity.However, combined treatments of ABA with JA or ABA with SA attenuated the ABA-induced activation of pStRD29, revealing a negative effect of both these phytohormones on ABA-dependent StRD29 transcription (Figure 3A).We subsequently constructed transgenic potato plants (cv.De ´sire ´e) carrying the pStRD29::fluc construct to confirm the negative effect of MeJA and SA on ABA-responsive promoter activity in planta (Figure 3B).The effect of MeJA on ABA activation of both RD29s was further analyzed in potato and A. thaliana by quantitative real-time PCR.The data revealed attenuation of ABA induction of RD29A/RD29 by jasmonates in both species (Figure 3C).
We first tried to explain the observed effect of jasmonates and SA on ABA-dependent RD29 activation through promoter motif analysis, but no SA-or JA-signaling-related motifs were identified in  New validated biological interactions (e.g., transcriptional and translational regulation of a target gene) from various sources (A) can be added to PSS through the guided contribution interface (B) and are consolidated according to the PSS schema.The contents of PSS can be explored through interactive search and visualization provided by both the PSS Explorer (C) and the PSS overview in Newt (D).Correspondingly, sources for CKN interactions (E) are integrated and consolidated into the CKN schema through batch scripts and are accessible for exploration through the CKN Explorer (F), which provides interactive search and visualization of CKN interactions.Data provenance and interoperability links (G) provide context for SKM contents.Exports of PSS and CKN (H) enable various additional analysis and modeling approaches, including through the Python functions provided in the SKM-tools resource (I).Links to specific external resources and tools are highlighted in red.HT, high throughput; PSS, Plant Stress Signaling network; CKN, Comprehensive Knowledge Network; TF, transcription factor; ncRNA, non-coding RNA (currently only miRNAs are included); DOT/SBGN/SBML/ SIF, systems biology data formats, see Table 3 for details.   Plant Communications 5, 100920, June 10 2024 ª 2024 The Author(s).5

SKM-tools Exports
Stress Knowledge Map

Plant Communications
the potato promoter sequence (Supplemental Figure 1).We therefore hypothesized that the signaling pathways interact upstream of actual transcriptional activation.Owing to the complexity of several phytohormone pathway interactions, this was a good case study for the hormone-centric and expertcurated PSS model.We performed a triple shortest-path analysis to identify potential mechanisms of studied crosstalk.This analysis revealed an intersection of JA signaling with the ABA pathway through a protein-protein interaction of the JAresponsive MYC-like transcription factor 2 (MYC2) with the ABA receptor PYRABACTIN RESISTANCE-LIKE 6 (PYL6; Figure 3D).This reaction entry (rx00459) is based on experimental in vitro and in vivo interaction studies of PYL6 and MYC2 in A. thaliana (Aleman et al., 2016).It is conceivable that this interaction depletes PYL, thereby limiting ABA perception (Aleman et al., 2016), which could explain the lower activation of the ABA pathway in the presence of jasmonates.The SA pathway was found to converge with the ABA pathway through the JA pathway with a protein-protein interaction between the SA receptor NPR1 and MYC2 (rx00432) (Nomoto et al., 2021), and this might influence the interaction of MYC with PYL.To verify the hypothesis of direct synergism between JA and SA in attenuation of the ABA response, we performed titration experiments of combined JA and SA treatment on ABAdependent StRD29 induction, which was confirmed (Figure 3E; Supplemental Table 3).
Case study 2: Effect of the Ca 2+ channel inhibitor LaCl 3 on proteome-wide peroxide responses Secondary messengers such as Ca 2+ and H 2 O 2 are important in the translation of many perceived environmental changes towards a cellular response (Kudla et al., 2010;Pirayesh et al., 2021).It is still a challenge to disentangle and understand the principles of specificity and information flow in such networks.Lanthanide ions are known to block anion channels and inhibit the flux of Ca 2+ across the plasma membrane (Knight et al., 1992;Tracy et al., 2008).Thus, they can be used to identify Ca 2+ -dependent plant responses.H 2 O 2 is known to induce Ca 2+ transients (Rentel and Knight, 2004).In this case study, we analyzed the proteome of A. thaliana rosettes treated with either H 2 O 2 or a combination of H 2 O 2 and LaCl 3 to identify the components of H 2 O 2 signaling that are Ca 2+ dependent.We initially identified 119 proteins whose abundance changed significantly in response to H 2 O 2 compared with the mock treatment after 10 or 30 min of treatment.Of these, 49 proteins did not respond significantly in the same manner upon pretreatment with LaCl 3 (Supplemental Table 4), indicating that a significant number of H 2 O 2 -induced changes in protein abundance required a Ca 2+ signal (Ca 2+ -dependent redox-responsive proteins).
In the quest to identify mechanistic explanations for these results, CKN provides a universal resource for large-scale hypothesis generation.The largest connected component of CKN contains 98% of the nodes and 99% of the edges, indicating its high connectivity; thus, the analysis was performed on this part of CKN only.Using CKN prefiltered to only leaf-expressed genes, we searched for directed shortest paths from known Ca 2+signaling-related proteins (source set) to the Ca 2+ -dependent redox-responsive proteins identified by the proteomics approach (target set).The final source set of 53 genes included mainly calmodulins, Ca 2+ -dependent protein kinases, and calcineurin B-like proteins (Supplemental Table 4).Of the 49 Ca 2+ -dependent redox-responsive target proteins, 41 were present in CKN.
All of these proteins either could be connected to the source set of Ca 2+ -signaling-related proteins, directly or through an up-to-four-step pathway (Figure 4A), or were in the source set themselves.Combining all the detected shortest paths (all sources to all targets) into a single network (Figure 4A) revealed major network hubs-connected to multiple known Ca 2+ signaling genes and potentially regulating multiple targets.For example, the analysis revealed an intricate network of calmodulin-dependent regulation of downstream targets in A. thaliana (CAM2,3,5,6,7, Figure 4B).Another example of such a hub is Floricaula/leafy-like transcription factor 3 (LFY3), shown in Figure 4C, which integrates paths originating from four source nodes and in turn potentially regulates four downstream targets.
The next step in the analysis would be confirmation of the identified mechanisms by functional analysis experiments, e.g., knockout experiments to confirm the role of the proposed regulatory network.The design of such experiments is, however, not always trivial; thus, we designed the CUT-tool within SKMtools to aid experimentalists.This analysis reveals the minimum interactions that must be severed (''cut-set'') to separate the upstream regulators from the downstream targets.The cut-set to disrupt the regulation of all targets is shown in Figure 4A.As an example, the cut-set of one target, glutamine-dependent asparagine synthase 1 (ASN1), is shown in Figure 4C, revealing that deregulation of ASN1 would require knockout of both LFY3 and A. THALIANA NAC DOMAIN CONTAINING PROTEIN 29 (NAP) genes.

Plant Communications
Stress Knowledge Map

DISCUSSION
Plant stress signaling pathways are connected by synergistic and antagonistic interactions in a complex network that checks and balances the plant's response to its environment and its growth/development (Eckardt, 2015;Bittner et al., 2022).To understand the functioning of these complex processes, novel approaches are required.Knowledge graphs, such as those provided by SKM, provide powerful and accessible tools to integrate and simplify interpretations within curated published knowledge, as well as providing a basis for a plant digital twin and all the advantages of in silico simulation experiments it enables.A number of tools have been developed within the SKM environment to support this and also enable efficient linking to complementary tools.
To showcase the applicability of SKM, we investigated two distinct experimental datasets.In the first, our experiments provided evidence that jasmonate and SA treatment attenuates ABA-activated transcription of RD29 in both the crop plant potato and the model plant A. thaliana through hormonal signaling cross-talk (Figure 3).A manual attempt to extract known information on the crosstalk between ABA and JA with a search in PubMed ((JA OR jasmon*) AND (ABA OR abscisic) AND (plant)) resulted in over 2000 published items.With the wealth of data generated these days, it would be laborious for an individual researcher to perform a thorough literature survey; instead, interrogation of SKM provided a mechanistic hypothesis that explained the experimental results within hours.The hypothesis was empirically supported by further experiments and provides an explanation for the synergistic action of jasmonates and SA that is sometimes argued for in the literature (Mur et al., 2006;Zhang et al., 2020) , 2022).This way, canonical principles of plant signaling networks can be assessed across species.
Our second case study showed that SKM is not only helpful in revealing mechanisms in complex pathways for a single target but also can be used to identify regulators using a large number of targets, as is commonly the case with interpretation of large omics datasets.Using network analyses, arguably the simplest qualitative modeling approach, we identified hubs involved in complex redox-Ca 2+ signaling interconnectedness.By identifying connections from known Ca 2+ -related proteins to our experimentally derived target list, we were able to prioritize certain processes and hypotheses in an informed manner.The majority of our targets were found to have only one or two intermediary nodes between them and the upstream Ca 2+related proteins.Paths with many intermediary nodes are less likely to be valid sources of regulation; however, examples of longer paths such as these are known to be functional in the cell, e.g., MAPK signaling cascades.One of the SKM-tools features, the CUT-tool, was designed to help in the next step of research: validation of generated hypotheses.It enables the design of complex functional validation experiments (e.g., gene knockout or overexpression) identifying the genes whose activity should be modulated to achieve a desired effect, taking network redundancy into account.
Overall, in both case studies, SKM proved to be a useful generator of potential mechanistic explanations for the observed data.As with any hypothesis, further validation is needed and some may not prove as valid.More likely hypotheses for further research can be prioritized by weighing the interaction reliability (edge ranks) and exploring the linked content in other resources.

Relative Expression
AtRD29A

Stress Knowledge Map
Plant digital twins, virtual replicas of physical systems, are expected to provide a revolutionary platform for modeling the effect of crop management systems and environmental changes in agriculture (Pylianidis et al., 2021).Digital twins can be used to perform in silico experiments that guide or replace lab and field experiments.The detail that digital twins provide, combined with fast computational methodologies, enables efficient planning of experiments and will thus speed up our understanding of plant function and provide information for more effective breeding.Aside from being a tool for the interpretation of experimental data, SKM also provides a starting point for the integration of stress signaling and growth trade-offs in digital twins.
SKM will be continuously updated, keeping abreast of the latest developments in the field.Future plans include extending the repertoire of stressors to include additional factors such as cold, salinity, or nutrient deficiencies.We believe the integrated knowledge in SKM will help in understanding plant interactions with the environment by enabling exploration of knowledge and by supporting diverse mechanistic modeling approaches.This is of interest to the wider plant scientific community, enabling the informed design of experiments and, in the long term, contributing to the breeding of improved varieties and precision agriculture.

PSS construction
From the predecessor model (PIS v.2; Ram sak et al., 2018), numerous improvements, additions, and reformulations were carried out, resulting in the current PSS.In addition to intracellular pathogens (potyviruses), we extended PSS to also contain perception of extracellular pathogens (Pseudomonas sp.) and insect pests, as well as heat, drought, and waterlogging stress.Downstream of perception, PSS now includes Ca 2+ signaling, ROS signaling, and the MAPK signaling cascade, as well as the synthesis and signaling of all major phytohormones.We also added the synthesis of actuator molecules and processes, as well as known regulators of growth and major processes leading to growth.
PSS is implemented as a Neo4j graph database.The types of nodes and edges (relationships) in the database are summarized in Supplemental Table 5. Genes and gene products are represented by functional cluster nodes, including protein and non-coding RNA nodes.Functional clusters enable the representation of genetic redundancy.These groups were defined using sequence similarity among genes (orthologs and paralogs) and experimental data that confirmed functional overlap.The functional cluster concept includes groupings of enzyme-coding genes (similar to the EC number system), as well as genes involved in transcriptional and translational regulation.Users can access the same information in PSS at the gene level by utilizing the gene-level representation of PSS interactions in CKN.Groups of metabolites with the same biological function are also represented as metabolite families.Nodes also include more abstract entities, such as known but unidentified gene products and plant processes.Finally, foreign entities, such as biotic or abiotic stressors, are also included as nodes.
In addition to biological entities, molecular interactions are also represented by nodes in PSS and are categorized into 10 formal reaction types (e.g., protein activation or catalysis, Supplemental Table 5).Reaction participant nodes are connected to the reaction nodes by relationships, with the type of relationship representing the role of the participant (e.g., SUBSTRATE, ACTIVATES), as demonstrated in Figure 2B.These relationships are annotated with the subcellular location and the form of the participant when involved in the reaction (e.g., ''cytoplasm'' or ''nucleus'' and ''gene'' or ''protein'').
All updates to PSS are immediately available in the various interfaces and in all download formats (https://skm.nib.si/downloads).A frozen version (PSS v.1.0.0) is also available in all export formats, and a database dump with detailed deployment instructions can be accessed at GitHub (https://github.com/NIB-SI/skm-neo4j).All sources and resources used to create PSS v.1.0.0 are available in Supplemental Table 6.

CKN construction
The second edition of CKN (CKN v.2) was created by merging pairwise interactions from 25 public resources (details in Supplemental Table 2).
Additional filtering was performed on the STRING v.11.5 network (Szklarczyk et al., 2023), where the requirement was to only include physical interactions confirmed by experimental data or existence in a database.As Table 2 summarizes, five reliability ranks were designed to describe the reliability of the interactions across the diversity of the various sources.All interactions were then integrated, resulting in a single network of 574 538 interactions.The network was then condensed by collapsing multiple interactions of the same type between a pair of interactors into a single edge.In this process, the highest ranked interaction took precedence to define the interaction type, but all sources that contained any interaction between the pair were retained in the edge attributes.
All gene loci nodes were annotated using Araport11 (Cheng et al., 2017) downloaded from TAIR in June 2023 (Berardini et al., 2015).Gene loci that had been merged or made obsolete were renamed or removed, respectively.Genes are also annotated with Plant Ontology annotations from TAIR (Berardini et al., 2015) 3 for complete response curves).The results show that SA and jasmonates indeed act synergistically on attenuation of ABA signaling, as the addition of SA and JA has a stronger effect than the addition of each hormone individually.
Plant Communications 5, 100920, June 10 2024 ª 2024 The Author(s).9 Stress (A) All shortest paths identified in CKN leading from known Ca 2+ -related proteins (sources, pink-bordered nodes) to Ca 2+ -dependent redox-responsive proteins identified by proteomics (targets, green-filled nodes) using rank 0, rank 1, and rank 2 edges (as described in the Table 1 legend

Plant Communications
Stress Knowledge Map

SKM-tools
SKM-tools (https://github.com/NIB-SI/skm-tools) is a collection of Python scripts and notebooks, incorporating network analysis and visualization tools, that facilitates interrogation of CKN and PSS with targeted questions beyond the scope of the web application.Included functionalities are described in Table 3.The tools are developed using the networkX (Hagberg et al., 2008) and py4cytoscape (Ono et al., 2021) libraries.
The CUT-tool makes use of the max-flow min-cut (Edmonds and Karp, 1972) algorithm, which determines the minimum edges that must be severed (''cut-set'') to separate the upstream sources from the downstream targets.A max-flow min-cut analysis of multiple sources to an individual target reveals the minimum cut-set needed to disrupt all signaling to the target.To calculate the max-flow min-cut across multiple sources, a dummy node connected with arbitrarily high capacity to all original sources is introduced, and the calculation is performed using the dummy node as the source.

Promoter analysis
Predicted cis-regulatory motifs within the 1-kb promoter sequences of AtRD29A and StRD29 were identified with the Atcis database of the A. thaliana Gene Regulatory Information Server (Lichtenberg et al., 2009).In addition, we used PlantPAN 3.0 (Chow et al., 2019) to identify StRD29-specific motifs that were not previously identified in AtRD29A.Plant material and growth conditions S. tuberosum (cv.De ´sire ´e) plants were propagated by cuttings from sterile-grown plants.After 7 days of sterile growth on 1 / 2 MS medium (pH 5.7, 2% [w/v] sucrose) to initiate root growth, plantlets were transferred to individual pots filled with soil (9 parts soil, 1 part Perligran). A. thaliana (ecotype Col-0) seeds were sown directly onto soil and transferred into individual pots after 4-6 days.All experiments used leaves from 18-to 21day-old plants grown in climate chambers (20 C ± 2 C) under long-day conditions (16 h light/8 h dark) with a light intensity of 120 mmol photons m À2 s À1 (Philips TLD 18W alternating 830/840 light color temperature).
For promoter reporter assays of transiently transformed N. benthamiana leaves, seeds were germinated on Profi substrate (Gramoflor).Five days after germination, seedlings were separated into pots of 15.5 cm diameter 3 12 cm height filled with substrate (3 parts Profi substrate, 1 part vermiculite, 1.5 kg Osmocote Start/m 3 ).Plants were grown in a greenhouse under long-day conditions (16 h light at 28 C/8 h dark at 22 C) with an average light intensity of $250 mE and 80% relative humidity.
Soltu.DM.03G017570 was identified as the orthologous locus of A. thaliana RD29A in S. tuberosum cultivar DM1-3 using the DM v.6.1 database (http:// spuddb.uga.edu/).To generate the gene reporter lines in the potato cultivar De ´sire ´e, 1158 bp of the 5 0 UTR directly upstream of the start codon region was amplified by PCR and either the firefly luciferase (fluc) or the mScarlet-I (mScar) gene in a custom variant of the pBIB Hyg vector carrying hygromycin resistance for selection in plants.The complete sequences of both vectors, including annotations, can be found in Supplemental Figure 4.Both constructs were introduced into the potato cultivar De ´sire ´e as described previously (Rocha-Sosa et al., 1989).Plate-reader-based luciferase assays Agrobacteria carrying the pBIN-StRD29::fluc or pBIN-AtRD29A::fluc plasmid were grown in LB liquid medium supplemented with the respective antibiotics.Overnight cultures were diluted to OD600 = 0.1 with fresh LB medium and grown to OD600 = 0.8.Cells were harvested by centrifugation (22 C, 15 min, 4000 g) and resuspended in 5% sucrose solution in H 2 O to OD600 = 0.2.The agrobacterium suspension was infiltrated into leaves 6, 7, and 8 of 4-week-old N. benthamiana plants.Care was taken that the N. benthamiana plants selected for infiltration and measurement were not suffering an obvious pathogen attack before infiltration and during the transformation period, hormone treatment, and measurement.After 48 h, leaf discs (ø 6 mm) of infiltrated plants were transferred into 96-well plates containing 100 ml buffered MS (5 mM MES [pH 5.8]) supplemented with 1% sucrose (w/v) and incubated for 2 h under greenhouse growth conditions.Immediately before measurement, luciferin, to a final concentration of 30 mM, and the hormones, to the final concentrations indicated in the text, were added to each respective well.For all combinatorial hormone treatments, the different hormones were applied at the same time to the indicated final concentrations.Fluc luminescence was recorded in a multi-mode microplate reader (TECAN Spark multimode microplate reader, serial no.2301004717) in a window from 550 to 700 nm for 2 s every 5 min for each well.During the measurement period, the leaf discs were kept in darkness at a constant temperature of 22 C.
For luminescence measurements of S. tuberosum StRD29::fluc plants, leaf discs (ø 6 mm) were placed in 96-well plates containing 100 ml of 30 mM luciferin dissolved in 1 / 2 MS.After 2 h of preincubation, the solution was replaced by 100 ml of 30 mM luciferin containing various effectors (50 mM ABA, 50 mM MeJA, or both).Since MeJA is rapidly hydrolyzed to JA (Stuhlfelder et al., 2002;Wu et al., 2008), JA and MeJA treatments are comparable when eliciting a jasmonate response.Luminescence was measured every 5 min for up to 12 h using a TriStar2 LB 492 multimode reader (Berthold Technologies, Germany).During the measurement period, the leaf discs were kept in darkness.All luminescence analysis was performed with at least five independent experimental replicates.Luminescence data are available in Supplemental Tables 3 and 7.

Transcript analysis
For analysis of StRD29 and AtRD29A transcripts, S. tuberosum or A. thaliana plants were treated with water (mock), 50 mM ABA, 50 mM MeJA, or a combination of both for 6 h in three or four independent biological replicates.Total RNA was extracted from 100 mg of leaf material using the Gene Matrix Universal RNA Purification Kit (Roboklon, Germany) according to the manufacturer's instructions.RNA integrity was assessed by agarose electrophoresis and RNA quantity and purity with a UV-vis spectrophotometer (Eppendorf, Germany).For quantitative real-time PCR, RNA was transcribed into cDNA using the RevertAid First Strand cDNA Synthesis Kit (Thermo Scientific, Germany).The reaction was stopped by a 5-min incubation at 75 C.
Where applicable, all primers were designed to span exon-intron borders using QUANTPRIME (Arvidsson et al., 2008) (gene identifiers and primer arrowheads indicate directed, regulatory interactions (see Table 1), whereas dashed edges indicate undirected binding.Red edges are part of the merged cut-set.Nodes with proteomics measurements are annotated with a heatmap indicating the change in protein abundance after 10 min (top row) and after 30 min (bottom row) between H 2 O 2 -and mock-treated samples (left column) and between Ca 2+ blocker treatment and H 2 O 2 and Ca 2+ blocker treatment (right column).Significant changes in abundance are marked with a white asterisk in the center of the square.Red, increase in treatment compared with control; blue, decrease in treatment compared with control.Nodes are labeled with their short names, where available.The complete network is provided in Supplemental Figure 3, and all source and target nodes are listed in Supplemental Table 4.

Stress Knowledge Map
Plant Communications sequences in Supplemental Table 8).Quantitative real-time PCR was performed with three technical replicates for each sample in 96-well plates using a CFX96 real-time thermal cycler system (Bio-Rad, Germany).Each reaction contained 13 SYBR Green master mix (Thermo Fisher), 2 ng/ml cDNA, and the respective forward + reverse primers at 10 mM each.The specificity of each product was assessed on the basis of melting curves after 40 cycles of amplification.All transcript levels were normalized against the geometric mean of the transcript abundances of the reference genes YLS8 and CYP5 for A. thaliana and YLS8 and ACT7 for potato.Target relative copy numbers were calculated using quantGenius (http://quantgenius.nib.si/;Baebler et al., 2017), provided in Supplemental Table 9.

PSS network analysis
We identified the pathway between ABA and RD29 by querying for all directed shortest paths from ABA to RD29 in the reaction participant bipartite projection of PSS.We then extracted all directed shortest paths from JA and SA to RD29 that partially overlapped with the ABA-to-RD29 path.For added context to these results, we expanded the network induced by the shortest paths to include the first neighbors of all nodes (Figure 3E).

Proteomic analysis
Complete rosettes of 3-week-old A. thaliana plants were incubated in 1 mM LaCl 3 solution or ddH 2 O for 1 h.Afterward, plants were transferred into either 20 mM H 2 O 2 or ddH 2 O and harvested after 10 or 30 min of incubation.Complete rosettes of 12 plants per treatment were pooled and immediately frozen in liquid nitrogen.Frozen plant material was homogenized using a precooled mortar and pestle and stored at À80 C. For peptide isolation, 500 mg of frozen plant material was mixed with 2 ml Lacus buffer (20 mM Tris [pH 7.7], 80 mM NaCl, 0.75 mM EDTA, 1 mM CaCl 2 , 5 mM MgCl 2 , 1 mM DTT, 1/200 mM NaF) containing 4 tablets of protease inhibitor (Roche cOmplete, EDTA-free, Protease inhibitor cocktail tablets) and 10 tablets of phosphatase inhibitor (Roche PhosSTOP) per 200 ml.Samples were incubated for 10 min on ice and then centrifuged at 15 000 g for 10 min at 4 C.The supernatant was transferred to a new tube, adjusted to 20% (v/v) trichloroacetic acid, and incubated overnight at À20 C. The precipitated samples were stored until preparation for mass spectrometry analysis.
Samples were centrifuged at 15 000 g, vacuum-dried, and eluted in urea lysis buffer (8 M urea, 150 mM NaCl, and 40 mM Tris-HCl [pH 8]).Protein concentration was determined via BCA assay (Thermo Fisher).In total, 3 mg of protein per sample was first reduced in 5 mM DTT and then alkylated in 15 mM iodoacetamide for 30 min at room temperature in the dark.The alkylated samples were quenched by adding DTT to a final concentration of 5 mM and mixed with 30 mg Sera-Mag carboxylate-modified magnetic beads (1:1 ratio of hydrophilic and hydrophobic beads, Cytiva, USA).
The peptides attached to the beads were washed four times with 80% (v/ v) ethanol and digested in a 30 mM ammonium bicarbonate buffer (pH 8.2) containing 30 mg trypsin (Promega, WI, USA).Tryptic digestion was performed overnight at 37 C under constant shaking.The digestion was stopped by the addition of formic acid (end concentration, 4%).In total, 100 mg of digested peptides per sample was transferred to a new reaction tube, vacuum-dried, and stored at À20 C until high-pressure liquid chromatography-tandem mass spectrometry (MS/MS) analysis.
The purified tryptic peptides were dissolved in 0.1% (v/v) formic acid in high-purity water.Approximately 1 mg of peptides was separated by an online reversed-phase high-pressure liquid chromatography apparatus (Thermo Scientific Dionex Ultimate 3000 RSLCnano LC system) connected to a benchtop quadrupole orbitrap (Q-Exactive Plus) mass spectrometer (Thermo Fisher Scientific).LC eluent was introduced into the mass spectrometer through an Easy-Spray ion source (Thermo Scientific) with the emitter operated at 1.9 kV.The mass spectra were measured in positive ion mode, applying a top 15 data-dependent acquisition.A full mass spectrum was set to 70 000 resolution at m/z 200 (automatic gain control target at 1e6, maximum injection time of 120 ms, and a scan range of 400-1600 [m/z]).The mass spectrometry scan was followed by an MS/MS scan at 17 500 resolution at m/z 200 (automatic gain control target at 5e4, 1.6 m/z isolation window, and maximum injection time of 80 ms).For MS/MS fragmentation, the normalized collision energy for higher-energy collisional dissociation was set to 27%.Dynamic exclusion was set to 40 s, and unassigned and +1, +7, +8, and >+8 charged precursors were excluded.The intensity threshold was set to 6.3e3, and isotopes were excluded.The analysis was performed with five independent experimental replicates for each sample.

Peptide identification and quantification
Identities and peptide features were defined by the peptide search engine Andromeda, which was provided by MaxQuant software (v.2.1.3.0,Max Planck Institute of Biochemistry), using standard settings (Tyanova et al., 2016b).In detail, trypsin-based digestion of the peptides with up to two missing cleavage sites was selected.Methionine oxidation as well as N-terminal acetylation was set as a variable modification for peptide identification.In total, up to three potential modification sites per peptide were accepted.The identified peptide sequences were searched and aligned against the Araport11 (Cheng et al., 2017) reference protein database.The false discovery rate cutoff for protein identification and side identification was set to 0.01.The minimum peptide length was 7 amino acids, and the maximum length was 40 amino acids.For each identified protein group, label-free quantitation intensities were calculated and used for further analysis (Supplemental Table 4).
Potential contaminants and reverse-sequenced peptides were removed before statistical analysis.Only proteins that were detected in at least three of five replicates in at least one treatment group were considered for statistical analysis, which was performed using Perseus (v.2.0.7.0) (Tyanova et al., 2016a).Missing values were replaced by sampling from a normal distribution using the default settings.Protein groups with an absolute fold change of greater than 1.5 compared with the control and a false discovery rate value below 0.05 were considered significantly regulated (Supplemental Table 4).
To filter for Ca 2+ -regulated proteins, significantly up(down)regulated proteins in La 3+ + H 2 O 2 -treated samples compared with La 3+ -only-treated samples were subtracted from the list of significantly up(down)regulated proteins in H 2 O 2 -treated samples.An additional filtering step was performed to ensure a compelling difference in abundance between the two contrasts.This required that abs(L 1 À L 2 ) R 1, where L 1 = log fold change for H 2 O 2 vs. mock and L 2 = log fold change for La 3+ + H 2 O 2 treatment vs. La 3+ only.For each of the protein groups that passed the filters, we extracted all identifiers in the group.For identifiers that occurred in multiple groups, we removed the identifier from the group where it occurred the least.

CKN network analysis
For each Ca 2+ -dependent redox-responsive protein group (target), we identified the closest nodes upstream that had a known Ca 2+ -signaling association (source).This was done by identifying all shortest paths in CKN with the source nodes set as all genes with Ca 2+ -signaling-related GoMapMan (Ram sak et al., 2014) annotations and the target set as the

Figure 1 .
Figure 1.Contents of the Plant Stress Signaling model represented as conceptual layers.

Figure 2 .
Figure 2. Stress Knowledge Map environment and features.

Figure 3 .
Figure 3. Elucidating connections from JA and SA to ABA-mediated regulation of RD29 expression in potato.(A and B) Expression of firefly luciferase driven by the StRD29 promoter (pStRD29::fluc) in (A) transiently transformed tobacco leaves treated with the indicated hormones (25 mM JA, 50 mM ABA, and 50 mM SA) and (B) transgenic potato leaves treated with the indicated hormones (50 mM MeJA, 50 mM ABA, and 50 mM SA).Values are shown as mean ± SE.Data are provided in Supplemental Table 3. (C) Relative transcript abundance of StRD29 (left) and AtRD29A (right) 6 h after application of 50 mM ABA, 50 mM MeJA, or a combination of both, analyzed by quantitative real-time PCR.Bars represent mean values ± SE of three or four independent biological replicates.(D) PSS node-induced subnetwork of shortest paths and immediate neighbors.Paths are directed from the hormones (source) to RD29 (target).Nodes and edges are colored by the path source: ABA (brown), JA (green), and SA (blue).Edges to first neighbors, edges not on the directed shortest paths, and
), merged into a single network.The excerpts show (B) a subnetwork with a focus on calmodulins and (C) a subnetwork with a focus on LFY3 and ASN1.Solid edges with (legend continued on next page) 10 Plant Communications 5, 100920, June 10 2024 ª 2024 The Author(s).

Table 1 .
Counts of unique CKN interactions by type and reliability ranking.

Table 2 .
Supported exports of SKM knowledge graphs.

Table 3 .
graph objects for PSS or CKN, thus providing access to the multitude of graph analysis and graph operations available in the library Node filter For PSS and CKN, filter on node type or node origin (plant or foreign), and additionally for CKN filter nodes based on tissue specificity, creating a network specific to the biological question at hand data visualization Import of multi-omics experimental data tables (e.g., logFC and p values) as context to the networks and functionality to visualize experimental data associated with nodes in the network, through rendering of PNGs (e.g., heatmaps) per node in the Cytoscape view Link to DiNAR Instructions for the use of CKN or PSS as the prior knowledge network for integration and visualization of multiple-condition high-throughput data in the DiNAR application (Zagor s cak et al., 2018) Features of SKM-tools.
The interactive visualizations of PSS and CKN are based on Biomine Explorer (Podpe can et al., 2019) implemented using vis.js and open-source Python libraries (including networkX The separation was carried out on an Easy-Spray analytical column (PepMap RSLC C18, 2 mm, 100 A ˚, 75 mm i.d. 3 50 cm, Thermo Fisher Scientific) with an integrated emitter, and the column was heated to 55 C. The liquid chromatography (LC) gradient was set to a 140-min gradient method, with a flow rate of 300 nl/min.The LC gradient was set to 5%-50% buffer B (v/v) (79.9% ACN, 0.1% formic acid, 20% ultra-high purity H 2 O [MilliQ]) for 125 min and then to 80% buffer B over 5 min.