A total of 315 transcriptome datasets of tomato (Zouari et al., 2014; Du et al., 2015; Barad et al., 2017; Sarkar et al., 2017; Sugimura and Saito, 2017; Xue et al., 2017; Yang et al., 2017; Zheng et al., 2017; Chen et al., 2018; Shukla et al., 2018; Fawke et al., 2019; Pesti et al., 2019; Wang et al., 2019b) and potato (Goyer et al., 2015; Zuluaga et al., 2015; Dees et al., 2016; Gao and Bradeen, 2016; Kochetov et al., 2017; Levy et al., 2017; Li et al., 2017; Lysøe et al., 2017; Hao et al., 2018; Kumar et al., 2018) were obtained from the Sequence Read Archive ( Figure S1 ). These studies included treatments with potentially beneficial organisms (arbuscular mycorrhizal fungi (AMF) and biocontrol agents) as well as detrimental organisms (pathogenic bacteria, nematodes, fungi, viruses, viroids, insects and oomycetes). Only studies with at least one mock treatment were included. The collected tissues included roots, stems, leaves, fruits and tubers. The time points of sampling after infection range from 0 days post-infection (dpi) up to 42 dpi or until the end of the host’s life cycle ( Figure S1 ). Approximately 20% of all tomato and potato cultivars were denoted as resistant to the applied pathogens.

The lists of the R-gene repertoires of S. lycopersicum and S. tuberosum were retrieved from Jupe et al. (2013). R-genes were classified as “full-length” NBS-LRRs if they contained both NBS and LRR domains as identified using InterPro (Mitchell et al., 2019). A slightly modified pipeline as described by Jupe et al. (2013) was used to verify their novel R-genes ( Figure S2 ). These novel R-genes were designated by the authors as R gene discovery consortium (RDC) genes. Using AUGUSTUS (version 3.3.1), a gene-prediction tool developed by Stanke et al. (2008), we analyzed these RDC genes for coding regions and searched for NBS and LRR domains using InterPro. RDCs were classified as true R-genes if they possessed a coding region and an NBS-LRR domain. Otherwise they were excluded from further analysis. In cases in which multiple splice variants were identified, the longest splice variant was analyzed. The well-established R-genes Pto (Martin et al., 1993) and EDS1 (Hu et al., 2005) from tomato were included in the dataset. In total, the expression patterns of 359 R-genes of tomato and 581 R-genes of potato were analyzed.

R-genes were classified as belonging to a cluster when more than one R-gene was located in a region of 200 kilobases (kb) on a chromosome (Van de Weyer et al., 2019). Since Jupe et al. (2013) performed their analysis on an earlier release of the tomato genome assembly (ITAG2.4 release, Tomato Genome Consortium, 2012), the positions of all tomato R-genes had to be re-defined ( Table S1 ). Positions of RDCs were verified using Blastn v2.6.0 (Altschul et al., 1990; Camacho et al., 2009) against the tomato (ITAG4.0; Hosmani et al., 2019) and potato genomes (PGSC_DM_v4.03; Potato Genome Sequencing Consortium, 2011; Table S1 ). All R-genes without defined chromosomal positions (39 genes in tomato) were classified as R-genes with unknown clustering.

To predict potential regulation of R-genes by the miR482-superfamily (de Vries et al., 2015), we used psRNATarget (release 2017; Dai et al., 2018). We used the coding sequence (CDS) of our R-genes as the target library. To ensure a low rate of false-positives, the maximum expectation was set to ≤3, since higher expectation values represent less likely mRNA/miRNA interactions. We evaluated the likelihood of an R-gene being targeted by the miR482 superfamily and whether the R-gene encoded a full length NBS-LRR and or belonged to a R-gene cluster using a chi-square test (Greenwood and Nikulin, 1996).

The program Kallisto (v.0.46.0) was used to estimate the relative expression of genes in tomato and potato (Bray et al., 2016). As a first step, the raw sequence reads were compared to the transcript sequences. This step in Kallisto is designated as the pseudoalignment step. To improve the quality of the pseudoalignment, low-quality reads and adapters were removed from the transcriptomes using Trimmomatic under the following settings: seed mismatch = 2; palindrome clip threshold = 30; simple clip threshold = 10; LEADING = 3; TRAILING = 3; SLIDINGWINDOW= 4:15; MINLEN =36 (Bolger et al., 2014; Figure S2 ). Subsequent quality controls were performed using FastQC (Andrews, 2010). As Kallisto requires information on fragment length for single-end sequenced transcriptomes, the fragment length denoted by the authors was used. If this information was not available, the recommended fragment length of the reported RNA isolation kit was used. The standard deviation was set to ±17.5 bp. Kallisto indices (used for generating the pseudoalignments) were based on the tomato ITAG4.0 and the potato PGSC_DM_v4.03 genome releases. R-genes missing from the current genome releases were manually added to the list of transcripts (indices in Kallisto). Transcript abundance was calculated as transcripts per million (TPM; Wagner et al., 2012). We chose to use TPM since it normalizes the transcript abundance for gene length and library size, making TPM values comparable across experiments. Genes for which the TPM values were less than 1 were treated as “off” and for these genes, TPM was set to zero. All scripts and settings used for these analyses are available at the following site: https://github.com/LauraERose/LargeScaleTranscriptomeAnalysis.

To verify the overall consistency of our estimated TPM values in this metaanalysis, we performed qRT-PCR on twelve NBS-LRR-genes and three reference genes (de Vries et al., 2018). We evaluated the expression of these fifteen genes over six time points on the Moneymaker cultivar inoculated with Phytophthora infestans (P. infestans) isolate IPO-C. Three replicates were studied at each sampling time point and treatment type. Additional details of this experiment are reported in de Vries et al., 2018. The Bioproject L (PRJNA487149) from Fawke et al., 2019 is the most similar in design to our 2018 study, since that project sampled transcriptomes from leaves of the tomato cultivar ‘MicroTom’ inoculated with P. infestans isolate 88069. We evaluated the consistency between the average TPM of these 15 genes from Fawke et al. with our estimated Cq values at 72 hours post infection, the single overlapping timepoint between both data sets.

To compare the mean relative expression between R-genes (R-gene set size for tomato = 359 and for potato = 581) and non-R-genes (the rest of the genome) we generated 100 replicate datasets for each transcriptome by sampling the TPM values of 359 random genes from tomato and 581 random genes from potato. The average TPM of all expressed genes was calculated for each replicate dataset. To compare expression values, four reference genes were used: ubiquitin (Solyc09g018730.4.1) and actin4 (Solyc04g011500.3.1) for tomato (Müller et al., 2015) an importin subunit (PGSC0003DMG400007289) and elongation factor-1 (PGSC0003DMG400023270) for potato (Mariot et al., 2015; Tang et al., 2017). TPM values were tested for normality using the Anderson-Darling (>5000 data points; Thode, 2002) or Shapiro test (< 5000 data points; Shapiro and Wilk, 1965) and for equal variances using the test from Kendall (1938). Significant differences in expression were identified using a Mann-Whitney-U test (Mann and Whitney, 1947) for non-normally distributed data or a two-sample t-test for normally distributed data.

We visualized R-gene expression using heatmaps created in R (v. 3.6.1). Genes were classified as off (if TPM < 1) or on (if TPM ≥1). In the heatmaps, libraries were clustered by similarity in patterns of expression between libraries and R-genes were sorted by the number of libraries expressing the corresponding gene. Correlations between 1) the total number of expressed R-genes and the total number of expressed genes, 2) the total number of expressed genes and the number of pseudo-aligned reads, as well as 3) the number of libraries in which an R-gene was expressed and the average level of expression of each R-gene were performed using a Spearman’s rank correlation test (Hollander et al., 2013).

To investigate the extent to which expression patterns of R-genes were similar to wild close relatives of tomatoes, we evaluated additional transcriptomes of four wild tomato species: S. peruvianum, S. chilense, S. ochranthum, and S. lycopersicoides (Beddows et al., 2017). A subset of R-genes was further analyzed for their patterns of sequence variation within and between these wild species. Standard population genetic parameters including intraspecific variation (π) and interspecific divergence (K) were estimated using DNaSP v. 5.10 (Librado and Rozas, 2009).

To identify the factors associated with differences in expression of R-genes across transcriptomes, we performed an ANOSIM in Primer 7.0.13 (PRIMER-e; Figure S2 ). ANOSIM is a non-parametric statistical test similar to ANOVA. The starting point of the analysis is a pairwise dissimilarity matrix. In our case, the dissimilarity matrix was computed as follows: First the TPM values for each gene within each transcriptome were LOG (x+1) transformed. On the basis of these transformed TPM values, the dissimilarity in gene expression patterns between transcriptomes were calculated based on Euclidean distances. Ranking was applied to the distance matrix. The two libraries from potato (SRR6511453 and ERR791944) with exceptionally low expression of the entire R-gene repertoire were excluded in these analyses.

To determine if gene expression is more similar within groups than between groups (for example when groups are defined by infection status or tissue type) the R test statistic value was calculated. The R values can range from -1 to 1, with larger values corresponding to greater differences between groups. Statistical significance is calculated through permutation of the group labels and recalculation of the R value for each replicate. In our case, 999 permutations were generated. The following factors were evaluated: BioProject, tissue type, type of treatment, specific treatment organism, life cycle of the organism, type/kingdom of the organism, susceptible vs. resistant cultivar, relative sequencing depth, paired- or single-end sequencing and days post infection. The ANOSIM analysis was also applied to the differential gene expression data (see below).

Differentially expressed genes between microbe treatments and mock treatments were identified using Sleuth (Pimentel et al., 2017; Figure S2 ). The p-values were adjusted using the Benjamini-Hochberg correction (FDR ≤ 0.05; Benjamini and Hochberg, 1995). Since Sleuth relies on replicates within treatments, BioProjects without replicates were removed from this part of analysis. We evaluated differences between i) R-genes and all genes, ii) proportion of up- versus down-regulation iii) average absolute fold changes.

In total we analyzed 7.78 x10⁹ raw reads from 315 transcriptomes of tomato and potato of which 5.58 x10⁹ could be uniquely assigned to a transcript from tomato/potato (average proportion of assigned reads: 77.3% for tomato and 66.8% for potato; Figure S3 ). Both mock-inoculated plants as well as plants inoculated with pathogenic and beneficial organisms were investigated. In total, 359 R-genes from tomato and 581 from potato were examined for their expression levels and fold changes. In tomato, 62.1% of all R-genes possessed NBS- and an LRR-domains and in potato 89.2% did ( Figure S4 , Table S1 ). A large majority of the R-genes of both species formed physical clusters meaning that two or more R-genes were found in a span of 200kb along the chromosome (62.6% in tomato; 83.1% in potato).

Since the miR482-superfamily is a known regulator of NBS-LRR expression (Shivaprasad et al., 2012; de Vries et al., 2015; de Vries et al., 2018), we evaluated the targeting probability by members of the miR482 gene family for each R-gene. In tomato 17.6% of all R-genes were predicted to be targeted by the miR482-superfamily, while in potato 28.6% were predicted to be targeted ( Figure S4 , Table S1 ). It has previously been shown that miR482-members regulate R-genes by reverse-complementary binding to the mRNA region encoding NBS-domains (Shivaprasad et al., 2012). Full length R-genes were more likely to be predicted to be regulated by the miR482-superfamily compared to partial length NBS-LRR genes (χ² _tomato = 21.32, p-value < 0.001; χ² _potato = 14.69, p-value < 0.001; Figure S4 ; Table S2 ).

Proteins encoded by R-genes act as key regulators of plant immunity by recognizing plant pathogens and activating the plant immune response. However, the existence of growth-defense trade-offs implies that the constitutive expression of R-genes in the absence of pathogens might be costly (reviewed in Brown and Rant, 2013; Vos et al., 2013). In our study, a large proportion of the R-gene repertoire in tomato (67.6% ± 13.8%) is not expressed in a given library (or is below the threshold of detection) whether or not the plant was treated with an interacting organism ( Figure 1A ). A smaller proportion (~ 46%) of the non-R-genes are “off” or below the threshold of detection in tomato ( Figure 1A ). For potato, the proportion of the R-gene repertoire that is not expressed is 49.3% (± 11.7%); this is nearly equal to the proportion of genes that are not expressed in the rest of the genome ( Figure 1B ).

In both species, the average TPM of R-genes per library is significantly lower than the average TPM of an equal number of randomly selected genes per library (p-value <0.001; Figures 1C , S5 ). Of the R-genes that are expressed, most are expressed at very low levels within each library (between 1 and 10 TPM). Approximately one quarter of R-genes in tomato (25.9% ± 8.4%) and 44.0% (± 9.8%) of R-genes in potato are expressed at this level. Less than 1% of the R-genes fall into the medium (50≤ TPM <200) or high (200≤ TPM <1000) expression classes, a scant proportion for these two expression classes compared to non-R-genes ( Figures 1A, B ).

The distribution of the expression classes for R-genes varies greatly across libraries ( Figure S6 ). For example, 88.3% of R-genes are not expressed in library SRR7073605, while in library SRR442353, 47.0% are not expressed ( Figure S6 ). Although the relative transcript abundance of a few R-genes can be high, the average TPM of the top 10% (or even the top 5%) is still well below the average TPM across all other genes in the genome for a given library ( Figure 1C ; S5 ; p-value_tomato <0.001; p-value_potato <0.001). Taken together, most R-genes are typically expressed at low to extremely low levels across libraries.

In our study, the overall distribution of expression classes of R-genes is similar between plants treated with interaction partners versus untreated controls ( Figures 1A, B , S6 ). However, conditioning on only the expressed R-genes in each individual library, the average expression level (measured as TPM) of these expressed R-genes is significantly higher in tomato plants treated with microbes compared to mock treated controls (p-value <0.05; Figure 1C ). This effect was specific for treatment with pathogenic organisms: We observed that the average TPM-values for expressed R-genes (TPM >1) was higher in tomato plants exposed to pathogenic organisms compared to plants exposed to beneficial microbes (p-value <0.001; Figure S7A ). In contrast, in potato no difference in the average expression of R-genes between treated and untreated plants, nor between the types of treatments (pathogenic versus beneficial) could be detected (p-value >0.05; Figure S5 ; p-value = 0.58; Figure S7B ).

We observed that some R-genes were expressed (TPM ≥1) under both challenged and unchallenged conditions. Therefore, the question arose if these R-genes represent a “core set” of expressed R-genes across all libraries. Approximately 7.7% of all R-genes in tomato are expressed in > 90% of all analyzed libraries ( Figure 2A ). In potato, 16.6% of all R-genes were expressed in >90% of all libraries ( Figure S8A ). Among these expressed “core” R-genes in tomato are EDS1, Pto, NRC2, NRC3 and NRC4. Wu et al. (2017) identified these NRCs as part of a complex network in Solanaceae in which the NRCs (Solyc10g047320, Solyc05g009630, Solyc04g007070) interact with NBS-LRR sensors to activate resistance.

We evaluated whether this set of expressed “core” R-genes shared other characteristics. We found that the mean TPM-value of an R-gene within a library was positively correlated with expression breadth as defined as the number of libraries in which it was expressed (correlation factor rho_tomato = 0.39, p-value <0.000, rho_potato = 0.47, p-value < 0.000, Figures S9A, B ). Therefore, this set of expressed “core” R-genes has both higher relative expression within a library and broader expression across libraries than non-core R-genes.

The total number of R-genes expressed per library varied from 27 to 191 in tomato, with a mean proportion of ~30% of R-genes expressed in a given library ( Table S3 ; Figure S10A ). For potato, the number of R-genes expressed per library ranged from 1 to 421 R-genes, with a mean proportion of 50.8% of the R-genes expressed in a given library ( Table S4 ; Figure S10B ). We also evaluated whether the proportion of expressed R-genes correlated with the total number of expressed genes in a given library. In both potato and tomato, libraries in which a larger number of genes were expressed also had a higher proportion of expressed R-genes (rho_tomato = 0.84, rho_potato = 0.71, p-value <0.000; Figures S9C, D ). We investigated how the distribution of the proportion of R-genes expressed correlated with the proportion of assigned reads (as a proxy for sequencing quality). Overall, we detected a weak positive correlation between both factors (rho_tomato = 0.39, rho_potato = 0.42, p-value <0.001; Figures S9E, F ).

We applied an ANOSIM method to evaluate which factors were associated with variation in R-gene expression across the libraries ( Table 1 ; Table S5 ; Table S6 ). In the ANOSIM analysis, higher R-values indicate a larger influence of a factor on the patterns of gene expression. The factor with the highest R-value for R-genes was BioProject (R-value_tomato = 0.876, p-value <0.001; R-value_potato = 0.928, p-value <0.001, Table 1 ). Differentiation by BioProject is also apparent in the principal component analysis (PCA, Figures 2B, C , Figures S8B, C ). Libraries clustering closer together in the PCA indicate those with more similar expression patterns. In this study, the factor BioProject corresponds to the set of libraries submitted by a single lab group. In total, 13 BioProjects for tomato were studied and 12 BioProjects for potato. The number of libraries submitted as part of a BioProject ranged from as low as two and up to 36. In some cases, BioProjects sampled only a single tissue type; other BioProjects sampled multiple tissue types. Most BioProjects focused only on a single potato or tomato cultivar. Individual BioProjects typically included treatment with one main (micro-)organism, except for a handful which studied treatments with two or more organisms. Due to the diversity of projects in terms of plant genotypes, type of organismal challenge and time of sampling and since the sampling was not based on a nested design, the large effect of the BioProject is not unexpected. However, the value of such a meta-analysis is that robust and consistent patterns of gene expression that do emerge from this study, in the face of a large amount of experimental variation across labs, are likely to be highly reliable because a wide range of sampling conditions were included (different lab conditions, different cultivars, different time of sampling, different treatments, etc.). Furthermore, this type of analysis can be used to identify key experiments that are missing (such as tissue type, time of sampling, cultivar, or pathogen) that if included could provide the necessary cross-lab validation of patterns.

Despite a large effect of BioProject, gene expression was also strongly affected by tissue type (R-value_tomato = 0.527, R-value_potato = 0.758, p-value <0.001) and days post infection (R-value_tomato = 0.408, R-value_potato = 0.522, p-value <0.001; Table 1 ; Figures 2D, E ; Figures S8D, E ). All other evaluated factors (type of treatment, life cycle of the organism, susceptible vs. resistant cultivars, specific treatment organism, type/kingdom of the organism) were characterized by lower R-values ( Table 1 ). Library dependent parameters such as relative sequencing depth (R-value_{tomato/potato} = 0.297/0.119, p-value_max <0.001) and paired- or single-end sequencing (R-value_{tomato/potato} = 0.358/0.218, p-value_max <0.001) were also characterized by low R-values ( Table 1 ). Furthermore, the rank order of the factors according to R-values did not differ depending upon the classification of R-genes into the following categories: full-length versus partial, miR482-targeted versus not targeted or clustered versus not clustered ( Table S7 ). The ANOSIM analyses of all coding genes did not deviate significantly from the analyses of the R-genes alone ( Table 1 ; Figures 2C, E ; Figure S8C, E ).

In a sub-analysis, we performed ANOSIM on the mock-treated libraries only. For the mock-treated libraries, the BioProject (R-value_{tomato/potato} = 0.911/0.936, p-value_max <0.001) and the tissue type (R-value_{tomato/potato} = 0.561/0.789, p-value_max <0.001) remain the two dominant factors associated with differences in R-gene expression ( Table S8 ; Figure S11 ). In a separate sub-analysis of organism-treated libraries only, the R-values for multiple factors increased compared to the ANOSIM analyses of all libraries together ( Table S8 , Figure S12 ). For example, the R-values for the factor “specific organism” was R=0.264 in tomato and R=0.104 in potato when all available libraries were included. The R-value for this factor increased to R=0.889 in tomato and R=0.85 potato when only microbe treated libraries were analyzed. This was also true for the related factors “life cycle of the organism” and “type of organism”.

qRT-PCR was conducted on twelve NBS-LRR-genes and three reference genes (de Vries et al., 2018). The expression of these fifteen genes was assayed in the Moneymaker cultivar at six time points following mock-inoculation or inoculation with P. infestans, isolate IPO-C. The Cq values for the three reference genes (SAND/Solyc03g115810, TIP2/Solyc10g049850, and TIF3H/Solyc12g098680) were always lower (corresponding to higher transcript abundance) at all time points compared to the twelve R-genes, supporting our findings in this metaanalysis ( Figure S13 ; Table S9 ). The sampling design of the BioProject L (PRJNA487149) from Fawke et al., 2019 was the most similar to the qRT-PCR experiment. Therefore, we compared the estimates of gene expression using TPM and Cq values from these two studies. The three reference genes all had the highest TPM values and lowest Cq values, while all the R-genes had low TPM values and high Cq values. Furthermore, gene expression as assayed by TPM and Cq did not radically differ between samples inoculated with P. infestans and mock inoculated samples (indicated by contrasting colors in Figure S13 ).

In tomato, 27.5% of the R-gene repertoire is not expressed in any library ( Figure 2A ). Even under this wide range of experimental conditions and treatments, these R-genes seem to be “off”. In a previous study, we evaluated the transcriptomes of 38 individuals of wild close relatives of cultivated tomato, namely S. chilense, S. peruvianum, S. ochranthum, and S. lycopersicoides (Beddows et al., 2017). Using this dataset from wild species of tomatoes, we evaluated whether any of these R-genes that are “off” in cultivated tomato are “on” in the wild genotypes. Expression was detected for ~35% of these genes, although the expression was restricted to a few libraries ( Figure S14 , Table S10 ). Five R-genes which were “off” in the studies of cultivated tomatoes (Solyc01g102920, Solyc01g102930, Solyc06g065150, Solyc10g079020 and Solyc12g038890) were expressed in >30% of all libraries from the wild species, although their overall relative expression was still low (Ø1.8-6.6 TPM).

For these five R-genes which are “off” in all libraries from cultivated tomatoes, but “on” in a subset of wild genotypes, we evaluated whether these genes showed the genetic signatures of evolutionary constraint within the population sample from our earlier study (Beddows et al., 2017). A signature consistent evolutionary constraint (or purifying selection) may indicate that these R-genes are still functionally intact in wild tomato species and could be exploited for crop improvement in the cultivated tomato. The low π_a/π_s ratios within species and K_a/K_s ratios between species indicated that purifying selection is the dominant force acting on these R-genes in wild tomatoes ( Table S11 ).

We evaluated the differential regulation of R-genes in the presence and absence of biotic treatments ( Tables S12 , S13 ). This included 26 datasets in tomato and 29 datasets in potato. On average, 11.9% of R-genes were differentially expressed in the presence of pathogens in tomato and 8.6% in potato ( Figures 3A , S15A ). Of these significantly differentially expressed genes in tomato, a larger proportion were up-regulated (72.5%) compared to down-regulated (27.5%; p-value <0.05; Figure 3B ). In potato, the proportion of up- versus down-regulated genes was not statistically different (up = 54.1%, down = 45.9%, p-value >0.05, Figure S15B ). In tomato, the proportion of genes differentially up- or down-regulated was not statistically different between the class of R-genes and the rest of the genes in the genome (p-value >0.05; Figure 3B ). In potato the proportion of down-regulated genes is lower for the class R-genes compared to the rest of the genes in the genome (p-value <0.05; Figure S15B ). Of the differentially expressed genes, the mean of the absolute fold change did not differ between the class of R-genes and the rest of the genes in the genome ( Figures 3C , S15C ). However, the mean of the absolute fold change for differentially up-regulated R-genes is significantly larger than the fold change of differentially down-regulated R-genes in tomato ( Figure 3C ).

The patterns of differential expression of R-genes are shared across datasets ( Figure 3A , Figure S15A ). However, in contrast to the previous ANOSIM analysis based on expression investment in R-genes (as captured by TPM values), variation in differential gene expression is not associated with the same factors such as BioProject or tissue type ( Figure S16 , Tables S14 , S15 , S16 ). Likewise, the assignment of R-gene type in terms of full-length versus partial, miR482-targeted versus not targeted or clustered versus not clustered did not correlate with the likelihood of differential regulation ( Table S16 ). It is known that about 20% of R-genes in tomato are targeted by the miR482-superfamily (de Vries et al., 2015). In the presence of pathogens, microRNA processing is down-regulated (Shivaprasad et al., 2012; de Vries et al., 2018). This should lead to a release of the suppression and consequently up-regulation of R-genes targeted by miR482 members in pathogen-infected plants. We tested whether R-genes predicted to be regulated by miR482 were over-represented in the class of up-regulated R-genes in the presence of pathogens. This was not the case. The R-genes predicted to be targeted by miR482 were neither enriched nor depleted in the set of differentially regulated R-genes (p-value > 0.05; Table S17 ).

We evaluated the level of shared differential regulation between plants treated with pathogens versus treated with putatively beneficial microbes. In tomato, only three R-genes were differentially regulated in the presence of beneficial microbes, two of which were also differentially down-regulated in pathogen treated plants ( Figure S17 ). In potato, a larger number of genes were differentially regulated in the presence of beneficial microbes and a large proportion of these overlapped with the genes that are differentially expressed in pathogen treated plants ( Figure S17 ). Only a single R-gene (PGSC0003DMT400014280) was differentially up-regulated in plants treated with beneficial microbes and was not differentially expressed in plants treated with pathogens. For the set of R-genes that are exclusively up- or down-regulated in pathogen treatments, most are limited to specific pathogen treatments, showing a high degree of pathogen specificity. Therefore, although broad-scale, shared up-regulation or down-regulation of specific genes is not detected across experiments in which plants were inoculated with different pathogens, some genes do show pathogen specific regulation ( Tables S12 , S13 ).