Phenotyping of host-pathogen biological interactions
Cross-compatibility among rice / P. oryzae paired samples was assessed through cross inoculation experiments. One rice accession (HO-Q-F16) could not be included since seeds were lacking, and the corresponding P. oryzae isolate (CH1866) was also excluded. The remaining 45 rice accessions were separated in two trays, each tray containing 22 or 23 accessions plus the two highly susceptible rice accessions CO39 and Maratelli (Gallet et al., 2016) used as positive controls, with six rice seeds sown per variety (i.e. 144 or 150 plants per tray). As many batches as P. oryzaeisolates (i.e. 45) of such two trays were prepared (i.e. 90 trays in total). Trays were inoculated four weeks after sowing when plants had 4-5 leaves, each batch of two trays was inoculated with a singleP. oryzae isolate. The fungal inocula were composed of conidia suspensions at 50,000 spores/ml and 25,000 spores/ml for the first and second repetition of the experiment, respectively. Spore suspensions were supplemented with 0.5% gelatin (Gallet et al., 2016). An inoculation corresponded to the spraying of the spore suspension of one particular isolate on one batch of two trays. All inoculations were performed at the same date. Seven days after inoculation, symptoms were read on four plants for each blast genotype × rice genotype interaction; the four corresponding leaves were glued on sticky papers and scanned for subsequent scoring of symptoms. The experimental design was repeated twice. Twelve other rice accessions were removed from the analysis because they were difficult to multiply, did not grow well in our controlled conditions, could not be assigned to any rice genetic cluster (BJ-Q-B06, Fig. 3 left panel), or were of modern origin (HongYang accessions). We thus ended with a matrix of 33*33 rice / P. oryzae paired samples with complete results.
Qualitative interactions were noted on each leaf qualitative scale of 1-6: scores 1-2 corresponding to incompatible reactions showing no symptoms, scores 3-6 corresponding to compatible reactions (Gallet et al., 2014). The percentage of compatible / incompatible reactions for each interaction was estimated by counting the total number of compatible / incompatible leaves among the total number of leaves scored over the two experimental repetitions. When less than four leaves were available in total, the data was considered as missing. We verified that the qualitative scores of the two independent replications were positively correlated (Supplementary Information SI3, Fig. SI3.2). To obtain quantitative measure of host-pathogen compatibility phenotype, the scanned images were analysed with Ebimage package implemented in R statistical environment (Pau, Fuchs, Sklyar, Boutros, & Huber, 2010). In-house scripts were used for calibration and image analyses (https://github.com/sravel/LeAFtool). Briefly, calibrations were made according to discriminant analysis of RGB composition of pixels chosen and classified by the user as lesion, leaf and background, and the resulting discriminant functions were used to assign pixels of the entire image to these three categories. Statistical analyses were performed using the nlme package in R environment. The studied variable was the percentage of diseased leaf area. After log-transformation of this variable (y = log(x + 0.15)), we performed a two-step analysis. First we performed an ANOVA considering only the positive controls to evaluate the respective effects of the following factors: “repetition”, “tray”, “P. oryzaeisolate”, “rice accession”, and the interaction between the last two. We obtained significant effects for fungal isolate (F = 5.13, P = 2.4e-10, df = 32) and rice landrace (F = 416.9, P < 10-16, df = 1); the effect of tray was significant (F = 1.7, P = 0.003, df = 98) but neglectable compared to the effect of experimental replicate (F = 181.7, P < 10-16, df = 1), and was therefore ignored in subsequent data analyses. We then analysed the log-transformed variable for the rest of the matrix (excluding the positive controls) using an ANOVA considering two factors: “repetition” and “combination” (corresponding to each P. oryzae isolate × rice accession combination). Heatmaps of the adjusted value of the log-transformed variable were drawn using ggplot2 package in R environment.
We analyzed nestedness and modularity of the quantitative interaction matrix following Moury et al. (2021). Nestedness and modularity are quantitative properties of matrices that reveal non-random distribution of links between rows and columns. Nestedness measures the tendency of hierarchical organization between lines and columns. Modularity measures the tendency of such matrices to be organized in different modules, with highest probability of strong interactions between members of the same module. Briefly, quantitative trait values were transformed into integers from 0 to 9, by defining ten intervals with equal sizes spanning the range of quantitative values, so that “0” and “9” grades correspond to the minimal and maximal trait value, respectively. We used the Wine algorithm (Galeano et al., 2009) for nestedness estimation, and the spinglass algotithm (Newman and Girvan, 2004) for modularity estimation, since these methods were shown to be the most statistically powerful (Moury et al. 2021). The significance of both estimates was assessed by generating 100 random matrices using the null models R1 (random matrices generated row by row ensuring that the total sum of the cells and the number of zero-valued cells are the same as in the actual matrix) and R2 (random matrices generated row by row by shuffling the cell values of the actual matrix), since these two null models were shown to exhibit the lowest false negative rates (Moury et al. 2021).