Cucumber seedlings of D8 and SSSL508-28 were grown in a growth chamber under controlled conditions (65% relative humidity, 28°C/16 h light and 20°C/8 h dark). At 21 days (three-leaf stage), homogeneous plants were selected for inoculation by spraying with a P. xanthii conidia (harvested from naturally infected D8 leaves) spore suspension of 1 × 10⁶ conidia/mL containing 0.01% Tween-20 or sterile distilled water until runoff (Xu et al., 2017). The leaves sprayed with the sterile water served as mock controls. An average relative humidity of 90–100% was maintained after inoculation. The PM-inoculated and control leaves with three biological replicates were harvested 2 days after treatment, quickly frozen in liquid nitrogen, and maintained at −80°C until used for RNA isolation.

Total RNA was extracted from 10 pooled cucumber leaves using Trizol reagent (Invitrogen, Carlsbad, CA, United States). RNA quality was assessed using an Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, United States), and samples with RNA integrity ≥8 were used to generate the sRNA libraries. Twelve sRNA libraries were generated following the protocol of the TruSeq sRNA Sample Prep Kits (Illumina, San Diego, CA, United States). Briefly, the RNA fragments that were 18 to 30 nt were enriched by polyacrylamide gel electrophoresis and ligated to 5′ and 3′ adaptors using T4 RNA ligase. The ligated RNAs were reverse transcribed to generate the cDNAs libraries. Single-end sequencing (50 bp) was performed on an Illumina HiSeq 2500 platform (Illumina) by Biomarker Technologies (Beijing, China).

Clean reads were obtained from the raw reads by removing low-quality and adaptor-containing reads using the FASTAX-Toolkit. Clean reads that were 18- to 30-nt long were retained and aligned to the cucumber draft genome assembly (9930 V2.0¹), Rfam 14.1,² Silva,³ GtRNAdb,⁴ and Repbase⁵ to identify non-coding RNAs using Bowtie (Langmead et al., 2009). The remaining clean reads were mapped to the miRBase database (release 22⁶) to identify known cucumber miRNAs. Up to two mismatches were allowed in the alignments. The remaining unaligned sequences were analyzed using MTide, a probabilistic model-based miRNA prediction software especially designed for plant miRNAs (Zhang et al., 2014), to predict novel miRNAs. The cucumber genome (9930 V2.0) was used as the reference genome. The novel miRNAs that met the following stringent criteria were retained: (1) sequence length 18 to 25 nt; (2) maximal free energy of the miRNA precursor −20 kcal/mol; (3) presence of star sequences and 3′ overhang; and (4) ≤7 nucleotide mismatches between the miRNA and star miRNA. The potential novel miRNA precursors that aligned with tRNA, rRNA, snRNA, or snoRNA sequences were discarded.

The abundance of identified miRNAs in the different samples was normalized to transcripts per kilobase million (TPM). The R-based DESeq2 package (Love et al., 2014) was used to determine differential expression patterns of the identified miRNAs between PM-inoculated and control leaves. An miRNA was considered to be significantly differentially expressed if it exhibited a fold change ≥1.5 or ≤0.67 with p ≤ 0.05.

Four libraries were constructed for degradome sequencing with the RNA extracted from PM-inoculated D8, PM-inoculated SSSL508-28, and the control D8 and SSSL508-28 leaves. An Oligotex mRNA mini kit (Qiagen, Hilden, Germany) was used to separate poly(A) +RNA from the total RNA. The poly(A) +RNA was bound to the mRNACapture beads. A 5′-adaptor was ligated to the RNAs with 5′ monophosphates, followed by reverse transcription with 3′-random primers and polymerase chain reaction (PCR) amplification. The gel-purified products were sequenced using only the 5′-adaptor ligated sequences. Single-end sequencing (36 bp) of the degradome libraries also was performed on an Illumina HiSeq 2500 platform by Beijing Biomarker Technologies (Beijing, China).

After trimming, the clean reads that matched sequences in GenBank and Rfam 14.1 were removed. The remaining 20- or 21-nt high-quality reads were aligned to the cucumber cDNA sequences (9930 V2.0). Perfectly matched sequences were processed using the CleaveLand pipeline (v4.4) to predict potential cleavage sites (Addo-Quaye et al., 2008).

The psRNATarget server⁷ was used to predict the miRNA-mediated target genes. We used the cucumber miRNA sequences as query sequences. The cucumber draft genome assembly (cDNA, 9930 V2.0) was used as the reference database. BLASTn hits with fewer than three mismatches were selected as potential targets. Gene ontology (GO) slim classification analysis was performed using the gene classification toolkit in CuGenDB.⁸

To validate the differential expressions of miRNA identified by sRNA sequencing (sRNA-seq), quantification of selected miRNAs was performed by stem-loop quantitative real-time reverse transcription (qRT)–PCR analysis. Synthesis of first-strand cDNA and the qRT-PCRs were performed using the miRcute fluorescent quantitative detection (SYBR Green) kit (FP401; Tiangen, Beijing, China). Primers were designed using the Vazyme miRNA designer (V1.01; Nanjing, China). The primer sequences are listed in Table 1. The PCRs were performed on an iQ5 multicolor real-time PCR detection instrument (Bio-Rad, Hercules, CA, United States). The relative expressions of the miRNAs were calculated using the 2^–ΔΔCT method with U6 snRNA as the internal control. The PCRs were performed in triplicate. A correlation analysis between the expressions obtained by qRT-PCR and the expression levels obtained using the sRNA-seq data was performed using SAS 9.0 software (SAS Institute, Inc., Cary, NC, United States).

Representative symptoms to PM infection on D8 and SSSL508-28 leaves are shown in Figure 1A. The mean disease indexes of D8 and SSSL508-28 were 37.3 and 2.1, respectively (Figure 1B). The phenotypic differences between the PM-resistant and PM-susceptible lines were determined by observing the extent of PM growth on the leaf surface at 2 days post PM inoculation by scanning electron microscopy. Dense PM hyphae were seen on the surface of the D8 leaves, whereas no conidia were detected on the surface of the SSSL508-28 leaves (Figure 1C). This suggested that differences in the resistance mechanisms between SSSL508-28 and D8 may contribute to the observed difference in phenotype.

To detect miRNAs related to PM resistance, 12 sRNA libraries with three biological replicates were constructed for sRNA-seq. They contained the RNAs extracted from non-inoculated D8 control leaves, PM-inoculated D8 leaves, non-inoculated SSSL508-28 control leaves, and PM-inoculated SSSL508-28 leaves. The numbers of raw reads from each library were between 17.9 and 35.1 million (Table 2). After filtering low-quality reads, adaptor sequences, and reads <16 or >30 nt long, more than 13.5 million clean reads remained in each library (Table 2). After the non-coding RNAs (e.g. tRNAs, rRNAs, snRNAs, and snoRNAs) were removed, the remaining >2.49 million reads in each library had at least one match to the cucumber reference genome (Table 2). The length distribution of all the miRNAs that mapped to the reference genome is presented in Figure 2A. The 24-nt miRNAs were the most abundant in all 12 libraries, which is similar to previous findings in plant species such as Arabidopsis, rice, and soybean. Analysis of nucleotide bias in the 18- to 30-nt miRNAs showed that the 19- to 22-nt mature miRNAs preferentially began with uracil (U) (Figure 2B). In addition, analysis of specific nucleotide occurrence showed a dominant bias for U at the first two positions, and the frequency of cytosine (C) was always lowest at positions 2 to 24 in the mature miRNA sequences (Figure 2C).

To identify the known miRNAs, we aligned the 18- to 30-nt sRNA reads to the known plant miRNA sequences in miRBase release 22 using BLASTn. A total of 170 known miRNAs corresponding to 35 miRNA families, including miR156, miR160, miR171, miR396, and miR408, were identified from the 12 libraries (Supplementary Table 1). The miR156 family was the largest with 20 members, followed by miR166 with 16 members and miR171 with 13 members. The precursor miRNA sequences varied in length from 54 to 403 nt. Of the known miRNAs, 152 and 148 were common between the control and PM-inoculated D8 and SSSL508-28 leaves, respectively, suggesting that expression of the known miRNAs was stable in the two cucumber lines. We also found that members of the same miRNA family exhibited variations in expression levels. For example, Csa-miR156j and Csa-miR156h both belong to miR156 family, and TPM values of Csa-miR156j were 3,553 to 7,519, whereas the TPM values of Csa-miR156h were only 0 to 9.8 in the 12 libraries. These differences in expression levels among different members of the same miRNA family suggest their roles in the PM response are different. Further, six known miRNAs, Csa-miR156o-3p, Csa-miR156q-3p, Csa-miR156v, Csa-miR169k, Csa-miR172b-5p, and Csa-miR477a, were detected in only one of the 12 samples.

A total of 133 novel miRNAs were identified from the 12 libraries (Supplementary Table 2), and all had star sequences. The novel miRNAs were named as Csa_novel-number. The lengths of the novel miRNAs varied from 18 to 25 nt, with 24 nt being the predominant length (102/133). The minimum free energies of their pre-miRNA hairpin structures were −192.5 to −20.9 kcal⋅mol^–1 with an average of −53.2 kcal⋅mol^–1, which corresponds to those of other plant miRNA precursors (Zeng et al., 2018). Csa-novel_12 was the most abundant novel miRNA with a TPM value of >52,000 in all 12 libraries. Forty-nine of the novel miRNAs clustered with known miRNA families (Supplementary Table 2), suggesting they may be recently evolved members of these families.

The read counts for the detected miRNA varied from 0 to 224,965, normalized to 0 to 368,491 TPM, among the 12 libraries (Supplementary Tables 1, 2). Using a fold change ≥1.5 and p ≤ 0.05 as the threshold, we identified 32 differentially expressed miRNAs (DEMs), including 27 known and five novel miRNAs, by comparing the PM-inoculated D8 (ID) with the non-inoculated D8 (NID) libraries (Table 3). Only six DEMs, including three known and three novel miRNAs, were identified by comparing the PM-inoculated SSSL508-28 (IS) with the non-inoculated SSSL508-28 (NIS) libraries (Table 4). Interestingly, all of the 32 DEMs identified in the ID versus NID comparison were upregulated, whereas four of the six DEMs identified in the IS versus NIS comparison were downregulated. This result provides an important basis for further studies of the roles of miRNAs in cucumber PM resistance and confirms that miRNAs may have important roles during pathogen infection.

To verify the identified DEMs obtained by sRNA-seq, seven known miRNAs (Csa-miR156g, Csa-miR171g, Csa-miR160b, Csa-miR162a, Csa-miR396d-3p, Csa-miR395d-3p, Csa-miR398d), and one novel miRNA (Csa-novel_71) were selected, and their expression levels were analyzed by stem-loop qRT-PCR. As shown in Figure 3, the stem-loop qRT-PCR showed that Csa-miR156g, Csa-miR171g, Csa-miR160b, Csa-miR162a, and Csa-miR396d-3p were upregulated only in PM-inoculated D8; Csa-miR398d was downregulated only in PM-inoculated SSSL508-28, and Csa-novel_71 was upregulated only in PM-inoculated SSSL508-28 compared with the corresponding non-inoculated controls. The Pearson correlation coefficient revealed a strong positive correlation (r² = 0.611, p < 0.01) between the sRNA-seq and stem-loop qRT-PCR expression values, which confirmed the reliability of the sRNA-seq data. However, divergences between the two analyses were also observed.

For example, our stem-loop qRT-PCR experiment showed that Csa-miR171g was downregulated in PM-inoculated SSSL508-28, whereas no significant difference was found from the sequencing results.

The discrepancies were also previously described and may be ascribed to sequence bias introduced by sRNA libraries or profiling stem-loop qRT-PCR or to different normalization approaches employed in these two strategies (Pantaleo et al., 2010; Tian et al., 2017).

To identify miRNA targets, we combined target prediction with degradome sequencing. The in silico approach predicted 517 target genes for the 32 DEMs in the ID versus NID comparison and 20 target genes for the six DEMs in the IS versus NIS comparison (see Supplementary Table 3 for details). Ten predicted targets, Csa2G423580, Csa5G497010, Csa3G144230, Csa6G298490, Csa1G153530, Csa2G215520, Csa7G447980, Csa6G510860, Csa2G076520, and Csa3G872160, were common in the two comparisons. They were targeted by Csa-miR395a-3p and Csa-miR395d-3p in the ID versus NID and IS versus NIS comparisons, respectively. The highest number of targets was identified for Csa-miR172c-3p (139 genes), followed by Csa-miR156g (53 genes) and Csa-miR396d-3p (50 genes). Four degradome cDNA libraries (NID, ID, NIS, IS) were constructed and sequenced to detect miRNA-guided cleavage products. A total of 16,479,774 degradome tags were obtained and mapped to the cucumber genome (9930 V2.0), with approximately 59.7% (over 9.8 million) matching perfectly to the genome. Based on the strength of the degradome signal at the miRNA target sites, a total of 67 target genes corresponding to 101 miRNAs were identified (p < 0.05) (Supplementary Table 4). The 67 target genes fell into three categories (0, 1, and 3). Approximately 85% (57 genes) of them were in category 0, which represents the most abundant degradome tags corresponding to the cleavage site and matching cognate transcripts (Zhu et al., 2019). The highest number of targets (60) was identified for IS, and the lowest number was identified for NIS (35). All the DEM–target pairs identified by degradome sequencing were among those predicted by psRNATarget. These two methods also showed that one target gene could be cleaved by multiply miRNAs. For example, Csa3G020600 (a GRAS TF) was cotargeted by Csa-miR171f-3p, Csa-miR171a-3p, and Csa-miR171g at the same location (Figure 4).

Gene ontology slim classification analysis was carried out to elucidate the potential functions of the DEM targets in response to PM inoculation. We obtained 86 different GO annotations for the target genes (Supplementary Table 5). The GO terms including carbohydrate metabolic process (GO:0005975, 31 genes), lipid metabolic process (GO:0006629, 16 genes), photosynthesis (GO:0015979, five genes), signal transduction (GO:0007165, 28 genes), and response to endogenous stimulus (GO:0009719, 13 genes) were among others (Figure 5A). It has been suggested that miRNAs may provide genetic switch mechanisms to essentially inactivate the target genes by regulation of TF functioning and TF-mediated events (Chen et al., 2011; Arora et al., 2013). In total, 77 targets belonging to 19 TF families were detected, which accounted for nearly 15% of the target genes of the 32 DEMs in ID versus NID. These TFs included members of the MYB (20 genes), NAC (nine genes), and SBP (eight genes) families (Figure 5B; see Supplementary Table 6 for details).

Given that miRNAs negatively regulate their targets, we considered that true targets will be downregulated when the miRNA is upregulated under the same PM-inoculation treatment and vice versa (Xu et al., 2019). Thus, the expression of PM-responsive DEMs (from the current sRNA-seq data) and their target genes (from the published RNA-seq study that had the same experimental design, GSE81234) were compared. We identified 92 and three miRNA–mRNA interaction pairs in D8 and SSSL508-28, respectively. A complete list of these DEMs and their negatively regulated targets (p < 0.05) with the log2 fold changes and annotations can be found in Supplementary Table 7. The identified miRNA–mRNA pairs accounted for only a small proportion of the predicted targets. The remaining targets either showed no significant change in their expression levels or had similar expression patterns as the corresponding miRNAs.

Among the 92 miRNA–mRNA interaction pairs in D8, two were located on the substituted segment (Chr5:16,676,542–23,484,079 bp; Xu et al., 2017) that includes Csa-miR172c-3p and Csa-miR395a-3p. Eight targets of Csa-miR172c-3p that were negatively regulated upon PM inoculation were classified as TFs, namely, two Dofs (Csa1G009790 and Csa1G033250), one TALE (Csa6G426360), one AP2 (Csa2G279250), one LBD (Csa5G219380), one Trihelix (Csa2G350410), one bHLH (Csa1G612950), and one CO-like (Csa7G031530) (Supplementary Table 7). Among the three miRNA–mRNA interaction pairs in SSSL508-28, three of the DEMs were located on the substituted segment (Csa-miR395d-3p, Csa-miR398d, and Csa-miR398b-3p), and their targets included a polygalacturonase (PG, Csa7G433170) and eukaryotic translation initiation factor 2α (eIF2α, Csa4G011690). A gene encoding UDP-glycosyltransferase 1 (UTG, Csa1G526840) was the target of Csa-miR398b-3p. Only one negatively regulated target of Csa-miR162a, Csa5G524850 (L-aspartate oxidase, Chr5:18,665,821–18,670,912 bp), was located on the substituted segment.

We further investigated the expression dynamics of Csa-miR172c-3p, Csa-miR395a-3p, Csa-miR395d-3p, and Csa-miR398b-3p at 12, 24, 48, and 96 h after PM inoculation by stem-loop qRT-PCR. The details are presented in Figure 6. Csa-miR172c-3p and Csa-miR395a-3p shared similar expression patterns in response to PM inoculation in D8 as compared with SSL508-28. Csa-miR395d-3p was downregulated in SSL508-28 at 48 h and upregulated in D8 at 96 h after PM inoculation. Csa-miR398b-3p was repressed congruously in PM-inoculated SSL508-28. These results further confirmed that these DEMs may be involved in the cucumber–PM interaction.