Plant materials and plant inoculation
Cassava plants were grown from mature stem cuttings and kept in a greenhouse at 26-30°C, with 12 h day-light photoperiod and 80% relative humidity. Cassava variety MBRA685 (resistant to Xam-CIO151) was used for small RNA library construction and variety MCOL1522 (susceptible to Xam-CIO151) was employed for RT-PCR experiments.
Six-week-old plants were inoculated with 36 h-old cultures of the aggressive Xanthomonas axonopodis pv. manihotis strain CIO151 in both leaves and stems. Leaves were inoculated by piercing six holes in the mesophyll and placing a 5 μL drop of a liquid Xam-MgCl2 culture calibrated at OD600 nm = 0.002 (1 × 108cfu/ml). Two leaflets per leaf and three leaves per plant were inoculated. Stems were inoculated by puncture. At least three plants per collection time were inoculated for each experiment.
Small RNA library construction and sequencing
For the inoculated library leaves and stems were collected from inoculated plants (0 h post inoculation -hpi, 6 hpi, 24 hpi, 2 days post-inoculation -dpi, 5 dpi, 7 dpi and 15 dpi).. RNA extractions were made using a LiCl-acid phenol:chloroform method. RNA extractions from inoculated plants (at least six plants per time point) were pooled together in equal amounts. For the non-inoculated library a single RNA extraction from tissues from six untreated plants was used.
For library construction, adapters were added to total RNA (150 μg/mL) and cDNA was synthesized using a Superscript Double-Stranded cDNA Synthesis kit (Invitrogen). cDNA was enriched through PCR and the small RNA fraction (10-50 nt) was then separated on a denaturing polyacrylamide gel. Libraries were constructed at the BC Cancer Agency's Michael Smith Genome Sciences Centre http://www.bcgsc.ca/ and sequenced with Illumina SBS deep sequencing technology using an Illumina Genome Analyzer IIx http://www.illumina.com/.
Analysis of small RNA sequencing data
Various quality filters were applied to raw reads data. Sequences having less than 0.6 chastity value (measured by sequencing software) in the first 25 bases were removed. Adapter sequences were then removed with an in-house C++ program; this program removed any sequence fragments larger than 10nt from the original sequence matching the adapter sequence used in the libraries (5'-ATCTCGTATGCCGTCTTCTGCTTG-3') and sequence fragments shorter than 10nt were only removed if they started after the 15thnt in the original sequence. The program used the EMBOSS wordfinder tool  to find adapter fragments (parameters: minimum match score = 70, alignment width = 6, gap opening 0.0 + gap extension 10 or gap opening 500 + gap extension 10). Sequences shorter than 15nt, as well as low-complexity and low-quality sequences, were then removed using UNIX commands.
Reads were mapped against known snoRNAs, tRNAs and rRNAs obtained from the Rfam Database  and then removed. This was done by using Blastn, v. 2.2.21 (parameters:e-value < 0.0001, ungapped, word size = 4) . Unless indicated, these were the standard parameters used for all mapping analysis.
Reads were then mapped to the cassava genome v. 4 (Cassava Genome Project 2010, http://www.phytozome.net/cassava) (parameters: e-value < 1e-5, 100% identity) and the preliminary version Xanthomonas axonopodis pv. manihotis genome (Xam genome project, unpublished (parameters; e-value < 1e-5, 100% identity).
miRNA and pre-miRNA identification
To identify phylogenetically-conserved miRNAs, reads from both small RNA libraries were mapped to the set of all mature Viridiplantae miRNAs obtained from miRBase release 16, September 2010  and the complete plant miRNA set obtained from the Plant MiRNA Database PMRD, v. September 2010 . Reads having less than two mismatches with a known miRNA were considered conserved miRNAs . Conserved miRNAs (and their expression profiles) were also identified using the miRProf tool from the UEA sRNA toolkit (default parameters) . Reads from both libraries in the 20-24nt size range were mapped on the cassava genome (Cassava Genome Project 2010, http://www.phytozome.net/cassava) to identify possible novel cassava-specific miRNAs.
Pre-miRNA analysis used the adjacent region (-150, +150 nt) to mapped positions from a read of interest, extracted from the genome using fastacmd . These were then mapped onto annotated cassava genes and regions having large overlaps (> 25%) with genes were removed from further analysis.
Two main criteria were considered for real pre-miRNAs: structural features identified from predicted foldings and a secondary structure statistical test. Candidate pre-miRNAs were folded with RNAFold from the Vienna RNA package  and mFold v. 3.5 . Additional secondary structures for easier visualization were obtained using the RNAfolding utility in the sRNA toolkit . Structures were analyzed with in-house pearl scripts. Real pre-miRNAs were considered if they had less than six mismatches between predicted mature miRNA and miRNA*, few (maximum three) and short (less than 3nt) asymmetric bulges in the structure [38, 59]. Secondary structure minimum folding energy (MFE) significance was calculated using a statistical test; 1,000 random sequences were generated for each possible precursor, maintaining the same base composition and dinucleotide frequencies (k-let = 2) using ushuffle . Random structures were then folded using RNAfold  and the p-value was calculated as the percentage of random structures having an MFE equal to or lower than the original precursor . Real pre-miRNAs must had < 0.05 p-value.
Candidate novel pre-miRNAs were analyzed and filtered using miRcheck (default parameters)  MIREAP (parameters, B = 55, a = 19, b = 24, u = 1,000, e = -10 kcal/mol, d = 200, p = 7, v = 10, s = 100, f = 10) https://sourceforge.net/projects/mireap/ and miPred (random-forest prediction module not used)  to obtain additional prediction support.
miRNA targets prediction
miRNA targets were searched using a modified miRanda version v. 2.0 September 2008 , as previously described to meet plant miRNA:target pairings criteria . Targets were also searched using psRNAtarget, v. December 2010 (default parameters, http://plantgrn.noble.org/psRNATarget/), an update to the miRU software . Targets were searched in cassava coding sequences (Cassava Genome Project 2010, http://www.phytozome.net/cassava).
miRNA, miRNA* and transient siRNAs in silico quantification
Reads from both libraries were mapped strand-specifically to quantify miRNA and miRNA* expression using Blastn (e-value < 0.0001, 100% identity, S = 1 for miRNA and S = 2 for miRNA*). Expression values were assigned for each family instead of each loci due to high similarity in mature miRNA sequences. Expression values were normalized using this formula: matching reads/total reads × 1,000,000.
Reads ranging from 20 to 24nt from both sRNA libraries were mapped strand-specifically against all identified possible targets for all cassava miRNAs to quantify possible transient siRNAs and mRNA fragments generated by miRNA-mediated cleavage (e-value < 0.0001, 100% identity, S = 1 for miRNA and S = 2 for miRNA). Only the -100nt, +100nt region immediate to the predicted miRNA target gene cleavage site was used for mapping to avoid mapping to random RNA fragments not generated by miRNA-cleavage. Expression values were normalized as the number of reads mapping to target/total reads.
The 1,000 nt upstream region of identified cassava pre-miRNAs, predicted miRNA targets and randomly-chosen cassava genes were extracted for identifying promoters; only complete and high-quality sequences were extracted. Regions having overlaps with genes were delimited to exclude the gene, unless the overlap was longer than 600 nt; in this case, the promoter region was not used. TFBS were identified in these regions as described in .
Wilcoxon paired tests (two-tailed) were made with R version v. 2.12.0 (R Development Team) comparing TFBS frequencies for each desired group of genes obtained as: Number of genes with a given TFBS/Total of genes evaluated.
Semi-quantitative RT-PCR analysis
RNA was extracted from leaves and stem tissues from cassava MCOL1522 non-inoculated and inoculated (4 dpi and 8 dpi) with Xam CIO151, grown as described above, to quantify miRNA target expression. After DNAse I (Fermentas) treatment, cDNA was synthesized using First Strand cDNA Synthesis with oligo dTs (Fermentas). cDNA concentration was normalized after PCR using tubuline primers (F = 5'-GATCCTACTGGGAAGTACATTGG-3', R = 5'-GATCATTCTCCACCAACTGA-3'). Gene-specific primers for predicted miRNA targets were designed covering the predicted cleavage site. PCRs were performed with 32 cycles a 94°C for 30 s, 54-60°C for 30 s and 72°C for 30 s. The primers used for each target are listed in Additional file 3.