Structural Variant Detection in Maize Genomes

Leaves were used to prepare high molecular mass DNA and optical genome maps were constructed as described above for B73. Structural variant calls were generated based on alignment to the reference map B73 v4 chromosomal assembly using the multiple local alignment algorithm (RefSplit)^{32 (link)}. A structural variant was identified as an alignment outlier^{32 (link),49 (link)}, defined as two well-aligned regions separated by a poorly aligned region with a large size difference between the reference genome and the map or by one or more unaligned sites, or alternatively as a gap between two local alignments. A confidence score was generated by comparing the non-normalized P values of the two well-aligned regions and the non-normalized log-likelihood ratio^{50 (link)} of the unaligned or poorly aligned region. With a confidence score threshold of 3, RefSplit is sensitive to insertions and deletions as small as 100 bp (events smaller than 1 kb are generally compound or substitution and include label changes, not just spacing differences) and other changes such as inversions and complex events which could be balanced. Insertion and deletion calls were based on an alignment outlier P-value threshold of 1 × 10⁻⁴. Insertions or deletions that crossed gaps in the B73 pseudomolecules, or that were heterozygous in the optical genome maps, were excluded. Considering the resolution of the BioNano optical map, only insertion and deletions larger than 100 bp were used for subsequent analyses. To obtain high-confidence deletion sequences, sequencing reads from the maize HapMap2 project^{8 (link)} for Ki11 and W22 were aligned to our new B73 v4 reference genome using Bowtie2 (ref. 51 (link)). Read depth (minimum mapping quality >20) was calculated in 10-kb windows with step size of 1 kb. Windows with read depth below 10 in Ki11 and 20 in W22 (sequencing depths for Ki11 and W22 were 2.32× and 4.04×, respectively) in the deleted region were retained for further analysis.