*3.1. Plant Material*

Three *P. granatum* cultivars with distinct phenotypes were chosen to reconstruct the complete cp genome: 'Nana' is a dwarf pomegranate, which has a small and sour fruit with hard seeds. 'Tunisia' is a domesticated cultivar characterized as a normal-sized tree with sweet taste and soft seeds. 'Taishanhong' is a widely grown landrace in China, characterized as having a bright red peel with delicious taste and hard seeds. The materials used in this study were collected from the experimental orchard at Nanjing Forestry University. The voucher specimen was deposited in Nanjing Forestry University.

## *3.2. DNA Sequencing, Genome Assembly, and Annotation*

Total genomic DNA was extracted from mature leaves using a modified CTAB protocol. Firstly, 1.0 µg Genomic DNA was sheared into an average fragment size of 350 bp by a Covaris S220 sonicator (Woburn, Massachusetts, MA, USA). Then, the size distribution and concentration of the libraries were determined using an Agilent 2100 Bioanalyzer and qualified by real-time PCR (2 nM), respectively. DNA libraries were sequenced on Illumina Hiseq X Ten (Nanjing, China) for at least 150 bp reads. The raw sequence data reported in this paper were deposited in the Genome Sequence Archive in Big Data Center [56], Beijing Institute of Genomics (BIG) [57], Chinese Academy of Sciences, under the BioProject with the accession number PRJCA001313. After the fragments were filtered and trimmed by the fastp program [58], clean reads were obtained. Subsequently, the high-quality paired-end reads were used to de novo assemble the complete cp genomes using the GetOrganelle program [27] with a combined k-mer of 95,105,125. Genome annotation was performed using the online program GeSeq [59] for the pomegranate cp genomes previously reported. The annotation results were inspected using Geneious [60] and adjusted manually as needed. The cp genome map was drawn using the online tool OGDRAW [61]. The complete cp genomes have been submitted to Genbank with accession number MK603511-MK603513.

#### *3.3. Codon Usage*

The complete cp genome of the pomegranate cultivar 'Taishanhong' was selected to analyze the codon usage pattern. The protein-coding genes with more than 300 nucleotides were extracted according to the annotation file. The GC content of GC1, GC2, and GC3 was calculated using an in-house python script. The codon usage indices were calculated by CodonW v1.4.4, including the relative synonymous codon usage (RSCU), codon adaptation index (CAI), and the effective number of codons (ENC). RSCU values were close to 1 indicating that all synonymous codons encoding the same amino acid were used equally. CAI is used to measure the extent of bias towards preferred codons in a gene. A higher CAI value means a stronger codon usage bias and a higher expression level. ENC is used to measure codon usage evenness. Its value ranges from 20 (extremely biased) to 61 (totally unbiased) [62].

#### *3.4. RNA Editing Sites*

Prediction of the possible RNA editing sites in *P. granatum* protein-coding genes were performed using the online program predictive RNA editor for plants (PREP) suite [63] with 35 genes as reference. Only those sites which had a cutoff value of 0.8 were kept.

#### *3.5. Sequence Diversity*

Four cp genomes from Lythraceae were downloaded from GenBank, including *Lagerstroemia indica* (NC\_030484), *Heimia myrtifolia* (MG921615), *Sonneratia alba* (NC\_039975), and *Trapa maximowiczii* (NC\_037023). These four cp genomes together with that of our newly assembled 'Taishanhong' genome were used to detect the divergent hot spot. Intergenic and protein-coding regions from five Lythraceae cp genomes were extracted using an in-house python script. Multiple sequence alignment was performed using MAFFT [64] and the mean P-distances were calculated using R package 'ape' [65] with Kimura's two-parameter model.

#### *3.6. Structure Comparison*

IR expansion and contraction of cp genomes among the five Lythraceae species mentioned above were analyzed using IRscope (Helsinki, Finland) [66]. We also conducted a co-linear analysis. A pairwise alignment was achieved by the lastz program. The results were visualized using AliTV (Wurzburg, Germany) [67].
