Sample collection and DNA extraction
To develop the microsatellite markers and evaluate the polymorphism of the markers, three natural populations of R. purdomii were collected from Laojun Mountain (LJS, n = 11), Longyuwan (LYW, n = 15) and Laojieling (LJL, n = 17) in Henan province, China, respectively. Furthermore, to validate the selected microsatellite primers in other Rhododendron species, we sampled two populations of R. concinnum from Yao Mountain (CYS, n = 10) and Laojun Mountain (CLJ, n = 10) in Henan, respectively. The locality information of the sampled populations was detailed in Table S1. Fresh and healthy leaves of the investigated individuals were collected and sampled individuals were at least 10 m apart within one site. Permissions were obtained from the local nature reserve for collecting plant materials. Leaves were dried by silica gel and stored in plastic bags until DNA extraction, and voucher specimens were deposited in the Herbarium of Zhengzhou University (ZZU).
RAD sequencing and microsatellite mining
Genomic DNA of all the investigated individuals was extracted from dried leaf tissues using a modified CTAB method . A restriction-site associated DNA (RAD) library of one R. purdomii individual was constructed using the EcoRI (5′-GAATTC-3′) enzyme following Miller et al.  and Baird et al. . The library was sequenced at Novogene (Beijing, China) using the Illumina HiSeq 2000 platform with 150 bp paired-end reads. After filtering low-quality reads, de novo assembling of the clean reads was performed by Velvet  with default parameters. Microsatellite motifs with a repeat unit of 2-6 bp and a minimum number of four repeats were detected using MISA  with default settings. The program Primer3  was used to design microsatellite primers, the length of primers ranged from 20 to 28bp, the annealing temperature was 60-65℃.
Validation of microsatellite loci and cross-species amplification
The validation of microsatellite loci was performed through three steps as follows. Firstly, we randomly selected three individuals from three different populations of R. purdomii to test the success of amplification for the designed primer pairs. Then, the forward primers of microsatellites successfully amplified in three individuals were labeled with fluorescent dye, and fluorescent PCR products of the selected six individuals from three different populations were analyzed for polymorphism. Thirdly, microsatellites showing expected size range on agarose gels, clear peaks and polymorphism during capillary electrophoresis in six individuals were further amplified in all the investigated individuals of R. purdomii. In the above processes, PCR reaction mixture used was 50 μL in volume, consisting of distilled water (22 μL), 2 × PCR Mixture (25 μL, Beibei Biotech, Henan, China), 10 μM forward and reverse primers (1 μL for each primer), genomic DNA (1 μL). The PCR reaction conditions were: an initial denaturation at 94°C for 4 min; 35 cycles at 94°C for 30 s, annealing temperature for 30 s, 72°C for 30 s; a final extension at 72°C for 7 min. The success of amplification was determined by electrophoresis on a 1% agarose gel. The fluorescent PCR products were analyzed through capillary electrophoresis on an ABI 3730XL DNA Sequencer (Applied Biosystems, Foster City, CA, USA) at the Sangon Biotech Corporation (Shanghai, China), and the genotypes were obtained using GeneMarker (Soft Genetics).
Samples of R. concinnum were used to evaluate the transferability of the developed microsatellite markers in congeneric species. The screening of primer pairs suitable for R. concinnum was consistent with three steps described above, excepting that two individuals from two different populations were used for amplification validation and four from two populations respectively were for polymorphism test.
For population genetics analyses, number of alleles (NA), number of effective alleles (NE), Shannon’s information index (I), observed heterozygosity (HO) and expected heterozygosity (HE) were estimated using GenAlEx . Polymorphism information content (PIC) was calculated by CERVUS . Test for Hardy-Weinberg equilibrium and linkage disequilibrium between microsatellite loci was conducted using GENEPOP . Frequencies of null alleles were estimated by Micro-Checker .