The 10 selected primers amplified a total of 115 reproducible and distinct ISSR bands with a percentage of polymorphic markers of 97.39%. Among the 115 bands scored, 25 bands (21.7%) were found in 50–100% of samples, 27 bands (23.5%) were found in 30–49% of samples, and 63 bands (54.8%) were found in less than 30% of samples. A few population-specific bands were observed in the data set. Table 1 describes the polymorphism in this species revealed by ISSR in detail. The percentage of polymorphic loci (P_p) was 30.49% on average, ranging from 12.17% to 42.61% at the population level. Nei’s genetic diversity (H) was estimated to be 0.0923 at the population level and 0.2820 at the species level, while Shannon indices (I) were 0.1413 and 0.4338 at these levels, respectively. Among the 15 populations, population KM exhibited the highest level of variability (H = 0.1494; I = 0.2239), while population LX exhibited the lowest level of variability (H = 0.0364; I = 0.0565).

Differentiation among populations was significant in this species. The coefficient of genetic differentiation among populations (G_ST, estimated by partitioning of the total gene diversity) was 0.6728. The level of gene flow among populations (Nm) was estimated to be 0.2431, indicating that there was a low migration rate among populations. This finding was consistent with the type of genetic structure predicted by the Shannon’s diversity index analysis, which suggested that 67.42% of the total variation was partitioned among populations. Table 2 shows an estimate of Nei’s genetic identities (G_I) and genetic distance (G_D) for each pairwise comparison between two populations. Genetic identities between populations varied from 0.6936 to 0.9184 with a mean of 0.7764 ± 0.0437. Correlative analysis with SAS software showed that there was no significant correlation either between genetic differentiation and latitude (p = 0.9408), or between genetic differentiation and longitude (p = 0.3368). The AMOVA indicated that most (77%) of the molecular variation in H. cordata populations existed among populations, with lesser amounts within populations (23%). Permutation tests (based on 999 permutations) suggest that the overall ΦPT was significantly different from the null distribution (ΦPT = 0.769, p = 0.010) (Table 3), which indicates the differences among populations are significant.

A UPGMA dendrogram based on Nei’s (1972) genetic distance indicated that the 15 populations were clustered into four geographically distinct groups, including Sichuan basin (EM, YA and YQ), Yungui Plateau (KM and HX), central south China (HH, YX and DW) and southeast China (LX, LC, ZJ, XM and LYG) (Figure 1). The Mantel tests indicated that there was no significant associated relationship between genetic distance and geographic distance among populations in H. cordata (r = 0.0400, p = 0.3760).

The genetic structure of plant populations reflects interactions amongst a range of different processes, including the long-term evolutionary history of the species (such as shifts in distribution, habitat fragmentation and population isolation), mutation, genetic drift, mating systems, as well as gene flow and selection [15]. In this study, the genetic structure of H. cordata was generally characterized by high genetic variation at species level (Shannon indices of 0.4338 and Nei’s gene diversity of 0.2820) and low genetic variation within population (0.1413 and 0.0923 on average, respectively). These results were similar to those obtained using other molecular markers (such as RAPD and SSR) previously in a much smaller sample size [16]. This herb survived and adapted as a result of agamic reproduction in the central and southern areas of the Northern Hemisphere during late tertiary to early quaternary [1]. At the glacial maxima, H. cordata was restricted to refugia with an endurable temperature, and such areas were expected to be small. One of the major genetic consequences of long-term habitation at a small population size is high levels of genetic drift [17,18]. This factor could be, at least partially, responsible for the low genetic variation within population and high genetic variation among populations.

The individuals in different regions were not from the same migratory routes derived from distinct sister groups that once lived in the refuge. In the present study, 15 sampled populations were collected from 13 provinces in China. Latitude and longitude of the areas vary from 21°12′ (N) to 34°36′ (N) and from 101°31′ (E) to 119°36′ (E), respectively. Different geographical distribution ranges, the intricate landform and the weather conditions of the sampled populations might have limited migration. In this study, the coefficient of genetic differentiation (G_ST) of the species was 0.6728, the proportion of diversity among populations based on the Shannon indices was similarly high (67.42%); correspondingly, the estimate of gene flow (Nm) is low (0.2431); and the AMOVA test also showed that there was high genetic diversity among populations (77%) and low genetic diversity within the population (23%). H. cordata is a relic of its agamic race [1], and the plant is small and short. This constrains gene flow via seed and pollen dispersal, and thus the plant fails to reproduce extraneously in each highly isolated population. Cytological examinations revealed that H. cordata is a polyploidy complex with 36 to 126 chromosomes [19], which hinders gene exchanges among populations harboring different numbers of chromosomes. Therefore, these patterns of variation are due not to ongoing gene flow, but rather to historical association among populations. According to the geographical distribution and the affinities shown in the UPGMA dendrogram (Figure 1), H. cordata might have survived in Southwest China during the last Pleistocene glaciation. The route extended southeastward of Southwest China along the Yungui Plateau Range, where the present distribution pattern appeared. The strong differentiation among populations may be mainly due to subsequent genetic drift and natural selection. Genetic drift changes the distribution of genetic variation either by reducing variation within a population or by increasing the differentiation among populations [18]. Although the UPGMA dendrogram (Figure 1) indicated that the distribution of H. cordata in China took on an overall pattern of eastward expansion, the lack of significant correlation between genetic distance and geographical distance supports the hypothesis of historical relationships and the subsequent action of genetic drift accounts for current-day patterns of variation.

A new species H. emeiensis Z.Y. Zhu and S. L. Zhang in the genus Houttuynia was collected at E’mei mountain in Sichuan province, China, and classified by Zhu and Zhang according to the color of the stem, the width and length of the inflorescence, and the number of flowers and bract [20]. However, in this study, Nei’s original measures of genetic distance (Table 2) and the UPGMA dendrogram (Figure 1) failed to provide support for their conclusions. The largest and the smallest genetic distances between EM and other populations was 0.3659 (KM) and 0.1679 (YA), respectively, with a mean of 0.2882 ± 0.0608. The overall correlation between genetic differentiation and geographical distance was weak. Cluster analysis showed strong correlations between EM and YA, YQ and NC, KM and HX, as well as RC and LX. EM was not classified as one group based on ISSR markers. Since H. emeiensis (EM) did not stand alone in these analyses, it is necessary to deliberate in classifying it as a new species.

Our results also provide theory for further protection of the germplasm resources. The low levels of genetic diversity within a population and the low gene flow among populations point towards the possibility of the possession of unique genotypes in a single population that are not found in other populations. It is therefore imperative for conservation planners, when designing conservation strategies for wild populations of H. cordata in China, to ensure that as many as possible separate populations are targeted for conservation rather than a few selected populations. Ex situ conservation may also be appropriate, because the total genetic diversity in a population of H. cordata may be adequately captured in only a few plants from the wild, which would not be the case for species with high levels of genetic diversity within a population. It would be beneficial to find ways to strengthen the gene flow between populations to maintain the natural genetic variation of H. cordata. Considering the high genetic differentiation among the wild populations, preservation of only a few populations may not adequately protect the genetic variation within the species in central China. Therefore, several populations throughout the entire range of the species in the country should be considered for conservation.

Leaf samples were collected from 266 individuals of H. cordata Thunb from 15 populations located in the natural distribution of the species in 13 provinces of China (Table 4 and Figure 2). The samples were taken from individuals at least 2 km apart in each population to avoid sampling clonal relatives. Leaves were dried, and preserved in silica gel until DNA extraction.

Genomic DNA was extracted from the leaves of each individual using CTAB protocol [21], dissolved in double distilled water, and quantified using a spectrophotometer comparing band intensities with known standards of lambda DNA on 1.5% agarose gels. In a preliminary experiment, 100 ISSR primers (designed by the University of British Columbia (UBC), Vancouver, Canada) were tested on five random individuals from all the populations. Ten primers (USB #807(AG)₈T, #808(AG)₈C, #810(GA)₈, # 811(GA)₈C, #812(GA)₈A, #825(AC)₈T, #826(AC)₈C, #848(CA)₈RG, #850(GT)₈YC, and #873(GACA)₄) that produced clear bands were chosen for all the 266 individuals in subsequent PCR reactions. For each primer analyzed, amplification was repeated once in the same situation to test the reproducibility of the ISSR bands, with a sample composed of two or three individuals selected randomly from each population.

PCR amplification was performed in a 10-μL reaction volume containing about 20 ng of template DNA, 1.0 μL 10× PCR buffer, 2.0–2.2 mM MgCl₂, 0.21 mM dNTPs, 200 nM primers, 1.0 unit of Taq polymerase and double-distilled water. Amplification was performed in a PTC-100 thermal cycler (MJ Research, Inc.) programmed for an initial step of 3 min at 94 °C, followed by 35 cycles of 1 min at 94 °C, 1 min at 50 °C or 52 °C and 1.5 min at 72 °C, and a final 8 min extension at 72 °C. A negative control reaction without template DNA was also included to verify the absence of contamination. PCR products were separated in 1.2% agarose gels buffered with 1× TBE. Only bands reproducible in different PCR reactions were further analyzed. Bands with similar migration distance amplified from different individuals were considered homologous. A positive control (PCR products amplified from a standard individual) and a Gene Ruler DNA 100 bp ladder (Bio Basic Inc.) were run on each gel for homology assessments of PCR products on different gels.

The presence or absence of a PCR band was scored as 1 or 0, respectively. Two people scored the bands together in order to minimize the scoring of artificial bands. Only bands that were consistently amplified were considered, while smearing and weak bands were excluded. We refrained from pruning the loci that fulfilled Lynch and Milligan’s [22] criterion because that might lead to significant bias in the estimation of population genetic parameters [23]. This bias was substantially eliminated if a high number of polymorphic dominant markers were generated [24]. Statistical analyses of ISSR patterns were based on the following assumptions: (1) ISSR fragments behave as diploid, and the dominant marker alleles are either present (amplified) or absent (non-amplified); (2) co-migrating fragments represent homologous loci; (3) polymorphic loci are inherited in a nuclear (Mendelian) fashion [25]; and (4) populations are in Hardy-Weinberg equilibrium (HWE; Fis = 0).

Four genetic diversity parameters were analyzed using POPGENE (version 1.31) [26]: effective number of alleles (Ne), Nei’s gene diversity (H) [27], Shannon’s Information index (I) [28], and percentage of polymorphic loci (P_p). To examine population genetic structure, the proportion of genetic divergence was estimated within and among populations as described by Nei [27], and G_ST was designated as the coefficient of gene differentiation among populations. The genetic identity (G_I) and the genetic distance (G_D) among populations were also computed using the model presented by Nei [17]. Gene flow estimates (Nm) were calculated as Nm = (1 – G_ST)/4G_ST [29]. Differentiation among population was analyzed with an analysis of molecular variance (AMOVA) using Genalex 6.1 based on 999 permutations. The AMOVA procedure in Genalex follows the methods of Excoffier et al. 1992, which estimates and partitions total molecular variance within and between populations and then tests the significance of partitioned variance components using permutational testing procedures. When the data are binary, AMOVA calculates the ΦPT value, which is analogous to Fst representing the proportion of the total variance among populations [30–32].

An unweighted pair group method using an arithmetic average (UPGMA) dendrogram was performed on the data matrix of mean character difference between pairs of samples with software MEGA version 2.1 (downloaded from the website at http://www.oup-usa.org/sc/0195135857). The relationship between latitude and longitude and the molecular indices was calculated as a two-polynomial regression. Correlations were calculated with software SAS version 6.12.

Finally, a Mantel correlation test [33] was applied, by using GenAlEx 6, to a matrix of pairwise geographic and pairwise genetic distances, with 999 random permutations to determine significance.