Variety composition and potential native origin
We started our analysis by assessing the variety (coastal or interior) and the potential geographic origin of the seed lots in Northwest America. Both the variety composition and the potential native origin were assessed using multilocus genotype data of the studied seedlings and the reference data set developed by van Loo et al. (2015). Genotypes were not corrected for null alleles. The reference data set represents genotypes of 746 individual Douglas-fir trees from 36 reference populations and covers the natural distribution range in Northwestern America (Figure S1 in the supplementary materials). The reference populations were genotyped with identical nuSSRs to those used in our study.
The variety composition was assessed by applying the software STRUCTURE v.2.3.4, which uses a Bayesian clustering approach by applying the Markov chain Monte Carlo (MCMC) algorithm to allocate individuals to clusters (K) that are genetically similar (Falush et al. 2003, 2007; Pritchard et al. 2000). We used the multilocus data of the reference data set to pre-define 2 clusters (K = 2), representing the two varieties, and probabilistically assigned all individual seedlings to these clusters. Twenty independent runs were applied, one for each cluster (K) with a burn-in period of 50,000, followed by a run length of 100,000 iterations. As recommended by Falush et al. (2003), we used an admixture model that allows the seedlings to have admixed ancestries, permitting it to detect inter-varietal admixed individuals. In addition, we used correlated allele frequencies, as we expected similar frequencies due to the common ancestors of the two varieties. For each independent run, the software STRUCTURE estimates a membership coefficient (Q), which corresponds to the probability of an individual belonging to each cluster. An individual was declared as coastal with Q > 0.90, as Rocky Mountain with Q < 0.10 and inter-varietal admixed with 0.90 > Q > 0.10. To find the optimal alignment of the 20 independent runs, the software CLUMPP v.1.1.2. was used. The estimated individual membership coefficients (Q) were averaged using the greedy algorithm in CLUMPP to correct for discrepancies between runs (Jakobsson and Rosenberg 2007). Then, the average Q values were plotted using DISTRUCT v.1.1 (Rosenberg 2004).
The potential geographic origin of the studied European seedlings in Northwestern America was estimated using genetic assignments of the software GeneClass2 (Piry et al. 2004). Although we knew the origin of the American seedlings, we also applied the analyses of these seedlings to test the accuracy of the applied assignment methodology. Seedlings of all seed stands (here S01-S10) were assigned to the reference populations of the reference data set. Three types of methods for likelihood estimation are available in GeneClass2. In our study, we applied two of them: (1) the frequency-based method after Paetkau et al. (1995) and (2) the distance-based method using Nei’s (1972) standard genetic distance-based criteria according to Takezaki and Nei (1996). The frequency method after Paetkau et al. (1995) assigns an individual or group of individuals to the reference population with the highest likelihood (score) according to its multilocus genotypes and the allele frequencies (Hauser et al. 2006). We applied a threshold likelihood value for assignment of p < 0.05. The distance-based method (Nei 1972; Takezaki and Nei 1996) assigns an individual or group of individuals to the reference population according to the smallest genetic distance (Piry et al. 2004). We use Nei’s (1972) standard genetic distance D
S to calculate a rank with the corresponding scores (%) to all reference populations. Commonly, frequency-based methods are more powerful than distance-based methods (Hauser et al. 2006). The distance-based methods tend to be less sensitive to violations of the Hardy–Weinberg Equilibrium (HWE) (Hauser et al. 2006), which could be the case in European populations as they might originate from small, fragmented but also cultivated variety mixed stands. This suggests using both methods.
All seedlings of the stands S03, S04, S05, S07, S08, S09 and S10 were assigned to the coastal variety (Fig. 3). For the Austrian stand S02, three of 30 individuals were assigned to the interior variety and four individuals were allotted to be inter-varietal admixed seedlings. The remaining 23 seedlings were assigned to the coastal variety. For population assignments, S02 was analysed with 30 individuals to test if it is derived from the admixture zone in the natural distribution range. Furthermore, S02 was separated into the 2 varieties (coastal and interior Douglas-fir) to investigate if it was established by two different seed sources. One inter-varietal admixed individual was detected in the US stand S06 and the Austrian stand S01 (Fig. 3).
The assignment results for the potential native origin are given in Table 2 and Fig. 4. The frequency-based assignment method of Paetkau et al. (1995) showed score values of correctly assigned populations of 100% in almost all studied populations. Accordingly, seedlings of the European stands (S01, S02, S03, S05) matched the reference populations (R) of the coastal variety R11, R15, R16 and R11 in the Western Cascades in Washington, respectively. The seedlings of S04 were allocated to R32, a coastal variety from Vancouver Island in British Columbia, Canada. Three seedlings of the Rocky Mountain variety of S02 referred to R18 in British Columbia. Results of S02 showed that the population with 30 variety mixed and admixed individuals, and if separated by variety (27 coastal individuals), was allotted to the same reference populations (Table 2). Hence, we conclude that S02 includes two different seed sources and does not originate from the admixture zone. The seedlings of the native US populations were assigned to reference populations in the Western Cascades in Washington and Oregon with the following results: S06–S08 (Darrington) to R15, the seedlings of S09 (Randle) to R11 and those from S10 (Trout Lake) to R05.
When using the second method based on Nei’s distance approach, the same reference populations for S01, S03 and S05-S10 were identified. However, the populations S02 and S04 were assigned to R30/R27 and R30, respectively. The likelihood score values, a measure for the probability of a studied population belonging to the reference population, ranged between 5.2 and 11.1%. We built 10 small subsamples with 3 randomly selected individuals of the reference population R18 and R27 to analyse the reliability of group assignments with 3 interior individuals for S02, (Table 3). The subsamples were assigned to reference populations following identical steps as for the other assignments. The randomly selected individuals were left out in the corresponding reference population. Results showed that the subsamples (1–10) were correctly assigned to R27 and R18 (Table 3).
Using the software GenAlEx v. 6.5 (Peakall and Smouse 2006, 2012), we assessed the genetic diversity of the seedlings using the following diversity parameters: (1) allelic diversity (Na), (2) effective number of alleles (Ne), (3) observed frequency of heterozygotes (Ho), (4) expected frequency of heterozygotes (He) and the (5) inbreeding coefficient (Fis). Fis values were tested for significance (p 0.05) of their deviation from the HWE (Hardy–Weinberg Equilibrium), also using the software GENEPOP v.4.1.4 (Raymond and Rousset 1995; Rousset 2008). We further calculated the allelic diversity parameters (6) allelic richness (As) and (7) private allelic richness As(p) (the number of distinct alleles private (specific) to a population based on a standardized population size by rarefaction using ADZE v. 1.0) to allow a comparison of populations with different sample sizes (Szpiech et al. 2008). We applied a standardized population size of 8 individuals, which is the same as used in the reference data set (van Loo et al. 2015).
The procedure allowed us to compare the allelic richness of the European seedlings with their corresponding native reference population since it accounts for differences in the sample size. If a population consists of a mixture of both varieties (as was the case with the Austrian S02), the genetic diversity indices by variety were calculated as well.
An independent t test was applied (with PASW Statistics for Windows Version 18.0) to evaluate the differences between the mean of the two independent groups. We compared the genetic diversity parameters between seedlings from the US and the European stands and the allelic richness parameter (As
8) of the seedlings from European stands with the reference populations from van Loo et al. (2015).
The results for the calculated genetic diversity indices (Na, Ne, Ho, He, Fis, As
8 and As
(p)) of the European and American seedlings are given in Table 2. For S02, these indices were additionally calculated for the coastal Douglas-fir variety. The t test revealed that Na (p = 0.021), Ne (p = 0.001), H
E (p = 0.023) and As
(p = 0.015) were significantly higher within the US populations versus the European populations. Other diversity parameters (Ho, Fis and As
(p)) of the European seedlings were within the range of the American seedlings. The Fis were significantly positive, indicating deviations from HWE. This could confirm previously published data on the Douglas-fir. The comparison with the assigned native reference populations R11, R15, R16, R32 (Table S1 in the supplementary materials) of the frequency-based approach, revealed that the allelic diversity parameter As
of the European seedlings was significantly lower (p = 0.008).
For the statistical analyses of the morphological characteristics, the PASW Statistics for Windows Version 18.0 was used. The Levene test was applied to verify the assumptions of equal variances. For equal variances, an analysis of variance (one-way ANOVA) using the post hoc test Scheffé for multiple comparisons was used at a significance level of α = 0.05. If the assumption of homogeneity of variances was violated, a one-way ANOVA with Welch’s F-test using the post hoc test Games-Howell for multiple comparisons of unequal variances was performed at a significance level of α = 0.05.
The adjusted Welch ANOVA determined a statistically significant difference between seedlings of the 10 seed populations (p < 0.001). Multiple comparisons (see Table 4; Fig. 5) of the studied populations indicated a significantly higher growth performance of the Austrian seedlings from S02 (µ 16.9 cm) when compared to seedlings from all other studied seed populations, followed by the seedlings of the US population S09 (µ 15.7 cm). The seedlings from the US population S10 (Fig. 6) showed the lowest growth performance with a mean value of 11.5 cm (highly significant). The seedlings from the remaining populations from Europe (S01, S03–S05) and the USA (S06–S08) indicated similar height growth rates ranging from 13.1 to 14.1 cm.
Root collar diameter
The one-way ANOVA determined a statistically significant difference between seedlings of the 10 seed populations (p < 0.001) of the morphological feature root collar diameter (range 1.8 to 2.2 mm) (Table 4; Fig. 5). Again the seedlings of the Austrian population S02 showed the highest growth in diameter (µ 2.2 mm) compared to seedlings from all other populations (significantly, except S01 and S09). The seedlings of the US population S07 revealed the lowest mean diameter at 1.8 mm. The German population S03 (µ 2.1 mm) indicated a significantly better growth performance in diameter compared to the US populations S06, S07 and S10.
The adjusted Welch ANOVA determined a statistically significant difference between the seedlings of the 10 stands (p < 0.001). Multiple comparisons (see Table 4; Fig. 5) indicated the highest ratio in height to diameter in seedlings from the US population S09 (µ 8.4), which is significantly higher compared to seedlings of all other populations, except S02 (µ 7.9). The seedlings of the US population S10 showed the lowest mean value of the sturdiness quotient at 6.3, followed by S03 at 6.6. In all other populations, similar values ranging between 7.0 and 7.4 were found.
Timing of bud burst
The investigated bud burst development stages showed that within 15 days (15.04.2013 to 29.04.2013), all seedlings completed stage number 3, when the first green needles of the terminal bud were visible. The separate examination of lateral and terminal bud burst indicated that 70% of the auxiliary buds flushed earlier than the terminal buds. Following this, the timing of the terminal buds was evaluated. The same statistical analysis, as described in “Morphological characteristics” section, was applied to compare the timing of bud burst. For estimating the effect of the population, blocks and their interaction, a general linear model (GLM) was applied. Therefore, each population was treated as fixed, and blocks as a random variable. Results of the GLM analysis showed that the main effect of population is significant (p 0.015) (see Table 5). Random block effects and population times block interactions were not significant (p 0.167 and 0.234, respectively) (Table 5). The timing of the terminal buds was further evaluated according to Welch’s one-way ANOVA and the statistically significant difference between seedlings of the 10 seed populations was determined (p < 0.001). The post hoc test Games-Howell for multiple comparisons was used to determine which specific seed lots differ. Results of the multiple comparisons showed the earliest bud burst of seedlings from the European stands S01 and S04 with a mean of 8.3 days, which is significantly earlier compared to seedlings from the German population S05 (µ 9.8 days) and the US population S08 (µ 10.1 days) (Table 6; Fig. 7). Bud flushing within the native US populations (S06–S10) ranged between 8.6 and 10.1 days (Table 6). Multiple comparisons showed a significantly earlier timing of bud burst of S09 (µ 8.6 days) as compared to S08 (µ 10.1 days) (Table 6; Fig. 7). In all other populations, no significant differences were found with mean values ranging between 8.9 and 9.6 days.