Life span variability
Total variance for life span (s2 = 7·24, 100 %) could be subdivided into individuals within populations (s2 = 3·70, 51·1 %), populations within regions (s2 = 0·76, 10·5 %) and between regions (s2 = 2·78, 38·4 %). The differences between populations within regions and the differences between regions were both highly significant (F95,1580 = 4·47, P < 0·001 and F4,100 = 50·92, P < 0·001, respectively). The variance among individuals in the synthetic population (‘1995’ synthetic population and ‘1998’ generation pooled) was 6·45, indicating that the synthetic population covered a substantial proportion (89 %) of the total variance. When observations ceased (i.e. early 2008), 29 plants were still alive: 24 out of 495 from region E, one out of 852 from region D and four out of 120 in the two generations of the synthetic population. Two plants from the base population sown in 1990 were still alive in 2008.
Substantial differences in median life span were found between the geographical regions (F4,95 = 56·35, P < 0·001) as estimated by a one-way ANOVA with a Tukey post hoc test (Table ). Survival curves per geographical region provide a more detailed picture (Fig. ).
Mean life span in the five geographical groups of populations and in the synthetic population (‘1995’ synthetic population and ‘1998’ generation pooled)
Fig. 2. Survivorship curves for the synthetic population S (pooled generations ‘1995’ and ‘1998’) and the combined populations from five regions: A (French Mediterranean); B (inland south-west France); C + D (French Atlantic); (more ...)
Since plants that do not require vernalization (almost all plants from region B and roughly half those from region A) were excluded, the synthetic population was principally derived from regions C, D, E and F and, to a lesser extent, from region A. The median life span of the synthetic population was, as expected, within the range of life spans determined for regions C to F (Table ). The survival curve appeared to be less steep compared with the geographical regions (Fig. ), which is a logic consequence of the combining of groups with different life spans. The variability of life span in the synthetic population had a significant genetic component: heritability (h2 ± s.e.) was estimated to be 0·48 ± 0·13 (P = 0·003).
Flowering date in relation to age
The comparison of plants from the same populations but with a 1-year age difference showed a highly significant, gradually later flowering of 1·31 d per year over all age classes (‘All’ in Fig. ). This value was significantly different from zero (n = 646; t = 3·115; P = 0·002). Each age class showed the same trend (Fig. ): there were no significant differences between them (one-way ANOVA, F6,2071 = 0·853; P = 0·529). It should be noted that variances were relatively high due to the comparison of different plants, often from different maternal origin, although from the same population. It was not possible to compare the same genotype at different ages in this outcrossing, non-clonal species. This approach would have required pure lines, or a comparison of ramets of the same genet.
Number of additional days for 1-year-older plants to flower (mean ± 95 % confidence interval). 1–2 = 1-year-old plants compared with 2-year-old plants, etc.; 7–8+ includes 8–9, 9–10, etc.
Two checks for the validity of this result were made. First, the five available pairs of plants with a 9-year age difference (the maximum in this experiment) showed on the average 12·0 (s.e. = 2·8) d later flowering of the older ones compared with their younger counterparts from the same populations. This illustrates that iteroparous plants can indeed flower substantially later when much older.
The second check was made to exclude the possibility that the effects of later flowering were exclusively or principally brought about by plants in their last year before dying (see the next section). This type of phenology could have been more frequent in the older group, or in short-living ecotypes. Indeed, mortality was slightly lower in the first years of life, although in the whole data set the correlation between life span and the number of years to death was not significant at the 5 % level (n = 912; r = –0·058; P = 0·082). A subset of plants that flowered in at least five consecutive years was analysed for the first four years. With an overall value of 1·21 d later flowering per year (n = 303; t = 2·066; P = 0·040), the results were very similar to those for the whole set of plants, although the confidence intervals were higher due to the lower sample sizes.
A few plants stopped flowering all together after reaching a certain age. They were no longer included in the data set from that age on. This created a slight bias and the effects of ageing may have been slightly underestimated, but only with respect to flowering date as no plants stopped flowering during the first three years of evaluation of seed production and root growth.
The relationship between root size before flowering and the subsequent flowering date was evaluated by calculating the correlation between both traits which was found to be significantly positive (n = 497, r = 0·120, P = 0·007). Excluding plants in their last year of life reinforced the correlation (n = 400, r = 0·185, P < 0·001).
Effects near the end of life
Effects associated with the number of years yet to live (‘years to death’) were only examined in the synthetic population ‘1995’ and later generations and not in the original populations. Plants in their last year of life systematically flowered later than plants that had at least 1 more year to live (0·287 vs. –0·062 standard deviations; n = 908, t = 4·09, P = 0·0005). This corresponds to a difference of about 3·3 d as the average standard deviation over years and generations was found to be 9·4 d. Figure A shows a more detailed picture of the plants that had 1, 2 and 3 or more (‘3 +’) years to live. Flowering date varied significantly among the different classes (ANOVA, F3,904 = 5·427, P = 0·0008) but multiple comparisons (Tukey's HSD test) indicated that there were no significant differences between the three categories with at least 1 more year to live and reproduce. The gradually later flowering with age inferred from Fig. is independent of the later flowering in the last year of life reported here. Here, plants of the same age were compared, only differing in future lifetime.
Fig. 4. Flowering date, annual seed production and annual root growth (data transformed into standard deviations, mean ± 95 % confidence interval) as a function of the number of years yet to live (‘years to death’; 3 + , ≥3 years). (more ...)
Annual seed production was less influenced by the number of years to death. The difference between plants that died or not before the next flowering season (–0·165 vs. 0·030 standard deviations) was significant at the 5 % level (n = 849, t = –2·079, P = 0·038), but no significant heterogeneity among the four classes was revealed (Fig. B; one-way ANOVA, F3,845 = 1·979, P = 0·116).
Annual root growth was considerably lower in the last year of a plant's life. Here a highly significant difference of –0·353 vs. 0·056 standard deviations (n = 868, t = –4·197, P = 0·0003) was found. A more detailed analysis (Fig. C) showed a similar pattern as for flowering date: root growth varied significantly among all four classes (one-way ANOVA, F3,864 = 5·695, P = 0·0006) but no heterogeneity was revealed among the three classes of persisting plants.
For all three variables, tests looking for differences between age classes 1, 2 or 3 were carried out. Age did not have a significant effect on the difference between plants in their last year of life vs. persisting plants (two-way ANOVAs: flowering date P = 0·987; seed production P = 0·394; root growth P = 0·723), nor was there any significant interaction between persistence and age (flowering date P = 0·726; seed production P = 0·067; root growth P = 0·918).
Correlations between annual seed production, annual root growth and flowering date were calculated over all measured plants of the synthetic population and later generations. The plants were subdivided into the same four classes of ‘years to death’ as used in the previous section. For the overall values, only the correlation between flowering date and seed production was strongly negative at a high significance level (n = 845, r = –0·428, P < 0·001). The detailed analysis showed no differences among the number of years to death: χ23 = 3·96, P = 0·265 (Fig. A).
Fig. 5. Correlation coefficients between flowering date, annual seed production and annual root growth (± 95 % confidence intervals) as a function of the number of years yet to live (‘years to death’; 3 + , ≥3 years). Values with (more ...)
The correlation between seed production and root growth was not significant when all plants were considered together (n = 805, r = 0·026, P = 0·465), but was heterogeneous among the years-to-death classes χ23 = 8·63, P = 0·035 (Fig. B).
Finally, the overall correlation between flowering date and root growth was not significant (n = 864, r = 0·004, P = 0·901), while further analysis showed a strongly significant heterogeneity χ23 = 16·81, P = 0·0008 (Fig. C).
Heritabilities and genetic correlations
Combining all available generations, 22 midparent–offspring pairs could be used for the regression analysis only using first-year data for both parents and offspring. All three characters showed significant h2 values (Table ). Phenotypic correlations rP were limited to the first age class and roughly corresponded to the overall values in Fig. but with a lower contribution of the soon dying plants as is to be expected in the first year. Only flowering date and seed production showed a significant genetic correlation; this trait combination also showed the only significant environmental correlation (Table ).
Heritabilities of flowering date, seed production and root growth obtained by midparent–offspring regression and the phenotypic (rP), genetic (rA) and environmental (rE) correlations between all combinations of these traits