The study was conducted in a Scots pine (Pinus sylvestris L.) full-sib progeny test #S23F 711261 “Grundtjärn” (lat. 63.5556° N, long. 17.4139° E, alt. 320 m, area 3.5 ha) located in central Sweden. The test was established for research purposes by The Forestry Research Institute of Sweden (Skogforsk) in 1971 on silty moraine using completely randomized single tree plot design. It comprised of 7,240 trees representing 179 full-sib families that were generated with a partial diallel mating design using 45 parents. Geographic origin of the parents (plus trees) and mating design for this progeny test was described in [3]. The test site was divided into 181 post-blocks to ease orientation and remove environmental effects during data analyses, and each post-block consisted of 40 trees (4 columns by 10 rows with spacing of 2.2 m in both directions). In total 622 trees representing 175 families of 44 parents were included in this study.

Skogforsk, an organization that has the authority to grant permissions to access and maintain this progeny test as well as collect, analyze and publish any data generated therein, approved this study; no other permissions were required. No protected or endangered species were involved.

The micro-drill Resistograph IML-RESI PD300 (Instrumenta Mechanic Labor, Germany) and penetrometer Pilodyn 6J Forest (PROCEQ, Switzerland) were used for estimating wood density of standing trees. The Resistograph was used for drilling trees bark to bark in two mutually perpendicular directions (from southeast and from southwest) at height of ca 1.2 m and special attention was paid to avoiding drilling through knots or visible stem damages. Each profile was checked immediately after drilling on the tool’s screen and the measurement was repeated when necessary. Pilodyn was applied on each stem with bark and without bark in the southwest direction at ca 1.3 m above ground. Both the Resistograph and Pilodyn (with bark) were applied on fissures between bark scales to ensure that the tools’ tips were stable during measurements and to avoid high proportion of bark in the records. The reciprocal of Pilodyn’s penetration depth and an average value of adjusted Resistograph’s records were used as indirect wood density estimates.

The Resistograph IML-RESI PD300 is capable of measuring trees up to 30.2 cm in diameter with a drilling resolution of 100 points per centimeter. Each radial measurement generates a resistogram that includes lower-density bark on both sides and a pith in the center (Fig 1). The overall shape of a resistogram also reveals differences in juvenile and mature wood densities, resulting in a bowl-like shape in the center passing into a plateau or slightly decreasing at the edges, which is typical for conifers [21]; a smooth curve fitted through a scatter plot of SilviScan mean ring density and cambial age makes the natural shape of wood density variation more clearly visible (Fig 2). Aside from the natural shape, most of the Resistograph profiles exhibited an increasing trend caused by accumulated needle friction (Fig 1A).

In order to eliminate bias in wood density estimates, each profile with this issue was detrended, assuming a linear trend. A slope of increase was calculated for each profile separately as tanα=DNRtail−DNRheadLmax (1) where DNRtail and DNRhead are Resistograph density numbers at the tail and head of a raw profile, respectively, and L_max is the total length of the profile. Then every density number point (100 cm^-1) along the whole profile was adjusted as DNTi=DNRi−(Li×tanα) (2) where DNRi and DNTi are Resistograph density numbers of the raw and detrended profiles at point i, respectively, and L_i is the length of the profile from the beginning till position i (Fig 1B). When a Resistograph profile was incomplete, whether due to the stem diameter being too large or due to an abrupt profile ending, the density number was detrended based on the overall-site average slope. Bark, together with the first unfinished growth ring, was removed from each detrended profile in order to obtain a more accurate wood density estimate (Fig 1C). The average density number (Resistograph wood density, RES) was calculated as the arithmetic mean of all density numbers DN_i along a profile. Resistograph wood density was calculated for each raw, detrended, and detrended & debarked profile and for each of the two drilling directions separately (the southeastern and southwestern sides being marked as A and B, respectively) and combined.

Bark-to-bark increment cores with diameter of 10 mm were taken at 1.3 m from southeastern side of each stem in 2011. Prior to the SilviScan (CSIRO, Australia) analysis, pith-to-bark radial strips (2 mm thick and 7 mm wide) were sawn from the cores, soaked in acetone to remove extractives and air-dried under laboratory conditions (23°C and 43% relative humidity). The following traits, obtained from SilviScan analysis, were used in this study: mean wood density (DEN); density of earlywood (EWD), transition wood (TWD) and latewood (LWD); proportion of earlywood (EWP), transition-wood (TWP) and latewood (LWP); microfibril angle (MFA); static modulus of elasticity (MOEs); and fiber wall thickness (FWT), fiber coarseness (FCS) and fiber width in radial (FRW) and tangential (FTW) direction. DEN served as a benchmark for evaluation of the Resistograph and Pilodyn for indirect wood density measurement. Earlywood, transition wood and latewood of each growth ring were defined as sections with densities ranging from 0-20%, 20-80% and 80-100% of the total density range within the annual ring (minimum to maximum), respectively. Since mean values weighted by the ring areas were shown to more accurately represent average wood properties [23], area-weighted values (AWV) for fiber and wood quality traits were calculated as: AWV=∑(αidi)∑αi (3) where d_i is a value for annual ring i with an area α_i [24].

Height (HGT) and diameter at breast height (DBH) were measured in summer 2011 and stem volume (VOL) was calculated according to [25] as a function of height and diameter.

The response variables were fitted into the following linear mixed model using statistical package ASReml 4 [26]: yijkl=μ+Bi+Gj+Gk+Sjk+eijkl (4) where y_ijkl is the lth observation for an offspring of jth and kth parent growing in ith block, μ is the overall mean of a given variable, B_i is random effect of ith block, G_j and G_k are random general combining ability effects of the jth and kth parent, respectively, S_jk is random specific combining ability effect for the cross between parents j and k, and e_ijkl is random error term.

Individual-tree narrow-sense heritabilities (hi2) for each variable were estimated using variance components from the univariate analysis as hi2=σA2σP2=4σG22σG2+σS2+σe2 (5) where σA2,σP2,σG2,σS2, and σe2 are variances for additive genetic, phenotypic, general combining ability, specific combining ability, and residual components, respectively. Standard errors were obtained using Taylor series expansion [26]. Phenotypic and genetic correlations (r_xy) were calculated as rxy=σxyσx2×σy2 (6) where σx2 and σy2 are phenotypic or additive genetic variance components for traits x and y, respectively, and σ_xy is phenotypic or additive genetic covariance component between traits x and y estimated by fitting a bivariate mixed model (Eq 4) [26]. Significance of the correlation coefficients was examined using t-test t=rxyn−21−rxy2 (7) where t is a t-value with n-2 degrees of freedom, n is the number of pairs and r_xy is the Pearson product-moment correlation. Furthermore, hierarchical cluster analysis based on dissimilarity matrix of additive genetic correlations between traits was performed using hclust function in R program [27] in order to construct an illustrative dendrogram. Genetic gain (GAx) for direct selection was calculated as GAx=ihx2CVx (8) where i is selection intensity (1% = 2.665), hx2 is individual-tree narrow-sense heritability for trait x, and CV_x is coefficient of variation for trait x calculated as phenotypic standard deviation divided by the mean. Correlated response (CR_y) to selection for a target trait y was calculated as CRy=ihxhyrxyCVy (9) where h_x and h_y are square roots of narrow sense heritabilities for selection trait x and target trait y, respectively, r_xy is the genetic correlation between trait x and y, and CV_y is the coefficient of variation for target trait y.

Descriptive statistics for wood density estimated by Resistograph, Pilodyn and SilviScan as well as for other wood, fiber and growth traits are summarized in Table 1. SilviScan area-weighted wood density (DEN), used as a benchmark trait in our evaluation study, varied between 365 and 566 kg.m^-3, with mean value of 448 kg.m^-3. On average, wood contained a substantially higher proportion of earlywood (55%) than latewood (15%). The Resistograph unadjusted wood density (RES) ranged from 1242 to 2403 density units from side A, and from 1427 to 2492 density units from side B. Measurements from side A showed a slightly higher variation (10.7%) compared to side B (9.7%). Linear detrending of resistograms lowered wood density estimates, while debarking increased them since the lower-density bark was excluded (RES_T = 1644 < RES_TB = 1848 < RES = 1857). Pilodyn penetration depth measured with bark (PIL) ranged from 15 to 31 mm with an average of 21.7 mm, and it decreased after bark removal (PIL_B) to only between 11 and 22 mm with an average of 16.4 mm. Variation of Pilodyn penetration depth was slightly higher after bark removal (11.4%). VOL exhibited the highest variation among the measured traits (34.5%), followed by MFA (23.8%).

Individual-tree narrow-sense heritability of Resistograph wood density substantially increased after detrending of raw density profiles (from 0.35 to 0.44), reaching a similar value as that of the benchmark variable DEN (0.46) obtained from SilviScan (Table 1). Removing bark however did not bring any further improvement. Pilodyn density heritability PIL (0.32) was close to that attained by the Resistograph using raw data (0.35), but since cutting off bark prior to the measurements did not bring any improvement to the estimate either (PIL_B remained at 0.32), the Resistograph proved to be a superior tool in this regard. Fiber traits exhibited highest heritabilities, particularly FCS (0.57), FTW (0.55) and FWT (0.53), followed by LWD (0.51). Compared to wood and fiber traits, growth traits had lower heritabilities (0.22–0.35).

Phenotypic correlations of unadjusted density RES with the benchmark DEN were moderate (0.53, 0.56 and 0.59 for side A, B and mean of the two sides, respectively) (Fig 3A, Table 2) but they considerably improved following detrending (RES_T; 0.63, 0.68 and 0.69, respectively) and even slightly more after subsequent debarking (RES_TB; 0.66, 0.71, and 0.72, respectively; Fig 3B). Compared to phenotypic correlations, genetic correlations showed only minor differences between the two sides and their mean values. Genetic correlations of the benchmark DEN with mean RES, RES_T and RES_TB increased from 0.89 through 0.93 up to 0.96, respectively. Phenotypic and genetic correlations between DEN and PIL_B (0.44 and 0.74, respectively) were higher than between DEN and PIL (only 0.38 and 0.59, respectively) (Fig 4, Table 2).

Phenotypic and genetic correlations of the benchmark DEN and the indirect wood density variables (RES, RES_T, RES_TB, PIL and PIL_B) with other wood, fiber and growth traits are summarized in Table 3. In most cases, the adjusted Resistograph wood density gave stronger correlations associated with lower standard errors compared to unadjusted measurements (rRES<rREST<rRESTB). Strong positive genetic correlations between Resistograph wood density and the three wood density components (EWD, TWD and LWD) reflected those with benchmark DEN, although they were a little lower. Genetic correlations of the three components with Pilodyn measurements were moderate, but PIL_B had a stronger relationship than PIL. Phenotypic correlations of the components were strong with DEN, moderate with Resistograph measurements and weak with Pilodyn measurements. Correlations with FWT followed a similar pattern. The indirect wood density estimates, including DEN, had strong to moderate positive correlations with LWP but their correlations with EWP and TWP were negative, whereas genetic correlations were in many cases non-significant and phenotypic correlations were associated with high standard errors. DEN showed moderate negative genetic and weak negative phenotypic correlations with MFA. Both genetic and phenotypic correlations of MFA with Resistograph and Pilodyn measurements were very weak. Moderate genetic and weak phenotypic correlations between MOE_s and Resistograph and Pilodyn measurements were lower than those with the benchmark DEN. Growth traits generally correlated negatively with DEN as well as with the indirect wood density estimates (Fig 5). Genetic correlations with DBH and VOL were moderate, while correlations with HGT were weak. Relationships among growth and wood quality traits based on additive genetic correlations are visualized in Fig 6.

Genetic gain and correlated genetic response to indirect selection for target traits using selection intensity of 1% is presented in Table 4. Selection based on VOL and DBH resulted in positive gain for all growth traits VOL, DBH and HGT as well as for MFA (ca 20%, 9%, 4% and 5%, respectively) and in negative gain for DEN and MOE_s (ca -3% and -6%, respectively). On the other hand, selection based on wood density traits (DEN, RES_TB, PIL and PIL_B) generated negative gain for growth traits and MFA and positive gain for wood quality traits. Selection for DEN led to its moderate genetic gain (9%) and a relatively big correlated response for MOEs (18%), MFA (-12%) and VOL (-13%), while selection based on RES_TB, PIL and PIL_B had a lower impact on MOE_s (12%, 8% and 8%, respectively), MFA (-5%, -4% and -2%, respectively) and DEN (8%, 4% and 6%, respectively).

The relevance of inclusion of wood quality traits (such as density, stiffness or strength) into breeding programs of different conifer species, especially when genetic correlations between growth and wood quality traits are unfavorable, has been intensively studied in a number of species (e.g. [4, 8, 12, 15, 22, 28]), particularly in radiata pine [29–31]. Breeding strategies aiming at overcoming such adverse correlations have been designed [32, 33] and economic weights for optimal combination of growth and stiffness have also been applied in real breeding selection [34]. However, the time and cost demanding assessment of wood quality on standing trees, along with a low market pressure, often prevents wood quality traits from being taken into account during selection [15]. Nevertheless, in spite of the lack of interest from the current wood processing industry, breeders should anticipate prospective needs already now, as long-term breeding of forest trees is not capable of flexible responding to rapid technological advances and market changes [22]. Wood density is generally considered to be the most important wood quality trait because it greatly influences wood properties and performance and thus significantly affects wood suitability for different end uses. Integration of wood density into a breeding program, however, requires non-destructive screening of a large number of standing trees.

Pilodyn penetrometer was successfully applied for wood density estimation in some studies (e.g. [15, 35]) whereas it failed in other studies (e.g. [6, 36]). Application of Pilodyn is quick and simple but it has a shallow penetration, low sensitivity and gets affected by thick bark [19]. In this study, Pilodyn was applied for Scots pine wood density estimation of standing trees both with and without bark. Phenotypic correlations of inverse Pilodyn penetration (PIL) with SilviScan wood density (DEN) were weak, while the genetic correlation was moderate (r_G = 0.59) before and relatively strong (r_G = 0.74) after bark removal. Correlations of a similar magnitude were found in black spruce [18] or loblolly pine [37]. Nevertheless, Pilodyn does not seem to be a suitable tool for wood density assessment in Scots pine compared to e.g. Norway spruce or Eucalyptus nitens, for which strong genetic correlations were estimated (r_G = 0.96) [15, 35].

In contrast to Pilodyn, Resistograph micro-drill measures a whole stem profile with a high accuracy. Its main drawback is, however, a need for adjustment of drilling profiles that show an increasing trend as an effect of accumulated needle friction. Approaches for adjustment of Resistograph measurements vary among authors. For example, [38] and [39] applied centered moving averages and centered moving minima to smooth and detrend profiles, while [22] and [18] used only a few first centimeters of the profile and thus eliminated part of the profile exhibiting a steep increase.

In the present study, the whole Resistograph drilling profiles were detrended, assuming that their increasing trend was linear, and bark was excluded from both sides because bark can reach considerable thickness in Scots pine and thus can constitute a substantial portion of the drilling profile. A strong genetic correlation (0.89) between Resistograph and SilviScan densities was attained already with raw data, and an even stronger correlation (0.96) was reached after profiles’ adjustment. The effect of detrending & debarking was more pronounced at phenotypic correlations (0.59 → 0.72). The results indicate that an adjustment of raw drilling profiles is an important procedure to obtain more accurate wood density estimates and that the Resistograph is a reliable tool for non-destructive wood density assessment of standing Scots pine trees. Similar results were also reported e.g. for loblolly pine (r_P = 0.75, r_G = 0.92) [38] or maritime pine (r_P = 0.77–0.79, r_G = 0.96 – 0.98) [22].

Resistograph measurements taken in the southeast direction (denoted as A) showed a higher variation and a lower minimum value compared with measurements taken from the southwest direction (denoted as B). This could be caused by wood distortion and/or wood deterioration in some of the assessed trees, as the ten-millimeter-cores extracted for SilviScan analysis were taken in the southeast direction four years earlier. The Resistograph was applied in a reasonably close proximity of healed boring holes with the intention to obtain profiles comparable to those from SilviScan but, at the same time, some distance (ca 5–10 cm) was kept to avoid drilling through wood affected by the increment core extraction.

Narrow-sense individual-tree heritability of Resistograph wood density varies between 0.14 and 0.64 [8, 18, 40]. In our study, heritability for adjusted Resistograph density was moderate (0.43), similar to that reported by [22] for maritime pine. Estimates of heritability for Pilodyn penetration range from very low to high. [41] reported individual narrow-sense heritabilities between 0.11 and 0.90 for Scots pine growing at different sites, while [37] between 0.06 and 0.46 for loblolly pine measured by different types of Pilodyn. A high heritability (> 0.6) was determined by [35] for Eucalyptus nitens. In this study, individual narrow-sense heritability of 0.32 was estimated for both Pilodyn penetration with bark and without bark. The results are similar to those reported for Norway spruce by [42] and [15].

We have determined negative genetic correlations between wood density and growth traits (-0.47 and -0.44 for DEN vs. DBH and VOL, respectively), which is well in line with earlier studies in Scots pine [3, 24, 43]. Genetic and phenotypic correlations of DBH and VOL with indirect wood density measurements (RES_TB, PIL, PIL_B) followed the same pattern as those with benchmark SilviScan. The only exceptions were phenotypic correlations of PIL, which were double compared to phenotypic correlations of PIL_B or DEN. We assume that trees with larger diameter have thicker bark and thus Pilodyn’s penetration is deeper as the pin has to penetrate through lower-density bark first before it reaches wood. That is the most plausible explanation of Pilodyn’s higher scores in larger trees, which, using the inverse of Pilodyn penetration depth, resulted in a stronger negative phenotypic correlation with DBH than for Pilodyn without bark (PIL_B). Nevertheless, such a big difference between PIL and PIL_B did not emerge in genetic correlations. The effect of Resistograph profiles’ adjustment was clearly visible in the magnitude of genetic correlations and their standard errors. Genetic correlations of DBH and VOL with RES were weak, whereas they were moderate with RES_TB and very close to the benchmark values. A similar pattern between genetic correlations of growth traits with wood density estimated by Resistograph or Pilodyn and genetic correlations of growth traits with benchmark wood density were also published in other studies; either the correlations were moderately negative [12, 15, 18 for a mixed forest only] or near zero [38].