Metabolic dynamics during autumn cold acclimation within and among populations of Sitka spruce (Picea sitchensis)


  • Rebecca Dauwe,

    1. Department of Wood Science, Faculty of Forestry, The University of British Columbia, 2424 Main Mall, V6T 1Z4, Vancouver, BC, Canada
    Search for more papers by this author
  • Jason A. Holliday,

    1. Department of Forest Sciences, Faculty of Forestry, The University of British Columbia, 2424 Main Mall, V6T 1Z4, Vancouver, BC, Canada
    Search for more papers by this author
  • Sally N. Aitken,

    1. Department of Forest Sciences, Faculty of Forestry, The University of British Columbia, 2424 Main Mall, V6T 1Z4, Vancouver, BC, Canada
    Search for more papers by this author
  • Shawn D. Mansfield

    1. Department of Wood Science, Faculty of Forestry, The University of British Columbia, 2424 Main Mall, V6T 1Z4, Vancouver, BC, Canada
    Search for more papers by this author

Author for Correspondence:
Shawn D. Mansfield
Tel: +1 604 822 0196


  • Autumnal cold acclimation in conifers is a complex process, the timing and extent of which vary widely along latitudinal gradients for many tree species and reflect local adaptation to climate. Although previous studies have detailed some aspects of the metabolic remodelling that accompanies cold acclimation in conifers, little is known about global metabolic dynamics, or how these changes vary among phenotypically divergent populations.
  • Using untargeted GC-MS metabolite profiling, we monitored metabolic dynamics during autumnal cold acclimation in three populations of Sitka spruce from the southern, central, and northern portions of the species range, which differ in both the timing and extent of cold acclimation.
  • Latitudinal variation was evident in the nature, intensity, and timing of metabolic events. Early development of strong freezing tolerance in the northern population was associated with a transient accumulation of amino acids. By late autumn, metabolic profiles were highly similar between the northern and central populations, whereas profiles for the southern population were relatively distinct.
  • Our results provide insight into the metabolic architecture of latitudinal adaptive variation in autumn acclimation and show that different mechanisms are the basis of early October cold hardiness and autumn-acclimated cold hardiness.


For many temperate and boreal tree species, cold acclimation exhibits genetic variation in timing and extent along latitudinal gradients, reflecting local adaptation (Howe et al., 2003; Savolainen et al., 2007). These clinal patterns are likely the result of nucleotide variation in both protein-coding genes and their cis-regulatory regions resulting from natural selection (Neale & Savolainen, 2004; Gonzalez-Martinez et al., 2006; Eckert et al., 2009), and may also reflect epigenetic factors (Yakovlev et al., 2010). At the cellular level, population differentiation is a function of variation in adaptive transcriptional responses that likely contribute to the timing and extent of cold-adaptive metabolic remodelling, as well as the expression of cryo-protective proteins (Holliday et al., 2008).

Sitka spruce (Picea sitchensis (Bong.) Carr.) provides a particularly dramatic example of local adaptation to climate across its range, which spans 3500 km along the Pacific Coast of North America, from Northern California to Southern Alaska (Mimura & Aitken, 2007, 2010; Holliday et al., 2010). Strong genetic differentiation in cold hardiness-related phenotypic traits, such as the timing of bud set and the timing and extent of cold acclimation, has been documented in Sitka spruce, wherein as much as 90% of genetic variation in climate-related adaptive traits is distributed among-population (Mimura & Aitken, 2007). At the transcriptional level, substantial variation has been observed among populations along latitudinal gradients in genes with putative roles in seasonal temperature-related adaptation (Holliday et al., 2008). Many of these transcripts appear to be involved in different aspects of primary as well as secondary metabolism, such as carbohydrate, lipid, and anthocyanin metabolism, suggesting that changes in metabolism during cold acclimation may reflect genetic adaptation to local climate. Several recent studies have focused on the metabolic mechanisms that underlie freezing tolerance or cold acclimation in Arabidopsis (Maruyama et al., 2009; Korn et al., 2010). However, seasonal acclimation to temperatures well below freezing is more critical to woody perennials than to annual species. Studies in the model tree, Populus, have shown that metabolic reprogramming occurs in the apical buds and cambial meristematic tissue during the autumn in preparation for cold hardiness and dormancy (Druart et al., 2007; Ruttink et al., 2007). In both Arabidopsis and poplar, major metabolic reprogramming was shown to be induced by cold temperatures and long nights, respectively. Following a critical night length cue, poplar accumulates a range of compatible solutes, including a variety of sugars such as the raffinose family of oligosaccharides (RFO) that are known to be induced by cold temperature and act as cryo-protectants (Druart et al., 2007; Ruttink et al., 2007). The synthesis of the RFOs is strongly associated with plant freezing tolerance (Maruyama et al., 2009; Korn et al., 2010). Although in a previous study we identified functional classes of genes for which the transcription is climate-adapted and modulated during autumn in Sitka, detailed modulation of the primary metabolism and accumulation of compatible solutes during autumn in Sitka spruce have not been thoroughly studied to date, and it remains unknown which metabolic mechanisms underlie latitudinal clines in the capacity to develop freezing tolerance during autumn. To address these questions, we used untargeted metabolite profiling in combination with previously collected transcript profiling data (described in Holliday et al. (2008)) of three populations of Sitka spruce that showed strong phenotypic differences in cold hardiness-related traits. This combined functional genomics approach facilitated the tracking of a temporal sequence of metabolic events and processes during autumnal cold acclimation in the different populations. Furthermore, we were able to estimate genetic variation and population-level adaptation in metabolism related to autumn cold acclimation of Sitka spruce.

Materials and Methods

Plant material and tissue sampling

Sampling was carried out in parallel with the sampling for transcript profiling, and the plant material sampled is described in Holliday et al. (2008). In a common garden on the University of British Columbia (UBC, Vancouver, Canada), three populations of Sitka spruce (Picea sitchensis) were sampled: Valdez, Alaska, USA, 61°N (AK), Prince Rupert, British Columbia, Canada, 54°N (BC), and Redwood, California, USA, 41°N (CA). Four-yr-old seedlings were sampled for foliage at four different time-points, namely, just after budset (T1) (30 August 2004), and then well before (T2), just before (T3), and just after the first frost (T4) (18 October, 22 November, and 1 December 2004, respectively). From the geographically central population (BC), an additional sampling at a fifth time-point, later in December (T5) (13 December 2004), was carried out in order to capture any late responses. At each time-point, samples from 7 to 11 distinct trees per population were collected and analyzed by GC-MS, so that the study comprised, in total, foliar samples of 110 distinct individuals.

Sample preparation and GC-MS analysis

Approximately 100 mg of frozen ground foliage was accurately weighed into a pre-chilled 2 ml lock-cap centrifuge tube. To this, 1 ml of HPLC-grade methanol (CH3OH) containing 0.5 mg of ribitol internal standard was immediately added and vortexed for 10 s to halt biological activity and minimize degradation. The sample was then incubated for 15 min at 70°C with constant agitation, and centrifuged at 14 000 g for 5 min. To a 300 μl aliquot of the methanol extract, 200 μl chloroform (CHCl3) was added, after which the sample was incubated for 5 min at 37°C with constant agitation. H2O (400 μl) was added, vortexed without heating, and then the sample was centrifuged for 15 min at 1350 g to permit the separation of polar (methanol/water) and nonpolar (methanol/chloroform) phases. A 160 μl aliquot of the polar (upper) phase was dried using a SpeedVac (Eppendorf, Hamburg, Germany) (3–4 h, 30°C), and methoxymated by re-suspending the pellet in 50 μl of methoxyamine hydrochloride solution (20 mg ml−1 in pyridine) and incubating with constant agitation for 2 h at 37°C in order to protect carbonyl moieties. To allow the determination of retention time indices, 10 μl n-alkane mixture (C12, C15, C19, C22, C28, C32, and C36, 2 mg ml−1 or 2 μl ml−1 of each in pyridine) was added. Acidic protons were then trimethylsilylated by adding 70 μl N-methyl-N-trimethylsilyltrifluoroacetamide (MSTFA) and incubating at 37°C with constant agitation for 30 min.

GC-MS analysis was conducted as described in Robinson et al. (2007) using a ThermoFinnigan Trace GC instrument, equipped with an AS2000 auto-sampler and a low-bleed Restek Rtx-5MS column coupled to a PolarisQ ion trap system via an electron impact source (Thermo Electron Co., Waltham, MA, USA).

Chromatogram alignment and metabolite level quantification

Peak finding, peak integration, and retention time correction were performed with the XCMS R-package (Smith et al., 2006). Integrated peaks of the mass (m/z) fragments were normalized for sample dry weight and internal standard (ribitol) signal. The XCMS output of integrated peaks was tested for robust integration based on the assumption that, due to the fragmentation, each metabolite detected by MS analysis is represented by at least two highly correlated m/z signals. Only m/z peaks that showed high intensity correlation (Pearson correlation coefficient > 0.95) and highly similar retention time (difference in median retention time after XCMS retention time correction < 0.03 s) with at least one other m/z peak, were maintained. Based on these criteria, groups of m/z peaks that were assumed to originate from the same metabolite were formed and the m/z signal with the highest intensity of such a group was selected as representative signal of the corresponding metabolite and used for further statistical analyses. The corresponding deconvoluted mass spectra and retention time indices of these representative m/z signals were obtained using AMDIS (Automated Mass Spectral Deconvolution and Identification System, National Institute of Standards and Technology (NIST), Gaithersburg, MD, USA; Stein, 1999). Annotation was done using NIST MS-Search software equipped with an in-house mass spectra and retention time library containing data of 513 known trimethylsilyl (TMS) derivatized compounds, in combination with the Q_MSRI_ID library from the Max Planck Institute of Molecular Plant Physiology (Potsdam–Golm, Germany) (Schauer et al., 2005), containing 574 mass spectral metabolite tags (MSTs), of which 306 have been identified. Each metabolite trace was attributed the identification code (ID), created by XCMS for the corresponding representative mass fragment, and following the pattern MxTy, where x is the mass of the representative fragment and y is the retention time in seconds.

Statistical analyses

All statistical analyses were performed using the open source statistical package R, version 2.10.1 ( Integrated m/z traces in GC-MS chromatograms were analyzed by means of paired t-tests to reveal differences in abundance between populations at each time-point, and within each population, between the first time-point and all subsequent later time-points.

Sparse Partial Least Squares Discriminant Analysis (sPLS-DA) was performed on centered and scaled m/z traces. Centering was done by subtracting the m/z trace means from their corresponding m/z trace values. Scaling was done by dividing the centered metabolite traces by their standard deviations. sPLS-DA was performed using the splsda function of the mixOmics R-package, with population or time-point as response categories (Lê Cao et al., 2009; González et al., 2011). Two components were included in the model and per component, 10 variables (m/z traces) were selected. The graphical representations of the sPLS-DA were obtained using the plotVar and plotIndiv functions of the same R-package.

Time- and population-effects on each of the analyzed m/z traces, were initially tested using the full linear model with interaction factor (Model 1), where metabolite ijk is the integrated m/z trace at latitude i, and date j. While only three populations were sampled from across the latitudinal range of Sitka spruce, these populations were representative of the well-documented steep latitudinal cline in bud set timing and cold acclimation in Sitka spruce (quantitative population differentiation Qst = 0.89; Mimura & Aitken, 2007, 2010; Holliday et al., 2010). Latitude of the population of origin was expressed in degrees north. Date was expressed in days past the first harvest time-point. Significance of the model was set at < 0.01. If the interaction effect (Latitudei : Datej) was not significant (> 0.01), the reduced Model 2 was applied, and if, in Model 2, the effect of latitude or date was not significant (> 0.05), the model was further reduced to Model 4 or Model 3, respectively. If the interaction factor in Model 1 was significant (< 0.01), Model 3 was applied separately for each of the time-points, and Model 4 was applied separately for each of the populations.

image( Model 1)
image( Model 2)
image( Model 3)
image( Model 4)

Agglomerative hierarchical clustering was performed employing the hclust algorithm from R, and ‘average’ was specified as agglomeration method. Heatmaps were constructed using the heatmap.2 function from the gplots package in R.


Metabolite profiling

A total of 189 distinct compounds were revealed in the GC-MS profiles of the foliar tissue, and quantified in all samples. Forty-seven compounds could be identified using a comparison of the retention time indices and mass spectrum with existing standard library data, while a molecular class could be assigned to 11 additional compounds (seven carbohydrates, two sugar alcohols, and two organic acids).

To obtain an overview of the temporal dynamics of the autumn metabolome, within each population (AK, Alaska; BC, British Columbia; CA, California), differential accumulation at each time-point relative to the first time-point was assessed for all metabolites. To evaluate population divergence for metabolites, differential accumulation between the populations was also assessed. Fifty compounds were found to accumulate differentially for at least one time-point as compared to the initial time point (P < 0.01), in at least one population (Table 1, Supporting Information Table S1). The number of differentially accumulating compounds in each of the populations was similar (24, 31, and 25 in CA, BC, and AK, respectively), but the fraction of these metabolites which showed decreasing levels during autumn cold acclimation was higher in the AK samples relative to the other populations (25%, 19%, and 48% in CA, BC, and AK, respectively) (Tables 1, S2). Sixty-eight compounds accumulated differentially between at least two populations for at least one time-point, and 32 compounds showed both temporal and population variation. Among the differentially accumulating compounds, the best represented class consisted of those involved in carbohydrate metabolism. Other compounds identified included amino acids, citric acid cycle intermediates, and aromatic amino acid biosynthesis intermediates.

Table 1.   Compounds accumulating significantly differentially during cold acclimation in a given population, or between populations with different latitudinal origin at a given time-point (P < 0.05)
IDCompoundRatios of metabolite levels at different time-pointsRatios of metabolite levels in different populations
CaliforniaBritish ColumbiaAlaskaT1T2T3T4
  1. ID, identification code; T1, time-point 1, 30 August 2004; T2, time-point 2, 18 October 2004; T3, time-point 3, 22 November 2004; T4, time-point 4, 1 December 2004; AK, Alaska; BC, British Columbia; CA, California. Only metabolites with known identity or metabolic class are shown. Fold-changes with P-value < 0.01 are indicated in bold

Amino acids
 M116T459Alanine2.0 2.4     0.5 2.22.2  2.3    0.5  
 M304T894GABA3.34.03.3        0.4 
 M246T975Glutamic acid2.2 4.0       1.83.1          
 M227T955Glutamine  7.0    9.4      6.8       
 M174T679Glycine 0.20.3 0.30.4            
 M158T666Isoleucine0.5      2.9      
 M142T669Proline   1.7                  
 M156T890Pyroglutamic acid1. 0.6 0.60.5 
 M278T736Serine     0.60.5       2.62.7      
Aromatic amino acid biosynthesis
 M345T1182Quinic acid     0.70.6 0.70.7       0.70.7 0.70.8
 M204T1134Shikimic acid 0.7      0.8       1.7 0.6 0.70.7
 M217T1017Arabinose0.  2.1         
 M73T1161D-Pinitol 1.81.8 1.61.7 
 M217T1192Fructose0.2          2.2    
 M433T1878Galactinol 3.15.6 5.2115.6  2.6 2.84.1 3.82      
 M333T1283Gluconic acid lactone0.4           1.8 4.62.5      
 M292T11072-Keto-L-Gluconic acid     1.4   
 M73T1212Glucose0.20.4 0.4   0.3        2.62.7  1.7 
 M387T1506Glucose-6-phosphate     1.7         
 M159T1557Gulose 2.3            1.8       
 M305T1349Myo-inositol  2       
 M361T2067Raffinose 1221 3045464.91116 3.63.7 
 M361T2122Stachyose  0.1 407   217        8.1    
 M361T1694Sucrose  1.4  1.3     1.5 1.62.4       
 M275T1605Unknown (Carbohydrate)     2.8        3.3       
 M289T1113Unknown (Carbohydrate)     2.5        2.2   1.8   
 M73T647Unknown (Carbohydrate)0.6            1.63.1       
 M135T1639Unknown (Carbohydrate)          0.5         0.4 
 M204T1636Unknown (Carbohydrate)            1.5 2.8       
 M73T1259Unknown (Sugar alcohol)    0.70.6
 M217T1087Unknown (Sugar alcohol)0.7 1.4
Citric Acid Cycle
 M273T1145Citric acid  1.6  1.8     1.6      
 M147T859Malic acid 1.52.0 2.42.7 
 M172T683Succinic acid   1.9 0.50.6    
 M368T1809Catechin1.94.64.6 3.4 3.42.1 4.22.0      
 M559T1904Kaempferol            2.1         
 M735T1979Myricetin  1.9                   
 M647T1956Quercetin            2.7         
 M91T15411-Benzylglucopyr-anoside           2.0          
 M371T16401-rac-glycerol        0.4         0.3   
 M254T1654Arbutin          0.5      0.5    
 M173T1169Ascorbic acid   0.4         0.6 1.9      
 M292T706Glycolic acid   2.5 1.9       1.72.8       
 M241T545Monomethylphos-phate 3.62.6   2.6   
 M75T1296Oleic acid        0.5         0.4   
 M232T1104Sinigrin0.6             2.4       
 M341T1429Steric acid ester                  0.3   
 M102T966Tartaric acid     2.9   0.40.4  0.30.4
 M147T930Threonic acid1.74.42.6 2.32.2 2.2  1.62.5  3.3       
 M151T748Threonic acid-1.4-lactone        0.7 0.10.1  0.2       
 M240T688Trigonellin 1.7      2.21.8          2.62.0
 M292T1285Unknown (Organic acid)        
 M333T1097Unknown (Organic acid)           2.62.0 3.7       

The clearest effects that were common to all populations during autumn cold acclimation were increases in raffinose, pinitol, galactinol, malic acid, and catechin, a decrease in glycine, and a temporary drop in glucose levels at the second time-point (T2). The northern population (originating from Valdez, Alaska, USA (AK)) exhibited an increase in stachyose, and a decrease in myo-inositol, an unknown sugar alcohol, GABA, and quinic acid, as well as the temporary peak in glutamine and isoleucine levels at T2. The southernmost population (originating from northern California, USA (CA)) exhibited an increase in GABA, pyroglutamic acid, and citric acid. The increase in stachyose and the decreases in the unknown sugar alcohol and citric acid were also observed in the central population (originating from Prince Rupert, British Columbia, Canada (BC)).

Patterns were generally clinal, showing a strong association with population latitude of origin. Between populations, the strongest differences in metabolite levels were generally found between the AK and CA populations, with the levels of the central population mostly being intermediate. The number of metabolites for which the levels were higher in AK as compared to CA was much larger at T1 (30) than at T4 (15). At the first time-point, raffinose and galactinol levels were similar in the two most southerly populations, but significantly higher in AK. The increase in raffinose and galactinol levels during autumn was however, more substantial in the two southerly populations than in the AK population. Pinitol levels detected at the first time-point were similar in the two northern populations, but significantly lower in the CA population, where the strongest increase in pinitol over the time course was observed. Most compounds that clearly increased during cold acclimation (pinitol, raffinose, and malic acid) had higher levels in populations from northern locations at each time-point. However, for galactinol, no difference between the populations was observed from the third collection time-point onward. Likewise, catechin levels were higher in the AK and BC populations relative to the CA population only at the first two time-points. The levels of GABA, pyroglutamic acid, succinic acid, and alanine evolved during the process of cold acclimation, from an initial higher concentration in more northern populations at the first time-point, to significantly higher levels in more southern populations at later time-points. While concentrations of arabinose and glycine changed over the course of study in all populations, no differences among populations were observed in the levels of these metabolites at any time-point.

Effects of latitude and autumn cold acclimation on individual metabolites

We assessed the extent to which accumulation of individual metabolites was influenced by the latitudinal origin of the population and by the time-point during cold acclimation, using linear models (Table S3). For example, a model for raffinose was the strongest, explaining 71% of its variation. Other significant models, explaining 25–50% of the variation, were obtained for malic acid, tartaric acid, pinitol, catechin, galactinol, pyroglutamic acid, two unknown sugar alcohols, and one additional unknown compound. Among these compounds, both latitude and time-point contributed positively and independently (no interaction) to the levels of malic acid, pinitol, galactinol, an unknown sugar alcohol, catechin, and the unknown compound. Of the 38 metabolites that showed a significant positive effect of date (but no latitude–date interaction effect), 28 metabolites (74%) also showed a positive latitude effect.

A significant interaction between latitude and time-point indicates that the effect of the latitude on the metabolite abundance is different at different time-points, or that the effect of the time-point on the metabolite abundance is different at different latitudes. This was observed for raffinose, tartaric acid, pyroglutamic acid, glutamic acid, GABA, an unknown sugar alcohol, and five unidentified compounds (Table 2). For raffinose abundance, a positive effect of latitude was significant from the first time-point onward, but increased at the subsequent time-points, and the positive effect of the time, during cold acclimation, was higher in populations from higher latitudes (lowest in CA, highest in AK). Whereas the time-effect on raffinose abundance was strongest in the AK population, several metabolites were induced specifically in the southern population. For example, the levels of some metabolites that initially showed a positive latitude effect (higher toward the north) were subject to a positive time-effect (increase during autumn) specifically in the CA and to a lesser extent in the BC population, while their levels in the AK population remained unaffected by time. Together, this resulted in a dissipation of the latitude effect (glutamic acid, M84T895) or even an inversion to a negative (pyroglutamic acid) latitude effect (lower levels toward the north) by late autumn. The abundance of three other metabolites did not initially differ among populations, but increased specifically in the CA and to a lesser extent in the BC population during acclimation (tartaric acid and two unknowns, M135T1773 and M107T1712). In contrast, one metabolite decreased in the AK and to a lesser extent in the BC population (unknown sugar alcohol M73T1259), to become more abundant toward the south by late autumn. The linear models for GABA and the unknown metabolites M248T798 and M433T1523, for which the date effect was negative in the AK but positive in the CA population, suggest that the levels of these metabolites show ephemeral peaks early during sampling of the AK population, but later in the CA population, and between these two extremes in the BC population.

Table 2.   Metabolites that showed a significant (< 0.01) interaction effect between latitude and date in the full Model 1
IDCompoundLatitude coefficients in the reduced models for the different time-pointsDate coefficients in the reduced models for the different populations
  1. ID, identification code; T1, time-point 1, 30 August 2004; T2, time-point 2, 18 October 2004; T3, time-point 3, 22 November 2004; T4, time-point 4, 1 December 2004; AK, Alaska; BC, British Columbia; CA, California. Values in bold indicate < 0.01.

M248T798Unknown1.39E+01 − 1.47E+01− 1.96E+014.28E+00 − 2.83E+00
M102T966Tartaric acid  − 5.10E+01− 4.70E+011.41E+019.06E+00 
M84T895Unknown8.30E+015.92E+01  2.36E+01  
M433T1523Unknown2.25E+02   5.47E+01 − 4.39E+01
M107T1712Unknown − 1.23E+02 − 4.41E+026.75E+01  
M135T1773Unknown  − 3.55E+02− 4.49E+027.59E+013.49E+01 
M246T975Glutamic acid5.79E+02   2.19E+02  
M156T890Pyroglutamic acid9.76E+021.88E+03− 7.84E+02− 1.36E+034.25E+022.14E+02 
M73T1259Unknown Sugar alcohol  − 1.44E+04− 1.50E+04 − 1.53E+03− 2.76E+03
M304T894GABA8.99E+01  − 56.247091.70E+01 − 1.21E+01

Dimensionality reduction using sPLS-DA

In order to assess and visualize genetic clines related to climate adaptation on one hand, and progressive changes in metabolite composition associated with autumn cold acclimation on the other hand, we employed sPLS-DA to jointly select the metabolites that are most strongly associated with temporal and among-population variation. For each of the populations, a two-dimensional sPLS-DA was performed with the time-points as response categories, and for each of the time-points, a two-dimensional sPLS-DA was performed with the populations as response categories.

In AK, sPLS-DA latent variable ξ2 separated the four time-points chronologically, and latent variable ξ1 separated T2 from the other time-points (Fig. 1a). A closer look at the metabolites selection in each dimension and their graphical representations reveals metabolites that are closely associated with either the chronological separation or the specificity of T2 in the AK population: latent variable ξ2 illustrates how raffinose, galactinol, catechin, and malic acid levels increased with later time-points, and latent variable ξ1 indicated a temporary elevation in glutamine, alanine, glycine, and isoleucine, as well as pyroglutamate and succinate levels, along with a temporary decrease in glucose and fructose levels at T2. In both BC and CA, the chronological separation between the time-points was associated with increased levels of raffinose, galactinol, pinitol, tartaric acid, and pyroglutamic acid, and by decreasing levels of glycine. Whereas in the BC population, separation between the time-points by latent variable ξ1 was gradual, in CA latent variable ξ1 clustered the first two time-points and the last two time-points. Interestingly, malic acid levels contributed positively to the chronological separation of the time-points in the northern populations AK and BC, but not in CA, whereas succinic acid levels contributed to the chronological separation of the time-points in CA.

Figure 1.

Graphical representation of two-dimensional Sparse Partial Least Squares Discriminant Analysis (sPLS-DA) models. (a) sPLS-DA models for the populations Alaska (AK), British Columbia (BC), and California (CA), with time-points as response categories. (b) sPLS-DA models for the time-points 1, 2, 3, and 4, with populations as response categories. In both panels the upper rows are scatterplots of the individuals in function of the two latent variables (ξ1, ξ2); the lower rows are variables selected on the two sPLS-DA dimensions. The coordinates of each variable (metabolite) are obtained by computing the correlation between the latent variable vectors (ξ1, ξ2) and the levels of the metabolite in question in the original data set. Two circumferences of radius 1 and 0.5 are plotted to reveal the correlation structure of the variables. Highly correlated metabolites cluster together. Only variables with known identities are shown. Time-point 1, 30 August; time-point 2, 18 October; time-point 3, 22 November; time-point 4, 1 December; time-point 5, 13 December. Ala, alanine; Ara, arabinose; Arb, arbutin; CA, citric acid; Cat, catechin; CH1, unidentified carbohydrate M73T647; CH2, unidentified carbohydrate M388T1646; Frc, fructose; Glc, glucose; GlcLac, gluconic acid lactone; Gol, galactinol; Gln, glutamine; Gly, glycine; Ile, isoleucine; MA, malic acid; meP, monomethylphosphate; OA, unidentified organic acid M292T1285; OlA, oleic acid; Palm, monopalmitoyl-rac-glycerol; pGlu, pyroglutamic acid; Pol, D-pinitol; Raf, raffinose; SA, succinic acid; SAlc1, unidentified sugar alcohol M217T1087; SAlc2, unidentified sugar alcohol M73T1259; ShA, shikimic acid; StA, stearic acid methyl ester; TA, tartaric acid; ThrALac, threonic acid-1,4-lactone.

The sPLS-DA models with populations as the response categories indicated better population separation toward the later time-points (Fig. 1b). On the first sampling date, the two more southern populations clustered together and were partially separated from AK along latent variable ξ1, whereas from T3 onward, CA became strongly separated from the cluster of the two more northern populations. Among metabolites contributing to the separation of the populations were pinitol, an unknown sugar alcohol (M217T1087), and malic acid. Catechin and galactinol specifically separated the populations in the early time-points, whereas raffinose separated the populations from T2 on, and an unidentified sugar alcohol (M73T1259) separated the populations by accumulating more strongly in the southern populations from the second time-point on. Interestingly, pyroglutamic acid levels contributed to the separation of the populations by positive correlation with latitude during the early time-points, and by negative correlation with the latitude at T4. From T3 onward, lower levels of shikimic acid contributed to a partial separation of the BC population at T3, which was associated with lower levels of several fatty acids (stearic, palmitic, and oleic). The loadings of the selected metabolites on the latent variables are given in Table S4.

Hierarchical clustering of metabolite profiles and of transcript profiles

The degree of similarity among the metabolite profiles and among the transcript profiles (using data obtained from Holliday et al., 2008), obtained from different populations and at different time-points during cold acclimation was assessed by hierarchical clustering (Fig. 2). Clustering of the transcript profiles resulted in two main clades, separating the early transcript profiles of summer and early autumn (time-points 1 and 2) from the late transcript profiles (from T3 onward) for all populations. Hierarchical clustering of the metabolite profiles resulted in a similar separation between early and late profiles for the two most southern populations, but all AK metabolite profiles clustered together with the late metabolite profiles of the other populations. The overall metabolite profiles in the AK population in late summer were therefore already more similar to the cold adapted metabolite profiles of the late autumn and winter time-points than to the late summer profiles in the two more southerly populations. Among the transcript and the metabolite profiles of the late time-points, the BC and AK profiles clustered together and were distinct from the CA population profiles.

Figure 2.

Hierarchical clustering of the transcript and metabolite profiles corresponding to the different time-point–population combinations. The transcript profiles comprise the 6070 clones chosen for clustering in Holliday et al. (2008) (fold change > 1.3 mean at any time-point, < 0.025). The metabolite profiles comprise the total set of 189 GC-MS detected metabolites. CA, California; BC, British Columbia; AK, Alaska; T1, 30 August; T2, 18 October; T3, 22 November; T4, 1 December; T5, 13 December.

Integration of metabolite and transcript data on the metabolic map

In order to establish a detailed and integrated overview of the primary and secondary metabolism remodelling occurring throughout this investigation, including the transcriptional basis of the observed changes in metabolite pools, we selected a set of 126 differentially accumulating transcripts (Holliday et al., 2008) that encoded enzymes involved in primary metabolism and in metabolic pathways that showed differentially accumulating metabolites (Table S5). We plotted metabolite intensities over time in the three populations (four time-points in CA and AK and five time-points in BC), and the intensities of the selected transcripts from Holliday et al. (2008) over the time-course in the BC population (five time-points), on a metabolic map (Fig. 3). Specifically, in primary metabolism, transcript levels of the key enzyme of the pentose phosphate pathway (PPP), glucose-6-phosphate dehydrogenase, as well as the levels of other PPP transcripts (6-phosphogluconolactone and transaldolase) were up-regulated during cold acclimation. By contrast, the transcript abundance of phosphofructokinase, which catalyzes a rate-limiting step in glycolysis, decreased during cold acclimation, a trend that was also observed for other enzymes participating in glycolysis (triose phosphate isomerase, phosphoglycerate kinase, phosphoglycerate mutase). Immediately upstream of the citric acid cycle (CAC), however, transcript abundance of pyruvate kinase and pyruvate dehydrogenase increased during cold acclimation, as did the transcript levels of five citric acid cycle enzymes and all of the identified CAC intermediates (citric acid, succinic acid, and malic acid). Two transcripts encoding phosphoenolpyruvate (PEP) carboxykinase were strongly down-regulated, suggesting a decrease in gluconeogenesis, and four transcripts encoding NADP-malic enzyme (NADP-ME) were up-regulated. The metabolic map suggests that the up-regulation of flavonoid biosynthesis led to increased abundance of catechin and to the biosynthesis of anthocyanins, with no significant effect on the intensities of flavonols such as quercetin, kaempferol, and myricetin, and as such is likely due to the observed concomitant decrease in flavonol synthase transcripts.

Figure 3.

Metabolic pathways. Metabolites that were identified in the GC-MS spectra are shaded in blue. These metabolites are accompanied by graphs with the x-axis representing the time-points 1 (30 August) to 5 (December 13), the y-axis representing the average peak-intensity in the GC-MS chromatograms, and the color indicating the population: California (CA), red; British Columbia (BC), green; Alaska (AK), blue. Enzymes for which the transcript levels during autumn cold acclimation have previously been monitored are shaded in green. These enzymes are accompanied by graphs in which the x-axis represents the time-points and the y-axis represents the expression level fold-change as compared to time-point one in the BC population, as reported in Holliday et al. (2008), and as mentioned in Table S5.

Transcripts and intermediates involved in photosynthesis and photorespiration were down-regulated during cold acclimation. By contrast, we observed (in the BC population) correlated increases in the transcript abundance of both phosphoglycerate dehydrogenase and phosphoserine aminotransferase, two enzymes catalyzing the biosynthesis of serine and glycine from glyceric acid, during cold acclimation, indicating a compensation reaction in response to the decreasing photorespiration levels to maintain glycine and serine pools.

The transcript abundances of some genes encoding enzymes involved in starch biosynthesis decreased temporarily during cold acclimation. For example, ADP-glucose pyrophsophorylase small subunit and starch branching enzyme decreased from the first to the fourth time-point, and starch synthase 2 decreased from the first to the third time-point. In turn, the abundance of starch biosynthetic enzymes increased in later autumn, from the third time-point onward (starch synthase 2), or from the second time-point on (ADP-glucose pyrophosphorylase large subunit and starch synthase 3). Increasing starch degradation from late summer onward is implied by the increasing abundance of three transcripts encoding alpha-amylase as well as a transcript encoding beta-amylase during the first half of the cold acclimation period. Interestingly, the levels of two other transcripts encoding beta-amylase decreased across the entire period of cold acclimation.

Correlation of metabolite and transcript levels with phenotypic variation in cold hardiness

Raffinose dynamics exhibited a striking correlation with freezing injury data presented in Holliday et al. (2008), where its accumulation closely tracked an increase in cold hardiness. The most important discrepancy between the raffinose plot and the cold hardiness plot is for the AK population at T2, where cold hardiness already exceeded the final level of cold-hardiness in the BC population, but the raffinose level was still significantly lower than the final levels in the BC population. These raffinose dynamics led us to investigate correlations between the metabolite and transcript levels, and cold hardiness. Hierarchical clustering that included both the freezing injury data and the metabolite data showed that the metabolite for which the accumulation pattern in the three populations was most similar to patterns of cold hardiness was malic acid. Other known metabolites that clustered with cold hardiness were catechin, stachyose, and raffinose. In a hierarchical clustering of the 22K transcripts represented on the microarray and the freezing injury data, flavanone-3-hydroxylase, an important enzyme in the biosynthesis of catechin, clustered closely with the cold hardiness data. Other annotated transcripts that clustered with the cold hardiness data encoded enzymes involved in taxol biosynthesis and lipid biosynthesis (taxane 13-alpha-hydroxylase, glycerophophoryl diesterase, phosphatidyl glycerophosphate synthase).


Metabolism shifts during cold acclimation

Following the first time-point, the expression of genes encoding components of the photosynthetic apparatus and metabolite pool sizes of two intermediates of the photorespiratory cycle, glycine and serine, continuously decreased. It has been previously noted that in photosynthetic tissue, pool sizes of glycine and serine vary with conditions affecting the rate of photosynthesis (Rawsthorne & Hylton, 1991). These observations suggested a decrease in photosynthetic activity, contributing to a lower energy supply. Additionally, during the cold acclimation period, day length decreases and starch remobilization (during the longer nights) results in the depletion of the starch granules. A transient shortage of available sugars was suggested by the decrease in the abundance of soluble glucose and fructose during early autumn. Such ephemeral shortage in glucose and fructose has been reported previously in the bark and needles of red pine (Pinus resinosa) during late summer (September), in Ontario, Canada (Pomeroy et al., 1970), as well as in the apices of poplar (Populus tremula x Populus alba) grown in a controlled growth chamber, shortly after the onset of a short-day treatment (Ruttink et al., 2007). Thus, it appears that such ephemeral responses represent a component of the transition to short days during winter priming of trees. Following the initial decrease, glucose and fructose concentrations gradually recovered. Concomitantly, transcript levels suggest that starch biosynthesis was up-regulated after a transient decrease in early autumn. An increase in starch degradation was suggested by the induction of several alpha-amylase genes and a specific beta-amylase over the autumn cold acclimation period. Conversion of starch to the soluble hexose sugar pool is a key metabolic process associated with autumn cold acclimation, as starch-derived sugars serve as a source of energy, as well as cryo-protectants (Schrader et al., 2004). In Arabidopsis, one specific beta-amylase has been shown to be crucial for the cold-temperature-dependent increase in soluble sugars and the associated protection of the photosynthetic electron transport chain during freezing stress (Kaplan & Guy, 2005). A similar role in the development of freezing tolerance could be suggested for the single up-regulated beta-amylase gene observed in this study.

The biosynthesis and accumulation of oligosaccharides such as raffinose and stachyose, known to function as osmoprotectants, was increased during the autumn cold acclimation. The sugar alcohol galactinol, the immediate precursor for raffinose synthesis, and other sugar alcohols, including pinitol, accumulated, suggesting that sugar alcohols might by themselves play a role in the protection of the cell from freezing damage. Such a role has previously been proposed for galactinol in a knockout mutant that fails to accumulate raffinose (Zuther et al., 2004).

Cold acclimation also shifted metabolism in the leaves toward the production of reductive agents and scavengers of radicals. A major adjustment of the central primary metabolism during cold acclimation was reflected in the combined increase in glucose-6-phosphate dehydrogenase and a decrease in phosphofructokinase transcript abundance in the BC population. This observation implies a shift toward a greater contribution of the oxidative PPP in glucose catabolism, at the expense of glycolysis. A similar shift has been reported in rape seed plants (Brassica napus) exposed to low temperatures (Maciejewska & Bogatek, 2002). In contrast to glycolysis, the oxidative PPP generates NADPH, and therefore may offer an important mechanism to supply reduced co-factors during cold acclimation. The transcript abundance of NADP-ME was also elevated, which may also contribute to the overall reductive power by providing NADPH. The increased demand in reduced co-factors could at least partially be attributed to an increase in flavonoid biosynthesis, as several steps in the flavonoid biosynthetic pathway require reductive power. The increased flux through the PPP may in turn be channelled towards the biosynthesis of catechin and anthocyanins via the general up-regulation of the genes involved in the shikimate pathway, phenylpropanoid biosynthesis, and flavonoid biosynthesis. It is well known that low temperatures induce the production of hydrogen peroxide that can damage membrane lipids via lipid peroxidation (Foyer et al., 1997; Kocsy et al., 2001). Moreover, in conditions where the capacity of the photosynthetic carbon metabolism is reduced, over-excitation of the photosynthetic apparatus leads to the formation of reactive oxygen species, which may result in damage to the photosystem. Accumulating catechin and anthocyanins during autumn are therefore likely to play a role in photoprotection and in the prevention of oxidative damage.

Taken together, our results have shown that with the cessation of growth and onset of dormancy, carbohydrate metabolism was modified toward the accumulation of storage compounds, cryo-protective or dehydration-protective solutes, and reductive co-factors, resulting in increasing cold tolerance.

Inter-population variation in metabolism, associated with phenotypic cold hardiness variation

Latitudinal clines in autumn cold hardiness and timing of bud set are well established for Sitka spruce, with more northern populations setting bud earlier, initiating cold acclimation earlier, and reaching higher levels of cold hardiness than southern populations in common garden experiments (Aitken et al. 2008; Savolainen et al., 2007). For Sitka spruce, these clines have been characterized for 17 populations, and have been shown to correspond closely with provenance latitude and climate (Mimura & Aitken, 2007). Although these clines are well characterized at the phenotypic level, little is known about how such changes relate to the underlying metabolic processes. We included three highly divergent populations in the current study that represent most of the latitudinal range of the species and we clearly observed latitudinal variation in the metabolism of these populations, at several levels.

In late summer, the overall metabolite profile of the AK population was similar to the autumn-acclimated winter profiles of the BC and CA samples, which explains the stronger pre-autumn-acclimation freezing tolerance observed in the AK population. This could be the result of earlier cold acclimation, developed before the start of the survey, or could reflect a constitutively increased freezing tolerance in the AK population with respect to the populations of more southern origin. In Arabidopsis, such constitutive freezing tolerance in a northern ecotype has been shown to be metabolically similar to the cold acclimated plants of other, more southern ecotypes (Hannah et al., 2006). Early freezing tolerance, even during the growing season, may be needed for the individuals of the AK population in their natural environment. The risk of encountering frost by the end of September is high in the northern environment from where the AK population originates: estimates from climate normals for 1971–2000 of the first date in the autumn with a 50% chance of frost, were 30 September in the northern environment (AK), 3 November in the central environment (BC), and 11 December in the southern environment (CA) (Mimura & Aitken, 2010).

During autumn acclimation, both the BC and the CA populations underwent major changes in both metabolite and transcript profiles between 18 October and 22 November, a period characterized by low, above-freezing temperatures (< 5°C) (Holliday et al., 2008). The trigger for the metabolome changes during this period in the BC and CA populations may therefore be a threshold non-freezing low temperature or a threshold day-length. In addition to the strong shift between 18 October and 22 November, the BC population exhibited a gradual change in the metabolome that was initiated earlier and spans the complete survey time, suggesting a response to short days earlier in the fall that was accelerated by low temperatures. By contrast, the CA metabolome remained fairly constant both before and after the period between 18 October and 22 November, and we thus hypothesize that this population may only be responding to low temperatures during this period. Though we cannot exclude the possibility that day-length plays a role in these changes, as the shortest day of the year for the native CA environment (9 h 15 min) occurs during this fall period at the Vancouver common garden site, it is unlikely that a day-length cue for cold acclimation would come so close to the shortest day of the year.

The abrupt change in the CA metabolome was characterized by increased levels of a large number of metabolites, but was not associated with a significant increase in freezing tolerance. In the BC population, metabolic remodelling was associated with a marked increase in cold hardiness between 18 October and 22 November and thereafter. That both the transcript and the metabolite profiles of the autumn-acclimated BC and AK profiles were similar and distinct from the CA profiles was consistent with the observation that the CA population did not achieve substantial freezing tolerance. Indeed, in their mild native environment the CA population rarely encounters temperatures much below freezing. The need for metabolic adaptations, adaptations in cell structure, and production of metabolites that specifically allow a plant to survive a freezing event were thus specific to the more northern BC and AK populations. The clear adaptation of the metabolism of the individuals of the CA population may reflect their need to adapt their metabolic function to low non-freezing temperatures in addition to seasonal changes in parameters such as light quality and duration.

With respect to individual metabolites, the accumulation patterns malic acid, catechin, raffinose, and stachyose, over the different time-points in the three populations, was highly correlated with levels of freezing tolerance, suggesting an important and direct functional involvement of these metabolites in the timing and extent of cold hardening. The strongest effects of latitude of origin and time-point during autumn acclimation on metabolite levels were observed for raffinose. Raffinose levels by themselves could, however, not explain the early freezing tolerance acquired by the plants of the AK population. This early freezing tolerance in the AK population was stronger by October than the freezing tolerance reached by the BC population at the end of the survey. The October metabolite profiles of the AK population were characterized by an ephemeral decrease in soluble sugars (glucose, fructose), which was more obvious than in the other populations, combined with an ephemeral increase in amino acid levels (glutamine, proline, isoleucine, alanine, serine), which was restricted to the AK population. These amino acids may, like raffinose, function as compatible solutes, which, at high concentrations, have no adverse effect and decrease the freezing point of the cytoplasm. The unique characteristics of the October metabolite profile in the AK population suggests that the early development of strong freezing tolerance in this population was based on distinct metabolic mechanisms as compared to the development of freezing tolerance obtained after longer autumn acclimation in the other populations.

Increases of certain metabolites in the more southern population created latitudinal clines that were not apparent during late summer, as was observed for tartaric acid. Tartaric acid is a product of ascorbic acid catabolism (De Bolt et al., 2006). Increasing tartaric acid levels during autumn in the CA population may therefore reflect the development of a foliar antioxidant system as a protection against photo-oxidative stress due to decreasing photosynthetic activity in the face of more direct sunlight during winter at the lower latitude from which this population originates.

For other metabolites, stronger accumulation in the more northern populations was apparent in late summer but this latitudinal cline was diminished later in the fall (e.g. galactinol, citric acid), or even inversed, (e.g. alanine, glutamic acid, GABA, succinic acid, and pyroglutamic acid), due to increasing levels in the more southern populations. Levels of GABA, glutamate and succinate are metabolically tightly related since GABA is synthesized from glutamic acid by decarboxylation and may be further metabolized to succinic acid. The inversion of certain latitudinal clines in metabolite levels was in some cases due to a peaking of these metabolites at different times in the various populations. For example, GABA, glutamate and succinate peaked in early summer in the AK population, while in the CA population it peaked in late autumn. GABA has previously been implicated in plant stress responses, and has also been shown to increase in cambial cells in the same time-frame as the induction of cold hardiness-related genes (Bouche et al., 2003; Druart et al., 2007). A role for GABA as a compatible solute has been hypothesized (Shelp et al., 1999), but GABA may function as a signalling molecule as well (Kang & Turano, 2003). The sequential peaks in GABA during autumnal cold acclimation in the different populations imply that in this case it likely functions as a signal in the induction of cold hardiness metabolism.


We have shown that cold-hardiness associated metabolites exhibit genetically determined variation among geographic populations distributed along a strong climatic gradient. The combination of metabolite and transcript profiling provided insight into the molecular architecture of adaptive variation in cold hardiness. A thorough integration of these results examining the natural genetic variation in the relevant genes and their upstream regulatory elements should enhance our understanding of the genomic underpinnings of phenotypic variation in climate-related traits such as cold hardiness. This goal is particularly important as anthropogenic climate change is substantially altering adaptive landscapes. A better understanding of the molecular genetic determinants of adaptive trait variation will facilitate predictions of the distribution and species composition of future forests and their ability to adapt to new climatic conditions, and enhance genetic conservation strategies aimed at preserving the natural adaptive capacity of our forests.


The authors acknowledge the technical assistance of Russell Chedgy for his efforts in sample preparation. The authors acknowledge financial support from Genome Canada grant held by S.D.M. and S.N.A.