The availability of Scots pine seeds (Pinus sylvestris L.) with high germinability is necessary for artificial forest regeneration. In this work, Scots pine seed orchard seeds were magnetic resonance (MR) imaged to noninvasively investigate the association of the anatomical images and quantitative relaxation times with the structure and germinability of the seeds. Relaxation time differences compared to the germination day were also investigated. The average whole seed relaxation times T1 (two methods), T2, and were 430 ± 59, 660 ± 20, 14 ± 1.7, and 0.83 ± 0.33 ms, respectively. It was observed that the seed structures had statistically significant (p < 0.05) differences in relaxation times, while no differences could be observed in relation to the rate of seed germination. Furthermore, the obtained data were compared to radiographs. Empty seeds were observed to provide a minimal MRI signal, whereas intact and mechanically damaged seeds provided a profound signal with distinguishable structures. The mechanically hardest region, i.e., the seed coat, was not visible in MRI as opposed to radiographs. Some seeds determined to be mechanically damaged by radiography were able to germinate, and mechanical faults could be distinguished in MRI. As such, MRI can be seen complementary to the currently used methods to optimize seed sorting and to interpret germination potential.
La disponibilité des semences de pin sylvestre (Pinus sylvestris L.) à haut pouvoir germinatif est nécessaire pour la régénération artificielle des forêts. Dans le présent ouvrage, les semences de verger de la semence de pin sylvestre ont été captées en image par RM (résonance magnétique) afin d’examiner de manière non invasive l’association des images anatomiques et des temps de relaxation quantitatifs avec la structure et le pouvoir germinatif des semences. La différence des temps de relaxation comparativement aux jours de germination a également été examinée. Les temps de relaxation complets des semences T1 (deux méthodes), T2 et étaient 430 ± 59, 660 ± 20, 14 ±1,7 et 0,83 ± 0,33 ms respectivement. On a observé que les structures des semences avaient des différences statistiques importantes (p < 0,05) dans les temps de relaxation alors qu’aucune différence ne pouvait être observée relativement au taux de germination des semences. De plus, les données obtenues ont été comparées aux radiogrammes. On a observé que les semences vaines fournissaient un signal d’IRM minimal alors que les semences intactes et mécaniquement endommagées fournissaient un signal profond avec des structures reconnaissables. La région la plus dure mécaniquement, le tégument, n’était pas visible en IRM contrairement aux radiogrammes. Certaines semences déterminées comme étant mécaniquement endommagées par radiographie étaient en mesure de germer et les défauts mécaniques pouvaient être distingués en IRM. Ainsi, l’IRM peut être considérée comme étant complémentaire aux méthodes actuellement utilisées pour optimiser le triage des semences et pour interpréter le potentiel de germination. [Traduit par la Rédaction]
The availability of Scots pine seeds (Pinus sylvestris L.) with high germination capacity and energy is crucial for successful artificial forest regeneration. Approximately 9000 kg of Scots pine seeds are used annually in Finland. In container seedling production, a germination capacity of 95% is typically regarded as the limit for single seed sowing (Edwards and El-Kassaby 1996). Germinability of conifer seeds is affected by multiple factors, including the maturity of the seeds at the time of harvest, mechanical damage incurred during seed extraction, seed-wing removal, and sorting (Tillman-Sutela and Kauppi 2014; Nygren et al. 2016), as well as developmental malfunctions (Kolotelo 1997). Empty and mechanically damaged Scots pine seeds can be separated from full and intact seeds with pneumatic sorting and with the PREVAC method (Bergsten and Wiklund 1987). Although these methods are generally effective and suitable in the scale of commercial Scots pine seed production, in some cases they are unable to produce seed lots that meet the requirements for marketable seed lots. Occasionally, for example, the discarded fraction separated in the PREVAC process contains seeds that appear intact using the commonly available seed-quality analysis methods.
For Scots pine, the so-called empty seeds are in most cases caused by the homozygosity of recessive lethal factors (Koski 1971). In empty seeds, the seed coat develops normally, but the seed contains a varying degree of dead remnants of the embryo, megagametophyte, and membrane structures. The self-fertilization rate and the proportion of empty seeds are positively correlated (Kärkkäinen and Savolainen 1993). Furthermore, the viability of Scots pine seeds is moisture dependent, as seeds at 30% water content can withstand only –10 °C, while at a water content of 10% they can survive –40 °C (von Schönborn (1964) according to Nygren (2003, 2016)). The water content of mature Scots pine seeds shows annual and clonal variation. Seed moisture content varying between 13% and 40% (fresh weight basis) has been reported at the end of the maturation phase (Nygren and Pulkkinen 1994; Nygren et al. 2016), with total lipid contents of 20%–30% (Tillman-Sutela et al. 1995; Tammela et al. 2005). To ensure optimal storability in commercial seed production, the seeds are artificially dried to 5%–7% moisture content. In mature conifer seeds, most tissues are alive and remain as such until germination (Vuosku et al. 2009). The main structures in Scots pine seeds include the embryo, megagametophyte storage tissue, and the seed coat (Fig. 1a). The megagametophyte is surrounded by membranes that regulate the water intake into the inner structures. Currently, two-dimensional (2D) radiography is used in seed research and production to assess seed maturation and the quantity of empty and abnormal Scots pine seeds. The seed maturation is interpreted (Fig. 1b) through anatomical potential (the proportion of embryo and megagametophyte to other structures) (Simak 1980). Other essential methods to test the seed quality include seed water content analysis and laboratory germination testing (Nygren 2003). However, as embryo development has been observed to be more temperature-dependent than the outer structures (Tillman-Sutela and Kauppi 2014), the evaluation of germination potential through anatomical potential and size differences, and by extension, the use of radiography, may be problematic.
A further understanding of seed structures, composition, and defects is required, as these could be associated with germination potential, and this information can be used to improve seed sorting methods. Magnetic resonance imaging (MRI) of hydrogen (1H) nuclei, mostly in water and lipids, is based on the phenomenon of nuclear magnetic resonance (NMR) and is widely used for noninvasive and nondestructive characterization of tissue and material. In addition to clinical imaging, MRI has also found applications outside these settings, including, for example, paper materials, polymers, and various seeds, as well as plant shoots and roots (Harding et al. 2001; Koizumi et al. 2008; Leisen et al. 2012; van Dusschoten et al. 2016). Since the measurable signal of water molecules is greatly affected by their microscopic molecular environment, MRI provides an exceptional soft-tissue contrast (unlike, for example, X-rays, which provide almost no soft-tissue contrast). The measurable signal is governed by the relaxation properties of the sample, which are dependent on the molecular environment of the nuclei, and also by the radiofrequency (RF) excitation scheme applied, termed a pulse sequence. T2 and relaxation times describe the rate at which the measurable signal diminishes, while T1 relaxation time describes the rate at which the system returns to its original equilibrium state prior to the RF excitation. For samples with an extremely rapidly diminishing signal, which are difficult to MRI (typically tissues with very low water content, or solid-like tissues), methods have also been developed, including zero echo time (ZTE), ultrashort echo time (UTE), and sweep imaging with Fourier transform (SWIFT) pulse sequences (Garwood 2013). These methods can detect extremely short-lived NMR signals, i.e., samples and regions with very short T2 and relaxation times. Furthermore, quantitative mapping of T1, T2, and relaxation times (and other quantitative MRI (qMRI) parameters) can be utilized for a more meaningful comparison of the results and values between studies, as these values depend more on the sample characteristics than on the device properties.
Previously, MRI has been used to study imbibition of western white pine seeds (Pinus monticola Dougl. ex D. Don) (Terskikh et al. 2005b), resin vesicles on seed coats of conifer seeds (Keeling et al. 2018), and development of white spruce seeds (Picea glauca (Moench) Voss) (Carrier et al. 1999). Beyond MRI, spectroscopic NMR methods and nuclei other than 1H have also been occasionally utilized for the compositional assessment of conifer seeds (Terskikh et al. 2005a, 2008). However, MRI and particularly qMRI is not widely adapted for the assessment of anatomy, viability, and germination potential of conifer seeds. Therefore, the purpose of this study was to determine the germination potential and overall MR visibility of Scots pine seeds and their different anatomical structures using quantitative microscopic MRI and compare the findings with the results of standard projection X-ray. As the current methods are generally invasive or provide only 2D information, MRI could be used to assist in the assessment of germination potential and overall viability of Scots pine seeds by observing differences in water distribution of the seed and seed structures. With different imaging parameters and various pulse sequences, MRI provides a plethora of different contrasts, image weightings, and quantitative data, not easily achievable with other imaging modalities, such as density- and photon-absorption-dependent computed tomography (CT) and X-ray radiography. Furthermore, the noninvasive nature of MRI provides a safe alternative for the structural interpretation of seeds (Fig. 1c) as well as assessment of their chemical composition without affecting their viability by radiation dose or other destructive procedures.
Materials and methods
Preparation of samples
The studied Scots pine seeds were collected from a 1.5-generation clonal seed orchard sv404 (Suhola; 62°14′50.2″N, 27°41′50.8″E), located in central Finland. The study material was chosen to present typical commercial seed material used in artificial forest regeneration in Fennoscandia. Cones were harvested from three ramets from three clones (K393, K682, and K942). The seeds were extracted from the cones, de-winged, X-ray radiographed, and germinated by the Natural Resources Institute Finland (Suonenjoki, Finland).
The harvest procedure and preparations prior to MRI were as follows. From each of the nine ramets, 12 cones were collected on 30 September 2019. The cones were kept in resealable plastic bags at +4 °C until the next day. The clonal cone water content (fresh weight basis) was determined from three cones from each ramet from 1/8 of a cone by drying the samples at 103 °C for 17 h. The average cone water contents were 45%, 43%, and 42% for K393, K682, and K942, respectively. The seed water content from one seed of each cone, used for water content measurement, was determined with the constant temperature oven method by measuring the weight difference before and after drying at 103 °C for 17 h (ISTA Moisture Committee 2007). The average seed water contents were 35%, 31%, and 31% for K393, K682, and K942, respectively.
The seeds from the remaining nine cones from each ramet were extracted in an oven at 45 °C for 5 days. The cones were watered twice during this period to enhance the opening of cone scales. After watering, a lower temperature (30 °C) was used for 2 h to avoid thermal damage. After this, the seeds were extracted manually and de-winged with water, after which they were stored in paper envelopes at ambient conditions until 17 October 2019. From each ramet, 70–135 seeds were obtained and altogether, 25 subsamples of five seeds from each ramet were counted using a seed counter (Elmor Unit Counter Model 600; Schwyz, Switzerland). From these, 13 sets of the subsamples were randomly chosen for further analysis. In total 10 sets of the subsamples from each ramet were recombined to be used in a germination test initiated on 18 October 2019. The seeds were germinated at +20 °C for 21 days on Petri dishes moistened with 5 mL of tap water in germination cabinets with 16 h daylength (model GC 10/11; FLOHR Instruments, Nieuwegein, the Netherlands). The germination percentage was counted in relation to full seeds observed with the projection radiography. In the case of ramet K682b, only 70 seeds were obtained. Thus, the initial germination analysis was not performed.
Two of the subsamples, forming a seed batch of 10 seeds, were subjected to further analyses: the samples were first radiographed with the highest magnification (Faxitron MX-20; Faxitron Bioptics LLC, Tucson, AZ, USA) with 18 kV and 4 s exposure time so that the individual seeds could be identified in the further analyses, i.e., MRI and evaluating the germinability of the seeds. Based on the projection radiographs, the seeds were classified as (i) empty, (ii) healthy, or (iii) mechanically damaged. Seeds were kept individualized throughout the experiment. Each individualized seed was then placed into an aluminum-plastic-layer bag and hot sealed with an impulse sealer.
Microscopic MRI was conducted at the facilities of the Kuopio Biomedical Imaging Unit of the A.I. Virtanen Institute for Molecular Sciences at the University of Eastern Finland (Kuopio, Finland) between 27 February and 29 April 2020. The seed batches were MR imaged with an 11.7 T vertical magnet (500 MHz Bruker Ultrashield) with microimaging capabilities (Bruker BioSpin MRI GmbH, Ettlingen, Germany) and controlled with the accompanied software (ParaVision 6.0.1. and TopSpin 3.1.). MRI was conducted at room temperature using commercial saddle-shaped volume 1H RF coils (5 mm and 10 mm in cross-sectional diameter) housed inside a Micro5 probe (Bruker) with a maximum gradient strength of 3 T/m (at 60 A) in all three spatial directions. Imaging parameters are presented in Table 1 with a more detailed description in supplementary Table S11.
Prior to actual imaging, the RF power was manually calibrated for each pine seed batch using NMR spectra with varying reference power. The automatic calibration scheme was bypassed because it was found to be unreliable with the low water and low lipid contents, and thus, low obtainable signal from the sample. The center reference frequency (resonance frequency) was automatically selected as the central frequency of the largest observed peak in the NMR spectrum (by reasonable assumption the lipid peak); however, it should be noted that an adjacent smaller peak was consistently observed downfield (positive frequency direction, most likely the water peak) as shown in an example of a calibration spectrum (Fig. 2).
A 10 mm 1H RF coil was used for each seed batch for all seed batches K393, K682, K943 with ramets a–c each containing 10 seeds. Samples were placed in a 3D-printed compartmentalized sample holder (using Tough PLA (polylactic acid) filament; Ultimaker, Utrecht, the Netherlands) for a total of 10 seeds (supplementary Fig. S1a1). The PLA sample holder was housed in a 10-mm-diameter borosilicate NMR tube (Kimble Kontes; DWK Life Sciences LLC, Millville, NJ, USA). As a precaution for maintaining seed identification, each seed was marked with a small “MRI-invisible” individualized pattern using gel paint pen (supplementary Fig. S1b1). The seeds were kept identified prior, during and after microimaging.
The seeds were MR imaged three-dimensionally with zero echo time (ZTE) and multi-echo gradient echo (MGE) pulse sequences as well as two-dimensionally with multi-slice multi-echo (MSME) and rapid acquisition with relaxation enhancement, variable repetition times (RAREVTR) sequences, collecting multiple axial imaging slices throughout the seed structures (from the micropyle to the chalaza of the seed; Fig. 1a). Commonly, RARE sequence is also known as a fast spin echo (FSE) pulse sequence. For the multi-slice spin echo sequences, a gap of 0.1–0.3 mm between consecutive slices was employed. The pulse sequences were used to obtain both quantitative and anatomical images. RAREVTR data were acquired with eight repetition times (TR) ranging from approximately 300 to 3000 ms and were used to obtain T1 maps for the seed regions. T2 and maps were calculated from the MSME and MGE data, respectively. For MSME, 10 echo times (TE) of between approximately 3 and 30 ms were used, whereas for MGE, the approximate range was from 0.8 to 8 ms with eight TEs.
ZTE was utilized for anatomical images with the largest obtainable flip angle (approximately 8°) as well as for quantitative T1 mapping by applying variable flip angles (VFA). ZTE-VFA was used to obtain estimates for T1 in regions with low signal intensity, i.e., from components and regions not necessarily visible with RAREVTR, i.e., from different spin pools and corresponding T2 times in the sample. The estimation of T1 with ZTE-VFA was limited due to the low achievable flip angles (because of the requirement of short RF pulse duration) and RF pulse inhomogeneity. ZTE-VFA was obtained using sequential but separate ZTE scans with varying flip angles, but otherwise fixed imaging parameters and gains. The individual seeds could be identified from the 3D images, as the asymmetric sample holder was visible with the ZTE pulse sequence (please see, e.g., Fig. 3d). The holder was not observable with any of the other imaging sequences with longer echo times, and as such, artifacts arising from the signal of the sample holder were avoided in all sequences except the ZTE.
A single seed (K393b-4) was imaged with the 5 mm 1H RF coil to provide a high resolution and high signal-to-noise ratio (SNR) example of microscopic MRI. The seed was placed into a 5 mm N51A borosilicate NMR tube (SP Wilmad-LabGlass, Vineland, NJ, USA), secured in place with a solid PTFE cylinder on top of the seed, and MR-imaged.
Radiography and germinability of imaged seeds
After MRI, the seeds were repacked into the plastic-coated aluminum bags, impact sealed, and kept in room temperature for approximately 7 months (April 2020 to November 2020) prior to germination tests conducted by the Natural Resources Institute Finland (Suonenjoki, Finland). Before the germination tests, the seeds were X-ray radiographed (Faxitron MultiFocus, Faxitron Bioptics LLC, Tucson, AZ, USA) using 18 kV tube voltage and 4 s exposure time. The tube current of the device was 1 mA and the pixel pitch 48 µm. The germinability of the individual seeds was evaluated as with the initial germination test. A seed was classified as germinated when the length of the radicle exceeded four times the length of the seed coat (15 mm on average). The sprouts were investigated on days 7, 10, 14, and 21 of the germination test.
The data analysis, comprising of image reconstruction and quantitative relaxation time mapping, was accomplished with MATLAB (version 2017b; The MathWorks, Inc., Natick, MA, USA). The data from MGE, RAREVTR, and MSME were reconstructed from the raw signal files using a MATLAB GUI (Aedes, University of Eastern Finland, Kuopio, Finland, http://aedes.uef.fi/), which was also used for further analysis including manual segmentation of appropriate regions of interest (ROIs). ZTE-VFA data were reconstructed with algebraic reconstruction (AR) (Froidevaux et al. 2018) and nonuniform Fourier transform (NUFFT) using Michigan Image Reconstruction Toolbox for MATLAB (Fessler 2021) to ensure identical preprocessing for all image sets in the VFA series. For anatomical ZTE images, the vendor-provided on-line image reconstruction was used. From the reconstructed 3D MGE and ZTE images, new image slices were recalculated from healthy, empty, and damaged seeds using Horos Dicom viewer (The Horos Project, https://horosproject.org/).
Three separate ROIs were defined on each individual seed and image slice, consisting of the embryo, megagametophyte, and cell membrane (see example ROIs in Fig. 3i). These areas were chosen because of their distinct biological roles and clearly observable intensity differences in the MR images. The cell membrane (Fig. 1a) surrounds or is a part of the outer region of the megagametophyte and is observable as a high signal intensity region. T1, T2, or curves were fitted for each voxel in the ROI. The arithmetic mean of each voxel-wise fit within an ROI was used to represent the quantitative parameter. Equations S1–S4 (in the supplementary material1) were used to estimate the relaxation time by fitting the signal model to the measured signal intensities. For T1 (RAREVTR), T2, and estimation, two-parameter fitting (proton density and relaxation time) was utilized. For T2 and , the background noise was subtracted prior to fitting. Using the ZTE-VFA data, T1 in eq. S41 was solved nonlinearly using Gauss–Newton method with no additional corrections for RF pulse inhomogeneity. With MSME, the data from the first TE was omitted on all T2 fits. The possibly erroneous measured data on the first TE (shortest TE) can originate from so-called stimulated echoes not present with the first TE (Matzat et al. 2015) potentially lowering the signal intensity obtained with the shortest TE with respect to the rest of the echo train.
Statistical analysis was conducted with MATLAB, using built-in functions. Normal distributions of the investigated data groups (relaxation times of the different structures) were verified using the one-sample Kolmogorov–Smirnov test (kstest in MATLAB). Megagametophyte and cell membrane structures in the MGE data exhibited non-normal distributions likely caused by poor SNR in these regions (supplementary Figs. S2–S41). For the seed regions and relaxation data, statistical analysis was conducted with one-way analysis of variance (ANOVA, using anova1 in MATLAB) and was performed between the germination groups (days of germination) and relaxation times as well as between structures (embryo, megagametophyte, and cell membrane) and their corresponding relaxation times. All empty seeds were excluded from the investigations, as the majority of their relaxation times could not be defined due to the low signal intensity. For statistically significant results obtained with one-way ANOVA, multiple comparison of means test was performed as a post hoc analysis to evaluate the differences between individual groups (using multcompare in MATLAB).
A single high-resolution MR image of a pine seed exhibits both similarities with the X-ray image, but distinct differences as well. The X-ray projection image (Fig. 1b) contains all the structures overlaid on top of each other along the path between the X-ray source and the detector while MRI allows creation of both an image from a selected slice within the sample (Fig. 1c) or a 3D volume rendering (Fig. 1d). The seed coat, observable in the X-ray image, is not visible with MRI. On the other hand, in the chalazal end of the seed, signal loss caused by mechanical damage is visible in the MR image, whereas such obvious differences are not clearly distinguishable in the radiograph. The majority of the NMR signal from the pine seeds may originate from lipids, as seen from the NMR spectrum; the lipid peak is larger than the water peak (Fig. 2).
Differences in the signal intensities of different seed structures, i.e., in the image contrast are clearly visible with the different pulse sequences (Figs. 3a–3d). Using the 10 mm 1H RF coil, the embryo, megagametophyte, and cell membrane can be distinguished using SE sequences (RAREVTR, MSME), governed by the T2 relaxation, whereas low -weighted signal (MGE) was obtained in some regions, mainly the megagametophyte. ZTE pulse sequence allowed the detection of not only the embryo, megagametophyte, and cell membrane but also the 3D-printed sample holder (Fig. 3d). From these data, quantitative relaxation time parameters T1, T2, , and T1, measured with RAREVTR, MSME, MGE, and ZTE-VFA sequences, respectively, were calculated for each seed within the defined ROIs (Fig. 3i). The T1 values appeared to be generally homogeneous throughout the seed structures with both RAREVTR and ZTE-VFA (Figs. 3e and 3h), although a difference of approximately 300 ms in the values between the two sequences was observed. Conversely, the T2 and relaxation times seemed to vary more between the structures (Figs. 3f and 3g). The signal intensity obtained with the MGE sequence, and by extension the estimated values, was much lower in the megagametophyte compared with the corresponding T2 relaxation times attributed to relatively long TEs in the utilized MGE sequence. As such, the obtained image set with varying echo times will contribute to the performance of the fitting observable as low SNR in the maps. Overall, no significant differences in the relaxation times of the damaged and healthy seeds could be observed (supplementary Figs. S5 and S61). Similar uniformity of the T1 values was observed using high-resolution magnetic resonance microscopy (using a 5 mm 1H RF coil), although these measurements demonstrated a larger spread in the T2 relaxation times in the megagametophyte regions with higher values near the embryo (Fig. 4).
The statistics of T1, T2, and relaxation times obtained with 10 mm 1H RF coil for different seed regions are presented in Table 2, excluding empty seeds, as their relaxation times could not be determined due to low signal intensity and their obscured structure. The average T1 values for all regions (embryo, megagametophyte, cell membrane, and whole seed) varied between 650 and 680 ms, with the longest value in the megagametophyte and the shortest value in the embryo. For T2 relaxation times, the values were between 13 and 15 ms, with the shortest values in the megagametophyte. value of the megagametophyte had larger variation (Table 2). The low SNR in the megagametophyte region with the MGE pulse sequence is likely to contribute to the increased spread observed in the values. Relaxation time data of the individual seed embryos and whole seeds are presented in supplementary Figs. S5 and S61, respectively.
Of all the imaged seeds (n = 90), 10.0%, 73.3%, and 80.0% had germinated by days 7, 10, and 14, respectively (Table 3). The nongerminating seeds (20.0%) were all classified with radiography as mechanically damaged or empty. With one-way ANOVA and grouping the seed relaxation times with corresponding germination days (excluding empty seeds as their relaxation times could not be estimated), no statistically significant differences could be observed (p values of 0.49, 0.21, 0.10, 0.81 for T1 RAREVTR, T2 MSME, MGE, and T1 ZTE-VFA, respectively) (Fig. 5). It should be noted, however, that categories other than 10-day germination were significantly underrepresented, limiting statistical interpretation (Fig. 5, 10% and 6.7% for days 7 and 14, respectively). When comparing the relaxation times between the seed structures, the one-way ANOVA showed statistically significant differences; by comparing the relaxation times among the embryo, megagametophyte, and cell membrane, the p values were well below 0.01 (p = 4.3 × 10–25, 7.8 × 10–27, and 1.0 × 10–12 for T1 RAREVTR, T2 MSME, and MGE, respectively) (Table 2). From the ANOVA statistics, statistically significant differences (p < 0.05) among the three regions (embryo, megagametophyte, and cell membrane) were observed for T1 RAREVTR among all regions, for T2 MSME between embryo and megagametophyte as well as between megagametophyte and cell membrane, and for MGE between embryo and cell membrane as well as between megagametophyte and cell membrane. These differences are presented in Table 2 as well. It should be noted that the arithmetic mean and median of T1 RAREVTR and T2 MSME in the embryo, megagametophyte, and cell membrane were equal, indicating a possibility for a normal distribution, whereas for MGE data with lower SNR this was not the case (Table 2). This is further emphasized by the results of the one-sample Kolmogorov–Smirnov test (supplementary Figs. S2–S41).
The seed structures can be more readily differentiated and interpreted by reorienting the 3D datasets from ZTE and MGE pulse sequences. This procedure also illustrates the differences between intact, mechanically damaged, and empty seeds (Fig. 6). Altogether five seeds were interpreted as mechanically damaged using radiography (K393a-4, K393b-4, K682a-2, K682b-7, and K682c-1). Of these seeds, however, three germinated. These seeds were studied by recalculating the sagittal and axial slices from larger 3D datasets containing multiple seeds (Fig. 7). Two nongerminating seeds had areas of more significant signal loss within the megagametophyte and embryo. For K393a-4, the cotyledons appeared to be entirely detached from the embryo stem, deduced from a signal gap between the structures. For K393b-4, the loss of germination was obvious, as the radicle was entirely detached, and an overall mechanical fracture was evident. Germination of seeds K682a-2, K682b-7, and K682c-1 was not affected. For K682a-2 and K682c-1, the fractures observable with MRI were minimal, with only small areas of possible signal loss near the embryo cavity. For K682b-7, the interpretation of mechanical damage in the X-ray images may have been erroneous, as the axial MRI slices revealed a possible developmental malformation of the megagametophyte, which could have caused the corresponding intensity changes in the radiograph.
The purpose of this study was to investigate the possibilities of MRI and quantitative relaxation time mapping to determine structural characteristics and germination potential of Scots pine seeds. The investigated Scots pine seeds were collected from three ramets of three clones of Scots pine from a commercial clonal seed orchard located in central Finland (K393, K682, and K942). These seeds were MR imaged with pulse sequences capable of providing both anatomical and quantitative relaxation time information. MRI allows the alteration of image contrast by using different pulse sequences and consequent imaging parameters (Fig. 3). By understanding the relaxation properties of various tissues and seed regions, different image weightings and thus application-specific contrasts can be obtained. In this study, the relaxation time maps were calculated for the embryo, megagametophyte, and cell membrane of intact and mechanically damaged seeds that can further elucidate their compositional characteristics. Anatomical MR images provided complementary information on radiographs with some structures more visible while others remained undetectable. The 3D MRI acquisitions yielded volumetric information on seed structures and provided insight into mechanical faults and even possible developmental malformations. Scots pine seeds seem to exhibit a prominent NMR signal from lipids, a similar property previously described with western white pine seeds (Terskikh et al. 2005b).
The differences between X-ray radiography, a widely applied method to estimate seed structures, and MRI used in this work are obvious; the two-dimensionality of the radiographs provides a significant limitation in structural interpretation, and the inflicted radiation dose, albeit small, could inhibit its use in some circumstances. Similar restrictions may occur with CT; a modality more comparable with MRI due to its three-dimensionality, although the dose inflicted by CT may be significantly larger than in radiography due to multiple projections. Conversely, MRI was not observed to affect germination, as all the healthy seeds germinated, even though the germination rate was reduced due to a year in room temperature prior to the germination testing. But this was expected. MRI also offers the possibility to interpret seed structures further through quantitative relaxation time mapping. Some of the mechanically hardest structures and regions void of free 1H (low water or lipid content), mainly the seed coat, are not visible in MRI at all, even with ultrashort echo time pulse sequences (i.e., the ZTE pulse sequence here). Thus, with MRI, the interpretation of the seed regions was restricted to the internal structures. As such, the two modalities can be seen complementary to each other.
In MR images, the megagametophyte and embryo exhibit vastly different signal intensities, while the X-ray radiographs demonstrate near uniform intensities throughout the structures (Fig. 1b). The MR images further reveal that the megagametophyte has nonuniform intensity with larger intensities near the embryo and lower intensities near the seed coat, suggesting regional differences in the composition. The differences in the intensities in the MR images provide novel research questions on the distribution of constituents (i.e., water and lipids), as these are typically evaluated based on the whole seed concentrations for conifer seeds (Tillman-Sutela et al. 1995; Tammela et al. 2005) rather than on the seed substructure level.
Relaxation times illustrate the dynamic properties of the seed regions, and as such these material- and sample-specific parameters also affect the choice of appropriate anatomical MRI parameters. Such user-controllable parameters are the echo time (TE) and repetition time (TR). Roughly, for exemplary anatomical contrast, the TE imaging parameter should be adjusted with respect to the T2 or relaxation times of the sample such that the desired regions are visible and differ. If the TE is much longer than the T2 of a given region, the region produces a low signal intensity, and remains invisible in MRI. Conversely, if the TE is short, or ultra-short as with, for example, ZTE, the signal is produced from virtually all possible regions, but the contrast between the regions may also be lost. Similarly, the TR is generally intertwined with the T1 relaxation time of the sample regions. By adjusting the TR, image weighting dependent on the T1 can be achieved. The resulting contrast for different MRI pulse sequences can be predicted from the respective signal eqs. S1–S4 presented in the supplementary material1. The MRI contrast is broadly tunable by adjusting the imaging parameters; for optimal visibility and contrast, the relaxation times of the target tissues should be known.
It was observed that the relaxation times did not greatly vary between the clones or ramets and individual full seeds. Although the signal intensities of the embryo were higher than those of the surrounding megagametophyte with all the presented pulse sequences, the T1 values obtained with RAREVTR did not vary greatly between the structures (Fig. 3). A similar property for T1 relaxation times was observed with the ZTE-VFA sequence, which can detect the signal from almost all structures. The T1 relaxation times obtained with the ZTE-VFA had a larger range (max–min), indicative of broader detection of signals and the possibility of RF pulse inhomogeneity affecting the obtained pixel-wise values over the whole acquisition volume. Furthermore, although the T1 relaxation times were homogeneous throughout the structures, the T2 and values varied, with generally higher values in the embryo than in the surrounding structures. It was observed that the signal intensity was lower, and present in fewer structures with the MGE sequence when compared to MSME (Figs. 3b and 3c). For the megagametophyte region, could not be accurately estimated due to the minimal signal intensity even with the shortest possible echo times (0.8 ms) with the MGE sequence ( effect), whereas with the spin echo sequences, the signal from this region was prominent even with significantly longer echo times of 40 ms (T2 effect). The short values in the megagametophyte could originate from its porous structure or composition, causing rapid signal loss due to local magnetic field inhomogeneities and magnetic susceptibility differences. A UTE approach with shorter TEs could provide a better estimate of the . For all anatomical images, the embryo was the most visible structure in terms of MR signal intensity. It should be noted, however, that although relaxation times can be considered to be more comparable between MR scanners, differences in relaxation time values between pulse sequences (Matzat et al. 2015) as well as RF coils (Chang et al. 2012; Pachowsky et al. 2013) have been demonstrated. Furthermore, T1, and to an extent T2 and values characterized in this work are dependent on the magnetic field strength.
With the ZTE pulse sequence, the inner anatomy of the pine seeds could be differentiated, although blurring on some individual seeds was observed (supplementary Fig. S71). This is most likely caused by the difference in the resonance frequencies between the water and lipid protons, which is due to the chemical shift (see Fig. 2), causing blurring in the radial acquisition scheme of ZTE (Froidevaux et al. 2018). This blurring was more pronounced in some seeds; it is likely that the water content in those seeds was higher than in the seeds that demonstrated minimal blurring, i.e., the blurred seeds may exhibit a larger off-resonance peak. Overall, the two-peak NMR spectrum (Fig. 2) provides artifacts, i.e., chemical shift and blurring, and can hinder the use of 3D ZTE and MGE datasets for volumetric interpretation. This effect is also evident in the multi-slice MR images as a visible spatial shift of the water vs. fat voxels in a specific direction (along the frequency encoding of the MRI pulse sequence, affects the signal also in the slice selection direction). In the future, MRI techniques supporting water–lipid separation could provide more optimal volumetric interpretation and beneficial information on the compositional aspects of Scots pine seeds, and for lipid-rich seeds in general. Alternatively, spectroscopic measurements of the relaxation times of the fat and water peaks separately could provide further information on the relative contributions of changes in either of the components.
The recalculated slices from the 3D MRI data (Fig. 6) provide interesting opportunities for more detailed analysis and identification of seeds and their structures from larger imaging batches. Furthermore, in the recalculated slice orientations, differences in the signal intensities between the seed categories (intact, damaged, and empty) can be more readily appreciated. Although the resolution can generally be higher with smaller RF coils (for example, Fig. 1c vs. Fig. 7), the reduced overall scan time per seed when imaging multiple seeds simultaneously should also be considered when planning and initializing such MRI experiments. The mechanical damage of Scots pine seeds could be distinguished and localized with the help of 3D MRI. Small fractures in the megagametophyte do not seem to affect germination, but conversely, larger, extensive signal loss in the megagametophyte extending to or near the embryo appeared to correspond with the loss of germination. Furthermore, the presented relaxation time values provide valuable information on the MRI characteristics of Scots pine seeds and can be used in imaging parameter selection as well as for the development and application of MRI pulse sequences. With the MRI of larger seed batches combined with, for example, a real-time surveillance of the germination protocol post-MRI, potential associations between qMRI parameters and germinability may be evaluated.
Previous studies related to MRI of conifer seeds utilizing SE pulse sequences provide similar signal intensities than the ones presented in this work, albeit obtained with a variety of magnetic field strengths between 7.1 and 9.4 T (Carrier et al. 1999; Terskikh et al. 2005b; Keeling et al. 2018), compared to the 11.7 T in this work. For conifer seeds, use of GRE sequences has not been described previously and MRI studies on conifer seeds are in general quite limited.
Some limitations exist for the anatomical MRI and relaxation time mapping of the Scots pine seeds. While potentially more feasible for pine seeds due to the broader signal detection, the T1 maps acquired with the ZTE sequence may be biased, as the RF excitation pulse applied, and its spatial inhomogeneity affect the obtained flip angles, and thus interfere with the T1 quantification (supplementary Fig. S51; Table 2). Thus, a more reliable estimate of the T1 values for the defined ROIs can be obtained with the RAREVTR sequence, as the RF pulse inhomogeneity is not as significant factor with this sequence. The T1 relaxation times estimated from RAREVTR data exhibit a homogeneity throughout the seed structures. However, similar spatial distribution of T1 relaxation times was observed with the ZTE-VFA, although fluctuations in the values were evident. Besides potential accuracy issues with the ZTE-VFA sequence, the pools of spins (1H) contributing to the observed signal and thus relaxation times are different between the sequences, explaining the bulk differences in the values.
Furthermore, as can be seen from Figs. 3 and 4, the calculated relaxation times are not numerically equivalent between slightly different measurement setups, as for example, the T2 values of an individual pine seed obtained with the 5 mm RF coil are 5–10 ms longer than the corresponding values obtained with the 10 mm RF coil, due to slight differences between the acquisition parameters in addition to differences in the physical measurement setup. This further illustrates the importance of standardized imaging parameters, as these can affect the obtained signal. The higher SNR of the smaller RF coil can contribute to these differences as well which is produced due to the closer fit of the seed sample material to the sensitive region of the RF coil assembly.
Another limitation of the study is the small sample size of the imaged seeds as well as only applying the method for seeds collected from a single location and in a single year. Further evaluation of MRI for the study of Scots pine or other conifer seeds would benefit from larger populations and from a broader geographical area. In addition, although the present method utilized a stationary device in a laboratory, smaller, mobile MRI devices could potentially be used for similar studies in the field (Blümich 2017).
Another aspect regarding, and in some cases even limiting the use of MRI, is the scan time required for appropriate image quality. The presented images and their respective pulse sequences required imaging times from 20 min to more than 6 h (supplementary Table S11). Although a large part of the scan time in the present study is due to extended signal averaging to obtain a better SNR, it is still substantial when compared to the time required for 2D radiography (4 s exposure time in this work). The 3D ZTE acquisition took 6 h and 18 min for anatomical acquisition with 16 averages. While some reduction of the scan time is certainly feasible by for example reducing the number of signal averages (i.e., number of excitations, NEX), or by collecting less data per single image (less k-space data), it will also reduce the SNR and thus image quality.
The definition of the ROIs is not straightforward. The selection of the highest intensity regions in the center of the seed cavity can be easily conducted with all the sequences, while the megagametophyte, although homogeneous in the radiographic images, exhibits considerable intensity gradients in the MR images, with higher intensities near the embryo (Fig. 3c, seed 3). Thus, due to possible problems in segmentation, the megagametophyte may not have been accurately identified in all cases. The observed cell membrane, a one- or two-voxel-wide intensity region with selected imaging parameters, could be detected at the outer boundary of the megagametophyte, although with some pine seeds and pulse sequences this differentiation could not be accurately achieved. Another consideration regarding the ROIs is the possibility of fractures and mechanical faults in the megagametophyte and other regions: these dark, low signal areas were not selected as they could interfere with the assessment of the ROI-average relaxation times.
Microscopic MRI provides insight into the structural and compositional characteristics of Scots pine seeds with a plethora of different contrasts that can elucidate the 1H concentration and distribution. The location and severity of mechanical damage in the studied seeds were detectable in more detail compared to the traditional projection radiography. The method thus shows promise in increasing understanding the reasons for poor germination in some seed lots and in finetuning existing seed sorting methods, such as PREVAC. In the future, study of the growth of seedlings could provide further insight into the utility of the MRI findings. The calculated T1, T2, and relaxation time maps represent the behavior of the NMR signal in different regions of the seed and assist in the selection of appropriate MRI parameters in the future. Seed germination potential could not be directly evaluated with the obtained relaxation time maps although the anatomical images of the seed structures did provide insight into to the closely related anatomical potential similar to that of radiographic results. In terms of relaxation times, the differences between certain seed regions were statistically significant although the small sample size does limit the evaluation on the whole population of Scots pine seeds. No statistical significance was observed between the relaxation times and the day of germination. When compared to the X-ray radiographs, the differences in the observable regions and the obtained contrasts vary extremely compared with the uniform signal in X-ray images, with some regions more distinguishable (embryo and megagametophyte) while others remain undetectable (the seed coat). Thus, MRI and radiography provide complementary support for each other in the analysis of seed structure and germination potential.
The authors declare there are no competing interests.
The MRI acquisitions for this work were carried out in the facilities of the Kuopio Biomedical Imaging Unit, University of Eastern Finland, Kuopio, Finland. MRI experiments were conducted as a part of the European Regional Development Fund and the Pohjois–Savo Regional Council project for MicroMRI (grants A73998 and A74016). The authors gratefully acknowledge further financial support for the project from the Niemi Foundation, Academy of Finland (grant Nos. 285909 and 325146) as well the project Improving Seed Orchard Management (SITKE), funded by the Ministry of Agriculture and Forestry of Finland. X-ray radiography and related information used in this work was made possible by the funding of European Regional Development Fund for a new X-ray device (project A75285).
Carrier D.J., Kendall E.J., Bock C.A., Cunningham J.E., and Dunstan D.I. 1999. Water content, lipid deposition, and (+)-abscisic acid content in developing white spruce seeds. J. Exp. Bot. 50(337): 1359–1364.
Chang G., Wiggins G.C., Xia D., Lattanzi R., Madelin G., Raya J.G., et al. 2012. Comparison of a 28-channel receive array coil and quadrature volume coil for morphologic imaging and T2 mapping of knee cartilage at 7T. J. Magn. Reson. Imaging, 35(2): 441–448.
ISTA Moisture Committee. 2007. Constant temperature oven method. In ISTA Handbook on Moisture Determination. 1st ed. Edited by H. Nijënstein, J. Nydman, R. Don, and C. McGill. The International Seed Testing Association (ISTA), Bassersdorf, Switzerland. pp. 72–84.
Kärkkäinen K. and Savolainen O. 1993. The degree of early inbreeding depression determines the selfing rate at the seed stage: model and results from Pinus sylvestris (Scots pine). Heredity, 71(2): 160–166.
Pachowsky M.L., Trattnig S., Apprich S., Mauerer A., Zbyn S., and Welsch G.H. 2013. Impact of different coils on biochemical T2 and T2* relaxation time mapping of articular patella cartilage. Skeletal Radiol. 42(11): 1565–1572.
Tammela P., Salo-Väänänen P., Laakso I., Hopia A., Vuorela H., and Nygren M. 2005. Tocopherols, tocotrienols and fatty acids as indicators of natural ageing in Pinus sylvestris seeds. Scand. J. For. Res. 20(5): 378–384.
Terskikh V.V., Feurtado J.A., Borchardt S., Giblin M., Abrams S.R., and Kermode A.R. 2005a. In vivo 13C NMR metabolite profiling: potential for understanding and assessing conifer seed quality. J. Exp. Bot. 56(418): 2253–2265.
Terskikh V.V., Feurtado J.A., Ren C., Abrams S.R., and Kermode A.R. 2005b. Water uptake and oil distribution during imbibition of seeds of western white pine (Pinus monticola Dougl. ex D. Don) monitored in vivo using magnetic resonance imaging. Planta, 221(1): 17–27.
Terskikh V.V., Zeng Y., Feurtado J.A., Giblin M., Abrams S.R., and Kermode A.R. 2008. Deterioration of western redcedar (Thuja plicata Donn ex D. Don) seeds: protein oxidation and in vivo NMR monitoring of storage oils. J. Exp. Bot. 59(4): 765–777.
Tillman-Sutela E. and Kauppi A. 2014. Maturity of surface structures in northern Pinus sylvestris L. seeds: a key to improved prediction of germination potential. Flora Morphol. Distrib. Funct. Ecol. Plants, 209(1): 45–53.
Tillman-Sutela E., Johansson A., Laakso P., Mattila T., and Kallio H. 1995. Triacylglycerols in the seeds of northern Scots pine, Pinus sylvestris L., and Norway spruce, Picea abies (L.) Karst. Trees, 10(1): 40–45.
van Dusschoten D., Metzner R., Kochs J., Postma J.A., Pflugfelder D., Bühler J., et al. 2016. Quantitative 3D analysis of plant roots growing in soil using magnetic resonance imaging. Plant Physiol. 170(3): 1176–1188.
This research was supported by the European Regional Development Fund and Pohjois-Savo Regional Council (grants A73998, A74016, and A75285), Niemi Foundation, Academy of Finland (grants 285909 and 325146) and Ministry of Agriculture and Forestry of Finland (Improving Seed Orchard Management, SITKE).
Metrics & Citations
Teemu V. Tuomainen, Katri Himanen, Pekka Helenius, Mikko I. Kettunen, and Mikko J. Nissi. Quantitative magnetic resonance imaging of Scots pine seeds and the assessment of germination potential. Canadian Journal of Forest Research.
52(5): 685-695. https://doi.org/10.1139/cjfr-2021-0273
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.