Three-dimensional Magnetic Resonance Imaging of fossils across taxa

The visibility of life forms in the fossil record is largely determined by the extent to which they were mineralised at the time of their death. In addition to mineral structures, many fossils nonetheless contain detectable amounts of residual water or organic molecules, the analysis of which has become an integral part of current palaeontological research. 5 The methods available for this sort of investigations, though, typically require dissolution or ionisation of the fossil sample or parts thereof, which is an issue with rare taxa and outstanding materials like pathological or type specimens. In such cases, non-destructive techniques could provide an interesting methodological alternative. While Computed Tomography has long been used to study palaeontological specimens, a 10 number of complementary approaches have recently gained ground. These include Magnetic Resonance Imaging (MRI) which had previously been employed to obtain three-dimensional images of pathological belemnites non-invasively on the basis of intrinsic contrast. The present study was undertaken to investigate whether 1 H MRI can likewise provide anatomical information about non-pathological belemnites and 15 specimens of other fossil taxa. To this end, three-dimensional MR image series were acquired from intact non-pathological invertebrate, vertebrate and plant fossils. At rou-tine voxel resolutions in the range of several dozens to some hundreds of micrometers, these images reveal a host of anatomical details and thus highlight the potential of MR techniques to e ﬀ ectively complement existing methodological approaches for palaeon- 20 tological investigations in a wide range of taxa. As for the origin of the MR signal, relaxation and di ﬀ usion measurements as well as 1 H and 13 C MR spectra acquired from a belemnite suggest intracrystalline water or hydroxyl groups, rather than organic residues.

The methods available for this sort of investigations, though, typically require dissolution or ionisation of the fossil sample or parts thereof, which is an issue with rare taxa and outstanding materials like pathological or type specimens. In such cases, nondestructive techniques could provide an interesting methodological alternative. While Computed Tomography has long been used to study palaeontological specimens, a 10 number of complementary approaches have recently gained ground. These include Magnetic Resonance Imaging (MRI) which had previously been employed to obtain three-dimensional images of pathological belemnites non-invasively on the basis of intrinsic contrast. The present study was undertaken to investigate whether 1 H MRI can likewise provide anatomical information about non-pathological belemnites and 15 specimens of other fossil taxa. To this end, three-dimensional MR image series were acquired from intact non-pathological invertebrate, vertebrate and plant fossils. At routine voxel resolutions in the range of several dozens to some hundreds of micrometers, these images reveal a host of anatomical details and thus highlight the potential of MR techniques to effectively complement existing methodological approaches for palaeon-
Such observations led to the suggestion that part of the organic material detected 15 in fossils might actually represent the most stable portion of the molecules originally constituting the individual at the time of its death (e.g. Abelson, 1954;Florkin, 1965;Westbroek et al., 1979;Eglinton and Logan, 1991;Engel et al., 1994;Schweitzer et al., 2007), which opened the door for palaeobiochemical investigations (Blumer, 1965;Albrecht and Ourisson, 1971;Niklas and Gensel, 1976;Weiner et al., 1976; West-

EGU
In this study, we concentrated on in saxo MRI whose signal intensity had been found to co-vary with pathological alterations of biomineralisation in the belemnite specimens mentioned above. Specifically, we addressed the issue whether the method would equally be applicable to non-pathological rostra and to fossils other than belemnitesas previous studies of fluids in porous rocks (de Swiet et al., 1998) and of extant biomin-5 eralised samples (Majumdar et al., 1998;Borah et al., 2001;Tsai et al., 2004;Müller et al., 2006) would suggest -and used MRI to investigate the internal morphology of non-pathological fossils of invertebrate, vertebrate and plant origins from different geological settings.
The fossil material figured here is housed at the Museum für Naturkunde der 10 Humboldt-Universität zu Berlin, Germany (acronym MB.) and in the Museum für Natur und Umwelt in Lübeck, Germany (acronym MNU).

Diagenesis of biomineralised structures
When considering the use of new methodologies like MRI for the structural investigation of fossils, it is necessary to reflect on the material characteristics of the skeleton, 15 the extent to which the original morphologic structures have been diagenetically altered and how this could affect the acquisition and interpretation of the data. The following sections will hence review biomineralisation and diagenetic processes in selected taxonomic groups, with a focus on preservation of organic matter in the mineral matrix (the preservation of water will be discussed in Sect. 5.4). Several reviews are available 20 that treat taphonomy, particularly chemical and microbial degradation of the organic and mineral fractions, in more detail (Behrensmeyer, 1978;Behrensmeyer et al., 2000;Collins et al., 2002;Dauphin, 2002;Kidwell and Holland, 2002;Briggs, 2003).
2963 monly been dissolved during diagenesis or replaced by other minerals, mainly calcite. The organic matrix has been attributed several functions in biomineralisation, including nucleation of the mineral, determination of the mineral phase, control of the orientation and growth of carbonate crystals, and enhancement of mechanical properties of the crystals (e.g. Crenshaw, 1990). Biochemically, it consists of proteins, poly-saccharides, 10 and water. A major portion of the soluble fraction of organic matrix macromolecules are aspartic acid-rich glycoproteins, whereas important constituents of the insoluble fraction are glycine, alanine and chitin (for details, see Lowenstam and Weiner, 1989). The properties of organic matrix components are subject to often severe postmortem alterations -proteins, for instance, decompose to individual amino acids which can un-15 dergo isomerisation and further decomposition down to simple hydrocarbons (Collins and Gernaey-Child, 2001). Several studies have documented the presence of amino acids in fossil shells as old as about 360 Ma (summarised in Weiner, 1979). However, the amino acid composition of a Late Cretaceous (80 Ma before present) ammonoid shell bears no resemblance to those found in a close extant relative, the cephalopod 20 Nautilus (Weiner, 1979).

Belemnites
Another group of cephalopods, now extinct, are the belemnites. They had an endoskeleton whose most distal part -the rostrum or guard that helped to maintain a horizontal swimming posture (Naef, 1922) -is frequently preserved. It consists of the 25 rostrum cavum with the alveolus -a conical cavity at its anterior end -and the rostrum 2964 periodical accretions of radial structures that resulted in spatial variations of the organic content. This variation is often subtle, so that growth rings may be difficult to define (Saelen, 1989).
Apart from the primordial rostrum and very early growth stages, the original mineralogy of belemnite rostra was low-Mg calcite (Veizer, 1974). As such they are relatively 10 stable even under freshwater influence (meteoric diagenesis). The microstructure consists of regular, fine prisms with parallel crystal axes, arranged in well-ordered prismatic layers (Bandel and Spaeth, 1988) which can still be discerned in even strongly recrystallised specimens. Diagenetic exchange between rostra and enclosing rock appears to be limited (up to about 10% by weight, according to Veizer, 1974) and may be caused 15 by solution-precipitation phenomena or filling of the pore space. The latter was either primary or diagenetically generated by the decay of organic matter. Growth lines are frequently preserved.
With respect to organic matrix macromolecules in belemnites, Westbroek et al. (1979) analysed rostra of two late Cretaceous taxa, Gonioteuthis and Belemnitella. 20 In the soluble macromolecular fraction of well-preserved Gonioteuthis rostra, these authors identified components with peptidic and saccharidic properties as well as an amino acid composition very similar to that of Nautilus, dominated by glycine and alanine. Even original antigenic properties of certain fractions were still preserved, suggesting that the biochemical materials derived from Gonioteuthis were original belem- 25 nite compounds which only experienced minor alterations during diagenesis. The observed enrichment in polyphenols may be due to reactions between peptides and carbohydrates during diagenesis. In contrast to the exceptionally well-preserved rostra of Gonioteuthis, those of Belemnitella were strongly recrystallised. Although the 2965 EGU amino acid composition is similarly dominated by glycine (12.6 mol%) and alanine (11.5 mol%), less stable (threonine, serine, arginine -5.5, 6.2 and 4.9 mol%, respectively) and even very labile amino acids (methionine, 2.2 mol%) were also present. Westbroek et al. concluded thus that the primary organic composition of the Belemnitella rostra was contaminated during or after recrystallisation by percolating ground 5 water.

Crinoids
As in all echinoderms, the crinoid stalk ossicles belong to the mesodermal endoskeleton and are composed of magnesium calcite which is arranged in the typical form of a stereom (a meshwork of anastomosing trabeculae and pillars). In the living crinoid, the interspace in the stereom is filled with soma which contains living cells, mainly sclerocytes and phagocytes (Heinzeller and Welsch, 1994). Columnals bearing cirri are nodals, those without cirri are internodals. Connections of the ossicles are nonmuscular and exclusively by elastic ligaments of mutable collagenous tissue. Short, intercolumnar ligaments connect each pair of adjacent columnals. Longer, continuous 15 ligaments connect each set of internodals and one associated nodal (Ausich et al., 1999).
Columnals of Isselicrinus buchii are cylindrical, rarely subpentagonal, with a smooth surface. The corresponding articulations of adjacent ossicles bear a characteristic pattern: Interlocking grooves and ridges of densified stereom occur in a petaloid pattern, 20 giving the stem a certain flexibility. A crenulation is also visible at the margin of most ossicles, and the stem possesses a small central canal which represents a tubular cavity with extensions of the coelom and nervous system (Ausich et al., 1999).
Skeletal particles of echinoderms are usually preserved as large single crystals of calcite. During diagenesis, the metastable high Mg-calcite of the echinoderm skeleton 25 is replaced by low Mg-calcite. This transformation occurs by means of a solid-state process known as incongruent dissolution. Unidirectional growth of calcite crystals gives rise to syntaxial cements, the formation of which starts with the infilling of the 2966 Diagenetic processes acting upon vertebrate remains are far from being completely understood but for most practical purposes, they can be reduced to the diagenesis of bone and its collagenous protein matrix. Although this is still a complex, multistage process, notoriously dependent on external biochemical, hydrological and taphonomical factors of the embedding environment and internal parameters such as bone size, 15 histological structure and collagen content (Martill, 1991;Schweitzer et al., 2007), it has been relatively well-studied in archaeological sciences, especially in relation to collagen decomposition . Similar to most other proteins, collagen is composed mostly of carbon, hydrogen, nitrogen and a minor content of sulphur, and its initial decay during early diagenesis is of great 20 importance for the final preservation and chemical composition of the bone Pfretzschner, 1998Pfretzschner, , 2000Schweitzer et al., 2005Schweitzer et al., , 2007. Generally, microbial degradation of collagen accelerates destruction (Collins et al., 1995) and therefore hinders fossil preservation of bone. However, microbial activity depends on a number of chemical parameters -including temperature, pH and the availability of oxygen and water -whose combination might lead to reduced microbial degradation (Bocherens et al., 1997;Nielsen-Marsh and Hedges, 2967 water and marine environments. Finally, even if two samples have been taken from the same fossil specimen, the circumstances of their excavation, handling, storage and analysis can severely influence the residual biochemical information contained therein, as has been shown for DNA contents (Pruvost et al., 2007) but can be assumed to be the case for other macromolecules and water as well.

Bone preservation and taphonomy
Bone material consists of the biomineral hydroxy apatite -Ca 5 (PO 4 ) 3 OH -which can be modified during fossilisation by various chemical interactions with the environment and the sediment covering the carcass (Rottländer, 1979). Depending on the environmental distribution of reactants (Pfretzschner, 1998), the mobile ions within the apatite 15 crystal -H + , Ca 2+ , HPO 2− 4 and OH − -can be substituted, as Piepenbrink (1989) has shown for subfossil human bone. Ca 2+ , for example, is often replaced by UO 2+ 2 Pfretzschner, 1997) (Newesely, 20 1989;Pfretzschner, 2000;Trueman, 1999). The histological structure of bone is important for the interaction with the ambient fluid. The pore structure, which often follows a trajectory pre-defined by in vivo biomechanical load, determines the internal surface area and therefore the capacity for surface reactions .

2968
Printer-friendly Version Interactive Discussion

Fossil ear bones
With the exception of some phylogenetically early species (cf. Luo et al., 2007), the mammalian ear bones typically comprise the tympanic bulla, the middle ear with the three ossicles malleus, incus and stapes, and the periotic bone with the cochlear portion containing the spiral cavity which forms the organ of sound reception (Berta and Sumich, 1999). The cochlea is coiled and divided lengthwise into three parallel tubular channels (Liem et al., 2001, Fig. 12-22). In cetaceans, the periotic is the most compact skull bone and of a comparatively stable and chemically resistent nature. As an adaptation to underwater hearing, it is displaced from the skull by pneumatic spaces (e.g. Purves and Pilleri, 1983) and only 10 loosely connected to it via ligaments to the mastoid process (Pilleri et al., 1987). The detachment allows separate reception of sound and isolated vibrations of the ear bones (Miller, 1923;Fleischer, 1978;Pilleri et al., 1987). The anatomy of the cetacean organ of hearing is well described (Pilleri et al., 1987) but was usually studied by producing serial sections by grinding the petrosals (e.g. Luo and Eastman, 1995;Luo and Marsh, 15 1996), resulting in the loss of the unique fossil specimens. To our knowledge, noninvasive imaging of fossil ear bones has not been reported so far, though petrosals obtained from contemporary animals (e.g. Ketten and Wartzok, 1990;Nummela et al., 1999) have successfully been imaged by CT, as have entire fossil cetacean skulls (at resolutions not suitable for petrosal investigations, cf. Marino et al., 2003). 20 Incidentally, the first MR imaging of fossils that we know of (Sebes et al., 1991) has been performed on the vertebrae of a Miocene dolphin, though the specimen was immersed in water which provided the contrast. The specimen was not specified in that report but determined as Xiphiacetus bossi (Kellogg) USNM 10480 (D. Bohaska and B. Rothschild, personal communication). At the time of writing, it was not known whether the water immersion had had any impact on the state of preservation of these vertebrae, as compared to non-immersed vertebrae of the same specimen. 2969 plant parts exhibit preserved morphological patterns down to the cellular scale, and they often contain organic and carbon compounds. Silicification depends on the thermal conditions (Sigleo, 1978) and can occur over a wide pH range and sometimes very rapidly (in vitro as fast as in 24 h, according to Drum, 1968). The most probable mechanism for wood silicification is hydrogen bonding between silicic acid, Si(OH) 4 , and the hydroxyl groups in cellulose. When studying Araucarioxylon arizonicum fossils from the Triassic Chinle Formation of Petrified Forest National Park in Arizona for lignin derivates, Sigleo (1978) was able to identify various organic compounds in silicified wood by sequential high vacuum pyrolysis and gas chromatography mass spectrometry. These organic components included phenols, methylphenols, alkyl substituted As for noninvasive imaging, x-ray tomography was applied successfully to pyritized fossil fruits from the Lower Eocene London Clay flora, namely to the visualisation of internal structures, including small seeds, within Myrtaceaen fruits (DeVore et al., 2006), and further applications are about to emerge, e.g. for the study of internal structures 25 (down to the cellular level) within charcoalified three-dimensionally preserved Cretaceous flowers (Friis et al., 2006). MR imaging has, however, found multiple applications to extant plant specimens (Chudek and Hunter, 1997), including conifer cones (e.g. Mill  The analysed belemnite rostra, identified as Belemnopsis sp. by Dietrich (1933), were 5 collected by the German Tendaguru Expedition (1909( -1912( , cf. Janensch, 1914. The fossil locality of Tendaguru, famous for its diverse dinosaur assemblages, is located approximately 60 km northwest of the seaport of Lindi in southeastern Tanzania. The Late Jurassic to Early Cretaceous (ca. 155-130 Ma before present) Tendaguru Beds reach a thickness of 110 m and consist of three fine-grained dinosaur-bearing sequences 10 which are intercalated with sandstone-dominated sequences containing a predominantly marine fauna (Aberhan et al., 2002). The rostra were embedded in a mediumto coarse-grained sandstone of Late Jurassic (Tithonian) age at the transition between the so-called Trigonia smeei Bed and the base of the Upper Saurian Bed at Tendaguru site IX, about 1.4 km northeast of Tendaguru Hill. Most specimens are fragmented and 15 the outer surface appears pitted due to intense weathering. Recent sedimentological and palaeoecological analyses of the Tendaguru Beds (Aberhan et al., 2002) suggest that deposition of the Trigonia smeei Bed took place in lagoon-like, shallow marine environments above fair weather wave base and with evidence of tides and storms. Sediments of the Upper Saurian Bed represent extended siliciclastic tidal flat environ-20 ments including brackish coastal lakes and ponds. The Late Jurassic palaeoclimate of the Tendaguru area was subtropical to tropical, characterised by seasonal rainfall alternating with a pronounced dry season.
The analysed rostrum of Belemnella (Pachybelemnella)  pre-existing upright stem columns as anchorage (Fujiwara et al., 2004). The long stem raised the cup highly above the sea floor into the water current. As a passive suspension feeder, this crinoid was dependent on currents transporting plankton to the catch apparatus formed by the crinoid's arms (Oji, 1985(Oji, , 1990. We used only columnals (stem elements) for MR imaging.

Vertebrates
The bone used for the current study was a periotic originating from a partial skeleton of a kentriodontid dolphin (Cetacea: Odontoceti) belonging to the genus Atocetus de Muizon 1988 of the subfamily Pithanodelphininae Barnes 1985. This Atocetus fossil was discovered in a commercial gravel pit near Groß Pampau in Schleswig-Holstein, 20 northern Germany, and is stored since then in the Museum für Natur und Umwelt in Lübeck (MNU-071-18). The site is famous for several whale remains (e.g. Höpfner, 1991;Hampe, 1999). The horizon containing the whale fossils belonged to a dark mica clay ("Oberer Glimmerton" in regional stratigraphy; cf. Hinsch, 1990) that was deposited between 10.6 Ma and 11.8 Ma (Spiegler and Gürs, 1996)  EGU silicate muscovite and occasionally in pyrite, glauconite, and carbonate (Gripp, 1964).

Plants
For this study, two silicified fossils have been chosen that exhibited some wellpreserved anatomical details under the light microscope. They were collected by M. Wehrfeld in North Patagonia, Argentina, in 1937 and later donated to W. Gothan (cf. 5 Gothan , 1950). The material comes from the "classic" locality Cerro Cuadrado, where silicified conifer remains are preserved in a volcanic ash. The age of these fossils is considered to be mid-to late Jurassic (ca. 150-160 Ma;Calder, 1953;Menéndez, 1960). Cones in the same state of preservation have been described earlier (Spegazzini, 1924;Gothan, 1925;Darrow, 1936;Calder, 1953;Stockey, 1977Stockey, , 1978. A com-10 prehensive book on Cerro Cuadrado conifer fossils has been published by Dernbach et al. (1992). The cones are completely silicified by alpha-quartz, as demonstrated by x-ray diffraction (Stockey, 1975). This quartz, known as chalcedon, spots a wide variety of colours. In case of the conifer cones, abundantly interspersed hematite renders the fossils reddish brown. However, it seems that different seed tissues of these cones tend 15 to show differently coloured quartz (Darrow, 1936). Consequently, the parts of these cones can be easily distinguished by light microscope when cut. Two fossils were used for this study. The first one (MB.Pb. 2006/511) is a silicified twig of about 9 cm length and approximately 1.3 cm diameter from an araucarian conifer and appears poorly preserved. The piece is broken off at the base and at the top. The Brachyphyllum-type 20 foliage is dense and seems to be helically arranged. The leaves show rhomboidal leaf cushions and acute apices. In most cases, only the leaf base is preserved, and no obvious wood or leaf structures are visible under the light microscope. The second piece (MB.Pb. 2006/512), a cone of a conifer of uncertain affinities, Pararaucaria patagonica (Wieland, 1935), is approximately 3.3 cm in length and 1.5 cm in maximal diameter 25 and exhibits a good three-dimensional preservation under the microscope (cf. Fig. 5a). This cone belongs to a group of three specimens in the collection that had already partly been sectioned horizontally and vertically (MB.  (Stockey, 1977). Numerous helically arranged cone-scale complexes are arranged around a slender cone axis. Each cone scale complex is composed of a large woody ovuliferous scale sub-5 tended by a smaller, flattened woody bract. Externally, cone-scales resemble those of some members of the Taxodiaceae, e.g., Sciadopitys and Sequoiadendron but the bract and ovuliferous scale of Pararaucaria are not fused and instead free for almost their entire length. Seed tissue is partly preserved, too: one flattened, winged, roughly heart-shaped seed is embedded in the upper surface of each ovuliferous scale.

Magnetic Resonance Imaging and Spectroscopy
As for MR imaging, the same methodology as in Mietchen et al. (2005)  EGU quickly spun around an axis inclined to the static magnetic field, such that anisotropic interactions between MR-sensitive nuclei average out (reviewed in Andrew, 1981;Garroway et al., 1981;Veeman, 1997). For MAS NMR spectroscopy, the powdered sample was placed inside a standard ZrO 2 MAS rotor of 4 mm outer diameter. The data were acquired at a spinning speed of 10 kHz and a repetition time of 3 s. For the 1 H spectra, 5 32 transients were averaged, and 22 200 for the 13 C spectra.

Belemnites
In cross-sectional MR images of the Belemnopsis samples (cf. Fig. 2C,E), concen-10 tric circles reflecting radially oscillating signal intensity can easily be identified -with higher signal intensity indicating a higher number of mobile 1 H nuclei, and lower signal indicating lower 1 H contents, lower mobility or a combination of both. As for the molecular affiliation of these nuclei (see also Sect. 5.4), water would perhaps provide the most parsimonious explanation but organic remains of the original material would 15 also be compatible with recently published findings (Florek et al., 2004): organic and inorganic signals in electron microprobe measurements showed very similar oscillatory patterns in belemnite rostra, yet with opposite sign, along a radial line extending from the central channel. The signal oscillations correspond to growth rings reflecting a layered microstructure similar to the one described for nacre (Jackson et al., 1988;Feng 20 et al., 1999Feng 20 et al., , 2000, where inorganic layers (aragonite in bivalves, calcite in belemnites) basically alternate with organic layers, with each of both phases forming a continuum through small bridges (cf. Fig. 10ff in Jackson et al., 1988). The latter, however, can not be identified in our MRI data. In the longitudinal sections, the apical line or central channel is marked by low signal intensities over its entire preserved length, from the Visible are the small central canal and the grooves and riches serving as attachment surfaces for connective ligaments as well as crenellae along the margin. The spatial resolution achieved with this sample was 108 µm in the image plane depicted here (subsequently, it was artificially increased by a factor of two via 10 zerofilling of the data set before Fourier transformation) and 50 µm perpendicular to it -the highest resolution achieved in the MR images described in this study.

Vertebrates
The MR images of the whale periotics (cf. Fig. 4) reveal a strong signal in the posterior cochlear part and the posterior process of the bone. At the site of Groß Pam- 15 pau, the fossil bones are associated with diagenetically produced minerals like glauconite and pyrite. The latter is a product of a reaction between iron (Fe released by haemoglobin degradation) and sulphur (H 2 S released by protein degradation, cf. Pfretzschner, 2000). FeS 2 (pyrite) develops under alkaline conditions, e.g. in the presence of water-dissolved NH 3 produced during collagen decay. It is less clear, how-20 ever, what the destiny was of the hydrogen and carbon of the decomposed protein.
It appears possible, though, that part of these molecules still reside in the bone and contribute to the high signal intensity.

2976
In the MR image depicted in Fig. 5b, several anatomical details of the fir cone (MB.Pb. 2006/512) are visible: The cone axis is situated in the centre, surrounded by scales. The embedded seeds on each ovuliferous scale show a high signal intensity. This even allows to discern seed-internal patterns, though it remains unclear whether these 5 reflect part of the former biological structures (like seed integuments or embryos) or diagenetic alterations thereof. The MR images (not shown) obtained from the Jurassic araucarian twig (MB.Pb. 2006/511) exhibited a much lower overall signal intensity than the Pararaucaria cone. One possible explanation for this could be long-term waterlogging, by means of which 10 the original organic matter would already have been decayed to a high degree before the onset of silicification -it is known that the physical appearance of fossil conifer remains is a function of the residence time either in a terrestrial or aquatic setting (Gastaldo, 1991). Experiments on extant leaf litter have demonstrated degradation in floating plant material: The breakdown starts already with the leaching of water-15 soluble substances and is accelerated by higher water temperatures. This phase is followed by invasion of micro-organisms, mainly fungi and bacteria (Ferguson, 1985). The fossil conifer twigs used for this study have most likely undergone both types of degradation while floating in the water column, perhaps for as long as several months. If the different fossilisation history of the cone, on the other hand, led to a complete 20 silicification shortly after abscission from the tree, it would appear plausible that some organic compounds might still be in place and possibly contributing to the MR image contrast.

Origin of the MR signal
The molecules that give rise to these signals can be characterised in multiple ways, 25 e.g. by using a Pulsed Gradient Spin Echo MR experiment (Stejskal and Tanner, 1965) which measures their self-diffusion coefficient, an estimate of their translational mobil-EGU ity. The highest self-diffusion coefficients we found in fossil samples (in some belemnites) reached 10 −11 m 2 s −1 , which is comparable to that of water bound to cell membranes (Volke et al., 1994), and more than two orders of magnitude smaller than the 2×10 −9 m 2 s −1 typically found in free water (Mills, 1973). In most of the fossils we tested, no translational mobility was detectable. With the 5 given equipment and experimental parameters, it can thus be estimated that the selfdiffusion coefficients of the 1 H-containing molecules within our fossil samples is generally below 10 −13 m 2 s −1 . Interestingly, diffusion determines decay in a recent physical model of kerogenesis (Rothman and Forney, 2007).
The transverse relaxation times (T 2 ) in the fossil samples ranged in the order of 10 1 ms, which is three orders below free water but still several orders above those in solid crystals (Mansfield, 1965), indicating that the MR-visible molecules retain a certain rotational mobility despite the absence of translational motion. This would be compatible with the idea of organic contributions to the signal, as the oscillating biomineralisation patterns described by Florek et al. (2004, see slso Sect. 5.1.1) would suggest. 15 To further test whether free water contributes to the MR signal, we froze one of the non-pathological Tendaguru belemnites with a self-diffusion coefficient near 10 −11 m 2 s −1 (Belemnopsis sp.; MB.C. 3700.11) down to -20 • C, which did not bring about any significant signal change indicative of a phase transition, nor did the signal change when the sample was heated up to 70 • C. We thus heated it in an oven for 8 h at 20 200 • C, which did not result in any observable weight loss on a milligram scale, and the MR spectra and images obtained thereafter showed no difference to those obtained before the temperature experiments.
In order to address the issue of the chemical nature of the MR signal in more detail, a piece of this belemnite was powdered and subjected to 1 H and 13 C NMR MAS 25 spectroscopy. The 13 C spectrum (cf. Fig. 6) shows a peak (at about 180 ppm) reflecting C=O or C=S double bonds, while no other signal can be clearly identified, and namely no CH x groups which would hint at organic material. These observations contrast with 13 C MAS spectra previously obtained from fossil pollen or Baltic amber,

EGU
where CH x peaks were readily observable (Hemsley et al., 1995;Lambert et al., 2000) but they fit nicely with the results of organochemical analysis of the belemnite's powder by successive extractions with dichlormethyl/methanol, dichlormethyl/ethanol and ultrasonication -which did not reveal any traces of lipids (P. Albrecht and A. Charrié, personal communication) -and with the finding that total organic contents in (even 5 recent) echinoid calcite was less than 0.2 wt% (Gaffey, 1995). The 1 H spectrum (cf. inset in Fig. 6), on the other hand, is dominated by the water and hydroxyl peak at 4.8 ppm but also shows a small peak between 0 and 1 ppm, which is indicative of cyclopropyl, metal-bound methyl groups or mineral hydroxyl groups (Kalinowski et al., 1984;Gaffey, 1995). So, at least in this belemnite, water or hydroxyl 10 groups (the latter perhaps bound to the mineral matrix) seem to have been the major signal contributor to our 1 H MR images, and the question about the origin of the signal translates into a question about the origin of these water or hydroxyl groups. We are not aware of any studies focusing on this issue specifically but given that temperaturestable water was found to exist in dental enamel (Myers, 1965, using NMR) as well as 15 in recent and fossil shells (Hudson, 1967, using gas chromatography), it was proposed that hydroxyl groups sometimes take the place of oxygen atoms in the crystal grid (Martin and Donnay, 1972), which has since been affirmed by infrared and NMR techniques (Aines and Rossman, 1984;Gaffey, 1988Gaffey, , 1995Rossman, 2006). Analyses of spots with high signal intensity (e.g. "3" in Fig. 4) could provide further insights into the signal 20 origins in other fossils. The strong but narrow peak in the 1 H spectrum (cf. Fig. 6) can thus perhaps best be explained in terms of rigidly bound H 2 O molecules or OH groups being close enough to H 2 O molecules in a more liquid state, so as to allow for proton exchange. This fits well with MR spectroscopic observations in echinoid calcite where liquid-like protons were 25 found to amount to about 60% of the total H 2 O content (Gaffey, 1995). The latter varied around 2-3wt%, both between individuals and species but even more so between ossicles of the same specimen, which would suggest that total H 2 O might be indicative of ossicle growth or structure (Gaffey, 1995). Nonetheless, a preliminary analysis of Introduction

Perspectives for fossil MRI
Perhaps the most obvious limitation for 1 H MRI of fossils is their typically low contents 5 of soft-bound 1 H. A possible strategy to deal with low intrinsic signal intensities is to fill cavities in the specimen of interest with solutions containing sufficient amounts of MR-visible 1 H (Sebes et al., 1991;Doughty and Tomutsa, 1996;Steiger et al., 1997;de Swiet et al., 1998;Steiger, 2001;Clark et al., 2004) but such treatment might impede future chemical analyses, especially of biological macromolecules (Pruvost et al., 10 2007). In this respect, it would be interesting to find out whether samples that experienced such liquid treatment show any peculiarities in their state of preservation if compared to non-treated samples from the same specimen, though the replacement of the liquids by inert gases (cf. Seeley et al., 2004) could eventually alleviate these concerns.

15
Besides 1 H, both MR spectroscopy and MR imaging are in principle possible with all isotopes that possess a net nuclear magnetic moment, i.e. those exhibiting odd numbers of protons or neutrons (e.g. 2 H, 11 B, 13 C, 14 N, 15 N, 17 O, 19 F, 23 Na, 25 Mg, 29 Si, 31 P and 39 K). Since many of these elements play important roles during diagenesis (cf. introduction), such non-1 H constituents might be of special interest for a variety of 20 palaeontological and related studies. The major limiting factor here is sensitivity, whose upper limit in a given static magnetic field depends on the gyromagnetic ratio of the target isotope and on its abundance within the sample, while the practically achievable value is further dependent upon acquisition parameters (for details, see Abragam, 1961;Callaghan, 1991). 13 C MR spectroscopy (as in Fig. 6) has already found multiple 25 applications in archaeological research (Lambert et al., 2000), and 31 P MRI has been applied to fresh bone and teeth (Li, 1991;Wu et al., 1999).
Furthermore, all states of matter can in principle be investigated but since the MR 2980 Introduction EGU signal correlates with spin (and thus mass) density, the MR signal obtainable from solid samples (like most fossils) will normally be above that from gaseous and -due to chemical binding strength -below that from liquid samples of similar chemical composition. From this perspective, it is perhaps surprising that we know of only one report of an MRI examination of a mummy and that imaging was not even attempted after examining the induction decay signal (Notman et al., 1986). It seems reasonable for us to assume that the technological advances over the two decades since then justify a reconsideration of MRI for the study of mummified or otherwise preserved ancient tissue samples, e.g. soft tissue recovered from within dinosaur bones (cf. Schweitzer et al., 2005Schweitzer et al., , 2007Asara et al., 2007), which we expect to yield a stronger MR signal 10 per voxel than the completely mineralised samples shown here.
Mildly frozen specimens like mammoths uncovered from permafrost soil often show remarkable states of soft tissue preservation (Ezra and Cook, 1959;Zimmerman and Tedford, 1976;Cooper, 2006), which sometimes even allows for genome-level genetic analyses (Poinar et al., 2006). Such samples, despite being frozen, still contain con-15 siderable amounts of unfrozen water (Koop, 2004) and thus represent another window of opportunity for MR techniques, as demonstrated in sea ice (Eicken et al., 2000) or permafrost samples (Kleinberg and Griffin, 2005).
Another important aspect to be considered is sample size. First, at a given resolution, imaging a larger sample means acquiring more data points and thus longer scanning. 20 Second, larger samples generally require larger coils (which will reduce the achievable resolution under otherwise identical conditions), and the ultimate coil size within a given imaging setup is given by the inner diameter of the gradient coils (in our case 38 mm, in human MRI scanners typically around 80 mm). Third, although an increase in field strength generally provides for an increase in signal, noise and signal-to-noise ratio, 25 a number of artifacts also get more pronounced then (e.g. susceptibility distortions near interfaces with different magnetic susceptibilities). Fourth, there are a number of further parameters relevant to MR measurements (e.g. field homogeneity, sample and coil temperature, gradient strength, filling factor, pulse sequence), and so the choice of Printer-friendly Version Interactive Discussion EGU the experimental conditions is vital (for details, see Abragam, 1961;Callaghan, 1991;Ernst et al., 1997).
In terms of materials, the only major restriction is that specimens should not exhibit a permanent magnetisation (due to ferro-or paramagnetic constituents) beyond about 10 −7 (i.e. 0.1 ppm) of the static magnetic field, as this would distort the latter and thus 5 interfere with the way spatial or spectral information are encoded. Fulfilment of this criterion will probably show (like the apical lines in Fig. 2c, d) some taxonomically relevant bias due to biomineralisation or diagenetic environment but this bias is unlikely to extend to higher taxonomic units, and so we suggest that the fossil invertebrate, vertebrate and plant taxa described here should not be seen as a limit but rather taken 10 as starting points for a more detailed screening of fossil lineages by MR techniques.
Comparative evolutionary studies -especially between closely related species, of which only one exhibits a particular trait of interest, while the other does not -have proven useful in and now form the core of most if not all biological disciplines. Palaeontological investigations, specifically, could profit from comparative studies including extant species, thereby complementing the existing knowledge with anatomical and dynamic details concerning processes like post mortem tissue decay (Weigelt, 1927) and in vivo cellular metabolic activities like biomineralisation (e.g. Freytet et al., 1996;Levi et al., 1998;Kolo and Claeys, 2005) as well as embryological development (Xiao, 2002) or locomotion (Gatesy et al., 1999). For all these applications, suitable MR techniques 20 are now in place (e.g. Ciobanu et al., 2003;Manz et al., 2003;Müller et al., 2006;Lee et al., 2006Lee et al., , 2007Honda and Hata, 2007). This versatility of MR techniques, along with their non-invasiveness, renders them a very promising tool for such comparative investigations (e.g. Hopkins and Rilling, 2000;Spoor et al., 2000;Glidewell et al., 2002).
Ongoing developments in MR technology (Glover and Mansfield, 2002) and the ever 25 wider availability of high-field imaging facilities suggest that MRI will continue to help bypass and circumvent current methodological limitations and to generate more applications in the geosciences (Carlson, 2006) and neighbouring fields. As an example, consider portable NMR devices (Eidmann et al., 1996;Prado, 2003;Blümich et al., 2982 sation tool when screening rock samples for embedded fossils.

Conclusions
The data presented here demonstrate that MR imaging allows micromorphological details within intact fossils to be studied non-invasively at resolutions down to 50 µm, i.e. comparable to those of CT images and well below the 100 µm that "would be satisfac-10 tory [. . .] for a majority of researchers and for most applications" (Lyons et al., 2000), at least after the Cambrian explosion. The intrinsic MR signal used to acquire the images appears to originate from mobile yet not diffusible water molecules, while the range of specimens suitable for MRI has been extended beyond liquid-filled mouldic or cavernous fossils (Sebes et al., 1991;Steiger, 2001;Clark et al., 2004) and pathological 15 belemnites (Mietchen et al., 2005), so that it now comprises invertebrate, vertebrate and plant specimens obtained from a variety of sites. Moreover, after MRI scanning, all other palaeontological investigations still remain possible -which is not necessarily true in the opposite case. It should be noted that the digital availability of MRI data (as with CT and other modalities; cf. Zollikofer and Ponce de Léon, 2005) renders them 20 ideal for applications like computational morphology (Bookstein, 1996) or rapid prototyping (Zollikofer and Ponce de Léon, 1995) which have a high potential not only in research but also in science education at school or in museums. Taken together, the microscopic resolution currently achieved with MR techniques, their non-invasiveness, the possibility to obtain taxonomically relevant 3-D spatial as well as chemical informa-   process; 2, canal for facial nerve; 3, useful site for future analyses of signal origin; 4, inner auditory passage; 5, groove for tensor tympani; 6, anterior process; 7, transversal septum; 8, fenestra rotunda; 9, cochlea. MR parameters: FOV: 20×20×40 mm 3 , MTX: 64×64×128 pxl 3 , T R : 900 ms, T E : 0.6 ms, N A : 16, T exp : 32 h. The corresponding original image slice series (in steps of 313 µm) is supplied as Movie 4 (http://www.biogeosciences-discuss.net/4/2959/2007/bgd-4-2959-2007-supplement.zip). The image plane runs from ventro-lateral to dorso-medial, starting at the anterior process, passing the groove for tensor tympani, which divides in the movie the anterior (top) from the posterior half (bottom) of the periotic, followed by the appearance of the cochlea (left hand) with its coiled inner auditory canal (labyrinth). Subsequently, the canal for the facial nerve appears below the cochlea in a "slash-like" manner. The movie ends with the sigmoidally shaped tympanic canal (scala tympani) which opens in the fenestra rotunda (bottom).