Chemical staining of particulate organic matter for improved contrast in soil X-ray µCT images

Staining of POM in presence of fine sand and coarse silt

Compared to the unstained control (unstained sample mean peak grey value ± standard deviation: 3686 ± 1557), all staining treatments caused a slight shift of the central grey value peak to the right, towards 3757 ± 1272 and 4421 ± 1646 for the PMA and PbAc treatments, respectively. This central peak represents X-ray attenuation by coarse silt and POM (Fig. 2a). However, no increased X-ray attenuation was observed in coarse silt in the reconstructed CT-sections (Fig. 3). Instead, the grey value distributions of pin-pointed POM particles (Fig. 3) confirmed that staining did cause a clear shift in POM’s grey values, viz. from 1000–4000 to 6000–18,000. This contrast enhancement of POM was more intense for AgNO₃, PbAc and Pb(NO₃)₂ treatments compared to PMA (Fig. 3) and caused the very clear increase in the histogram’s right tail (grey values > 7500), which was absent with PMA treatment (Fig. 2a). Staining with Pb²⁺ or Ag⁺ also resulted in broad grey value peaks of POM that overlapped those of the coarse silt (grey value range: 2500–5000) and fine sand particles (grey value range: 5000–7500), represented by the second and third peak on the bulk soil histograms (Fig. 2a). However, the histogram peak maxima of the AgNO₃ (11,118), PbAc (13,829), Pb(NO₃)₂ (14,255) and PMA (9678) stained POM particles (Fig. 3) clearly exceeded the grey value range of both the coarse silt and fine sand, which suggests that CT grey value based discrimination of POM from these mineral fractions should be feasible. The smaller peak shift in the POM histograms for PMA (Fig. 3) shows that PMA was less efficient in increasing attenuation of POM and this could be ascribed to the lower atomic mass (AM) of Mo (95.9) compared to Pb (207.2). Likewise, the effect of Ag (AM 107.9) was closer to that of Mo. Next to atomic mass, the affinity for binding to OM likely also differs between the contrast agents. While Ag⁺ and Pb²⁺ bind with the organic functional groups of POM via ionic bonds, PMA is known to interact with conjugated unsaturated bonds³⁷.

There was no shift towards higher grey values in the third peak in the soil core histograms, representing fine sand (Fig. 2a). In addition, we did not find traces of Pb, Ag or Mo on any SEM–EDX scanned fine sand or coarse silt particles (Supplementary Fig. S1). Visual inspection of horizontal reconstructed slices revealed a clear discrimination of the stained POM particles vs. the soil mineral phase. SEM–EDX scans confirmed substantial Mo, Pb or Ag EDX-peaks on randomly chosen POM particles, directly confirming the successful selective staining of POM by all agents (Supplementary Fig. S1). Smaller mineral patches on the surface of POM were visible in the SEM-images (Supplementary Fig. S2), but our analysis demonstrates that these were not responsible for Mo, Pb or Ag-staining of the POM. Stained POM could also be discriminated from several other high density particles such as glauconite, since the latter were not only characterized by higher grey values³² but also a different grey value pattern. The SEM–EDX spectra of some dense mineral particles moreover revealed Zr peaks, suggesting these to be ZrO₂ or Zr-silicate. In conclusion, the experiment demonstrates that Pb(NO₃)₂, AgNO₃ and PbAc are able to selectively bind with POM in fine sand + coarse silt mixtures. However the potential to raise POM contrast in such soil mixtures by treatment with PMA is smaller.

Staining of POM in presence of fine silt and clay

PMA did not appear to stain fine silt, with no shifts of peaks in the bulk soil histogram (Fig. 2B), no visible impact on contrast of the soil mineral phase (Supplementary Fig. S3) and again no discernible Mo peaks in the SEM–EDX spectra of mineral particles (Supplementary Fig. S4 and Supplementary Fig. S5). In contrast, AgNO₃, PbAc and Pb(NO₃)₂ appeared rather non-selective for POM (Fig. 2B). Indeed, the staining increased the grey values of histogram peaks corresponding to fine silt (Fig. 2b), which can be seen from shifting fine silt histogram peaks from a grey value of 4303 to 4702–5577. The increase of the fine silt grey values following staining even resulted in a partial (Pb(NO₃)₂ and AgNO₃) or complete (PbAc) overlap with the sand fraction histogram peak. The grey value histogram right tail was also much enlarged by Pb(NO₃)₂, AgNO₃ and PbAc treatment and increasingly overlapped with the grey value range of stained POM (6000–18,000) (Fig. 3). This overlap was clearly an effect of staining of fine silt, as was also apparent in the CT-sections (Supplementary Fig. S3). The shift in histogram grey values following staining in the fine sand + clay mixtures was similar to that in the sand + fine silt mixtures, but with a larger shift in the right tail of the Pb²⁺ and Ag⁺-histograms to grey values between 7500 and > 12,000 (Fig. 2c). However, we did not detect Pb or Ag EDX peaks on pinpointed fine silt or clay particles (Supplementary Fig. S4). But as inspection of CT-sections (Fig. 4) demonstrated that large patches of the mixtures were stained by AgNO₃, PbAc and Pb(NO₃)₂, it is well possible that such discrete stained areas were coincidently not sampled in our ancillary unsystematic SEM-analysis of subsamples. Regardless, overlap in grey value-ranges of Pb²⁺ and Ag⁺ stained POM and stained clay particles clearly impedes proper segmentation of both phases. As was the case for fine silt, PMA treatment did not cause any shift in histogram peaks of mineral particles (Fig. 2c) and there was no visible impact on mineral phase attenuation in the CT-sections (Fig. 4). This also suggests that Mo has a higher potential than Pb²⁺ and Ag⁺ to bind to OM solely and not to mineral surfaces. Indeed, this may be a result of the neutral Mo(VI)O₃ in PMA, while the cations Pb²⁺ and Ag⁺ may adsorb to negatively charged mineral surfaces. However, Chenu and Plante²⁴ did not detect any sorption of both Pb and Ag on pure clay minerals (vermiculite, illite, kaolinite). Despite minerals identified in our clay fraction being kaolinite and smectite, sorption of Pb (as observed by Chenu and Plante²⁴) or Ag onto Al and Fe oxides and hydroxides in the clay fraction could have occurred.

While manual pin-pointing of the PMA stained POM was not impeded, automated POM extraction from the CT volume may still prove to be challenging due to the lack of image contrast between POM and mineral soil particles. Very recent work by Lammel et al.³⁸ applied a machine learning segmentation tool in synchrotron-based soil CT volumes but experienced limited success. Piccoli et al.³⁹ suggested that an operator-based ability for the selection of thresholds may still result in the most accurate segmentation of POM in soil.

Impact on sample structure

Bulk sample histogram evaluation (Fig. 2a) of the fine sand + coarse silt samples demonstrated that the grey value peak of the pore space (unstained sample mean peak grey value ± standard deviation: 1765 ± 609) had slightly shifted following staining with Pb(NO₃)₂ (1930 ± 386), AgNO₃ (1923 ± 392) and PbAc (2204 ± 314), and to a lesser extent with PMA (1801 ± 611). There was also a reduction in the pore space peak height for the Pb(NO₃)₂, AgNO₃ and PbAc stained treatments of 16%, 27% and 42%, respectively. In the PbAc treatment, the pore space peak disappeared nearly completely, whereas this drop was smaller for AgNO₃ and much smaller for Pb(NO₃)₂. In contrast, no such decline in pore space peak was observed in the PMA-stained samples. A decrease in pore volume (Fig. 5) for the Pb(NO₃)₂, AgNO₃ and PbAc treatments matched the trend in reduction of pore peak height. In addition, opposite trends of peak height increases existed for the second, coarse silt peak (+ 6%, 13%, 25%, respectively) and for the third, fine sand peak (+ 18%, 19% and 33%, respectively). Combined, these observations demonstrate that the observed changes in pore space and mineral phase peak areas are the result of the compaction of the samples by the Pb(NO₃)₂, AgNO₃ and PbAc treatments.

Given that peak height reductions were very different for Pb(NO₃)₂ and PbAc, the deterioration of soil structure could not be solely related to the heavy-element applied (both containing Pb²⁺). In addition, the decreased porosity is unlikely to have been predominantly due to the addition of NO₃⁻, as the addition of Pb(NO₃)₂ would have impacted structure to a greater extent than AgNO₃ which contains less (0.5×) NO₃⁻ (solutions were all at 1 M). Deterioration of the sample structure was therefore more likely to be a physical phenomenon. One possible mechanism could be that an increased underpressure, as a result of perfusion of a more viscous agent, could have more severely disrupted the structure of the soil mixtures. However, an increasing degree of structure distortion could not be related to a higher viscosity of the staining solutions (considered at 0.01 or 0.1 M^40,41,42). Hence, the results in this study did not allow identification of the exact origin of this artifact. As a consequence, caution has to be taken when perfusing soil samples with liquid staining agents since deterioration of soil structure would alter spatial location of the POM as well. Thus, these results indicate that soil structure validation is still required as long as the specific cause for deterioration is not identified.

The Pb(NO₃)₂, AgNO₃ and PbAc treatment histograms of both the fine silt (Fig. 2b) and clay (Fig. 2c) mixtures also had a smaller and broader pore space peak. It is likely that these are both derived from increased occurrence of partial volume effects (PVE) from chemical staining. It is thought that compaction following staining resulted in more voxels containing both pore space and mineral particles. With a voxel resolution of 7 µm this increased the number of pore space voxels with an intermediate grey value (2000–4000). The order in grey value increase magnitude for the fine silt and clay samples was similar as for the coarse silt mixtures: PbAc > Pb(NO₃)₂ > AgNO_3. Chemical staining had an increasingly stronger effect on pore space X-ray attenuation for the finer particle size mixtures, probably because of more intensive binding of Ag⁺ and Pb²⁺ on their much larger reactive surfaces.

In this study the combination of the structural degradation and the overlap in grey value ranges of POM and the mineral fractions render these staining agents unsuitable for staining natural soils. However, further development and testing of the staining method may reduce the impact on soil structure to a minimum. PMA treated silt and clay samples were not structurally degraded and no undesirable shifts of mineral particles’ grey value ranges were observed. However, the artificial soil samples are not fully representative of naturally structured soil and further testing could also rule out degradation of natural soil structure by perfusion with PMA solutions. We expect structural integrity of natural soil samples to exceed that of the ‘loose’ soil mixtures tested in this work.

Performance of PMA compared to gaseous OsO₄ staining

Inspection of horizontal CT-sections suggested that OsO₄ (Supplementary Fig. S6a–c) did not increase X-ray attenuation of mineral material. This is also suggested by the absence of a shift to the left of mineral fraction peaks (grey value 2500–7000), following OsO₄ treatment (Fig. 2d).

The grey value distribution obtained via manual pin-pointing of POM particles demonstrated that the grey value interval of OsO₄-stained POM (Fig. 6) corresponded to the right tail of the soil mixture histograms. This effect was larger but still comparable to the POM grey value shift obtained by PMA treatment. This outcome is a likely result of both staining agents targeting unsaturated bonds in e.g. hydrocarbons or proteins, and the higher atomic mass of Os (190.2) compared to Mo (95.9). Because of the significant health risks when using OsO₄, PMA appears to be a suitable alternative. The increase in POM grey values following both PMA and OsO₄ staining showed a sufficient differentiation of POM particles from the mineral fraction, at least for manual pin-pointing. However, single threshold-based segmentation of PMA- or OsO₄-stained POM in the X-ray CT volumes obtained with this lab-scale polychromatic X-ray µCT system did not appear possible. More sophisticated segmentation algorithms are required, as was also very recently suggested by Piccoli et al.³⁹. We propose to develop self-learning algorithms (e.g. incorporate machine learning) that consider the local grey value patterns of POM in combination with morphological characteristics for a more objective and faster segmentation of the stained POM. By using self-learning algorithms, the expert intervention to segment POM particles would be reduced to an absolute minimum and decrease further over time. In addition, technological development will very likely further enhance the X-ray CT resolution, which will strongly improve the morphological characterization of finer grained POM.

Alternatively, the occurrence of a K-edge in the X-ray absorption spectrum of molybdenum could be exploited to discriminate POM in CT images. Rawlins et al.³³ and Peth et al.³¹ have previously used the occurrence of a K-edge in the X-ray absorption spectrum of Os by scanning soil with synchrotron µCT at photon energies immediately below and beyond the K-edge successfully. However, the K-edge of Mo is situated at 20 keV, an energy level at which most X-rays may be attenuated by the soil mineral fraction, thereby probably making a dual energy approach similar to that used for Os challenging for non-synchrotron scanners and probably also for synchrotrons. Very recently, Lammel et al.³⁸ identified gaseous iodide (I₂) as a plausible candidate for selective staining of OM in soil for use with synchrotron scanners. However, they did not fully demonstrate its selectivity for OM versus silt and clay sized mineral particles, nor in X-ray µCT soil volumes obtained with non-synchrotron scanners.

The findings presented here demonstrate that laboratory based X-ray µCT scanners may also enable the segmentation of OsO₄-stained POM from mineral particles, provided that better segmentation tools are developed. This opens up new possibilities for a more widespread application of the OsO₄ staining technique due to much better availability of laboratory based X-ray µCT scanners and throughput time of samples.

Source: Ecology - nature.com