Development of novel 2D and 3D correlative microscopy to characterise the composition and multiscale structure of suspended sediment aggregates

Abstract Suspended cohesive sediments form aggregates or ‘flocs’ and are often closely associated with carbon, nutrients, pathogens and pollutants, which makes understanding their composition, transport and fate highly desirable. Accurate prediction of floc behaviour requires the quantification of 3-dimensional (3D) properties (size, shape and internal structure) that span several scales (i.e. nanometre [nm] to millimetre [mm]-scale). Traditional techniques (optical cameras and electron microscopy [EM]), however, can only provide 2-dimensional (2D) simplifications of 3D floc geometries. Additionally, the existence of a resolution gap between conventional optical microscopy (COM) and transmission EM (TEM) prevents an understanding of how floc nm-scale constituents and internal structure influence mm-scale floc properties. Here, we develop a novel correlative imaging workflow combining 3D X-ray micro-computed tomography (μCT), 3D focused ion beam nanotomography (FIB-nt) and 2D scanning EM (SEM) and TEM (STEM) which allows us to stabilise, visualise and quantify the composition and multi-scale structure of sediment flocs for the first time. This new technique allowed the quantification of 3D floc geometries, the identification of individual floc components (e.g., clays, non-clay minerals and bacteria), and characterisation of particle-particle and structural associations across scales. This novel dataset demonstrates the truly complex structure of natural flocs at multiple scales. The integration of multi scale, state-of-the-art instrumentation/techniques offers the potential to generate fundamental new understanding of floc composition, structure and behaviour.


Floc Capture and Stabilisation 149
Natural sediment collected from the Thames Estuary, SE England. These sediments 150 are typically fully saline, fine grained silty clays with organic content typically < 10% 151 (measured as % loss on ignition, e.g., O'Shea et al. 2018). Sediment was added to 152 an artificial seawater solution (Sigma sea salts 34 g L −1 ) and gently agitated using a 153 magnetic stirrer to induce flocculation. Fragile flocs were sampled following the 154 protocol outlined in Droppo et al. (1996), which involved settling flocs directly into 155 plankton chambers and immobilising flocs in agarose gel. µCT scans of a test 156 sample (FS0, see Fig. S1 of Supplementary Materials) were conducted in order to 157 assess potential artefacts associated with this technique (e.g., particle-particle 158 overlap, Droppo et al. 1996). Immobilised flocs were subsequently prepared for 159 imaging following the block staining protocol outlined in Wheatland et al. (2017). Floc 160 samples were rendered vacuum stable by resin embedding, which included the 161 addition of electron dense stains (e.g., uranyl acetate etc.) to improve the contrast of 162 organic constituents. Following resin embedding, fiducial markers (aluminium wire, c. indicate variability of floc constituents and structure at the sub-voxel scale. This information helps guide the selection of a suitable site for a cross-section within the floc which is exposed via ultramicrotomy. The precise location of the cross-section within the floc is then verified by re-scanning the sample using µCT and registering the two corresponding µCT datasets using the aluminium registration pin (d). 2D SEM-BSE image montages of the floc cross-section, obtained to identify suitable RoI for further analysis, are then registered to the µCT data (e). Following 2D SEM-BSE imaging, RoI are prepared for 3D FIB-nt (h). 2D SEM-BSE and STEM imagery and 3D FIB-nt data obtained from RoI can be registered to the image montage based on 'internal' fiducial markers (e.g., silt grains, cyanobacteria etc.) that can be identified in the corresponding datasets (g and h).

Description of the Correlative Workflow 170
The correlative workflow developed for investigating floc composition and multiscale 171 structure is shown in Fig. 1. Low-resolution µCT scans (3D pixel or 'voxel' size, c. 10 172 µm 3 ) were initially conducted to characterise floc size and morphology and identify 173 RoI for further analysis ( Fig. 1a and b). At this resolution individual floc constituents 174 <100 µm (e.g., bacteria and clay minerals) cannot be resolved. However, variations 175 in X-ray attenuation ( Fig. 1c and Fig. S2 of Supplementary Materials) indicate the 176 variability of floc constituents and structure at the sub-voxel scale. Subsequently, 177 selected RoIs were exposed by trimming the resin-block using an ultramicrotome 178 (Leica UCT ultramicrotome), creating a smooth cross-section suitable for 2D SEM 179 and 3D FIB-nt. During this process, ultrathin-sections (thickness, 70−100 nm) cut 180 directly adjacent to the cross-section were retained for STEM (Fig. 1). 181 The accurate co-registration of µm and nm-scale EM datasets with mm-scale 182 µCT scans relied on the location and characterisation of the cross-section created 183 within the floc. Therefore, samples were re-scanned using µCT following 184 ultramicrotomy to locate the floc cross-section within the original µCT data (Fig. 1d). imagery based on identification of 'internal' landmarks within the floc ( Fig. 1g and h). 193 Landmarks were selected that could be identified across scales and imaging modalities, e.g., silt grains and cyanobacteria (see Table 1

Co-Registration of Datasets 302
The process of aligning multiscale datasets (i.e. co-registration) is a critical aspect of 303 the correlative workflow, allowing information obtained using different imaging 304 modalities and different spatial scales to be directly related. Co-registration of the 305 multiscale 2D/3D datasets was achieved in the visualisation software Avizo (FEI 306 Visualisation Sciences Group, Berlin, Germany). The success of registration is 307 dependent upon the identification of fiducial markers in the different datasets. Co-308 registration of µCT datasets relied upon the identification of the aluminium wire 309 implanted within the resin block, whilst internal landmarks (e.g., silt particles, bacteria 310 etc.) were used in the co-registration of higher resolution 2D and 3D datasets. Fig. 1  311 shows the sequence of steps taken to co-register the correlative datasets. Coarse 312 alignment was manually conducted using the Transform Editor tool within Avizo, 313 while fine registration was conducted automatically using the Landmark Surface 314 Warp module applied using a rigid transformation algorithm.  Table 2). 331 Quantitative analysis of the floc samples showed FS2 to have the largest 332 volume, with a total occupied volume (i.e. voxel count) of 5.04 × 10 8 µm (Table 2). 333 Descriptions of floc diameter (D) and height to width ratios (H/W) were made using 334 the Feret diameter, i.e. the distance between two parallel planes enclosing an object. 335     Fig. S2 and Fig. S4 respectively). Similar to µCT, 356 greyscale contrast (8-bit pixel depth, e.g., 256 greyscales) was sufficient to allow 357 flocculated material to be segmented from surrounding resin. However, the higher 358 resolution of SEM also enabled the recognition of additional floc components, which 359 could be classified based on particle size, shape and greyscale value and further 360 validated by comparison with SEM-EDS elemental maps. Four additional materials 361 were identified; i) resin filled pore-space, ii) floc matrix (e.g., clays, unicellular 362 bacteria, organo-mineral debris), iii) individual non-clay mineral grains (e.g., quartz, 363 feldspar and mica) and iv) large bio-organic and organic structures (e.g., organic 364 detritus, diatoms, cyanobacteria) (Table 1 and Fig. 3d). Particles <10 µm (e.g., EPS, 365 RoIs selected for FS1 and FS2 respectively. RoIs were targeted to either 388 characterise further floc nm composition and particle-particle interactions via STEM, 389 or selected for SEM imaging to define submicrometre structure. Selected SEM-BSE 390 (resolution, c. 25 -30 nm 2 ) and STEM (resolution, c. 5 -10 µm 2 ) imagery and 391 corresponding SEM-EDS elemental maps are shown in Fig. 4 and Fig. S5 of 392 Supplementary Materials. 393 The four main floc constituents (pore-space, floc matrix, non-clay mineral 394 grains, and large bioorganic and organic structures) were also identified in SEM-BSE 395 and STEM imagery. However, the higher resolution enabled further distinction 396 between materials within the floc matrix: i) clay minerals, ii) microbial cells, iii) 397 organo-mineral debris, and iv) EPS (Table 1) were highly porous (Fig. 4a). STEM showed the nanometre pore space between 419 primary particles filled with exopolymeric material, whilst EPS was notably absent in 420 the larger micrometre pore channels (Fig. S6 of Supplementary Materials). In 421 comparison, high density areas consisted primarily of closely packed clay minerals 422 dispersed with pyrite (Fe+S) (Fig. 4c), and had a lower porosity and high organic 423 signal (Fig. 4d).  (Table 1 and Fig. 6 and 7). Additionally, the enhanced spatial 438 resolution of FS2-B enabled the segmentation of closely packed particles c. <2 µm 439 (e.g., clays within the floc matrix), enabling their reconstruction in 3D (Fig. 7e). This 440 is demonstrated in Fig. 7d in which individual clays can be discriminated in 2D slices 441 from the FIB-nt dataset. Several of the particle-particle associations identified within 442 SEM and STEM data (see Section 3.1.3 and Fig. S6 of Supplementary Materials) were also identified in FS2-B. Visualised in 3D these structures are revealed to be 444 discrete units, separable from surrounding floc matrix by nanopores (Fig. 7e and f). 445   .g., Fig. 6i). These structures usually 449 consist of several particle-particle associations and larger primary particles (e.g., silt 450 grains, organic detritus), loosely arranged in an open, card-house structure, linked 451 together by filamentous cyanobacteria. 452 Quantification reveals the occupied volume of FS2-A largely consists of 453 inorganic material, with clays accounting for c. 98% of occupied space and non-clay 454 minerals c. 0.5% (Table 3). In contrast, a larger proportion of FS2-B is occupied by 455 organics, which accounted for c. 34% of the occupied volume compared to inorganic 456 material, c. 66%. Within both FIB-nt datasets micrometre pore channels can be 457 identified together with elongated nanopores throughout the floc matrix ( Fig. 6 and  458   7). Combined, these give a total porosity of c. 95 % and c. 52 % for FS2-A and FS2-459 B respectively. The resolution and 3D nature of the datasets enabled the 460 classification of microbial cells based on morphotype (Dazzo & Niccum 2015) and 461 five categories were recognised: i) cocci, ii) regular straight rods (e.g., bacilli), iii) 462 curved/U-rods (e.g., vibrio), iv) spirals (e.g., spirilla) and v) unbranched filaments 463 (e.g., cyanobacteria) (Fig. 6d -h). Cocci were characterised as near spherical 464 (length/width, <2:1) with diameters <1.5 μm (Fig. 6h), frequently forming groups of 465 several cells. In comparison, bacilli exhibited a straight, rod-like morphology 466 (length/width, <16:1) and were larger (diameter, c. 2 μm). Cells with a crescent 467 curvature (comma-shaped) were identified as vibrio, and had similar dimensions to 468 bacilli (Fig. 5d -h). Although observed less frequently, spirilla were classified as 469 elongated cells displaying a distinctive repeated waveform (e.g., corkscrew-shaped). 470

Merits of the Correlative Workflow 527
This imaging workflow enables for the first time floc composition and 3D structure to 528 be investigated at all relevant spatial scales, from primary particles to entire flocs 529 several mm in size. This represents a significant advance in our ability to 530 characterise flocs, filling the resolution gap between traditional imaging techniques 531 (e.g., TEM, CLSM and COM) (Fig. 8). 532 The success of the workflow critically depends upon the quality (i.e. resolution, 533 signal-to-noise ratio) and degree of similarity (i.e. resolution and mechanisms for 534  in the corresponding µCT datasets. Within both pairs of µCT data the wire was easily 549 segmented from other material phases based on its high greyscales (e.g., see 550 Section 3.1.1 and Fig. 2). However, discrepancies in the size of the segmented 551 aluminium wire were observed -1.6% for FS1 and 3.8% for FS2 -which likely 552 resulted from scan artefacts, i.e. secondary edge effects due to partial volume effect. 553 Assuming an even distribution of the extraneous voxels around the surface area of 554 the aluminium fiducial marker, the minimum offset between co-registered datasets 555 can be estimated to be of the order of less than a voxel (c. 3 -6 μm) over a total 3D 556 size of 10 × 10 8 μm. With evidence of only minor peripheral misalignment of the 557 registered datasets, the co-registration of the 3D µCT scans has been successful. 558 2D SEM-BSE image montages of the floc cross-sections were critical for the 559 co-registration of 2D and 3D nm and µm datasets with the sub mm-scale 3D µCT 560 data. The trapezoidal shape of the cross-sections ( Fig. 3b and Fig. S3 of 561 Supplementary Materials) can be defined in the µCT datasets, enabling the SEM-562 BSE image montages to be tied to the surface within an accuracy of 3 -6 voxels (c. 563 30 -60 μm). However, further confidence in the accuracy of the co-registration can 564 be obtained by comparing the actual shape of the floc boundary depicted in the two 565 datasets. Reducing the pixel resolution of the SEM-BSE image montages (i.e. down-566 sampling) to match that of the µCT datasets (c. 10 μm) enables a direct comparison 567 between the SEM-BSE image montages and µCT data (Fig. 9), which indicates the 568 error to be less than a voxel (c. <10 μm). In addition, the features responsible for the 569 imaging (montaging and imaging of RoI) and FIB-nt allowed reference landmarks 577 within the corresponding datasets to be recognised with a high degree of certainty 578 (20 -60 nm). As the contrast mechanisms in both SEM-BSE and dark-field STEM 579 are similar (related to atomic number) fiducial markers internal to the floc (e.g., silt 580 grains and bacteria etc.) could be easily identified. However, inspection of the 581 overlaid STEM images following co-registration with SEM imagery revealed 582 discrepancies in the positions of these markers. These displacements are likely the 583 result of ultramicrotomy, as shear stresses imposed during sectioning are known to 584 cause thin-section compression (Peachey 1958). 585 586

Applications of 3D Floc Structural and Compositional Data 587
Providing such detailed 3D analysis of flocs is not readily applicable for field scale 588 quantification of suspended sediment aggregates. However, this technique has the 589 potential, through targeted experimental or field campaigns, to provide new 590 understanding of floc composition and controls on floc characteristics and structures. 591 For example, these datasets quantify 3D floc characteristics (e.g., size, shape and 592 porosity) that are critical input parameters to cohesive sediment transport models. 593 Additionally, the datasets demonstrate the complex structural associations and 594 shows a magnified subset from (a) that isolates a single cyanobacteria to 822 demonstrates how differential staining of subcellular structures has taken place. 823