Characterizing the structure of aerobic granular sludge using ultra-high field magnetic resonance

Despite aerobic granular sludge wastewater treatment plants operating around the world, our understanding of internal granule structure and its relation to treatment efficiency remains limited. This can be attributed in part to the drawbacks of time-consuming, labor-intensive, and invasive microscopy protocols which effectively restrict samples sizes and may introduce artefacts. Timedomain nuclear magnetic resonance (NMR) allows non-invasive measurements which describe internal structural features of opaque, complex materials like biofilms. NMR was used to image aerobic granules collected from five full-scale wastewater treatment plants in the Netherlands and United States, as well as laboratory granules and control beads. T1 and T2 relaxation-weighted images reveal heterogeneous structures that include highand low-density biofilm regions, waterlike voids, and solid-like inclusions. Channels larger than approximately 50 μm and connected to the bulk fluid were not visible. Both cluster and ring-like structures were observed with each granule source having a characteristic structural type. These structures, and their NMR relaxation behavior, were stable over several months of storage. These observations reveal the complex structures within aerobic granules from a range of sources and highlight the need for non-invasive characterization methods like NMR to be applied in the ongoing effort to correlate structure and function.


INTRODUCTION
Compared to conventional activated sludge systems, aerobic granular sludge (AGS) offers numerous benefits for wastewater treatment including compact design, lower energy costs, and excellent biomass retention (de Bruin et al. ). In the last two decades, research into the formation, structure, and metabolism of granular sludge has flourished Typical methods of analysis include use of fixatives, fluorescent in situ hybridization (FISH) probes, or staining with fluorophores followed by cryo-sectioning and imaging with confocal microscopy to identify the spatial arrangement of cells and biopolymers like proteins and polysaccharides within the granules. These types of studies have revealed microbial communities occupying niche environments within the granule architecture as well as the presence of voids and channels in the granule interior. This study, for the first time, applied ultra-high field nuclear magnetic resonance (NMR) to explore the structural heterogeneity of aerobic granular sludge from full-scale municipal wastewater treatment reactors, as well as from laboratoryscale reactors.
Time-domain NMR can provide spatially resolved data on the structure and water diffusivity present in intact granules under various hydrodynamic conditions (Van As & Lens ; Codd et al. ). Early NMR experiments on anaerobic granules provided evidence of a cluster structure (Gonzalez-Gil et al. ), while NMR imaging of methanogenic granules showed that the matrix of extracellular polymeric substances (EPS) is organized in concentric layers and the granules appear to have hollow centers (Lens et al. ). While the achievable spatial resolution is less than that typically obtained with optical methods, magnetic resonance is non-destructive and non-invasive. It can be used to image opaque and heterogeneous samples by exploiting the contrasts between the NMR signal intensity and decay rates associated with various granule components. In this research paper, the additional use of state-of-the-art ultra-high magnetic field strength (22.3 Tesla (T) or 950 MHz where MHz ¼ Tγ and γ is 2.675 × 10 8 rad/(s*T) for protons) allows for enhanced contrast between various biochemical macromolecules and fast, high-resolution imaging. Moreover, the sensitivity of NMR to molecular motion provides an opportunity to observe the inherent transport characteristics of a system in situ in relation to structure (Van As & Lens ). These capabilities represent significant advantages compared to conventional methods like microscopy and micro-electrode studies which are necessarily invasive or destructive. Thus, because NMR is sensitive to a different array of physical and chemical parameters than traditional microscopy methods and can be used to screen larger sample sizes non-invasively, NMR imaging has the potential to highlight structural features that might otherwise not be visible, providing complementary data to inspire new research questions.
The primary research questions addressed here are related to the internal granule structure. To what extent does the internal structure of aerobic granules used in practice for wastewater treatment conform to a layered (Lens et al. ) or cluster (Gonzalez-Gil et al. ) conceptual model? How consistent are the observed structures between samples from the same treatment plant, from different treatment plants, over time as the granules age, and between laboratory granules and those treating municipal wastewater? NMR imaging experiments and T 1 -T 2 relaxation correlation experiments (see Methods section) give insight to these structural questions. Complementary NMR studies have begun to explore how diffusive transport varies within a granule with respect to structural features.

Sample collection and preparation
Samples of aerobic granular sludge were collected during the aeration phase of the treatment cycle from the following sequencing batch reactors (SBRs): full-scale Garmerwolde (The Netherlands), full-scale Vroomshoop (The Netherlands), prototype Utrecht (PNU) (The Netherlands), and demonstration-scale Rockford (IL, USA). In addition, granules were sampled from the fullscale Utrecht wastewater treatment plant during the start-up phase, after seeding the reactor with Garmerwolde granules. Table 1 provides data regarding average influent characteristics for the treatment plants sampled. These granules will be collectively called 'full-scale' granules hereafter, unless otherwise noted.
Two types of granules were compared with the full-scale granules. Control beads were prepared by extracting structural EPS components from aerobic granules with acid and re-gelling them with CaCl 2 solution (Felz et al. ). The control EPS beads originate from AGS, but do not contain cells or the full suite of extracellular polymers present in full-scale granules. Second, saline laboratory granules, grown in synthetic wastewater with an added 5 g/L NaCl, were collected from a 3 L SBR at TU Delft (de Graaff et al. ). The saline granules contain microbes that would be expected to exist within the AGS samples from municipal wastewater reactors but were not exposed to complex substrates.
All samples, which ranged in volume from several hundred millilitres to several litres, were stored in airtight plastic containers in the refrigerator at 4 C without substrate addition until the NMR measurements were performed. For the NMR imaging experiments, several granules (∼5-10) were added to water in a 5 mm NMR sample tube. For the multidimensional correlation measurements, numerous granules (>> 10) were added to the 5 mm NMR sample tube without excess water to maximize the signal obtained from the granules relative to bulk water signal. Granules selected for NMR measurements were typically 2-4 mm in diameter to provide sufficient image pixels within the granule relative to the imaging field of view.

NMR measurements
1 H NMR is sensitive to hydrogen protons. In most biological systems, the dominant source of the NMR signal is water, although protons found in organic matter and extracellular polymers also contribute to the total signal. NMR image contrast depends on signal intensity and the signal relaxation rates of sample constituents (Van As & Lens ). Signal intensity is a function of the position r dependent proton density within a sample and applied magnetic field strength. The T 1 (or longitudinal) relaxation time and the T 2 (or transverse) relaxation time provide information on the physico-chemical environments in which different water populations exist (Callaghan ). T 1 relaxation is related to the timescale for the sample's net magnetization to return to thermal equilibrium following an excitation pulse. T 2 relaxation is related to molecular interactions that occur in the local magnetic field during the measurement. T 1 and T 2 relaxation occur on the order of seconds in bulk liquids and tens to hundreds of milliseconds in biofilms. Chemical exchange of protons between water and hydroxyl groups on polymer chains (Hills ) as well as the presence of rotationally restricted water in cells and the EPS matrix enhances T 2 relaxation in biofilms relative to bulk water. Controlling the timing of the radio-frequency (rf) pulses which make up magnetic resonance measurements can accentuate contrast between the various water populations without addition of chemical tracers. Further discussion of NMR theory is provided in Gjersing et al. () and Callaghan ().
T 1 relaxation-weighted images show signal intensity at a single time, while T 2,eff maps, which are derived from a series of T 2 relaxation-weighted images, show rates of signal relaxation over a period of time. T 1 relaxationweighted images will look different for the same granule depending on when the image was acquired during the measurement and are considered qualitative images. T 2,eff maps capture the entire signal decay quantitatively and are therefore more representative of the structure than any single T 1 relaxation-weighted image. T 1 relaxation-and T 2 relaxation-weighted (T 2,eff ) images were collected on an NMR spectrometer operating at 22.3 T (950 MHz) using an Avance III HD console, with a Micro5 imaging probe, maximum gradients 3 T/m, and a 5 mm birdcage coil (Bruker Biospin, Ettlingen, Germany) at the uNMR-nl national facility (Utrecht, The Netherlands) using parameters found in Table 2 for AGS samples from PNU and Garmerwolde, as well as the EPS and saline control granules. The matrix size was 128 × 128 and receiver bandwidth was 50 kHz. Note that parameters used for NMR images of Utrecht start-up granules, acquired at a later time on the 22.3 T system and shown in Figure 3, are different and are given in the caption to Figure 3. The NMR images of PNU granules were collected on both fresh granules (several days after sampling) and on aged samples (approximately 2 months after sampling). Granules from the Vroomshoop and Rockford treatment plants were not available at the time of image acquisition on the 22.3 T system. Measurement parameters for images of Vroomshoop, Garmerwolde, and Rockford AGS samples made on a Bruker Avance III system with a 5.9 T (250 MHz for protons) superconducting magnet are found in Table 3.
Multidimensional NMR experiments combine two pulse sequences such that the induced signal is encoded in multiple independent dimensions to explore how two parameters are related. In the T 1 -T 2 relaxation correlation experiment, T 2 relaxation is measured for protons experiencing a given T 1 relaxation rate over a range of T 1 relaxation times. The measurement can be used to separate signal which overlaps in one domain or, as in this case, can provide insight into the mechanism behind enhanced signal relaxation. T 1 -T 2 relaxation correlation experiments were performed on a 5.9 T superconducting magnet (250 MHz for protons) networked to a Bruker Avance III spectrometer and using a high rf power probe, a 5 mm rf coil, and a 5 mm sample tube. The correlation measurements were performed using an inversion recovery-CPMG sequence at 20 C with a dwell time of 10 μs, an echo time, t E , of 100 μs, 30,000 echoes, 32 inversion times between 1 ms and 50 s, and 16 averages.

Other microscopy
Aged granules from Garmerwolde were also imaged using transmission electron microscopy (TEM). Granules were fixed in 3% glutaraldehyde for 24 hours, cut into quarters, then returned to the glutaraldehyde solution. The granule quarters were subsequently treated according to Brumfield et al. (). Images were made using an LEO 912AB TEM (Zeiss) operated at 100 KV accelerating voltage and a 2,048 × 2,048 Proscan CCD camera.

Data analysis T 2,eff maps
Multi-slice multi-echo (MSME) imaging produces twodimensional (2D) T 2 relaxation-weighted images showing the signal echo amplitude per voxel, or volume element, in each sample slice for a given t E. An image is collected at times equal to n*t E , where n is the echo number, such that the stack of images for each slice shows the attenuation of signal in each voxel with time. Fitting the echo attenuation in each voxel as an exponential decay produces a 2D relaxation map of the sample where the pixel intensity corresponds to the effective relaxation rate, R 2,eff , or the effective relaxation time, T 2,eff . Since signal from protons experiencing restricted rotational mobility or chemical  . T 2,eff relaxation times are significantly shorter than T 2 relaxation times measured without spatial resolution due, in part, to the influence of water diffusion across magnetic field gradients applied for imaging. Higher resolution images produce shorter T 2,eff relaxation times. In addition, higher magnetic field strength and longer t E also enhance relaxation (Edzes et al. ). Thus, T 2,eff values are only comparable between images when the same magnetic field strength and measurement parameters are used across samples. T 2,eff map data were analysed using Prospa ((v3.13) Magritek Ltd, Wellington, NZ).

RESULTS AND DISCUSSION
The goal of this study was to identify internal structural features within aerobic granules from full-scale municipal wastewater treatment plants and determine how consistent those features are across granule sources and over time. T 1 and T 2 relaxation-weighted images of granules from wastewater treatment plants reveal heterogeneous internal structures where water molecules and biopolymers experience different degrees of rotational mobility and variable opportunities for chemical exchange as reflected in the T 1 and T 2 relaxation behavior. The NMR images presented here show an axial 'image slice' through a 5 mm NMR sample tube with water surrounding the intact granule. The NMR parameters, as defined in Table 2, are identical in all T 1 -weighted images presented in Figure 1.
T 1 relaxation-weighted images EPS beads (Figure 1(a)) and saline laboratory granules (Figure 1(b)) showed significantly different degrees of heterogeneity compared to the aerobic granules treating municipal wastewater (Figure 1(c) and 1(d)). The EPS beads were uniform and showed no discernible internal structure. This is attributable, in part, to the extraction method, which is selective for specific biopolymers. Saline granules showed a radial distribution of signal intensity with faster relaxation and more apparently dense regions on the outer surface and higher intensity, less dense regions in the center of the granules. Images of granules from the PNU reactor and Garmerwolde treatment plant reflected spatially heterogeneous NMR signal intensity and relaxation behavior, though the AGS samples from PNU appeared to have a higher relative volume of brighter regions with slower T 2 relaxation than the Garmerwolde granules. It was significantly more difficult to collect high quality images from the Garmerwolde granules at 22.3 T due to their rapid signal decay at the ultra-high magnetic field, suggesting a denser structure overall in the EPS matrix or greater abundance of relaxation-enhancing components like paramagnetic ions (e.g. Fe(III)) and biopolymers with a high proportion of exchangeable protons. Granules from Vroomshoop and Rockford were not imaged at 22.3 T, though imaging at lower magnetic field strength (5.9 T, see Figure 5) suggests that their internal structural properties are more similar to Garmerwolde granules than to PNU granules ( Figure 2). Figure 2 shows an assortment of fresh granules from the PNU reactor with characteristic dark, dense regions and brighter, less dense regions. Like Figure 1, these images are T 1 -weighted with a short repetition time (T r ) of 550 ms to enhance signal from the gel-like regions of the granule with shorter T 1 times. They are also T 2 -weighted since these images were collected with a t E of 15.9 ms (third echo image) rather than the 5.3 ms (first echo image) shown in Figure 1. The granule pictured in the upper left corner is the same PNU granule shown in Figure 1(c). The combination of T 1 -and T 2 -weighting provides excellent contrast between regions with differing relaxation behavior. While these images show some variation between the individual granules with respect to size, shape, and spatial arrangement of more and less dense biopolymer and cell mass regions, the general form of internal structures (i.e. clusters) in the granules show similarity indicating consistency in NMR-visible properties among granules from the same reactor. Consistency in structural forms was also observed in granules imaged at 5.9 T. These structural similarities among granules from the same reactor (Figure 2), in conjunction with larger variation across different reactors (Figure 1), raise questions about the operational conditions under which these structures form and the range of function they support.
Research on NMR relaxation in the presence of proteins and polysaccharides shows that the bound water mechanism for T 2 relaxation enhancement is small compared to the influence of chemical exchange of protons between water molecules and functional groups in biopolymer solutions (Hills ). Macromolecules like proteins and polysaccharides provide more abundant exchangeable protons than cell membranes, making T 2 relaxation more sensitive to EPS density than to cell concentration (Beuling et al. ). Thus, lighter regions are interpreted as lowerdensity EPS regions while darker areas suggest more concentrated or more tightly cross-linked EPS, with the caveat that EPS components are complex and various constituents may provide more or fewer exchangeable proton sites. The EPS comprising the control bead has not been analysed to characterize the number of exchangeable protons but comparison of structures published for the extracted EPS (Lin et al. ) and Granulan (Seviour et al. ), another gelforming component of structural EPS, suggests that there may be fewer exchangeable protons on the extracted EPS in the control bead than on other biopolymers known to exist in AGS. Moreover, since EPS is produced by active microbes, it can be surmised that the darker regions also indicate cell clusters. Lemaire and colleagues reported evidence suggesting that different bacterial populations produce different types of EPS and are associated with a range of cell densities (Lemaire et al. ). Black regions in the images indicate a lack of signal due to solid-like inclusions, gas bubbles, or signal relaxation, T 2,eff , on the timescale of the measurement (i.e. faster than 5.3 ms).
Images of granules from the new full-scale reactor at the Utrecht wastewater treatment facility sampled during the reactor start-up phase show what appears to be a core with fast T 2 relaxation typical of Garmerwolde structure overgrown by biomass (i.e. cells and EPS matrix) with relaxation characteristics and internal structures typical of PNU granules (Figure 3). Harvested Garmerwolde granules were used as seed sludge to inoculate the system at startup while the influent wastewater comes from the same municipal source as the PNU reactor. Verawaty and coworkers observed that flocs labelled with fluorescent microbeads aggregated around crushed seed granules labelled with contrasting fluorescent microbeads in both batch and SBR systems. (Verawaty et al. ). After 80 days of SBR operation, Verawaty's experimental granules reflected a distribution where granule-labelled beads were located near the core and floc-labelled beads were located near the surface of the aggregates. Figure 3 appears to support the finding that seed granules can act as a template for new growth and suggests that the seed granules may retain their structural form while new biomass develops under the influence of current operating conditions. The NMR images of these granules therefore confirm that influent wastewater characteristics and reactor operations govern the formation of the internal structures observed within AGS. However, the controlling parameter(s) cannot be identified with these NMR images.
T 2,eff maps T 2,eff maps of a control EPS bead and two granules from the PNU reactor are shown in Figure 4. One of the PNU granules was from a fresh, 4-day old sample (Figure 4(b)) while the other was 2 months old (Figure 4(c)), with both stored at 4 C. Both PNU granules show the same heterogeneous internal structure observed in Figures 1 and 2. The T 2,eff maps also show that neither the structure nor the T 2,eff relaxation behavior changes significantly over the timescale of approximately 2 months. In both PNU T 2, eff maps, the bulk water T 2,eff relaxation time is 18 ms, which is approximately equal to the maximum internal T 2,eff time in the granule voids. In the dense regions of the granules, the minimum T 2,eff times are approximately 6 ms with the apparent transition between voids and dense EPS occurring around a T 2,eff time of 12 ms.
T 2,eff maps of EPS beads collected with the same measurement parameters (Figure 4(a)) show low contrast between the effective relaxation times of bulk water and the EPS bead. The T 2,eff maps produced the same bulk water relaxation time as the PNU maps, as expected, and a relatively uniform effective relaxation time within the bead of approximately 14.7 ms. This T 2,eff is within the range of relaxation times found in the apparent voids in the PNU granules. The longer T 2,eff relaxation times observed in the EPS beads, compared to the dense regions of the full-scale granules, may be due to the absence of other exopolymers that enhance relaxation. Because of the sensitivity to biopolymer structure with respect to exchangeable protons, time-domain NMR may be useful in identifying differences in EPS components that are not apparent in confocal micrographs of granules stained with fluorescein isothiocyanate for proteins, concanavalin A for α-polysaccharides, or calcofluor white for β-polysaccharides. Within the field of NMR spectroscopy, characterization of protein structure is well established (Banci et al. ), while characterization of polysaccharides remains a challenge. Thus, time-domain NMR imaging alone is not sufficient to identify and elucidate details of EPS content and structure but may be a useful guide to motivate further investigation with other methods, including NMR spectroscopy.
Several researchers (Stoodley et al. ; Wilking et al. ) have reported the possibility of liquid flow within a biofilm. Channels within the biofilm would permit convection and could potentially increase the penetration of a substrate into the biofilm. Ultra-high field NMR imaging with a spatial resolution of 19.5 × 19.5 × 100 μm 3 revealed the presence of fluid channels in extra-large (1,500-2,500 μm diameter) algal Both of these microscopy methods involve sectioning and dehydration of the granule or granule slice. It can therefore be questioned if the observed channels are an artefact of the sample preparation or if they are truly present in the intact granules. NMR imaging has potential to identify channel morphologies in intact and representative granules, since the method is non-invasive and non-destructive, provided the achievable spatial resolution is greater than the channel diameter.
The T 2,eff relaxation maps of PNU granules in Figure 4(b) and 4(c)) reveal the presence of a large number of voids, distributed throughout the granule and a large cavity in the center of the aged granule (Figure 4(c)). The similarity of T 2,eff times in Figure 4(a) (bulk liquid and EPS) and granule voids in Figure 4(b) and 4(c) suggests that the voids are in fact filled with water and a small fraction of EPS or EPS with few exchangeable proton sites. In these images, the voids do not appear to be connected to the outside bulk liquid but instead are separated by biomass (cells and EPS) with distinctively different relaxation behavior. It should be noted that the spatial resolution of NMR images presented here (47 × 47 × 100 μm) is relatively low compared to traditional microscopy and is lower than that applied in the algal colonies where channels were observed (Schadewijk et al. ). A channel with a diameter smaller than the spatial resolution will be hard to detect with NMR, though ultra-high field magnetic resonance has potential to further improve resolution limits. The channels reported by Gonzalez-Gil & Holliger () are estimated to be roughly 50 μm in diameter, approximately the detection limit in this study. These NMR results suggest that channels with a diameter larger than 50 μm and connected to the bulk fluid are not present in aerobic granules imaged here and that convection may play a limited role in substrate transport in the granule interior.
T 2,eff maps of granules from Vroomshoop and Rockford were made using the 5.9 T (250 MHz) system. Figure 5 shows a representative longitudinal slice through a stack of granules in water in the 5 mm NMR sample tube and highlights the variability observed both within and across granule sources. These granules again show heterogeneous internal structure and appear to be more similar to Garmerwolde granules than to granules from the PNU reactor. In the Rockford granules there appears to be more of a concentric ring structure than is apparent in PNU granules, and the less dense EPS regions appear to be less numerous but larger in volume. There are few locations in both samples with relaxation times equal to the bulk water value. The Vroomshoop granules exhibit more of a cluster structure than concentric rings and most closely resemble Garmerwolde granules of all the granules imaged in this study. Despite the internal structural differences between the granule sources, a similar range of relaxation times is apparent in Vroomshoop and Rockford granules, ranging from less than 10 ms in the dense biomass to more than 40 ms in the less dense EPS regions. The bulk water relaxation in both cases is approximately 50 ms. Since the magnetic field strength, imaging gradient strengths, and spatial resolution are different between Figures 4 and 5, the values of T 2,eff are not directly comparable. The lower spatial resolution and lower magnetic field strength on the 5.9 T system produce a wider range of T 2,eff times within the biofilm, while the ultra-high field 22.3 T system provides high contrast and resolution of chemical exchange between T 1 -T 2 relaxation correlations In addition to imaging experiments, 2D T 1 -T 2 relaxation correlation measurements were collected. These measurements confirm the presence of multiple T 2 populations and a distribution of T 1 populations within the granules. Since these data were collected without spatial resolution, the resulting distributions reflect the range and relative proportion of relaxation times inherent in the sample and are not subject to enhanced relaxation related to imaging.
T 1 -T 2 relaxation correlations were measured for 1-month-old Vroomshoop granules without bulk water using the 250 MHz (5.9 T) system ( Figure 6). The measurements were made using intact granules (left) and the same sample after crushing the granules in the sample tube (right). The broad T 1 -T 2 correlation for the intact granules reflects the presence of spatial heterogeneities that exist across length scales many orders of magnitude larger than the length scale of molecular interactions between water and biopolymers. The distribution approaches the T 1 -T 2 parity line at longer relaxation times, indicating highly mobile water populations in the voids, and diverges from the parity line at shorter relaxation times. The short relaxation times indicate water in the pores of a highly crosslinked EPS matrix where the biopolymer signal may exhibit more solid-like behavior. Because of the spatial heterogeneity of the internal structural features, even highly mobile water in a void would not experience the dense EPS of a cluster over the timescale of the measurement, resulting in the broad T 1 -T 2 correlation observed. After crushing the granules, the T 1 -T 2 correlation shows the loss of the population with larger T 1 and T 2 relaxation times, corresponding to the very diffuse EPS regions in the granules where molecular interactions are limited on the timescale of the measurement. Homogenization of the biofilm minimizes the length scale of heterogeneity, allowing interactions between hydrogen-bearing species previously tube. The two samples contain a similar range of relaxation times and provide an example of the variability of relaxation times within granules from the same reactor and between treatment plants. The maps were produced from MSME images (8 echoes) made on the 5.9 T system. Spatial resolution is 94 × 94 × 500 microns. separated in space. Water molecules previously in the void regions can undergo chemical exchange of protons with EPS after homogenization, leading to a decrease in the maximum T 2 relaxation time. Physically breaking the structure of the granules may also have entangled polymers that were spatially separated, increasing the rotational correlation time of the biopolymers previously in the voids. This would also lead to a decrease in the maximum T 1 relaxation time, as was observed. These measurements provide further confirmation that the structure of the granule is responsible for the distribution of relaxation times observed in the NMR measurements. Thus, the observed relaxation times in the NMR images are a clear indicator of different local physical and chemical environments and point to the importance of chemical exchange of protons in the relaxation response.

Apparent 'boundary layer'
An apparent 'boundary layer' was observed as a thin dark layer on the surface of the municipal AGS samples imaged at 22.3 T but was not observed on the EPS beads. The layer was approximately 1 pixel wide (∼50 μm) in the NMR data. The darkness of the layer in the image could be attributed to a lack of signal (fewer protons), or very fast signal relaxation. To eliminate the possibility that the 'boundary layer' was simply an artefact of the measurement at the ultra-high magnetic field, a granule from PNU was cut in half with a razor blade and imaged in the 22.3 T system (Figure 7(a)). The layer is visible on the undisturbed outer surface of the granule but not on the cut edge, suggesting that the physico-chemical properties of the surface are indeed different from the center of the granule. An NMR Figure 6 | The T1-T2 correlation measured on the 5.9 T system on a stack of 1-month-old Vroomshoop granules (left), and the same sample after homogenization inside the NMR sample tube (right). The loss of the population with the longer T 1 and T 2 times, corresponding to water populations in the granule voids, confirms that the internal granule structures are responsible for the distribution of relaxation times observed.
artefact would be expected to be visible on the cut edge as well.
As part of the effort to identify the cause of the apparent boundary layer observed with NMR, images were also made using TEM to more directly observe the surface layer of aerobic granules from the Garmerwolde treatment plant. Inorganic debris and solid particulates that might be present in municipal wastewater would be visible under TEM. Figure 7(b) shows dense cell clusters at the surface of a granule in the top of the image, with the embedding resin outside of the granule at the bottom. There was no precipitate or solid particulate debris visible on any granule surface examined with TEM. Research is ongoing to determine the source of the apparent boundary layer and to determine if the same mechanism is responsible for the apparent ring structure that can be seen in the T 2,eff relaxation maps of Rockford granules shown in Figure 5 or in the Garmerwolde granule in Figure 1.

CONCLUSIONS
Even though the structure of aerobic granular sludge is directly linked to mass transport and conversion processes in the granules, our understanding of the granule structure and the mechanisms by which it forms is still quite limited. In this paper, state-of-the-art ultra-high field and non-invasive magnetic resonance methods were used to explore the structural characteristics of full-scale and laboratory-scale granules. NMR provides the ability to screen larger samples sizes of intact granules and can, therefore, help to provoke and answer questions related to structure and function, as well as the reliability and representativeness of data collected via more labor intensive and destructive microscopy methods. NMR results indicated that aerobic granules from municipal treatment plants exhibit a heterogeneous structure comprised of variable density EPS and cell clusters, and water-like voids. The structures observed were either similar to the cluster structure observed in some anaerobic granules (Gonzalez-Gil et al. ) or similar to a concentric ring structure with each granule source having a characteristic type. The structures in all granules were stable over a storage period of several months. The NMR data further revealed an apparent boundary layer, which appeared to be of organic origin and cannot be attributed to measurement artefact or accumulation of inorganic material on the granule surface. The granules did not contain channels larger than 50 μm, but rather apparent voids containing water and diffuse EPS or exopolymers with few exchangeable protons. These observations highlight the complex structure of aerobic granules and the need for non-invasive characterization methods such as NMR that can provide unique data on the granule structure.