Human lung virtual histology by multi-scale x-ray phase-contrast computed tomography

Objectives. As the central organ of the respiratory system, the human lung is responsible for supplying oxygen to the blood, which reaches the erythrocytes by diffusion through the alveolar walls and is then distributed throughout the body. By exploiting the difference in electron density detected by a phase shift in soft tissue, high-resolution x-ray phase-contrast computed tomography (XPCT) can resolve biological structures in a sub-μm range, shedding new light on the three-dimensional structure of the lungs, physiological functions and pathological mechanisms. Approach. This work presents both synchrotron and laboratory XPCT results of postmortem tissue from autopsies and biopsies embedded with various preparation protocols such as precision-cut lung slices, cryogenically fixed lung tissue, as well as paraffin and alcohol fixed tissue. The selection of pathological abnormalities includes channel of Lambert, bronchus-associated lymphoid tissue and alveolar capillary dysplasia with misalignment of pulmonary veins. Subsequently, quantification and visualization approaches are presented. Main results. The overall high image quality even of in-house XPCT scans for the case of FFPE biopsies can be exploited for a wide range of pulmonary pathologies and translated to dedicated and optimized instrumentation which could be operated in clinical setting. By using synchrotron radiation, contrast can be further increased to resolve sub-μm sized features down to the sub-cellular level. The results demonstrate that a wide range of preparation protocols including sample mounting in liquids can be used. Significance. With XPCT, poorly understood 3D structures can be identified in larger volume overview and subsequently studied in more detail at higher resolution. With the full 3D structure, the respective physiological functions of airways or vascular networks, and the different pathophysiologic mechanisms can be elucidated or at least underpinned with structural data. Moreover, synchrotron data can be used to validate laboratory protocols and provide ground truth for standardizing the method.


Introduction
The lung is a perfect example of how the function of an organ is enabled by its three-dimensional (3D) structure, here formed by intricate and intertwined networks of ventilation and vasculature. In radiology, the lung is known as a high contrast organ for radiography and computer tomography (CT), and quite a number of lung diseases can readily be diagnosed from radiographs or CT scans, such as infections, emphysema or fibrosis. The 3D structure of the lung cyto-architecture is again key to physiology and pathophysiology, i.e. the cellular and the histological level. Histological diagnostics at much higher resolution than radiology is advantageous for a range of non-neoplastic lung diseases, such as interstitial lung diseases, pulmonary vascular diseases, virus induced pneumonia, fibrosis or other displastic pathologies. To this end, histomorphological assessment of formalin-fixed, paraffin embedded (FFPE) thin tissue sections, stained with haematoxylin and eosin still represents the gold standard. However, the functional cyto-architecture of the lung parenchyma is often insufficiently captured in 2D, and 3D visualisation and quantification of the morphological changes associated with different pathologies and diseases could significantly improve and complete the diagnosis.
To this end, it is by now well appreciated that propagation-based x-ray phase-contrast tomography (XPCT) offers a unique potential to extend histology and pathohistology. It promises a scalable, isotropic resolution without destructive slicing of the specimen and quantitative density-based contrast (Khimchenko et al 2016  Recent work has significantly improved the classic phase retrieval approaches (Paganin and Nugent 1998, Cloetens et al 1999, Alloo et al 2022, for example by nonlinear generalizations and Tikhonov regularization (Lohse et al 2020, Huhn et al 2022. Furthermore, scans with highly coherent synchrotron radiation both in the parallel beam settings with high resolution scintillator-based detector systems, and in magnifying cone-beam settings now routinely reach the micron and sub-micron resolution range, respectively, down to voxel sizes around 100 nm, which can even be implemented in the same instrument (Frohn et al 2020), and at least partly in high throughput (by continuous scans and detector readout).
Lung imaging by XPCT has early on been implemented on macro-and microscopic scales (Parsons et al 2008, Leong et al 2014, Stahr et al 2016, Bayat et al 2020, Morgan et al 2020, Borisova et al 2021, Shaker et al 2021, Bayat et al 2022a, 2022b, O'Connell et al 2022, including even live animal models of respiratory diseases. At the histological level, we have recently used multi-scale XPCT to image the parenchymal architecture of unstained FFPE lung tissue of patients who succumbed to COVID-19 (Eckermann et al 2020). The 3D reconstructions provided visualisations of diffuse alveolar damage (DAD) with its prominent hyaline membrane formation, mapping the spatial distribution and density of lymphocytes infiltrating the tissue, vascular damage and intussusceptive angiogenesis, as well as distance histograms from tissue interior to the closest air compartment.
In this work, we extend virtual 3D histology and pathohistology of lung by XPCT to a wider range of pathologies. We investigate how the mounting and embedding as well as different setups and implementations affect the image quality. We provide benchmark reconstructions and illustrative examples, including both synchrotron radiation (SR) and compact μ-focus sources. The latter is particularly relevant for potential dissemination, as laboratory instrumentation is more accessible for pathology units. A central goal of this work is to support the assertion that integration of 3D morphological information with well established histology techniques helps to unravel the pathophysiology of lung diseases, as previously shown for COVID-19. Furthermore, future extension could possibly also cover advanced clinical diagnosis. As a fist step, we here provide XPCT results for postmortem tissue from autopsies as well as for biopsies in the context of different pathologies. Finally, we include first XPCT results on cultivated lung tissue, namely precision-cut lung slices (PCLS), in this study, as well as CT results of cryogenically fixed lung tissue.
The manuscript is organized as follows: after this introduction, methods of sample preparation and image acquisition both with SR and μ-CT are detailed, as well as the analysis workflow. In the following the results are presented for paraffin-and liquid embedded (channel of Lambert, bronchus-associated lymphoid tissue, alveolar capillary dysplasia with misalignment of pulmonary veins) samples as well as for precision cut lungs slices and cryogenically fixed tissue, respectively. Subsequently, quantification and visualization approaches are presented while the manuscript concludes with a discussion of the results.

Methods
2.1. Sample preparation 2.1.1. Fixation in liquid media To prevent freshly dissected tissue samples from deterioration, uncontrolled drying and deformation, chemical fixation is required after extraction. Formaldehyde (formalin) is commonly used as a highly protein-crosslinking fixative, but dehydration in alcohol can also be used as a non-carcinogenic and less hazardous substitute. Because it does not cross-link molecular groups, it also facilitates immunohistochemical studies (van Essen et al 2010). For liquid embedding in this work, freshly prepared samples, which were several cubic centimeters in size, were placed in 98% alcohol solution (2% Phosphate-buffered saline (PBS)), and wrapped with absorbent cotton to avoid sample movement during image acquisition ( figure 1(v)). An example can be found in the supplemental document.

Paraffin-embedding
Formalin-fixed paraffin-embedded (FFPE) tissue is the most common embedding and preservation procedure in clinical pathology, and has already been successfully used for 3D virtual histology of lung with synchrotron and laboratory radiation (Eckermann et al 2020). For this experiment, tissue fixation, embedding and mounting were performed according to existing protocols (Töpperwien 2018). Further details on the preparation protocol can be found in the supplemental document.

Precision cut lung slices
In this work, we include first results on PCLS by XPCT study, in order to explore if XPCT could potentially support PCLS research by 3D imaging. To this end, different sample preparation protocols and imaging conditions have been tested. More details on the preparation protocol and the selection of samples can be found in the supplemental document.

Cryogenic fixation
Immediately after tissue collection, the lung can be frozen in toto to preserve delicate structures (figure 1(vii)). By applying water-controlled air pressure, the lung is inflated and the pressure lowered to achieve an uniform recruitment of the lung specimen. The lung is then placed in a styrofoam box and frozen in liquid nitrogen vapors. At this point, the lung can be safely stored at −80°C until further use (Verleden et al 2022). While this technique was previously applied in combination with commercially-available in-house micro-CT scanners  (Campbell et al 2012, Vasilescu et al 2017, Maes et al 2022, to our best knowledge this study is the first to scan macroscopic cryo-fixed biological tissue with synchrotron radiation. 2.2. Experimental setup 2.2.1. Laboratory-based μ-focus sources Synchrotron radiation (SR) sources offer highly brilliant beams of high lateral and longitudinal coherence, ideally suited for XPCT. Monochromaticity on the order of Δλ/λ ; 10 −4 is achieved by single crystal monochromators; a wide range of beamline optics typically support collimated beams (parallel beam geometry) or a secondary focus with cone beam illumination of the sample. However, since the availability of SR is restricted to just a few sites and selective allocation of beamtime, translation of XPCT to setups operated at laboratory sources, namely μ-focus x-ray tubes (with spot sizes down to the sub-micron range) is of paramount importance for daily preclinical research and future clinical applications. Since spatial coherence is a prerequisite for phase sensitivity, the lateral coherence length ξ = λx 01 /σ s must at least exceed the effective pixel size. This is possible at high resolution (small pixel size), if the source-sample distance (x 01 ) and the effective wavelength λ representing the spectrum are sufficiently large, and the source size σ s is sufficiently small. This was first achieved by the advent of liquid metal jet anodes (Hemberg et al 2003), which enabled XPCT scans and reconstructions (Bartels et al 2013), and later by compact laser-driven synchrotron sources (inverse Compton sources) (Eggl et al 2015). More recently, more compact μCT instruments even with solid targets and spot sizes around a few microns or even in the sub-micron range (denoted as nanofocus sources by vendors) have reached sufficient partial coherence for XPCT of soft biological tissue (Eckermann et al 2019, Massimi et al 2019. In this work, differently prepared human lung tissue samples were scanned at a commercially μCT system (LS1-setup) as well as a custom built in-house CT instrument (LS2-setup). Details on instrumentation and acquisition parameters can be found in the Supplementary Document.

XPCT with synchrotron radiation
For applications requiring highly brilliant and monochromatic x-ray beams, SR is the best choice. Quantitative phase retrieval is possible based on the much higher coherence and multiple length scales can be covered down to sub-100 nm resolution. In this work we took advantage of a multiscale XPCT approach, using both the parallel (SR1) and cone beam (SR2) configurations of the GINIX endstation installed at the P10 beamline of the PETRA III storage ring (DESY, Hamburg) (Frohn et al 2020). The GINIX endstation is dedicated to propagation-based holographic tomography of biological samples, and operates in the medium to hard x-ray regime (5-14 keV). For high resolution (cone beam) recordings, the beam was focused by a pair of Kirkpatrick-Baez (KB) mirrors and further reduced by a germanium x-ray waveguide, which also increases the spatial coherence of the propagated x-ray wave by mode filtering . Two configurations can be used at the same beamtime and were used in this work to investigate lung tissue at multiple scales. Details on the setup and acquisition parameters can be found in the supplementary document.
For exploiting the multi-scale capabilities of XPCT, biopsy punches were taken from the paraffin embedded samples with diameters varying from 4 mm (overview of the sample) to 1 mm (high resolution scans). To achieve a large FOV, samples with diameters larger than 1 mm are examined with a parallel beam geometry (1.6 × 1.4 mm) or a cone beam geometry at the LS1-setup. Here, the magnification and therefore the FOV depends on the setup geometry. By increasing x 12 or decreasing x 01 (figure 2) a magnification resulting in a smaller effective pixel size can be achieved. Since the difference in attenuation between paraffin and soft tissue is about three magnitudes lower compared to the induced phase shift, paraffin tends to be a very suitable embedding material for XPCT when it comes to biological soft tissue. In the course of this work, different pathologies were investigated to illustrate the capabilities of XPCT for 3D imaging of lung tissue, and to suggest suitable experimental configurations, depending on preparation type and application.

Data reconstruction and post processing
For experiments at the synchrotron (SR) as well as the custom-built laboratory setup (LS2), data acquisition was immediately followed by phase-and tomographic reconstruction.
Phase reconstruction was carried out assuming a homogeneous, weakly absorbing object using the linearized single step CTF-approach (Cloetens et al 1999 or nonlinear Tikhonov (Huhn et al 2022b) at the synchrotron assuming a low Fresnel number (F = a 2 /(x 12 λ), where a is the characteristic size of the object and x 12 the propagation distance) resulting in a holographic regime. In the laboratory, to preserve image sharpness the transport-of-intensity equation based Bronnikov-aided correction (BAC) scheme is used (Witte et al 2009, Töpperwien et al 2016. For the tomographic reconstruction, the projections were flatand dark field corrected with the iradon function implemented in MATLAB using a Ram-Lak filter for backprojection. To increase the signal-to-noise ratio at in-house setups, several projections (# accumulations in supplementary table 1) were taken at the same rotation position and averaged directly after acquisition was completed. Fresnel numbers were verified and if needed refined by inspecting the radially averaged PSF of a hologram to the expected CTF and according rescaling. Additionally, a wavelet-based filtering approach was applied to further mitigate ring artifacts occurring during the reconstruction (Münch et al 2009). All used algorithms are implemented in our code package HoloTomoToolbox (Lohse et al 2020). At the commercial nano-CT device (LS1), reference images (additional images at different angles before and after the scan) were taken to account for source spot movement by autocorrelation with tomographic projections and the rotation axis was corrected. Ring filter correction and phase retrieval algorithm were applied before tomographic reconstruction. The phase retrieval parameter were set manually after eye inspection.
Fiji (Schindelin et al 2012) and MATLAB (R2020a) were used for further analysis and NVIDIA IndeX (NVIDIA Corporation, Santa Clara, CA, USA) (Schneider et al 2021) for rendering of the volumes. Video sequences and further renderings were created using the Avizo software (Thermo Fischer Scientific, Waltham, CA, USA). Figure 3 presents images of a channel of Lambert, an important bypass airway relevant for collateral ventilation. Images were obtained on an unused donor lung with mild emphysema which we examined by a parallel beam geometry as well as with waveguide optics using synchrotron radiation (GINIX, PETRA III, Hamburg). For the former, nine volumes were recorded in a plane with a sufficient overlap to stitch the volumes together (AVIZO, Thermo Fisher Scientific, Berlin) during post-processing. This resulted in an extended FOV of about 3.8 mm and thus allowed to investigate the 3.5 mm punch in its full diameter to observe an interbronchiolar channel (channel of Lambert) which usually has a diameter of around 30 μm and can be found between respiratory and terminal bronchioles of adjacent segments (Martin 1966). This morphological feature enables so-called collateral ventilation which serves to compensate for imbalances between ventilation and perfusion, as they occur in many diseases. It has been speculated that collateral ventilation can increase in response to emphysema, Figure 2. Optical layout of the GINIX instrument. The configuration can be switched between (a) a parallel beam geometry, and (b) cone beam beam geometry. For (a), polished single crystal attenuators are inserted in the unfocused synchrotron beam to avoid detector overexposure. KB-mirrors and the waveguide are moved out of the beam path. A microscope detector system (Optique Peter) equipped with an sCMOS camera (PCO.edge) was used. The detector system was used with a 50 μm Lu:Ag scintillator, and a 10x objective) resulting in px = 0.65 μm. For the configuration shown in (b), the beam is focused by KB mirrors down to about 300 nm × 300 nm. After passing an aperture (APY), the beam is coupled into an x-ray waveguide, aligned in the focal plane of the KB. After exiting the waveguide, the broadened beam with opening angle θ illuminates the sample positioned at distance x 01 behind the waveguide exit, and projection images are recorded in the detection plane at distance x 02 , with a fibre-coupled scintillation detector (Zyla HF 5.5, 15 μm Gadox scintillator, fiber optic plate coupled to sCMOS, px = 0.65 μm). either opening up new flow channels for flow or causing enlargement of existing channels, and thereby reducing their resistance (Cetti et al 2006). Emphysema is defined as an increase beyond the normal in the size of the air spaces distal to the terminal bronchiole and is associated with the destruction of septae (Cetti et al 2006). However, up to now, channels of Lambert were not yet depicted and quantified in its full three-dimensionality. Therefore, they could not be properly compared to other ways of collateral ventilation (figure 3) and their role in gas exchange both in normal and disease states still remains largely unknown (Terry and Traystman 2016). With the advent of XPCT which can now provide the required 3D data this situation has changed.

Bronchus-associated lymphoid tissue
In the next example, we show results for bronchus-associated lymphoid tissue (BALT) after a SARS-CoV-2 infection obtained in-house with the LS1 setup. Figure 4 presents a representative slice through the reconstruction volume, and a rendering of structures of interest. BALT describes a localized, intrapulmonary accumulation of lymphatic cells in the lungs and bronchus that participate in airway immune responses and are located on the airway mucosa. It is frequently found between airways and arteries (Meyerholz et al 2018) and often within the walls of large airways (Sminia et al 1990). While BALT is similar in most mammal species, it differs in maintenance and inducibility, and appears in mice or humans only after infection or inflammation (Tschernig and Pabst 2000). Favouring factors for the presence of BALT in human adults include infection, pathogen exposure, chronic pulmonary inflammation or autoimmune disease (Ak 2021). It can further harbor an overall rare type of extranodal marginal zone lymphoma, mostly present in primary non-Hodgkin lymphoma (NHL). In view of their role in immune response and pathology, investigations of BALT formation in human lung biopsies are highly relevant and could be enhanced by 3D information from XPCT. As we show here, BALT and lymphoid proliferation can be detected using the in-house LS1-setup (figures 4(i), (ii)). By scanning the same pathology using the SR1-setup (figure 4(iii)), we observe that the lymphoid nodes seem to be filled with lymphocytes which potentially entered the lumen by high endothelial venules facilitated by pulmonary inflammation or immunization (Sato et al 2000) caused by a SARS-CoV-2 infection. These structures are usually studied by transmission electron microscopy, at high resolution but only on twodimensional sections. In-house XPCT, however, can produce a high contrast three-dimensional dataset in just a few hours, providing the possibility of a fast segmentation of features of interest (figure 4(iv)), allowing to study the distribution of the lymphoid tissue as well as their shape, orientation and preferred area of accumulation. Figure 5 presents another example of human lung tissue, where we studied alveolar capillary dysplasia with misalignment of pulmonary veins (ACD/MPV), first again using paraffin embedded lung tissue, scanned with a parallel beam at the SR1-setup as well as in liquid media at an in-house setup (LS1-setup). The term ACD/MPV describes a rare, fatal lung developmental disorder, affecting the alveoli and their surrounding capillaries resulting in an impaired gas exchange. According to Bishop et al (2011), it is commonly distinguished by immature lobular development, reduced capillary density, abnormal air-blood barriers and anomalous veins in bronchovascular bundles. Historically, it was suggested that the fetal lung vascularization leads to a significant reduction of pulmonary capillaries forcing the blood entering an individual lung to drain via misaligned veins (Bishop et al 2011). However, in a recent work, Norvik et al were able to show that these suspected veins indeed represent dilated bronchopulmonary shunt vessels (in their designated position). Therefore, the term MPV seems to be a misnomer (Norvik et al 2020). Presumably, the clinical course is strongly influenced by the severity of disruption in capillary density and their lack of contact with the alveolar epithelium (Melly et al 2008). In severe cases, the sum of the pathologies result in respiratory failure with a nearly 100% mortality rate (Deutsch et al 2007). While chest x-rays may already give some indications for the pathology, they can easily be misinterpreted as normal. Biopsies, on the other hand, can be used to diagnose ACD/MPV. Further, in order to gain a more complete understanding of the disease, a three-dimensional imaging technique such as XPCT may add a significant benefit, by shedding light on the underlying histopathological manifestation of the condition.

Alveolar capillary dysplasia with misalignment of pulmonary veins
The samples used in the study were extracted from ethanol-fixed fresh human lung tissue of one ACD case. Bronchi and vasculature could be easily detected and segmented so that their course becomes observable. Notably, the artery and the shunt vessel show parallel, wavelike structures at the bottom of the rendered volume ( figure 5(ii)). The histological slice align almost perfectly with the virtual slice, allowing to study the immunochemical properties at specific regions of interest ( figure 5(iii)).

Precision cut lung slices
Respiratory diseases are among the leading causes of death, closely following stroke and cancer (Ritchie et al 2018). However, in contrast to cardiovascular diseases and tumors, respiratory pathologies suffer from a deficit in translatable animal models. This compromises the development of efficient therapies, and creates to use human cell lines to model the pathology. Despite many recently established techniques such as transwell systems (cell co-culture environment mimicking in vivo conditions) and 'organ-on-a-chip' systems, many drawbacks and problems with translatability still persist. More recently, Precision Cut lung slices (PCLS) were proposed as a potential solution, with slices being cultured from explanted or diseased human lungs, containing all cell types from the tissue of interest including the extracellular matrix . Until now, the PCLS system has already provided insights into various pulmonary diseases such as asthma  and mounted in different media, as described in section 2.1) were scanned at different modalities and in different configurations. The results show that high contrast is obtained even at in-house setups for critically point dried samples, see figures 6(i)-(iii). However, as a downside of this method, as expected the tissue morphology is strongly affected by dehydration, resulting not only in shrinkage but also rupturing of the tissue. Also, due to the very high phase contrast between dried tissue and air edges, the contrast even with laboratory sources, but certainly for synchrotron scans, may even be too strong to perform successful phase retrieval, at least for most of the conventional phase retrieval approaches. For the in-house scans, one can chose a configuration with large propagation distance and magnification with a detector of larger pixel sizes as shown in (ii) for a scan of larger  FOV, here obtained at the LS2 setup. Alternatively, at the same instrument one can chose a detector with very small pixel size, see (i), where the Argos detector with px = 0.54 μm is positioned very close to the sample, which allows us to have much higher resolution, at the cost of much longer acquisition time. Finally, at the commercially available in-house setup LS1 consisting of a solid target source (iii) and a CCD detector with medium sized pixel sizes px = 9 μm, appropriately chosen geometric parameters x 01 and x 12 (see supplementary  table 1). This allowed us to achieve a high contrast in the reconstructed slices while having a resolution comparable to the LS2 PCLS setup with the high resolution camera Argos. Finally, liquid embedded ex vivo samples scanned with synchrotron radiation (SR1) clearly reveal a well resolved and well-preserved alveolar network. It can be concluded that the specimen remained largely undamaged during the challenging sample preparation process of thinly sectioning the tissue with a microtome.
3.5. Imaging cryogenically fixed samples Following mountings in paraffin (FFPE-tissue blocks), liquid embedding, and critical point drying, we now turn to cryogenic fixation, which-if cryo-cooling is sufficiently fast-can be expected to preserve the tissue structure well and also preserves the natural tissue-air contrast . After cryogenic fixation of the lung in an inflated state, the sample was placed into the path of the parallel beam setup at the P10 beamline (LS1). To this end, a custommade box, consisting of a self-fabricated plastic body, a circular shaped polyimide (Kapton) foil and a perforated lid, was used. Directly after taking the samples out of the cooled chamber, they were placed on a metal pin (Huber Diffraktion-stechnik) and inserted into the box. Dry ice was put into the perforated lid, which was subsequently placed onto the Kapton foil to ensure that the sample was not exposed to room temperature for too long. The dry ice evaporated and thus cooled the sample directly underneath, see figure 7(i).
The results can be inspected in figure 7, where two slices from a lung with pediatric pulmonary hypertension are displayed with perpendicular orientations. The xy-slice appears to have parallel, horizontal stripes which are an indication for sample movement during the scan (iii). This can be attributed to shrinkage of the tissue due to a non-ideal setup and a temperature in the box of −15°C which might be not sufficient to maintain the sample frozen for the duration of the data acquisition. However, as expected, the strong contrast caused by a large difference in electron density between tissue and air allows to achieve a very high contrast, which is a challenge for phase retrieval and is likely to compromise the image quality regarding the detailed structure of the alveolar septae. Contrarily, the high contrast enables a straightforward, threshold based tissue segmentation of the septa, see (iv). A comparison with the same cryogenically fixed sample imaged with a commercially available in-house scanner shows the strong increase in resolution and contrast achieved with synchrotron radiation, compare (ii) and (iii). Importantly, in contrast to paraffin-or alcohol-embedded samples, the tissue appears to be barely damaged and may best represent the real alveolar and tissue structure in an intact organism.

Quantification of virtual histology datasets
After presenting examples of XPCT reconstructions of illustrative lung tissue samples, highlighting different pathologies imaged with advanced in-house and synchrotron setups, we now consider automated quantification by 3D metrics of tissue structure. By automated computation of morphometric parameters, tissue features can be assessed in an objective manner without human visual judgement. Given proper validation with current pathological assessment and interpretation, large volumes can be interrogated and patterns representing pathological manifestations could be identified. To this end, we want to put forward two approaches, which we consider particularly well suited to quantify lung tissue in a generic manner, namely firstly the so-called chord length distribution (CLD), and secondly the structure tensor approach. CLD which has already proven to be a valuable concept to characterize pore size distributions of concretes (Chung et al 2020) and ( 3.6.1. Chord length distribution Figure 8 presents the CLD analysis. As a proof of concept, the chord length distribution is calculated here for a young lung donor without any known preconditions and an old donor, also with no known preconditions apart from age-related decay (laboratory scan: figure 8(i); synchrotron scan: figure 8(iii)). In both cases, chemically fixed paraffin embedded samples were prepared and scanned. First, the volume recorded with LS1-CLD setup is binarized using a manually determined optimal threshold and morphological operations which are applied to filter out artifacts and to correctly distinguish tissue from paraffin. In this way, the volume was separated in two phases (tissue: phase 1; paraffin: phase 0). Secondly, a slightly modified, already available, MATLAB-function (MacIver and Pawlik 2017, Maclver 2022) was used to calculate a distribution of chord lengths by creating directional anisotropic lines on a number of slices in a ROI cropped from the entire volume. Each of the lines intercepts the interfaces between the two phases. Each interception can be considered as an end point of a line segment. For each of these segments (or chords), the pixel length is calculated using the Bresenham line algorithm (Bresenham 1965) and extracted in tables with chords in phase 0 (paraffin) and phase 1 (tissue). Input data is the logical volume obtained from the segmentation described above (division of the whole reconstruction volume into two phases). Input parameters are the number of line segments to sample, and the minimum length of chords that should be considered. Finally, these chords are displayed in histograms for both phases (LS1setup, figure 8(i)) see figure 8(ii). The young donor lung shows a smaller amount of short and larger amount of long chord lengths for phase 0, indicating thicker tissue segments in comparison to the old donor. In addition, the young donor biopsy exhibits a larger amount of medium sized chord lengths for phase 1, reflecting the fact that the alveolar spaces are similarly sized and well shaped. Contrarily, the CLD for phase 1 of the old donor lung is much broader, representing a more disordered alveolar network architecture. This interpretation of the quantitative metrics is corroborated by visual inspection, both in the laboratory and the synchrotron data, and also in a rendering of the laboratory dataset ( figure S3).

Shape measure analysis
Originally designed as a tool to determine fibre orientation in reconstructed CT datasets of composite materials (Krause et al 2010), shape measure analysis recently proved to be a valuable tool in the determination of orientation and geometrical features in biological tissue based on the three-dimensional electron density reflected by the gray values (Reichardt et al 2021). In particular, this method is used in this work to determine a fiber-like score which gives indications on the presence of fibrotic human lung tissue. Here, this method is exemplified on a ACD-MPV sample with the rendered results displayed in figure 9. The first step of such a shape measure analysis includes the calculation of a structure tensor which encodes the predominant orientation and degree of anisotropy. The tensor is determined locally for a small test volume and then reported as a 3D map for the entire reconstruction volume. According to (Krause et al 2010), the local structure orientation at point r is given by the vector  with ν ä R 3 and |ν| = 1, where I is a discrete, twice continuously differentiable function describing the grayvalue intensity at a point r on the tissue. Due to an ill-posedness of taking gray value derivatives in the computation of gradients for noisy data, the data is first replaced by its smoothed version using a Gaussian filter I → K σ * I. A second convolution K ρ is applied to define the length scale over which the orientation is averaged (i.e. essentially the size of the 'test volume'), resulting in the structure tensor J: with I x representing the partial derivative ¶ ¶ I x . The minimisation problem (1) can be approximated by its regularised form, which corresponds to the eigenvector of the smallest eigenvalue of J. Thus, the orientation vector w(r) can be obtained for each point r. Contrarily, the direction of the gradient is given by the eigenvector corresponding to the largest eigenvalue of J at any given point (Krause et al 2010). In general, the relations between the eigenvalues (λ 1 λ 2 λ 3 ) reflect the anisotropy and symmetry properties of the local tissue structure, and can be referred to as quantitative shape measures. Namely, the shape measured help to distinguish between local structures exhibiting fiber, planar or spherical symmetry. Here, we chose smoothing values K σ = 1 pixel and K ρ = 18 pixels, which seemed reasonably in view of the expected lateral extent of potential fibrotic structures in pixel units at the given magnification. Figure 9(iv) illustrates the results, after multiplication of the shape measure results with a logical mask of the tissue, such that the values are only represented for tissue and not for paraffin. In (iv), the coefficient representing the degree of fiber-like symmetry given by

Discussion
Our results indicate an overall surprisingly high image quality even of in-house (laboratory) XPCT scans for the case of FFPE biopsies. This data already allows for segmentation and detailed tissue characterization, including (iii) Shape measure map applied on the same slice as in (i). For this representation, the shape measures were computed from the tensor of gray value derivatives as described in the main text, and then multiplied by a logical tissue mask, such that the shape measures are only shown for points belonging to tissue. Since fiber-shaped fibrotic tissue is of specific interest, line shaped features were extracted by calculating the first eigenvector of the structure tensor, thus identifying the predominant orientation in a given volume. The color red reflects a high score for linear features and can thus be at least tentatively be associated with areas which are likely to be fibrotic. (iv) Three dimensional display of the line shape measure (again after multiplication with the logical tissue mask), where red indicates a high line shape score. Scale bar: 200 μm. automated morphometric analysis. As we have shown here, this can be exploited for a wide range of pulmonary pathologies, and can be translated to dedicated and optimized instrumentation which could be operated in clinical setting. By using synchrotron radiation and optimized beamline instrumentation and parameters, the contrast can be further increased to resolve sub μm sized features down to the cellular and sub-cellular level. Furthermore, a wider range of preparations including sample mounting in liquids can be used. Challenging and poorly understood 3D structures such as interbronchial channels can thus be first identified in larger volume overview and subsequently studied in more detail at higher resolution. As a result of having the full 3D structure at hand, the respective physiological functions of airways or vascular networks, and the different pathophysiologic mechanisms can be elucidated or at least underpinned with structural data. At the same time, the synchrotron data can provide ground truth for validation of laboratory data and protocols, important for future standardisation and translation of the method.
Future extension of this work should in particular target sample preparation methods, such as further improvements of PCLS-protocols, developments of novel tissue extraction techniques, or optimized staining agents for certain tissue components (Busse et al 2021, Reichmann et al 2022). The goal here would be to further improve the overall image contrast, while preserving delicate tissue structures. PCLS protocols and imaging workflows, in particular, must be further optimized for XPCT. It is necessary to avoid sample movement, maintain the native tissue structure at least in terms of the alveolar geometry, and at the same time achieve sufficient contrast between tissue and embedding media. If this is given, XPCT imaging of PCLS could benefit studies of the lung tissue response to viral exposure, as well as the assessment of therapeutics. Furthermore, we can conclude that cryogenic fixation of tissue is a promising approach to study human lung parenchyma in its well preserved original state, ideally with no deformation or damage during tissue extraction and preservation. To this end, cryogenic fixation during sample extraction, known as cryobiopsies (Raghu et al 2019), should be considered, as well as adapting more advanced cooling elements such as Peltier cooling, to the different setups used here. Concerning phase retrieval, cryogenically fixed lung tissue can be challenging due to the high contrast between tissue and air, resulting in large phase gradients, which in conventional phase reconstruction such as the CTF-approach result in poor reconstruction quality. Here, further progress by more sophisticated algorithms can be anticipated. Finally, regarding image processing and analysis, the CLD and structure tensor results proved to be highly suitable and further extensions of this explorative study towards standardized workflows and the creation of reference libraries and data for different pathologies could possibly serve automated diagnostic classification.
In summary, this manuscript demonstrates the suitability and versatility of XPCT, being able to image human lung tissue in different embedding media on multiple scales and using a variety of instruments, with high resolutions and large FOV. The technique proves to be well suited to gain further understanding about human lung physiology based on mapping of the three dimensional fine structure of tissue, as well on the highly relevant pathophysiology of a wide range of lung diseases. Quantification of 3D datasets by radiomic tools such as shape measure analysis (figure 9) and chord length distribution (figure 8) allows for an objective assessment of tissue and can help to identify structures not easily detectable, even by the trained eyes of pathologists. In future, this work could be extended towards automatic fibrotic and emphysema quantification. Due to its threedimensionality and non-destructive nature, virtual histology by XPCT can become a useful extension which can be implemented in a complementary or correlative manner, with respect to existing histological imaging modalities. This would benefit even routine diagnostic assessment in a clinical setting, given further systematic refinement towards this goal.