2.1. Sample preparation

The lung and the brain used for the experiment were obtained from two organ donors registered at the Laboratoire d'Anatomie Des Alpes Françaises (LADAF) in Grenoble, France. Dissections were performed respecting current French legislation for body donation. Body donation was based on free consent by the donors antemortem. All dissections respected the memory of the deceased. The post-mortem study was conducted according to QUACS (Quality Appraisal for Cadaveric Studies) scale recommendations (Wilke et al., 2015). The postmortem processing of organs (human brain and lung) follows the procedure involving formalin fixation and ethanol dehydration described in detail elsewhere (Walsh et al., 2021; Xian et al., 2022; Brunet et al., 2023). The human brain and lung samples were cut into ∼20 mm-thick slices with a diameter of ∼70 mm [see Fig. S1 of the supporting information (SI)]. The slices fit laterally into the cylindrical plastic containers such that the section plane is parallel with the lid to simulate the imaging geometry in tomography (Walsh et al., 2021) and to facilitate gas escape as gasses are formed on the tissue interior and surfaces. The tissue slices were immersed in 70% ethanol solution in water and immobilized with crushed agar (Walsh et al., 2021; Xian et al., 2022; Brunet et al., 2023). The non-tissue controls contain only the agar–ethanol mixture. Half of the containers were `highly' degassed using a diaphragm pump while the other half were not. During the vacuum degassing, the dissolved gas vaporized and bubbles appeared. The endpoint of vacuum degassing was determined by two observations: the emergence of bubbles after 90 s and a noticeable reduction in bubble formation upon visual examination. Details on the sample preparation and mounting are provided in SI Section S1.

2.2. In-line phase-contrast cineradiography

X-ray phase-contrast radiography and cineradiography experiments were carried out at the ESRF BM05 beamline using a polychromatic source. The X-ray white beam dimensions were 51.2 mm × 5.2 mm and was attenuated by 40 mm SiO₂ and 1.53 mm aluminium for an average energy of 82 keV. Only the field of view was illuminated; the rest of the X-ray beam was stopped with four-blade tungsten slits. The radiographs were directly recorded with the parallel beam configuration shown in Fig. 1(c) using a 2 mm-thin LuAG:Ce scintillator (lutecium-aluminium garnet doped with cerium, custom-made by Crytur, Czechia) and an sCMOS light sensor (PCO edge 4.2 Camera Link HS, PCO Imaging, Germany). The measurement configuration has a propagation distance (between the sample on the rotation stage and the detector) of 3475 mm, resulting in an isotropic pixel size of 25 µm for all radiographs. For each sample container, we started by locating the sample position with static phase-contrast radiographs. To trace the onset and evolution of X-ray-induced bubbles, cineradiographs were recorded with a data acquisition rate of three images per second for at least ∼10 min after the first bubble appeared within the field of view. The experimental hutch was kept at a constant temperature of 23°C.

2.3. Real-time gas chromatography

In the experiments, we used a portable commercial micro-gas chromatograph (ChromPix2, APIX Analytics, France) along with the data acquisition software DataApex Clarity v8.5. Four detection modules (MS5A, PDMS5, PDMS10, PPU), each containing a separate column as the stationary phase, were used to cover a wide range of detectable molecular species. We recovered the relative concentrations of EtOH and H₂O from the µGC module PDMS10, and those for N₂ and O₂ from the module MS5A. The modules do not detect H₂ due to the capacity limit, but the detected O₂ and volatile organic compounds were used to monitor the experiment. To improve gas detection, a reservoir of He gas was used as the carrier gas to continuously flush the volume within the cylindrical housing [see Fig. 1(d) and SI Fig. S2(a)], which functions as a dynamical headspace, to inject the escaped gas into the chromatograph. A pump was attached to the exhaust of the µGC to facilitate the gas extraction. During operando measurements, each round of gas elution through the four detection modules required about 2.4 min.

2.4. Computational analysis

Multiple flat-field images from static radiographs and cineradiographs have been used to estimate the bubble volume without explicit tomographic scans, which is possible after calculating and subtracting the absorption profile of the agar and tissue from the overall profile (see SI Section S2). The volume calculation used the direct inversion of the Beer–Lambert law. The bubble circularity in constrained environments is calculated using 4πA/C², where A and C are, respectively, the area and perimeter of the specific bubble obtained from image segmentation.

3.1. Degassing increases time to nucleation and reduces bubble load for a given X-ray dose

t^bbo shown in Fig. 3(a) indicates that the degassing procedure substantially increases the time before bubble nucleation for human brain and lung tissues, as well as for the non-tissue controls containing only the agar–ethanol mixture. The increase of t^bbo in degassed compared with non-degassed tissues demonstrates quantitatively that vacuum degassing is essential for avoiding bubble formation during prolonged X-ray scans. Over the course of a scan, at any given irradiation time t^rad, the degassed samples consistently have a lower cumulative gas volume (SI Fig. S5) and the excess gas volume is positive [Fig. 3(b)]. Moreover, the difference between degassed and non-degassed samples is more pronounced (in both nucleation time and difference between highly and non-degassed samples) in the tissue samples derived from the human lung than those from the brain, indicating a relation to their difference in gas solubility and tissue microstructures (Weathersby & Homer, 1980). These findings confirm that vacuum degassing enhances the sample resistance to bubble formation [hypothesis (i)] and that the bubbling process depends on the tissue type [hypothesis (ii)].

Figure 3 Quantitative characteristics of bubble onset and sample dependence after X-ray irradiation. (a) Bubble onset time [tbo in Fig. 2(a)] and X-ray dose deposited into different samples. Compared with non-degassed samples, the degassed samples show markedly delayed bubble onset and therefore higher dose threshold associated with bubble formation. (b) The excess in cumulative bubble volumes for three settings are calculated as a function of tabo, showing an increase until saturation, followed by a downward trend. (c) The initial ∼11 min of bubble growth measured against the time after bubble onset [tabo in Fig. 2(a)] shows a speed-up for highly degassed samples. The color choices in (a–c) are consistent to display their interrelationship.

3.2. Once formed, bubbles in highly degassed sample initially grow more quickly than in non-degassed sample

For all three pairs of tissue, the bubble time course dynamics can be adjusted for the time of bubble onset to facilitate comparison of the early bubble growth dynamics. After adjusting for the bubble onset times [see Fig. 3(b)], we observed a faster initial bubble growth rate in highly degassed than non-degassed samples [Fig. 3(c)]. To quantify the initial growth trend of X-ray-induced bubbles, we approximate the cumulative bubble volume by a power-law relation,

Here, rect is a window function, the terms t^bo represent the bubble onset (bo) time and α is the growth exponent. Fitting of equation (2) to the data up to 10 min following bubble onset yields a consistent growth exponent α ≃ 3.0 for highly degassed samples while in non-degassed samples α varies by at least 0.4, exhibiting strong dependence on the sample (see SI Table S2). Since dissolved gas modifies the liquid's thermodynamic landscape (Jones et al., 1999), the noticeable change in kinetics between highly degassed and non-degassed samples indicates the distinct metastable states that the embedding solution was in immediately before bubble onset. In addition, we calculated the difference in terms of cumulative bubble volume as shown in Fig. 3(b). For all control pairs, the excess increases until a certain threshold is reached before decreasing.

3.3. Bubble growth is geometrically constrained by sample microstructure

When bubbles nucleate in less constrained spaces, such as on the material (tissue and crushed agar) surfaces or in the solution phase of the embedding media (ethanol), they aggregate, coalesce and often move upwards out of the imaging field of view. These less spatially constrained bubbles tend to be more globular and grow more uniformly, whereas the bubbles formed within void spaces in the tissue interior such as empty blood vessels or the lung's air-contacting bronchi are more likely to be trapped in situ (bubble entrainment). The entrapped bubbles take on the shape of the surrounding material confining their growth and are generally highly nonspherical. These distinct bubble dynamics all contribute to the overall growth shown in Figs. 2 and 3. Once the X-ray beam is shut off, the bubbling process keeps developing but gradually subsides as the excess energy dissipates, which can take several minutes in our experiments, as checked by taking static radiographs at later times. The appearances of samples before and after X-ray irradiation are compared in SI Fig. S2, where bubbles are visible around and above the tissue sample up until the container lid.

Understanding the site-specific bubble dynamics can help elucidate the nature of the bubble formation process (Brennen, 2013). To this end, we selected representative bubbles from each sample that were spatially separated from other bubbles within the recorded radiographs (see Fig. 4). For each case, we calculated the bubble circularity (see Section 2) and area over time after image segmentation. A circularity of 1 indicates a perfect circle, whereas the closer it approaches 0 the more eccentric the shape becomes. From the shape analysis of the bubble growth, three distinctive scenarios emerge: (i) Unrestricted 3D growth of bubbles [Figs. 4(a)–4(c)] that nucleated within the gaps of crushed agar, showing a nearly constant circularity over time. The relationship between bubble area (A) and the time after bubble onset (t^abo, here referring to the specific bubble's timescale) is At^abo, indicating that its radial growth rate ≃ , which is consistent with vapor (Prosperetti, 2017) or diffusion-driven growth (Brennen, 2013). (ii) Growth within a large blood vessel [Figs. 4(d)–4(f)] in the brain tissue (∼100 µm diameter), showing a continuous decrease in circularity that correlates with bidirectional expansion within the tubular vessel, which resembles the behavior of intravascular gas embolism (Branger & Eckmann, 1999). (iii) Stepped growth of bubbles that indicates the distinct growth rates in a multi-compartmental structure like the alveoli and the lung's peripheral air spaces (Haefeli-Bleuer & Weibel, 1988) [Figs. 4(g)–4(i)]. The step-like feature in time-dependent circularity and bubble area correlate with the sequential expansion of the bubble into nearby alveoli further and further away from the nucleation site. Gas expansion within neighboring pulmonary alveoli is the fastest among the three scenarios [increasing by ∼4 × 10⁶ µm³ in less than 2 min as in Fig. 4(c)] due to their optimized anatomical design that facilitates alveolar gas exchange (Sapoval et al., 2002).

Figure 4 Constrained bubble dynamics in different microscopic void spaces, including (a–c) regions between crushed agar, (d–f) within brain vasculature and (g–i) within alveoli of the lung. (a, d, g) Schematics showing the local spatial geometry where the bubbles (white or light gray inside in phase contrast) are formed. (b, e, h) The extracted growth dynamics of the bubbles within these spaces are quantified using time-dependent circularity and bubble area. (c, f, i) Representative radiographic frames showing the morphological changes of the specific bubble dynamics. The respective time points are indicated by vertical dashed lines in (b), (e) and (h), respectively. The white scale bars represent 500 µm.

3.4. Gas analysis shows changes in relative solvent concentration during bubbling

The µGC-based gas detection setup is briefly described in Section 2. The detected gasses come from two sources: (i) the dynamical headspace within the plastic housing around the sample container, and (ii) the gasses generated within the sample container, including dissolved gas, vapor, and gaseous or volatile end products from potential photoreactions. Chromatographic monitoring provides a way to examine the bubble content as they escape the system on the second to minute timescale. Therefore, the detection scheme naturally filters out short-lived, highly reactive reaction intermediates as well as trace gasses since none of them is a main contributor to bubble formation macroscopically. In our case, four detection modules were used in the micro-gas chromatograph to cover different ranges of chemicals (see SI Section S4.1). As presented in Fig. 5, the chromatograms show clear signatures of evaporated solvent (EtOH and H₂O) and gasses from ambient air or dissolved gasses in tissue (primarily N₂ and O₂). Although dissolved O₂ and N₂ concentrations differ with tissue composition (Weathersby & Homer, 1980), we have observed that the relative concentrations of N₂ and O₂ detected by µGC are essentially unchanged in all cases before and during sustained bubbling under X-ray irradiation, as shown in Fig. 5. This contrasts with the behavior of evaporated solvents from µGC. Specifically, EtOH shows an increase in its relative concentration for all samples except the non-degassed agar, whereas it generally decreases for H₂O. For each sample, the changes took place around the corresponding observed t^bo, which is largely consistent for all pairs of samples, without a noticeable dependence on sample type. This indicates that the gas from X-ray induced bubbles largely comes from solvent vapor. Moreover, within our detection sensitivity limit, no significant signals of other chemical species, such as the radiolytic product acetaldehyde (CH₃CHO) from EtOH, have been detected, indicating the relative radiation stability of agar and solvents (see SI Section S4.2).

Figure 5 Gas evolution around the bubble onset time (tbo) during X-ray irradiation for (a–b) agar samples, (c–d) human brain tissue samples and (e–f) human lung tissue samples. Four dominant chemical species detected by the micro-gas chromatograph, including N2, O2, EtOH and H2O, are shown here in their relative concentrations (rel. conc.) at 6 min, 3 min before bubble onset and 5 min, 10 min after it. The relative concentrations are separated into two sets corresponding to the two modules: MS5A measured O2 and N2 concentrations (yellow), while PDMS10 measured EtOH, H2O and O2 + N2 (purple). The relative concentrations are expressed on a normalized scale, where 0 indicates 0% concentration and 1 corresponds to 100% concentration. For each type of gas, the corresponding tbo is drawn in the figures with vertical red lines. The general behavior before and after bubble onset is that the relative concentrations of dissolved gasses (N2 and O2) remain constant, while the solvent vapor (EtOH and H2O) shows noticeable changes.

3.5. Mechanisms driving bubble nucleation and growth

In addition to our initial hypotheses regarding sample preparation and tissue composition, our study enables the investigation of the mechanisms that underlie bubble formation and growth in the high-energy X-ray regime. Our quantitative observations for our experimental conditions led us to consider a few plausible underlying mechanisms, which can come from chemistry and heating (thermal or nonthermal). The mechanisms we considered were (i) accumulated photochemical effects, (ii) dose-dependent solvent heating effects, and (iii) phase transition in the gel embedding. Regarding scenario (i), the most likely reactions are the radiolysis of EtOH and H₂O, which produce gasses (H₂, O₂, CH₃CHO and H₂O as volatile primary products), radical ions, and oxidized solvent as end products (Meents et al., 2010; Le Caër, 2011; Freeman, 1974; Sazonov et al., 2015). Since the high-energy X-rays used for imaging are very far off resonance to sample composition, we expect the gas contribution from direct water and ethanol splitting with X-rays (Mao et al., 2006) to be low, albeit nonvanishing. This stems from the two-order-of-magnitude reduction of the photoelectric effect cross-section between X-rays in the energy range 80–100 keV and 10–30 keV for light elements which constitute soft tissues (Hubbell, 1999). In fact, an earlier experiment on ice showed that X-ray-induced H₂O decomposition is undetectable with X-ray energies above 30 keV (Mao et al., 2006). In our case, the reasoning is supported by the µGC results, which identify X-ray-induced vaporization (Weon et al., 2011) as a primary source of gas escape.

To evaluate (ii) and (iii), we performed dosimetry measurements with an ionization chamber dosimeter and numerical estimation (see SI Section S3). With a measured dose rate of 36.8 Gy s⁻¹, we found that the accumulated surface dose on the sample ranges between 15.5 kGy (non-degassed agar) and 46.4 kGy (highly degassed lung) just before bubble onset [see Fig. 3(a) and SI Table S3]. The dosimetry data are then used to estimate the temperature change (see SI Section S3): assuming no heat exchange, the thermal load imparted to the sample generally creates a temperature increase within the X-ray field of view of up to about 11°C, if assuming only H₂O, or 19°C, if assuming only EtOH. These estimations correspond to the maximum possible temperature change. Although detailed dosimetry calibrations are beyond the scope of this work, our estimation provides a sensible upper bound on these values. Therefore, the macroscopic metastable state at bubble onset differs from the two prominent examples of thermally induced bubble formation involving electromagnetic radiation: (a) resonant heating which produces plasmonic microbubbles from water vaporization observed in solvated gold nanoparticles (Wang et al., 2018); (b) the bubble-chamber phenomenon (Roy, 2001; d'Errico, 2001), which requires a superheated liquid medium to be maintained between its boiling and critical temperature for the effect to take place. For 70% EtOH, the boiling point is over 78°C at ambient pressure (https://webbook.nist.gov/cgi/cbook.cgi?ID=64-17-5), which is inconsistent with our experimental condition. However, rapid photoionization can produce heating effects through disruption of the coordination environment in the solvation shell (Beyerlein et al., 2018) or lowering of the vaporization enthalpy (Weon et al., 2008, 2011). In our case, it is likely that the exsolution of dissolved gas in tissue due to a moderate temperature change or trace gas generated through photochemical reactions seeded the bubble formation in the solvent, which is further promoted by a photoionization-disrupted local solvation environment to rupture the bulk liquid.

For scenario (iii), the working hypothesis is that the agar may undergo thermoreversible sol–gel transition due to X-ray-induced heating, resulting in the release of entrapped gas. This behavior has been observed at much lower X-ray energies (∼30 keV), resulting in melting of the embedding agar gel during imaging (Mittone et al., 2020). However, in our case, the crushed agar has been degassed after gelling and before its use in sample preparation. At the considerably higher X-ray energy we used for radiography, no change in the consistency of the agar gel was observed from before the X-ray beam is switched on and immediately after the beam is shut off. Moreover, due to the large thermal hysteresis of agar, its melting temperature is much higher than the gel-setting temperature, therefore the very moderate temperature change we estimated before t^bo would unlikely initiate melting.

3.6. Discussion

Our estimation of the bubble onset and volume estimation is only accounting for those within the field of view of the X-ray beam, hence we exclude the bubbles generated from scattered X-rays. Nevertheless, we observe that X-ray-induced bubble formation appears to have a dose threshold in the 10⁴ kGy range for in-line radiography. Vacuum degassing delays the bubble onset time, and reduces bubble load over the timescale of a typical scan. The degassing process also leads to faster bubble growth than non-degassed samples at this delayed time. These results, on the one hand, complement existing discussions of the dose-resolution trade-off in X-ray nanoimaging of single cells and thin tissues and in macromolecular crystallography (Howells et al., 2009; Jacobsen, 2020), where samples (at only up to millimetres in size) are typically kept in a frozen-hydrated or freeze-dried state. On the other hand, they relate to the emerging field of X-ray virtual histology (Walsh et al., 2021; Bravin et al., 2013; Busse et al., 2018; Töpperweien et al., 2018), where the balance of sample integrity and imaging throughput requires consideration of radiation–matter interaction across various scales determined by the biological questions. Since bubbles can cause local deformation which may lead to damage in soft materials including tissues (Barney et al., 2020; Hasan et al., 2021) from the pressure exerted on them during cavitation, collapse or subsequent bubble entrainment, our results provide the empirical foundation for developing precautionary measures, such as tissue degassing, pressurization, temperature stabilization, to mitigate or postpone bubble formation, thereby reducing the measurement failure rate. The methods described here may also be used to further investigate microbubble formation and growth phenomena in various tissues relevant for clinical models. The examples include air embolism (Muth & Shank, 2000; Branger & Eckmann, 1999) from positive pressure ventilation or after scuba diving accidents, where alveolar gas ruptures into neighboring capillaries due to an increase in gas volume (Russi, 1998), and decompression sickness, where the mechanism and progression of off-gassing from tissue and vasculature remains unproven (Papadopoulou et al., 2013). In both cases, constructing microscopically accurate physiological simulations requires empirical parameters that are often hard to obtain in parenchymal tissues by conventional means.

Beyond the context of bioimaging, spontaneous bubble formation during X-ray experiments has also been briefly reported in X-ray spectroscopy (Mesu et al., 2006) and real-time imaging of electrodeposition (Tsai et al., 2002) and batteries (Charalambous et al., 2021) involving liquid electrolytes. In these scenarios, the environment is also heterogeneous and bubbles likewise produce undesired and sometimes deleterious effects for the precise characterization and stable operation of devices. However, because of the complexity of the liquid environment and gasses produced during electrochemical operations (Rowden & Garcia-Araez, 2020), it remains exceedingly difficult to separate the sources of bubbles. Our observations in soft biological tissues and gels reaffirm the proposition that X-rays may not be treated solely as a passive probe under intense or prolonged irradiation (Bras et al., 2021), although isolating bubble sources will require subsequent studies in simplified model systems. The set of control experiments reported here demonstrates that bubble formation remains an issue even at high X-ray energies, when sufficient dose is deposited into the sample and its embedding environment, regardless of the initial condition.

3.7. Limitations of the study

Synchrotron experiments are typically strongly constrained by time, which precludes the exhaustive testing and optimization of many experimental factors. The current study focuses on examining the degassing time, the bubble growth characteristics, and the tissue type dependence on the bubble onset. Other parameters may influence bubble formation but were not tested include the dependence on X-ray energy (e.g. in the 30–80 keV range), the synchrotron electron bunch structure, sample and container size, and embedding medium characteristics (e.g. concentration, viscosity, pH, etc). Nevertheless, our analysis indicates that under similar synchrotron X-ray irradiation and environmental conditions, the vaporization enthalpy of the liquid embedding medium and the sample property may be key factors for determining t^bbo, the quantity we intend to maximize in uninterrupted imaging. These insights may help infer the behavior of other common embedding media: water has comparable vaporization enthalpy (Drisdell et al., 2010) to EtOH, but that value for formaldehyde (active ingredient in formalin) is significantly lower because it exists as a gas at room temperature and ambient pressure. Therefore, we expect that bubble formation to be more significant in formalin than EtOH or water under similar experimental conditions. Besides, the sizes of the sample and container may also be modified, albeit not necessarily easily, to reduce the radiation dose applied to the sample and thereby the tendency of bubble formation. Finally, we want to note that the temporal quantities such as t^bo, t^bbo and t^abo that are empirically determined here are contingent on the bubble size exceeding the resolution of the imaging system. At higher resolution, smaller bubbles may be observed, so their values may differ. However, the power-law growth characteristics of the bubbles indicate that the true t^bo differ by only a few seconds, which is a negligible time interval in our analysis of the bubble growth characteristics on the scale of tens of minutes.