Obstructive sleep apnea (OSA) is a very prevalent respiratory disorder affecting patients of all ages, from children to the elderly (Benjafield et al., 2019). Patients suffering from OSA exhibit an abnormally increased collapsibility of the upper airway during sleep and thereby experience recurrent events of upper airway obstruction, usually terminated by an arousal. In more severe instances, these patients can sustain more than one obstructive event per minute of sleep. In addition to the consequences directly caused by the disruption of sleep architecture (diurnal somnolence, fatigue, increased traffic and labor accidents, poor quality of life, cognitive deficits, and depression), patients with OSA are also at increased risk of both morbidity and mortality from cardiovascular, metabolic, neurocognitive and malignant diseases (Sánchez-de-la-Torre et al., 2013; Gileles-Hillel et al., 2016; Gozal et al., 2020; Brockmann and Gozal, 2022). These adverse outcomes are primarily caused by the recurrent events of hypoxemia induced by the upper airway obstructions (i.e., apneas and hypopneas). Indeed, during these events, the absence/reduction of pulmonary alveolar ventilation results in transient reductions in the partial pressure of O₂ in arterial blood, which is clinically assessed by non-invasively measuring arterial oxygen saturation (SaO₂) by pulse oximetry (Figure 1A) (Farré et al., 1998). The recurrently hypoxic blood leaving the lungs enters the systemic capillaries that perfuse all the patient tissues and organs (Almendros et al., 2011; Reinke et al., 2011; Torres et al., 2014; Moreno-Indias et al., 2015), and this reduced O₂ tension then diffuses into the surrounding extra capillary space, thereby subjecting parenchymal cells to intermittent hypoxia of varying degrees (Figure 1B). It has been well established that this noxious challenge triggers cascades of oxidative stress, inflammation, and immune and hormonal deregulations, which ultimately result in the increased end-organ and systemic adverse consequences of OSA (Sánchez-de-la-Torre et al., 2013; Gileles-Hillel et al., 2016; Gozal et al., 2020; Brockmann and Gozal, 2022).

Given that intermittent hypoxia is a major driver of the OSA-induced increase in morbidity and mortality, considerable experimental research efforts have been devoted to investigating how different types of cultured cells respond to these hypoxic cycles (Farré et al., 2018). However, a particular challenge in precisely achieving this research aim is to ensure that the cultured cells of interest are de facto subjected to the high-frequency events of hypoxia (up to 60 cycles per hour) characterizing OSA. Indeed, given the relatively slow process of O₂ diffusion in water or other liquids, using the conventional procedure based on cyclically modifying the O₂ partial pressure in the air on top of the culture medium is sub-optimal to achieve adequate fast changes of O₂ partial pressure at the cell culture level (Allen et al., 2001). Fortunately, optimized experimental settings based on culturing cells on a thin O₂-permeable membrane have been described to ensure that the cultured cells experience the desirable fast OSA-mimicking intermittent hypoxia events (Campillo et al., 2016; Minoves et al., 2017). However, these improved settings were designed and tested in cells cultured in 2D set-ups. Whereas such geometry is suitable for cells naturally living in monolayers (e.g., epithelial and endothelial cells), it is suboptimal when investigating the effects of intermittent hypoxia in most parenchymal (or tumor) cell types that naturally reside and evolve in a 3D environment within an extracellular matrix (ECM). To date, it has been a common approach to use 2D culture plates to investigate the biology of cells that naturally live in vivo in 3D microenvironments. However, there is solid emerging evidence that 2D and 3D cell microenvironments differently modulate essential cell functions (Duval et al., 2017; Jensen and Teng, 2020; Kusuma et al., 2022). Therefore, optimal study of parenchymal cell biology requires realistic simulation and replication of the native 3D configuration by culturing cells inside hydrogels or using ex-vivo precision-cut tissue slices (Tibbitt and Anseth, 2009; Caliari and Burdick, 2016; Dewyse et al., 2021). Accordingly, the present study aimed to devise and characterize a procedure for applying physiomimetic intermittent hypoxia to cells in a 3D microenvironment.

The experimental setting devised to apply intermittent hypoxia to cells in a 3D environment is schematically shown in Figure 2A. It is based on a previously described well with a bottom consisting of a membrane of O₂-permeable polydimethylsiloxane (PDMS) (Campillo et al., 2016). A flow of intermittently hypoxic air at a frequency characterizing severe OSA (e.g., 60 apneas/h: 30 s of normoxic gas (i.e., air) and 30 s of hypoxic gas (i.e., target FiO2 of interest) circulates through the compartment beneath the membrane. The 3D scaffold (thickness W) containing the cells subjected to intermittent hypoxia (hydrogel made of natural ECM or precision-cut tissue slices) is placed on the PDMS membrane. A micrometric fast-response O2 sensor is only used to characterize the setting to measure O₂ inside the 3D sample at different distances from the membrane (z, Figure 2A).

The time course of O₂ concentration measured at different distances from the membrane (Figure 2A) in a W = 500-μm lung ECM hydrogel when subjected to intermittent hypoxia (20% O₂-0% O₂) at a rate of 60 events/h is shown in Figure 3A. For z = 100 μm, the maximum O₂ concentration was virtually 20%, but the minimum was slightly greater than 0% O₂ because of the slight drop in O₂ concentration caused by diffusion across the PDMS membrane. As expected, increases in z decreased the amplitude of O₂ concentration swings. However, the amplitude of O₂ cycles was still considerable, even close to the top of the hydrogel (z = 400 μm). In contrast, when the thickness of the hydrogel was 800 μm, the amplitudes of O₂ oscillations were considerably reduced as z increased (Figure 3B), with a maximum of 13.1% O₂ and a minimum of 10.7% O₂ for z = 700 μm.

Figure 4 shows the excellent reproducibility observed (mean ± SD) of the maximum and minimum data shown in Figure 3A when the O₂ measurements were carried out in 4 different random areas of two different W = 500-μm hydrogel samples. This figure also reflects that, despite the slight decrease in maximum and increase in minimum observed when z increased, the variance of minimum and maximum O₂ concentrations across the hydrogel section was lower than 2% O₂ around the corresponding mean value.

The range of O₂ oscillations within the sample was readily modulated by modifying the O₂ concentration of the gas circulating beneath the membrane. For instance, as shown in Figure 5, lower amplitude of intermittent hypoxic gas (15% O₂–5% O₂ in the gas beneath the membrane) subjected the hydrogel to oxygenation values close to the ones in the arterial blood perfusing tissues in patients with severe OSA (Figure 1A). The maximum and minimum values were similar over the hydrogel section (z from 100 to 400 μm), with maxima and minima ranging within .5% O₂ and .2% O₂, respectively.

3D culturing MSCs in the hydrogel did not significantly change the O₂ diffusion across the sample as compared with the acellular hydrogel (Figure 3). Figure 6 shows the small differences in maxima and minima of O₂ concentration observed in W = 500-μm lung ECM hydrogels in three cases: acellular and seeded with MSCs at two concentrations (300 and 450 × 10³ cells/mL).

Figure 7 shows the O₂ concentration recordings in mouse brain slices for 500-μm and 300-μm thicknesses. This figure further illustrates that changing the gas concentration flowing beneath the membrane (in this case 95%/50% O₂) allowed modulation of the maximum and minimum of oxygenation in the sample. For W = 500-μm, O₂ profiles were less homogeneous across the sample as compared with acellular (Figure 3) or MSC-cultured hydrogels (Figure 6). Indeed, the maximum - minimum of O₂ were 80.8%–42.0% for z = 100 μm and 66.3%–34.0% for z = 400 μm (Figure 7A). In contrast, the intermittent hypoxia experienced across 300-μm mouse brain slices was nearly homogeneous, with maximum - minimum of O₂ of 90.5%–49.0% for z = 50 μm and 82.3%—44.3% for z = 250 μm (Figure 7B).

This study provides initial and compelling experimental evidence that it is possible to apply well-controlled fast intermittent hypoxia cycling to cultured cells in 3D environments mimicking severe sleep apnea, either when using an ECM hydrogel or when cells are maintained ex vivo in tissue slices.

In addition to its robust and reliable performance, one advantage of the experimental setting employed and tested in this study is its simplicity. Indeed, it is based on the fast diffusion of O₂ through a thin membrane of PDMS, a material with a coefficient of diffusion for O₂ (D ≈ 3.5 × 10⁻⁵ cm²/s) slightly higher than that of water (≈2.5 × 10⁻⁵ cm²/s) (Farré et al., 2018). Thus, the 3D diffusion time (∆t = L²/6·D) for O₂ corresponding to a membrane with a thickness L = 100–200 μm is .5–2 s, a time much shorter than the periods of intermittent hypoxia in severe OSA (60 apneas/h). In this study, we used a commercially available PDMS membrane that was not particularly thin (165 μm). Therefore, employing thinner membranes (≤100 μm) as the ones also used in the bottom of cell culture wells (Campillo et al., 2016) could further fasten the transmission of O₂ concentration changes to the 3D sample. Interestingly, as it does not require the use of complex microfluidic tools, this experimental setting can be easily employed by any cell biology laboratory: PDMS bottom wells are already commercially available or can be easily homemade (Campillo et al., 2016; Campillo et al., 2019). The other component required is a flow of intermittent hypoxic-normoxic gas beneath the membrane. It is of note that using a relatively expensive blender like the one employed in this work is not necessary. Indeed, the required square wave of O₂ concentration can be achieved in a much easier and cheaper way by simply connecting the inlet of the well to two conventional sources of gas. For instance, compressed gas bottles (with a low-pressure regulator) can be alternatively connected to the well by a simple on-off timer (either by two low-cost perfusion pumps or a 3-way controllable valve).

The O₂ microsensor was employed to characterize the specific levels and dynamics of hypoxia applied to the different 3D materials (hydrogels or tissue slices) and thicknesses. Therefore, once a given setting is defined, carrying out repeated and systematic experiments to study cell behavior should not require continued O₂ measurements across all experiments. However, using the sensor is advisable when designing a specific experiment given that different hydrogels/tissues, and mainly their thickness, modify O₂ transmission across the sample. It should be mentioned that an O₂ microsensor device similar to the one employed in this study (or one based on Clark micro-electrode (Almendros et al., 2011) is not expensive (Otero et al., 2021). It is also worth noting that although we focused only on the application of intermittent hypoxia, the setting could easily be used for also applying intermittent hypercapnia which parallels intermittent hypoxia in OSA and is another challenge to normal cell response. Alternatively, both intermittent hypoxia and hypercapnia could be also implemented easily by a pre-mixed gas source, thereby reproducibly simulating naturally occurring events in patients with OSA (Imamura et al., 2019; Tripathi et al., 2019). However, we did not test intermittent hypercapnia because there is no available CO₂ sensor with the size and time resolutions required for this application. Nevertheless, it can be anticipated that the setting would also work for intermittent hypercapnia, taking into account that the coefficients of diffusion for CO₂ in PDMS (2.2 × 10⁻⁵ cm²/s) and water (2.1 × 10⁻⁵ cm²/s) are only slightly lower than for O₂ (Hou et al., 2010; Kanehashi et al., 2012).

The results obtained in W = 500-μm hydrogels (Figure 3; Figure 5) show that O₂ diffusion through the 3D scaffold was sufficiently fast to allow high-frequency intermittent hypoxia requiring 30-s for raising and decreasing times in O₂ concentration across the 3D sample. Interestingly, ≈500 μm is a commonly used thickness for the research of 3D cell-seeded samples (De Hilster et al., 2020; Falcones et al., 2021; Marhuenda et al., 2022). The suitability of this thickness for optimal application of 60 hypoxic events/h was expected taking into account that the O₂ coefficient of diffusion D in this type of hydrogel is slightly lower than that of water. Indeed, data from O₂ diffusion studies in similar hydrogels (alginate, agarose, collagen, dense fibrin, reconstituted basement membrane (rBM/Matrigel)) indicate that D ranges around 75% of D in water (Ehsan and George, 2013; Colom et al., 2014). Also, it has been shown that in alginate and agarose hydrogels, the diffusion of small molecular weight species such as O₂ is not significantly affected by the characteristics of the matrix (Li et al., 1996). Hence, assuming D = 1.87 × 10⁻⁵ cm²/s (75% of D in water), the diffusion time would be ∆t = 22 s which fits with the apparent time-constant of the quasi-exponential variation pattern of O₂ concentration changes (Figure 3). However, it is remarkable that the diffusion rate and, thus, the amplitude of the high-frequency intermittent hypoxia cycles achieved largely depends on W, as illustrated by comparing the results of O₂ cycling across two hydrogels that only differed in thickness. Whereas for W = 500 μm the transmission of O₂ cycling was very satisfactory (Figure 3A), in the case of W = 800 μm the attenuation of O₂ cycles amplitude was so apparent (Figure 3B) that no application of uniform intermittent hypoxia could be ensured for cells residing across the hydrogel. This considerable difference in results when comparing 500 μm and 800 μm thicknesses is not surprising given the quadratic dependence of the diffusion time ∆t on L (∆t = L²/6·D). Indeed, ∆t would be 57 s for L = 800 μm (as compared with 22 s for L = 500 μm), a time similar to the period of oscillation for 60 events/h.

Theoretically, the coefficient of diffusion D is not the only factor affecting the rate of O₂ concentration change in cell-containing 3D scaffolds. Indeed, the rate of decreasing concentration C(z,t) depends on the coefficient of diffusion D according to the second Fick law and also on the cellular O₂ consumption rate per unit volume of the material. Assuming that cellular O₂ consumption follows the Michaelis-Menten kinetics: ∂ C z ,   t ∂ t = D ∂ 2 C z , t ∂ z 2 −   ρ c e l l ·   s O C R   ·   C z ,   t K m + C z ,   t (1) where ρ _cell is cell density, sOCR is the maximum rate of O₂ consumption per single cell, and K_m is the half-maximum rate of O₂ concentration (Magliaro et al., 2019). When considering Eq. 1 in the case of O₂ unidirectionally flowing through a membrane-like hydrogel with thickness W, the rate of O₂ consumption (V’_con) per unit area within the volume can be estimated as V ’ c o n = ρ c e l l   ·   s O C R   ·   C m e a n   ·   W   K m + C m e a n (2) where C_mean is the average O₂ concentration across the construct. The diffusion flow (V’_dif) across the membrane-like hydrogel is V ’ d i f = D · ∆ C W = D   ·   α   · ∆ P   W (3) where ∆C is the difference in O₂ concentration across the construct, α is the solubility of O₂ in the material (assuming that of water α = 1.1 × 10⁻⁶ mol/(cm³ · atm)) and ∆P is the difference of O₂ partial pressure across the membrane-like hydrogel. Therefore, the ratio between consumed and diffused O₂ flows is V ’ c o n V ’ d i f = ρ c e l l   ·   s O C R   ·   C m e a n   ·   W 2   D   ·   α   · ∆ P   ·   K m + C m e a n (4)

Given the quadratic dependence of this ratio on W, the role played by O₂ consumption is decreased as the hydrogel thickness is reduced. Indeed, assuming W = 500 μm, ∆P = 152 mmHg (corresponding to a difference from 20% O₂ to 0% O₂), and taking the parameters (D = 1.2 × 10⁻⁵ cm²/s, sOCR = 1.22 × 10⁻¹⁶ mol/(cell·s), and K_m = 4.1 × 10⁻⁶ mol/cm³) reported (Magliaro et al., 2019) for a collagen hydrogel seeded with hepatocytes with a cell concentration ρ _cell = 500 × 10³ cell/cm³, Eq. 4 results in a ratio V’_con/V’_dif = .015, indicating that, in this case, the consumption of O₂ can be neglected when compared with O₂ diffusion. This conclusion remains valid when using the parameters measured for a cell concentration one order of magnitude higher (5 × 10⁶ cell/cm³) (Magliaro et al., 2019) or when considering cell types other than hepatocytes since, together with neurons, hepatocytes are the highest O₂-consuming cells (McMurtrey, 2016). The fact that oxygen consumption is not relevant for applying O₂ cycling in thin hydrogel samples is reflected by the results in Figure 6 showing virtually no differences between acellular and cell-seeded hydrogels.

The systematic O₂ concentration measurements carried out in hydrogels have been complemented with preliminary data obtained when applying intermittent hypoxia to tissue slices. Figure 7 provides a proof-of-concept measurement in brain mouse slices having a thickness (300 and 500 μm) typical in studies using precision-cut tissue slices (Majorova et al., 2021). These results clearly suggest that, as in hydrogels, fast intermittent hypoxia can be applied to ex vivo tissue samples. We observed that for the same sample thickness of W = 500 μm O₂ diffusion across the brain slices was slightly slower than across the hydrogels (Figures 3, 7), resulting in a less homogeneous distribution of intermittent hypoxia as z varied. Based on the two components in Eq. 1, two potential reasons may account for this observed difference. On the one hand, the diffusion coefficient for O₂ in body tissues could be lower than in most usual hydrogels, as suggested by the few data available providing measured values of D for O₂ in native tissues (MacDougall and McCabe, 1967). On the other hand, owing to higher cell density (ρ _cell), the Eq. 1 component corresponding to the rate of O₂ consumption by cells could be greater in native tissues than in artificial hydrogels. Regarding this particular point, it is interesting to note that when considering specific O₂ consumption per cell (sOCR), brain tissue is most challenging for O₂ transfer given the very high metabolic rate of its cellular components (McMurtrey, 2016). Although requiring further systematic analysis to characterize reproducibility in different types of ex vivo slices, the results obtained in brain samples suggest that similar or even better results could be obtained in 300–500 μm tissue slices from different organs (e.g., lung, liver, kidney, heart, pancreas, spinal cord or tumors (Wang et al., 2015; Stribos et al., 2016; Wu et al., 2019; Dewyse et al., 2021; Dimou et al., 2022)) particularly affected by end-organ OSA morbidity (Sánchez-de-la-Torre et al., 2013; Gileles-Hillel et al., 2016; Gozal et al., 2020; Brockmann and Gozal, 2022).

In conclusion, we have characterized an experimental approach, based on existing and easily available tools, that allows for controlled and precise application of high-frequency intermittent hypoxia both to 3D cell culture scaffolds such as ECM hydrogels as well as to precision-cut tissue slices. This approach, which is an advance on previous models investigating continuous hypoxia (Svanström et al., 2021), opens the door for the study of the specific noxious effects of intermittent hypoxia in the 3D-residing cells of the organs and tissues of patients with OSA under physiomimetic conditions.