Key points: Changes in the level of inspired oxygen have dramatic effects on the pathophysiology of obstructive sleep apnoea (OSA): hyperoxia reduces the severity of OSA in some but not all patients, whereas hypoxia transforms obstructive events into central events. Given that OSA is likely to result from the interaction of key pathophysiological traits, including a compromised pharyngeal anatomy, inadequate upper airway muscle function, a large ventilatory response to a disturbance in ventilation (high loop gain) and a low arousal threshold, we examined how changes in oxygen levels alter these traits. Our study demonstrates that the beneficial effect of hyperoxia on OSA severity is solely based on its ability to attenuate loop gain, whereas hypoxia increases loop gain and the arousal threshold in addition to improving pharyngeal collapsibility. Such effects help to explain why oxygen therapy may not work in every patient with OSA and explain the disappearance of OSA and the emergence of central events during hypoxic conditions. Oxygen therapy is known to reduce loop gain (LG) in patients with obstructive sleep apnoea (OSA), yet its effects on the other traits responsible for OSA remain unknown. Therefore, we assessed how hyperoxia and hypoxia alter four physiological traits in OSA patients. Eleven OSA subjects underwent a night of polysomnography during which the physiological traits were measured using multiple 3-min 'drops' from therapeutic continuous positive airway pressure (CPAP) levels. LG was defined as the ratio of the ventilatory overshoot to the preceding reduction in ventilation. Pharyngeal collapsibility was quantified as the ventilation at CPAP of 0 cmH2O. Upper airway responsiveness was defined as the ratio of the increase in ventilation to the increase in ventilatory drive across the drop. Arousal threshold was estimated as the level of ventilatory drive associated with arousal. On separate nights, subjects were submitted to hyperoxia (n = 9; FiO2 ∼0.5) or hypoxia (n = 10; FiO2 ∼0.15) and the four traits were reassessed. Hyperoxia lowered LG from a median of 3.4 [interquartile range (IQR): 2.6-4.1] to 2.1 (IQR: 1.3-2.5) (P < 0.01), but did not alter the remaining traits. By contrast, hypoxia increased LG [median: 3.3 (IQR: 2.3-4.0) vs. 6.4 (IQR: 4.5-9.7); P < 0.005]. Hypoxia additionally increased the arousal threshold (mean ± s.d. 10.9 ± 2.1 l min-1 vs. 13.3 ± 4.3 l min-1; P < 0.05) and improved pharyngeal collapsibility (mean ± s.d. 3.4 ± 1.4 l min-1 vs. 4.9 ± 1.3 l min-1; P < 0.05), but did not alter upper airway responsiveness (P = 0.7). This study demonstrates that the beneficial effect of hyperoxia on the severity of OSA is primarily based on its ability to reduce LG. The effects of hypoxia described above may explain the disappearance of OSA and the emergence of central sleep apnoea in conditions such as high altitude.