In severe chronic obstructive pulmonary disease (COPD), carbon dioxide retention during exercise is highly variable and is poorly predicted by resting pulmonary function and arterial blood gases or by tests of ventilatory control. We reasoned that in patients with compromised gas exchange capabilities, exercise hypercapnia could be explained, in part, by the restrictive consequences of dynamic lung hyperinflation. We studied 20 stable patients with COPD (FEV(1) = 34 +/- 3 percent predicted; mean +/- SEM) with varying gas exchange abnormalities (Pa(O(2)) range, 35 to 84 mm Hg; Pa(CO(2)) range, 31 to 64 mm Hg). During symptom-limited maximum cycle exercise breathing room air, Pa(CO(2)) increased 7 +/- 1 mm Hg (p < 0.05) from rest to peak exercise (range, -6 to 25 mm Hg). We measured the change in Pa(CO(2)) after hyperoxic breathing at rest as an indirect test of ventilation-perfusion abnormalities. The change in Pa(CO(2)) from rest to peak exercise correlated best with the acute change in Pa(CO(2)) during hyperoxia at rest (r(2) = 0.62, p < 0.0005) and with resting arterial oxygen saturation (r(2) = 0.30, p = 0.011). During exercise, the strongest correlates of serial changes in Pa(CO(2)) from rest included concurrent changes in end-expiratory lung volume expressed as a percentage of total lung capacity (partial correlation coefficient [r] = 0.562, p < 0.0005) and oxygen saturation (partial r = 0.816, p < 0.0005). In severe COPD, the propensity to develop carbon dioxide retention during exercise reflects marked ventilatory constraints as a result of lung hyperinflation as well as reduced gas exchange capabilities.