When modelling the blood oxygenation level dependent (BOLD) response in the brain it is often assumed that arterial haemoglobin (Hb) is fully oxygenated, and that therefore the change in BOLD signal during functional hyperaemia is driven by changes in venous Hb saturation¹. However, recent work in mice has shown that Hb oxygen saturation (SaO₂) in cerebral precapillary arterioles can be as low as 78%², which suggests that arterial Hb saturation may affect measurement of the BOLD response. Here, we aimed to non-invasively measure SaO₂ along the cerebral arterial tree in humans, by assessing T₂* of arterial blood with short inversion time (TI) arterial spin labelling (ASL) in normoxia and mild hyperoxia.

5 healthy volunteers (2 F, 27.2 ± 1.6 years) underwent a MRI scan session (3 T, GE HDx, Milwaukee, USA). Single slice PICORE ASL images were acquired: TI = 450ms, TR = 600ms, 4 echoes = 3, 26, 49, and 72ms, number of time points = 3200, slice along the anterior-posterior cingulate line, resolution = 3x3x7 mm³, FOV = 198 mm, label thickness = 50 mm (10 mm below the imaging plane). There were 8 different levels of fixed inspired O₂ (FiO₂) ranging from 21% (normoxia) to 42% in 3% increments. Each level lasted 4 minutes and was created by mixing medical air and 100% O₂. Automated flow controllers driven by in-house Matlab (R2012b) scripts ensured a randomized order. Gases were delivered via a face mask. To determine baseline arterial blood volume (aBV), a multi-TI PICORE ASL scan was acquired at normoxia. Parameters: imaging and labelling planes identical to the above scan; 8 TIs (150ms to 850ms, spacing 100ms); variable TR (minimized); TE = 3ms; 320 time points (20 tag-control pairs per TI). End-tidal partial pressures of O₂ and CO₂ (P_ETO₂ and P_ETCO₂) were monitored throughout the scan.

A running subtraction was performed per echo to obtain 4x1600 subtraction (ΔM) images. A mono-exponential decay over TE was fitted to obtain time series of T₂* (in ms) per voxel. To obtain average T₂* per FiO₂, the parameter maps of the last minute of each hyperoxia level were averaged together, i.e. when P_ETO₂ was stable. Maps of arterial blood volume (aBV) were used to create masks of large (aBV_Large = aBV > 5 %_v), medium (aBV_Medium: 2 %_v < aBV < 5 %_v), and small arteries (aBV_Small: 0.5 %_v < aBV < 2%_v) (see Fig 1, where the anterior cerebral artery is disregarded because of field inhomogeneity above the sinus). The aBV maps were created by voxelwise fitting of an arterial input function³ to the average tag-control subtraction images from the multi TI scan at normoxia. Average T₂* for each hyperoxia level was calculated for aBV_Large, aBV_Medium, and aBV_Small. Repeated measures ANOVA (RM-ANOVA) was used to investigate the effect of Hyperoxia and Artery Size on the fitted T₂*.

The RM-ANOVA of the fitted T₂* showed no significant interaction between Hyperoxia and Artery Size (F(14,56) = 2.15, p = 0.056), but significant effects of Hyperoxia (F(7,28) = 6.32, p < .01) and Artery Size (F(2,28) = 10.6, p < .01). The T₂* for aBV_Large was significantly smaller than for aBV_Small (paired t-test, p < .01, see Fig 2). Separate RM-ANOVA per Artery Size showed that there was no effect of Hyperoxia on T₂* in aBV_Large (F(7,28) = 2.12, p = 0.10), but T₂* decreased in aBV_Medium (F(7,28) = 4.63, p < 0.02) and aBV­­_Small (F(7,28) = 4.69, p < 0.01). Absolute in vivo quantification of SaO₂ by T₂* measurements is difficult, because not only O₂ content, but also field inhomogeneity and blood flow velocity affect relaxation⁴. The low T₂* values in aBV_Large may therefore be due to intravoxel dephasing with higher blood flow velocities⁴, rather than indicating low SaO₂. The decrease in T₂* with hyperoxia as seen in aBV_Small­ may be caused by increased sensitivity to local susceptibility changes induced by O₂ dissolved in the blood plasma. Note that an increase in T₂* in aBV_Small would have been expected at low levels of FiO₂, should baseline SaO₂ in small arteries have been low. Moreover, assuming low blood velocity in aBV_Small, the baseline T₂* at normoxia (52.3 ± 8.2 ms) suggests that SaO₂ in aBV_Small is high (96%⁵), rather than low (78%²). This is possibly explained by insufficient resolution to be sensitive to arteriolar SaO₂, which is reported to be spatially heterogeneous². Given that a similar resolution is often used in functional imaging, partially oxygenated arterioles may not interfere with measurement of the BOLD response.